Compare commits

..

1 Commits

Author SHA1 Message Date
0470cac520 updaed
Signed-off-by: Robert Shaw <robshaw@redhat.com>
2025-08-14 02:14:03 +00:00
2658 changed files with 211968 additions and 340979 deletions

View File

@ -5,11 +5,11 @@ import os
import sys
import zipfile
# Read the VLLM_MAX_SIZE_MB environment variable, defaulting to 500 MiB
# Note that we have 800 MiB quota, please use it wisely.
# See https://github.com/pypi/support/issues/6326 .
# Read the VLLM_MAX_SIZE_MB environment variable, defaulting to 400 MiB
# Note that we have 400 MiB quota, please use it wisely.
# See https://github.com/pypi/support/issues/3792 .
# Please also sync the value with the one in Dockerfile.
VLLM_MAX_SIZE_MB = int(os.environ.get("VLLM_MAX_SIZE_MB", 500))
VLLM_MAX_SIZE_MB = int(os.environ.get("VLLM_MAX_SIZE_MB", 400))
def print_top_10_largest_files(zip_file):

View File

@ -8,8 +8,7 @@ template = """<!DOCTYPE html>
<html>
<body>
<h1>Links for vLLM</h1/>
<a href="../{x86_wheel_html_escaped}">{x86_wheel}</a><br/>
<a href="../{arm_wheel_html_escaped}">{arm_wheel}</a><br/>
<a href="../{wheel_html_escaped}">{wheel}</a><br/>
</body>
</html>
"""
@ -22,25 +21,7 @@ filename = os.path.basename(args.wheel)
with open("index.html", "w") as f:
print(f"Generated index.html for {args.wheel}")
# sync the abi tag with .buildkite/scripts/upload-wheels.sh
if "x86_64" in filename:
x86_wheel = filename
arm_wheel = filename.replace("x86_64", "aarch64").replace(
"manylinux1", "manylinux2014"
)
elif "aarch64" in filename:
x86_wheel = filename.replace("aarch64", "x86_64").replace(
"manylinux2014", "manylinux1"
)
arm_wheel = filename
else:
raise ValueError(f"Unsupported wheel: {filename}")
# cloudfront requires escaping the '+' character
f.write(
template.format(
x86_wheel=x86_wheel,
x86_wheel_html_escaped=x86_wheel.replace("+", "%2B"),
arm_wheel=arm_wheel,
arm_wheel_html_escaped=arm_wheel.replace("+", "%2B"),
)
template.format(wheel=filename, wheel_html_escaped=filename.replace("+", "%2B"))
)

View File

@ -1,12 +0,0 @@
# For hf script, without -t option (tensor parallel size).
# bash .buildkite/lm-eval-harness/run-lm-eval-chartqa-vllm-vlm-baseline.sh -m meta-llama/Llama-4-Maverick-17B-128E-Instruct-FP8 -l 100 -t 8
model_name: "meta-llama/Llama-4-Maverick-17B-128E-Instruct-FP8"
backend: "vllm-vlm"
tasks:
- name: "chartqa"
metrics:
- name: "relaxed_accuracy,none"
# TODO(zhewenl): model card is 0.90, but the actual score is 0.80.
value: 0.80
limit: 100
num_fewshot: 0

View File

@ -1,10 +0,0 @@
# For hf script, without -t option (tensor parallel size).
# bash .buildkite/lm-eval-harness/run-lm-eval-mmlupro-vllm-baseline.sh -m meta-llama/Llama-4-Maverick-17B-128E-Instruct-FP8 -l 250 -t 8 -f 5
model_name: "meta-llama/Llama-4-Maverick-17B-128E-Instruct-FP8"
tasks:
- name: "mmlu_pro"
metrics:
- name: "exact_match,custom-extract"
value: 0.80
limit: 250 # will run on 250 * 14 subjects = 3500 samples
num_fewshot: 5

View File

@ -0,0 +1,12 @@
# For vllm script, with -t option (tensor parallel size).
# bash .buildkite/lm-eval-harness/run-lm-eval-gsm-vllm-baseline.sh -m nm-testing/Qwen2-1.5B-Instruct-W8A16-Channelwise -b "auto" -l 1000 -f 5 -t 1
model_name: "nm-testing/Qwen2-1.5B-Instruct-W8A16-Channelwise"
tasks:
- name: "gsm8k"
metrics:
- name: "exact_match,strict-match"
value: 0.595
- name: "exact_match,flexible-extract"
value: 0.582
limit: 1000
num_fewshot: 5

View File

@ -1,5 +1,4 @@
# For vllm script, with -t option (tensor parallel size)
# bash .buildkite/lm-eval-harness/run-lm-eval-gsm-vllm-baseline.sh -m RedHatAI/Qwen2.5-VL-3B-Instruct-FP8-Dynamic -l 1319 -t 1
# bash .buildkite/lm-eval-harness/run-lm-eval-gsm-vllm-baseline.sh -m RedHatAI/Qwen2.5-VL-3B-Instruct-FP8-Dynamic -b auto -l 1319 -f 5 -t 1
model_name: "RedHatAI/Qwen2.5-VL-3B-Instruct-FP8-Dynamic"
tasks:
- name: "gsm8k"

View File

@ -1,12 +0,0 @@
# For vllm script, with -t option (tensor parallel size).
# bash .buildkite/lm-eval-harness/run-lm-eval-chartqa-vllm-vlm-baseline.sh -m Qwen/Qwen2.5-VL-7B-Instruct -l 2500 -t 1
model_name: "Qwen/Qwen2.5-VL-7B-Instruct"
backend: "vllm-vlm"
tasks:
- name: "chartqa"
metrics:
- name: "relaxed_accuracy,none"
value: 0.855
limit: 2500
num_fewshot: 0

View File

@ -1,14 +0,0 @@
model_name: "Qwen/Qwen3-235B-A22B-Instruct-2507-FP8"
tasks:
- name: "mmlu_pro"
metrics:
- name: "exact_match,custom-extract"
value: 0.82
limit: 250 # will run on 250 * 14 subjects = 3500 samples
num_fewshot: 5
enforce_eager: false # we use false to speed up the eval process
kv_cache_dtype: fp8 # we use fp8 to speed up the eval process
max_model_len: 40960
apply_chat_template: true
fewshot_as_multiturn: true
gen_kwargs: "temperature=0,top_p=1,top_k=0,max_gen_toks=5632,until=<|ENDANSWER|>"

View File

@ -1 +0,0 @@
Qwen3-235B-A22B-Instruct-2507-FP8.yaml

View File

@ -3,3 +3,4 @@ Meta-Llama-3-70B-Instruct.yaml
Mixtral-8x7B-Instruct-v0.1.yaml
Qwen2-57B-A14-Instruct.yaml
DeepSeek-V2-Lite-Chat.yaml
Meta-Llama-3-8B-QQQ.yaml

View File

@ -1 +0,0 @@
Meta-Llama-4-Maverick-17B-128E-Instruct-FP8-MM.yaml

View File

@ -1 +0,0 @@
Qwen2.5-VL-7B-Instruct.yaml

View File

@ -1,44 +0,0 @@
#!/bin/bash
# We can use this script to compute baseline accuracy on chartqa for vllm.
#
# Make sure you have lm-eval-harness installed:
# pip install lm-eval==0.4.9
usage() {
echo``
echo "Runs lm eval harness on ChartQA using multimodal vllm."
echo "This pathway is intended to be used to create baselines for "
echo "our correctness tests in vllm's CI."
echo
echo "usage: ${0} <options>"
echo
echo " -m - huggingface stub or local directory of the model"
echo " -l - limit number of samples to run"
echo " -t - tensor parallel size to run at"
echo
}
while getopts "m:l:t:" OPT; do
case ${OPT} in
m )
MODEL="$OPTARG"
;;
l )
LIMIT="$OPTARG"
;;
t )
TP_SIZE="$OPTARG"
;;
\? )
usage
exit 1
;;
esac
done
lm_eval --model vllm-vlm \
--model_args "pretrained=$MODEL,tensor_parallel_size=$TP_SIZE" \
--tasks chartqa \
--batch_size auto \
--apply_chat_template \
--limit $LIMIT

View File

@ -2,7 +2,7 @@
# We can use this script to compute baseline accuracy on GSM for transformers.
#
# Make sure you have lm-eval-harness installed:
# pip install git+https://github.com/EleutherAI/lm-evaluation-harness.git@206b7722158f58c35b7ffcd53b035fdbdda5126d#egg=lm-eval[api]
# pip install lm-eval==0.4.4
usage() {
echo``

View File

@ -3,7 +3,7 @@
# We use this for fp8, which HF does not support.
#
# Make sure you have lm-eval-harness installed:
# pip install git+https://github.com/EleutherAI/lm-evaluation-harness.git@206b7722158f58c35b7ffcd53b035fdbdda5126d#egg=lm-eval[api]
# pip install lm-eval==0.4.4
usage() {
echo``

View File

@ -1,50 +0,0 @@
#!/bin/bash
# We can use this script to compute baseline accuracy on MMLUPRO for vllm.
# We use this for fp8, which HF does not support.
#
# Make sure you have lm-eval-harness installed:
# pip install git+https://github.com/EleutherAI/lm-evaluation-harness.git@206b7722158f58c35b7ffcd53b035fdbdda5126d#egg=lm-eval[api]
usage() {
echo``
echo "Runs lm eval harness on MMLU Pro using huggingface transformers."
echo "This pathway is intended to be used to create baselines for "
echo "our automated nm-test-accuracy workflow"
echo
echo "usage: ${0} <options>"
echo
echo " -m - huggingface stub or local directory of the model"
echo " -l - limit number of samples to run"
echo " -f - number of fewshot samples to use"
echo " -t - tensor parallel size to run at"
echo
}
while getopts "m:b:l:f:t:" OPT; do
case ${OPT} in
m )
MODEL="$OPTARG"
;;
b )
BATCH_SIZE="$OPTARG"
;;
l )
LIMIT="$OPTARG"
;;
f )
FEWSHOT="$OPTARG"
;;
t )
TP_SIZE="$OPTARG"
;;
\? )
usage
exit 1
;;
esac
done
lm_eval --model vllm \
--model_args "pretrained=$MODEL,tensor_parallel_size=$TP_SIZE,add_bos_token=true,trust_remote_code=true,max_model_len=4096" \
--tasks mmlu_pro --num_fewshot "$FEWSHOT" --limit "$LIMIT" \
--batch_size auto

View File

@ -19,35 +19,21 @@ RTOL = 0.08
def launch_lm_eval(eval_config, tp_size):
trust_remote_code = eval_config.get("trust_remote_code", False)
max_model_len = eval_config.get("max_model_len", 4096)
batch_size = eval_config.get("batch_size", "auto")
backend = eval_config.get("backend", "vllm")
enforce_eager = eval_config.get("enforce_eager", "true")
kv_cache_dtype = eval_config.get("kv_cache_dtype", "auto")
model_args = (
f"pretrained={eval_config['model_name']},"
f"tensor_parallel_size={tp_size},"
f"enforce_eager={enforce_eager},"
f"kv_cache_dtype={kv_cache_dtype},"
f"enforce_eager=true,"
f"add_bos_token=true,"
f"trust_remote_code={trust_remote_code},"
f"max_model_len={max_model_len},"
f"max_model_len={max_model_len}"
)
results = lm_eval.simple_evaluate(
model=backend,
model="vllm",
model_args=model_args,
tasks=[task["name"] for task in eval_config["tasks"]],
num_fewshot=eval_config["num_fewshot"],
limit=eval_config["limit"],
# TODO(yeq): using chat template w/ fewshot_as_multiturn is supposed help
# text models. however, this is regressing measured strict-match for
# existing text models in CI, so only apply it for mm, or explicitly set
apply_chat_template=eval_config.get(
"apply_chat_template", backend == "vllm-vlm"
),
fewshot_as_multiturn=eval_config.get("fewshot_as_multiturn", False),
# Forward decoding and early-stop controls (e.g., max_gen_toks, until=...)
gen_kwargs=eval_config.get("gen_kwargs"),
batch_size=batch_size,
batch_size="auto",
)
return results

View File

@ -2,23 +2,40 @@
## Introduction
This directory contains a benchmarking suite for **developers** to run locally and gain clarity on whether their PR improves/degrades vllm's performance.
vLLM also maintains a continuous performance benchmark under [perf.vllm.ai](https://perf.vllm.ai/), hosted under PyTorch CI HUD.
This directory contains two sets of benchmark for vllm.
- Performance benchmark: benchmark vllm's performance under various workload, for **developers** to gain clarity on whether their PR improves/degrades vllm's performance
- Nightly benchmark: compare vllm's performance against alternatives (tgi, trt-llm and lmdeploy), for **the public** to know when to choose vllm.
See [vLLM performance dashboard](https://perf.vllm.ai) for the latest performance benchmark results and [vLLM GitHub README](https://github.com/vllm-project/vllm/blob/main/README.md) for latest nightly benchmark results.
## Performance benchmark quick overview
**Benchmarking Coverage**: latency, throughput and fix-qps serving on B200, A100, H100, Intel® Xeon® Processors and Intel® Gaudi® 3 Accelerators with different models.
**Benchmarking Coverage**: latency, throughput and fix-qps serving on A100 (the support for FP8 benchmark on H100 is coming!) and Intel® Xeon® Processors, with different models.
**Benchmarking Duration**: about 1hr.
**For benchmarking developers**: please try your best to constraint the duration of benchmarking to about 1 hr so that it won't take forever to run.
## Nightly benchmark quick overview
**Benchmarking Coverage**: Fix-qps serving on A100 (the support for FP8 benchmark on H100 is coming!) on Llama-3 8B, 70B and Mixtral 8x7B.
**Benchmarking engines**: vllm, TGI, trt-llm and lmdeploy.
**Benchmarking Duration**: about 3.5hrs.
## Trigger the benchmark
The benchmark needs to be triggered manually:
Performance benchmark will be triggered when:
- A PR being merged into vllm.
- Every commit for those PRs with `perf-benchmarks` label AND `ready` label.
Manually Trigger the benchmark
```bash
bash .buildkite/performance-benchmarks/scripts/run-performance-benchmarks.sh
bash .buildkite/nightly-benchmarks/scripts/run-performance-benchmarks.sh
```
Runtime environment variables:
@ -30,11 +47,14 @@ Runtime environment variables:
- `REMOTE_HOST`: IP for the remote vLLM service to benchmark. Default value is empty string.
- `REMOTE_PORT`: Port for the remote vLLM service to benchmark. Default value is empty string.
Nightly benchmark will be triggered when:
- Every commit for those PRs with `perf-benchmarks` label and `nightly-benchmarks` label.
## Performance benchmark details
See [performance-benchmarks-descriptions.md](performance-benchmarks-descriptions.md) for detailed descriptions, and use `tests/latency-tests.json`, `tests/throughput-tests.json`, `tests/serving-tests.json` to configure the test cases.
> NOTE: For Intel® Xeon® Processors, use `tests/latency-tests-cpu.json`, `tests/throughput-tests-cpu.json`, `tests/serving-tests-cpu.json` instead.
For Intel® Gaudi® 3 Accelerators, use `tests/latency-tests-hpu.json`, `tests/throughput-tests-hpu.json`, `tests/serving-tests-hpu.json` instead.
>
### Latency test
@ -118,17 +138,48 @@ The raw benchmarking results (in the format of json files) are in the `Artifacts
The `compare-json-results.py` helps to compare benchmark results JSON files converted using `convert-results-json-to-markdown.py`.
When run, benchmark script generates results under `benchmark/results` folder, along with the `benchmark_results.md` and `benchmark_results.json`.
`compare-json-results.py` compares two `benchmark_results.json` files and provides performance ratio e.g. for Output Tput, Median TTFT and Median TPOT.
If only one benchmark_results.json is passed, `compare-json-results.py` compares different TP and PP configurations in the benchmark_results.json instead.
`compare-json-results.py` compares two `benchmark_results.json` files and provides performance ratio e.g. for Output Tput, Median TTFT and Median TPOT.
Here is an example using the script to compare result_a and result_b with Model, Dataset name, input/output length, max concurrency and qps.
Here is an example using the script to compare result_a and result_b without detail test name.
`python3 compare-json-results.py -f results_a/benchmark_results.json -f results_b/benchmark_results.json --ignore_test_name`
| | results_a/benchmark_results.json | results_b/benchmark_results.json | perf_ratio |
|----|----------------------------------------|----------------------------------------|----------|
| 0 | 142.633982 | 156.526018 | 1.097396 |
| 1 | 241.620334 | 294.018783 | 1.216863 |
| 2 | 218.298905 | 262.664916 | 1.203235 |
| 3 | 242.743860 | 299.816190 | 1.235113 |
Here is an example using the script to compare result_a and result_b with detail test name.
`python3 compare-json-results.py -f results_a/benchmark_results.json -f results_b/benchmark_results.json`
| | Model | Dataset Name | Input Len | Output Len | # of max concurrency | qps | results_a/benchmark_results.json | results_b/benchmark_results.json | perf_ratio |
|----|---------------------------------------|--------|-----|-----|------|-----|-----------|----------|----------|
| 0 | meta-llama/Meta-Llama-3.1-8B-Instruct | random | 128 | 128 | 1000 | 1 | 142.633982 | 156.526018 | 1.097396 |
| 1 | meta-llama/Meta-Llama-3.1-8B-Instruct | random | 128 | 128 | 1000 | inf| 241.620334 | 294.018783 | 1.216863 |
| | results_a/benchmark_results.json_name | results_a/benchmark_results.json | results_b/benchmark_results.json_name | results_b/benchmark_results.json | perf_ratio |
|---|---------------------------------------------|----------------------------------------|---------------------------------------------|----------------------------------------|----------|
| 0 | serving_llama8B_tp1_sharegpt_qps_1 | 142.633982 | serving_llama8B_tp1_sharegpt_qps_1 | 156.526018 | 1.097396 |
| 1 | serving_llama8B_tp1_sharegpt_qps_16 | 241.620334 | serving_llama8B_tp1_sharegpt_qps_16 | 294.018783 | 1.216863 |
| 2 | serving_llama8B_tp1_sharegpt_qps_4 | 218.298905 | serving_llama8B_tp1_sharegpt_qps_4 | 262.664916 | 1.203235 |
| 3 | serving_llama8B_tp1_sharegpt_qps_inf | 242.743860 | serving_llama8B_tp1_sharegpt_qps_inf | 299.816190 | 1.235113 |
| 4 | serving_llama8B_tp2_random_1024_128_qps_1 | 96.613390 | serving_llama8B_tp4_random_1024_128_qps_1 | 108.404853 | 1.122048 |
A comparison diagram will be generated below the table.
Here is an example to compare between 96c/results_gnr_96c_091_tp2pp3 and 128c/results_gnr_128c_091_tp2pp3
<img width="1886" height="828" alt="image" src="https://github.com/user-attachments/assets/c02a43ef-25d0-4fd6-90e5-2169a28682dd" />
## Nightly test details
See [nightly-descriptions.md](nightly-descriptions.md) for the detailed description on test workload, models and docker containers of benchmarking other llm engines.
### Workflow
- The [nightly-pipeline.yaml](nightly-pipeline.yaml) specifies the docker containers for different LLM serving engines.
- Inside each container, we run [scripts/run-nightly-benchmarks.sh](scripts/run-nightly-benchmarks.sh), which will probe the serving engine of the current container.
- The `scripts/run-nightly-benchmarks.sh` will parse the workload described in [nightly-tests.json](tests/nightly-tests.json) and launch the right benchmark for the specified serving engine via `scripts/launch-server.sh`.
- At last, we run [scripts/summary-nightly-results.py](scripts/summary-nightly-results.py) to collect and plot the final benchmarking results, and update the results to buildkite.
### Nightly tests
In [nightly-tests.json](tests/nightly-tests.json), we include the command line arguments for benchmarking commands, together with the benchmarking test cases. The format is highly similar to performance benchmark.
### Docker containers
The docker containers for benchmarking are specified in `nightly-pipeline.yaml`.
WARNING: the docker versions are HARD-CODED and SHOULD BE ALIGNED WITH `nightly-descriptions.md`. The docker versions need to be hard-coded as there are several version-specific bug fixes inside `scripts/run-nightly-benchmarks.sh` and `scripts/launch-server.sh`.
WARNING: populating `trt-llm` to latest version is not easy, as it requires updating several protobuf files in [tensorrt-demo](https://github.com/neuralmagic/tensorrt-demo.git).

View File

@ -0,0 +1,184 @@
steps:
- label: "Wait for container to be ready"
key: wait-for-container-image
agents:
queue: A100
plugins:
- kubernetes:
podSpec:
containers:
- image: badouralix/curl-jq
command:
- sh .buildkite/nightly-benchmarks/scripts/wait-for-image.sh
- label: "Cleanup H100"
agents:
queue: H100
depends_on: ~
command: docker system prune -a --volumes --force
- label: "A100"
# skip: "use this flag to conditionally skip the benchmark step, useful for PR testing"
agents:
queue: A100
depends_on: wait-for-container-image
if: build.branch == "main"
plugins:
- kubernetes:
podSpec:
priorityClassName: perf-benchmark
containers:
- image: public.ecr.aws/q9t5s3a7/vllm-ci-postmerge-repo:$BUILDKITE_COMMIT
command:
- bash .buildkite/nightly-benchmarks/scripts/run-performance-benchmarks.sh
resources:
limits:
nvidia.com/gpu: 8
volumeMounts:
- name: devshm
mountPath: /dev/shm
env:
- name: VLLM_USAGE_SOURCE
value: ci-test
- name: HF_TOKEN
valueFrom:
secretKeyRef:
name: hf-token-secret
key: token
nodeSelector:
nvidia.com/gpu.product: NVIDIA-A100-SXM4-80GB
volumes:
- name: devshm
emptyDir:
medium: Memory
- label: "H200"
# skip: "use this flag to conditionally skip the benchmark step, useful for PR testing"
agents:
queue: H200
depends_on: wait-for-container-image
if: build.branch == "main"
plugins:
- docker#v5.12.0:
image: public.ecr.aws/q9t5s3a7/vllm-ci-postmerge-repo:$BUILDKITE_COMMIT
command:
- bash
- .buildkite/nightly-benchmarks/scripts/run-performance-benchmarks.sh
mount-buildkite-agent: true
propagate-environment: true
ipc: host
gpus: 4,5,6,7
volumes:
- /data/benchmark-hf-cache:/root/.cache/huggingface
environment:
- VLLM_USAGE_SOURCE
- HF_TOKEN
#- block: "Run H100 Benchmark"
#key: block-h100
#depends_on: ~
- label: "H100"
# skip: "use this flag to conditionally skip the benchmark step, useful for PR testing"
agents:
queue: H100
depends_on: wait-for-container-image
if: build.branch == "main"
plugins:
- docker#v5.12.0:
image: public.ecr.aws/q9t5s3a7/vllm-ci-postmerge-repo:$BUILDKITE_COMMIT
command:
- bash
- .buildkite/nightly-benchmarks/scripts/run-performance-benchmarks.sh
mount-buildkite-agent: true
propagate-environment: true
ipc: host
gpus: all # see CUDA_VISIBLE_DEVICES for actual GPUs used
volumes:
- /data/benchmark-hf-cache:/root/.cache/huggingface
environment:
- VLLM_USAGE_SOURCE
- HF_TOKEN
# Premerge benchmark
- label: "A100"
# skip: "use this flag to conditionally skip the benchmark step, useful for PR testing"
agents:
queue: A100
depends_on: wait-for-container-image
if: build.branch != "main"
plugins:
- kubernetes:
podSpec:
priorityClassName: perf-benchmark
containers:
- image: public.ecr.aws/q9t5s3a7/vllm-ci-test-repo:$BUILDKITE_COMMIT
command:
- bash .buildkite/nightly-benchmarks/scripts/run-performance-benchmarks.sh
resources:
limits:
nvidia.com/gpu: 8
volumeMounts:
- name: devshm
mountPath: /dev/shm
env:
- name: VLLM_USAGE_SOURCE
value: ci-test
- name: HF_TOKEN
valueFrom:
secretKeyRef:
name: hf-token-secret
key: token
nodeSelector:
nvidia.com/gpu.product: NVIDIA-A100-SXM4-80GB
volumes:
- name: devshm
emptyDir:
medium: Memory
- label: "H200"
# skip: "use this flag to conditionally skip the benchmark step, useful for PR testing"
agents:
queue: H200
depends_on: wait-for-container-image
if: build.branch != "main"
plugins:
- docker#v5.12.0:
image: public.ecr.aws/q9t5s3a7/vllm-ci-test-repo:$BUILDKITE_COMMIT
command:
- bash
- .buildkite/nightly-benchmarks/scripts/run-performance-benchmarks.sh
mount-buildkite-agent: true
propagate-environment: true
ipc: host
gpus: 4,5,6,7
volumes:
- /data/benchmark-hf-cache:/root/.cache/huggingface
environment:
- VLLM_USAGE_SOURCE
- HF_TOKEN
#- block: "Run H100 Benchmark"
#key: block-h100
#depends_on: ~
- label: "H100"
# skip: "use this flag to conditionally skip the benchmark step, useful for PR testing"
agents:
queue: H100
depends_on: wait-for-container-image
if: build.branch != "main"
plugins:
- docker#v5.12.0:
image: public.ecr.aws/q9t5s3a7/vllm-ci-test-repo:$BUILDKITE_COMMIT
command:
- bash
- .buildkite/nightly-benchmarks/scripts/run-performance-benchmarks.sh
mount-buildkite-agent: true
propagate-environment: true
ipc: host
gpus: all # see CUDA_VISIBLE_DEVICES for actual GPUs used
volumes:
- /data/benchmark-hf-cache:/root/.cache/huggingface
environment:
- VLLM_USAGE_SOURCE
- HF_TOKEN

View File

@ -0,0 +1,28 @@
# Nightly benchmark annotation
## Description
This file contains the downloading link for benchmarking results.
- [benchmarking pipeline](artifact://nightly-pipeline.yaml)
- [benchmarking results](artifact://results.zip)
- [benchmarking code](artifact://nightly-benchmarks.zip)
Please download the visualization scripts in the post
## Results reproduction
- Find the docker we use in `benchmarking pipeline`
- Deploy the docker, and inside the docker:
- Download `nightly-benchmarks.zip`.
- In the same folder, run the following code:
```bash
export HF_TOKEN=<your HF token>
apt update
apt install -y git
unzip nightly-benchmarks.zip
VLLM_SOURCE_CODE_LOC=./ bash .buildkite/nightly-benchmarks/scripts/run-nightly-benchmarks.sh
```
And the results will be inside `./benchmarks/results`.

View File

@ -0,0 +1,39 @@
# Nightly benchmark
This benchmark aims to:
- Provide performance clarity: Provide clarity on which one (vllm, tensorrt-llm, lmdeploy and SGLang) leads in performance in what workload.
- Be reproducible: one can run the exact same set of benchmarking commands inside the exact same docker by following reproducing instructions.
Latest results: [results link](https://blog.vllm.ai/2024/09/05/perf-update.html), scroll to the end.
Latest reproduction guilde: [github issue link](https://github.com/vllm-project/vllm/issues/8176)
## Setup
- Docker images:
- vLLM: `vllm/vllm-openai:v0.6.2`
- SGLang: `lmsysorg/sglang:v0.3.2-cu121`
- LMDeploy: `openmmlab/lmdeploy:v0.6.1-cu12`
- TensorRT-LLM: `nvcr.io/nvidia/tritonserver:24.07-trtllm-python-py3`
- *NOTE: we uses r24.07 as the current implementation only works for this version. We are going to bump this up.*
- Check [nightly-pipeline.yaml](nightly-pipeline.yaml) for the concrete docker images, specs and commands we use for the benchmark.
- Hardware
- 8x Nvidia A100 GPUs
- Workload:
- Dataset
- ShareGPT dataset
- Prefill-heavy dataset (in average 462 input tokens, 16 tokens as output)
- Decode-heavy dataset (in average 462 input tokens, 256 output tokens)
- Check [nightly-tests.json](tests/nightly-tests.json) for the concrete configuration of datasets we use.
- Models: llama-3 8B, llama-3 70B.
- We do not use llama 3.1 as it is incompatible with trt-llm r24.07. ([issue](https://github.com/NVIDIA/TensorRT-LLM/issues/2105)).
- Average QPS (query per second): 2, 4, 8, 16, 32 and inf.
- Queries are randomly sampled, and arrival patterns are determined via Poisson process, but all with fixed random seed.
- Evaluation metrics: Throughput (higher the better), TTFT (time to the first token, lower the better), ITL (inter-token latency, lower the better).
## Known issues
- TRT-LLM crashes with Llama 3.1 8B [issue](https://github.com/NVIDIA/TensorRT-LLM/issues/2105).
- TGI does not support `ignore-eos` flag.

View File

@ -0,0 +1,196 @@
common_pod_spec: &common_pod_spec
priorityClassName: perf-benchmark
nodeSelector:
nvidia.com/gpu.product: NVIDIA-A100-SXM4-80GB
volumes:
- name: devshm
emptyDir:
medium: Memory
- name: hf-cache
hostPath:
path: /root/.cache/huggingface
type: Directory
common_container_settings: &common_container_settings
command:
- bash .buildkite/nightly-benchmarks/scripts/run-nightly-benchmarks.sh
resources:
limits:
nvidia.com/gpu: 8
volumeMounts:
- name: devshm
mountPath: /dev/shm
- name: hf-cache
mountPath: /root/.cache/huggingface
env:
- name: VLLM_USAGE_SOURCE
value: ci-test
- name: HF_HOME
value: /root/.cache/huggingface
- name: VLLM_SOURCE_CODE_LOC
value: /workspace/build/buildkite/vllm/performance-benchmark
- name: HF_TOKEN
valueFrom:
secretKeyRef:
name: hf-token-secret
key: token
steps:
- block: ":rocket: Ready for comparing vllm against alternatives? This will take 4 hours."
- label: "A100 vllm step 10"
priority: 100
agents:
queue: A100
plugins:
- kubernetes:
podSpec:
<<: *common_pod_spec
containers:
- image: vllm/vllm-openai:v0.6.2
<<: *common_container_settings
- label: "A100 sglang benchmark"
priority: 100
agents:
queue: A100
plugins:
- kubernetes:
podSpec:
<<: *common_pod_spec
containers:
- image: lmsysorg/sglang:v0.3.2-cu121
<<: *common_container_settings
- label: "A100 lmdeploy benchmark"
priority: 100
agents:
queue: A100
plugins:
- kubernetes:
podSpec:
<<: *common_pod_spec
containers:
- image: openmmlab/lmdeploy:v0.6.1-cu12
<<: *common_container_settings
- label: "A100 trt llama-8B"
priority: 100
agents:
queue: A100
plugins:
- kubernetes:
podSpec:
<<: *common_pod_spec
containers:
- image: nvcr.io/nvidia/tritonserver:24.07-trtllm-python-py3
<<: *common_container_settings
env:
- name: VLLM_USAGE_SOURCE
value: ci-test
- name: HF_HOME
value: /root/.cache/huggingface
- name: VLLM_SOURCE_CODE_LOC
value: /workspace/build/buildkite/vllm/performance-benchmark
- name: HF_TOKEN
valueFrom:
secretKeyRef:
name: hf-token-secret
key: token
- name: TEST_SELECTOR
value: "llama8B"
- label: "A100 trt llama-70B"
priority: 100
agents:
queue: A100
plugins:
- kubernetes:
podSpec:
<<: *common_pod_spec
containers:
- image: nvcr.io/nvidia/tritonserver:24.07-trtllm-python-py3
<<: *common_container_settings
env:
- name: VLLM_USAGE_SOURCE
value: ci-test
- name: HF_HOME
value: /root/.cache/huggingface
- name: VLLM_SOURCE_CODE_LOC
value: /workspace/build/buildkite/vllm/performance-benchmark
- name: HF_TOKEN
valueFrom:
secretKeyRef:
name: hf-token-secret
key: token
- name: TEST_SELECTOR
value: "llama70B"
# FIXME(Kuntai): uncomment this after NVIDIA gives us their test docker image
# - label: "A100 trt benchmark"
# priority: 100
# agents:
# queue: A100
# plugins:
# - kubernetes:
# podSpec:
# <<: *common_pod_spec
# containers:
# - image: nvcr.io/nvidia/tritonserver:24.07-trtllm-python-py3
# <<: *common_container_settings
# FIXME(Kuntai): uncomment this after TGI supports `--ignore-eos`.
# - label: "A100 tgi benchmark"
# priority: 100
# agents:
# queue: A100
# plugins:
# - kubernetes:
# podSpec:
# <<: *common_pod_spec
# containers:
# - image: ghcr.io/huggingface/text-generation-inference:2.2.0
# <<: *common_container_settings
- wait
- label: "Collect the results"
priority: 100
agents:
queue: A100
plugins:
- kubernetes:
podSpec:
<<: *common_pod_spec
containers:
- image: vllm/vllm-openai:v0.5.0.post1
command:
- bash .buildkite/nightly-benchmarks/scripts/nightly-annotate.sh
resources:
limits:
nvidia.com/gpu: 8
volumeMounts:
- name: devshm
mountPath: /dev/shm
env:
- name: VLLM_USAGE_SOURCE
value: ci-test
- name: VLLM_SOURCE_CODE_LOC
value: /workspace/build/buildkite/vllm/performance-benchmark
- name: HF_TOKEN
valueFrom:
secretKeyRef:
name: hf-token-secret
key: token
- block: ":rocket: check the results!"

View File

@ -5,7 +5,7 @@
- Input length: 32 tokens.
- Output length: 128 tokens.
- Batch size: fixed (8).
- GPU/HPU Models: llama-3.1 8B, llama-3 70B, mixtral 8x7B.
- GPU Models: llama-3.1 8B, llama-3 70B, mixtral 8x7B.
- CPU Models: llama-3.1 8B.
- Evaluation metrics: end-to-end latency (mean, median, p99).
@ -16,7 +16,7 @@
- Input length: randomly sample 200 prompts from ShareGPT dataset (with fixed random seed).
- Output length: the corresponding output length of these 200 prompts.
- Batch size: dynamically determined by vllm to achieve maximum throughput.
- GPU/HPU Models: llama-3.1 8B, llama-3 70B, mixtral 8x7B.
- GPU Models: llama-3.1 8B, llama-3 70B, mixtral 8x7B.
- CPU Models: llama-3.1 8B.
- Evaluation metrics: throughput.
@ -28,7 +28,7 @@
- Output length: the corresponding output length of these 200 prompts.
- Batch size: dynamically determined by vllm and the arrival pattern of the requests.
- **Average QPS (query per second)**: 1, 4, 16 and inf. QPS = inf means all requests come at once. For other QPS values, the arrival time of each query is determined using a random Poisson process (with fixed random seed).
- GPU/HPU Models: llama-3.1 8B, llama-3 70B, mixtral 8x7B.
- GPU Models: llama-3.1 8B, llama-3 70B, mixtral 8x7B.
- We also added a speculative decoding test for llama-3 70B on GPU, under QPS 2
- CPU Models: llama-3.1 8B.
- Evaluation metrics: throughput, TTFT (time to the first token, with mean, median and p99), ITL (inter-token latency, with mean, median and p99).

View File

@ -0,0 +1,66 @@
# SPDX-License-Identifier: Apache-2.0
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
import argparse
import pandas as pd
def compare_data_columns(
files, name_column, data_column, drop_column, ignore_test_name=False
):
print("\ncompare_data_column: " + data_column)
frames = []
compare_frames = []
for file in files:
data_df = pd.read_json(file)
serving_df = data_df.dropna(subset=[drop_column], ignore_index=True)
if ignore_test_name is False:
serving_df = serving_df.rename(columns={name_column: file + "_name"})
frames.append(serving_df[file + "_name"])
serving_df = serving_df.rename(columns={data_column: file})
frames.append(serving_df[file])
compare_frames.append(serving_df[file])
if len(compare_frames) >= 2:
# Compare numbers among two files
ratio_df = compare_frames[1] / compare_frames[0]
frames.append(ratio_df)
compare_frames.pop(1)
concat_df = pd.concat(frames, axis=1)
return concat_df
if __name__ == "__main__":
parser = argparse.ArgumentParser()
parser.add_argument(
"-f", "--file", action="append", type=str, help="input file name"
)
parser.add_argument(
"--ignore_test_name", action="store_true", help="ignore_test_name or not"
)
args = parser.parse_args()
files = args.file
print("comparing : " + ", ".join(files))
drop_column = "P99"
name_column = "Test name"
data_cols_to_compare = ["Output Tput (tok/s)", "Median TTFT (ms)", "Median"]
html_msgs_for_data_cols = [
"Compare Output Tokens /n",
"Median TTFT /n",
"Median TPOT /n",
]
ignore_test_name = args.ignore_test_name
with open("perf_comparison.html", "w") as text_file:
for i in range(len(data_cols_to_compare)):
output_df = compare_data_columns(
files,
name_column,
data_cols_to_compare[i],
drop_column,
ignore_test_name=ignore_test_name,
)
print(output_df)
html = output_df.to_html()
text_file.write(html_msgs_for_data_cols[i])
text_file.write(html)

View File

@ -1,19 +1,17 @@
# SPDX-License-Identifier: Apache-2.0
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
import argparse
import json
import os
import shlex
from importlib import util
from pathlib import Path
from typing import Any
import pandas as pd
import psutil
import regex as re
from tabulate import tabulate
results_folder = Path("results/")
# latency results and the keys that will be printed into markdown
latency_results = []
latency_column_mapping = {
@ -44,30 +42,20 @@ throughput_results_column_mapping = {
serving_results = []
serving_column_mapping = {
"test_name": "Test name",
"model_id": "Model",
"dataset_name": "Dataset Name",
"input_len": "Input Len",
"output_len": "Output Len",
"tp_size": "TP Size",
"pp_size": "PP Size",
"dtype": "dtype",
"gpu_type": "GPU",
"completed": "# of req.",
"qps": "qps",
"max_concurrency": "# of max concurrency.",
"request_throughput": "Tput (req/s)",
"total_token_throughput": "Total Token Tput (tok/s)",
"output_throughput": "Output Tput (tok/s)",
# "total_input_tokens": "Total input tokens",
# "total_output_tokens": "Total output tokens",
"total_input_tokens": "Total input tokens",
"total_output_tokens": "Total output tokens",
"mean_ttft_ms": "Mean TTFT (ms)",
"median_ttft_ms": "Median TTFT (ms)",
"p99_ttft_ms": "P99 TTFT (ms)",
"std_ttft_ms": "STD TTFT (ms)",
"mean_tpot_ms": "Mean TPOT (ms)",
"median_tpot_ms": "Median",
"p99_tpot_ms": "P99",
"std_tpot_ms": "STD TPOT (ms)",
"mean_itl_ms": "Mean ITL (ms)",
"median_itl_ms": "Median ITL (ms)",
"p99_itl_ms": "P99 ITL (ms)",
@ -106,104 +94,7 @@ def get_size_with_unit(bytes, suffix="B"):
bytes /= factor
def _coerce(val: str) -> Any:
"""Best-effort type coercion from string to Python types."""
low = val.lower()
if low == "null":
return None
if low == "true":
return True
if low == "false":
return False
# integers
if re.fullmatch(r"[+-]?\d+", val):
try:
return int(val)
except ValueError:
pass
# floats (keep 'inf'/'-inf'/'nan' as strings)
if re.fullmatch(r"[+-]?\d*\.\d+", val):
try:
return float(val)
except ValueError:
pass
return val
def parse_client_command(cmd: str) -> dict[str, Any]:
"""Parse the client_command shell string into {executable, script, args}."""
toks = shlex.split(cmd)
if len(toks) < 2:
raise ValueError("client_command must include an executable and a script")
executable, script = toks[0], toks[1]
args: dict[str, Any] = {}
i = 2
while i < len(toks):
t = toks[i]
if t.startswith("--"):
# --key=value or --key (value) or boolean flag
if "=" in t:
key, val = t.split("=", 1)
if key == "--metadata":
md = {}
if val:
if "=" in val:
k, v = val.split("=", 1)
md[k] = _coerce(v)
else:
md[val] = True
args[key] = md
else:
args[key] = _coerce(val)
i += 1
continue
key = t
# Special: consume metadata k=v pairs until next --flag
if key == "--metadata":
i += 1
md = {}
while i < len(toks) and not toks[i].startswith("--"):
pair = toks[i]
if "=" in pair:
k, v = pair.split("=", 1)
md[k] = _coerce(v)
else:
md[pair] = True
i += 1
args[key] = md
continue
# Standard: check if next token is a value (not a flag)
if i + 1 < len(toks) and not toks[i + 1].startswith("--"):
args[key] = _coerce(toks[i + 1])
i += 2
else:
# lone flag -> True
args[key] = True
i += 1
else:
# unexpected positional; skip
i += 1
return {"executable": executable, "script": script, "args": args}
if __name__ == "__main__":
parser = argparse.ArgumentParser()
parser.add_argument(
"-r",
"--result",
type=str,
default="results",
help="Folder name for benchmark output results.",
)
args = parser.parse_args()
results_folder = Path(args.result)
if not results_folder.exists():
raise FileNotFoundError(f"results folder does not exist: {results_folder}")
# collect results
for test_file in results_folder.glob("*.json"):
with open(test_file) as f:
@ -211,6 +102,7 @@ if __name__ == "__main__":
if "serving" in str(test_file):
# this result is generated via `vllm bench serve` command
# attach the benchmarking command to raw_result
try:
with open(test_file.with_suffix(".commands")) as f:
@ -218,44 +110,12 @@ if __name__ == "__main__":
except OSError as e:
print(e)
continue
# Parse Server Command Arg
out: dict[str, Any] = {
"server_command": parse_client_command(command["server_command"])
}
parse_args = [
"--tensor-parallel-size",
"--pipeline-parallel-size",
"--dtype",
]
col_mapping = ["tp_size", "pp_size", "dtype"]
for index, arg in enumerate(parse_args):
if arg in out["server_command"]["args"]:
raw_result.update(
{col_mapping[index]: out["server_command"]["args"][arg]}
)
# Parse Client Command Arg
out: dict[str, Any] = {
"client_command": parse_client_command(command["client_command"])
}
parse_args = [
"--dataset-name",
"--random-input-len",
"--random-output-len",
"--request-rate",
]
col_mapping = ["dataset_name", "input_len", "output_len", "qps"]
for index, arg in enumerate(parse_args):
if arg in out["client_command"]["args"]:
raw_result.update(
{col_mapping[index]: out["client_command"]["args"][arg]}
)
# Add Server, Client command
raw_result.update(command)
# update the test name of this result
raw_result.update({"test_name": test_file.stem})
# add the result to raw_result
serving_results.append(raw_result)
continue
@ -345,10 +205,7 @@ if __name__ == "__main__":
columns=latency_column_mapping
)
if not serving_results.empty:
valid_columns = [
col for col in serving_column_mapping if col in serving_results.columns
]
serving_results = serving_results[valid_columns].rename(
serving_results = serving_results[list(serving_column_mapping.keys())].rename(
columns=serving_column_mapping
)
if not throughput_results.empty:
@ -370,7 +227,7 @@ if __name__ == "__main__":
# The GPUs sometimes come in format of "GPUTYPE\nGPUTYPE\n...",
# we want to turn it into "8xGPUTYPE"
df["GPU"] = df["GPU"].apply(
lambda x: "{}x{}".format(len(x.split("\n")), x.split("\n")[0])
lambda x: f"{len(x.split('\n'))}x{x.split('\n')[0]}"
)
# get markdown tables
@ -388,11 +245,9 @@ if __name__ == "__main__":
)
# document the result
md_file = "benchmark_results.md"
json_file = "benchmark_results.json"
with open(results_folder / md_file, "w") as f:
with open(results_folder / "benchmark_results.md", "w") as f:
results = read_markdown(
"../.buildkite/performance-benchmarks/"
"../.buildkite/nightly-benchmarks/"
+ "performance-benchmarks-descriptions.md"
)
results = results.format(
@ -405,7 +260,7 @@ if __name__ == "__main__":
f.write(results)
# document benchmarking results in json
with open(results_folder / json_file, "w") as f:
with open(results_folder / "benchmark_results.json", "w") as f:
results = (
latency_results.to_dict(orient="records")
+ throughput_results.to_dict(orient="records")

View File

@ -0,0 +1,26 @@
# SPDX-License-Identifier: Apache-2.0
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
import argparse
from transformers import AutoTokenizer
def main(model, cachedir):
# Load the tokenizer and save it to the specified directory
tokenizer = AutoTokenizer.from_pretrained(model)
tokenizer.save_pretrained(cachedir)
print(f"Tokenizer saved to {cachedir}")
if __name__ == "__main__":
parser = argparse.ArgumentParser(
description="Download and save Hugging Face tokenizer"
)
parser.add_argument("--model", type=str, required=True, help="Name of the model")
parser.add_argument(
"--cachedir", type=str, required=True, help="Directory to save the tokenizer"
)
args = parser.parse_args()
main(args.model, args.cachedir)

View File

@ -0,0 +1,97 @@
# SPDX-License-Identifier: Apache-2.0
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
import argparse
import json
from pathlib import Path
import numpy as np
import pandas as pd
from tabulate import tabulate
def parse_arguments():
parser = argparse.ArgumentParser(
description="Parse command line arguments for summary-nightly-results script."
)
parser.add_argument(
"--results-folder",
type=str,
required=True,
help="The folder where the results are stored.",
)
parser.add_argument(
"--description", type=str, required=True, help="Description of the results."
)
args = parser.parse_args()
return args
def get_perf(df, method, model, metric):
means = []
for qps in [2, 4, 8, 16, "inf"]:
target = df["Test name"].str.contains(model)
target = target & df["Engine"].str.contains(method)
target = target & df["Test name"].str.contains("qps_" + str(qps))
filtered_df = df[target]
if filtered_df.empty:
means.append(0.0)
else:
means.append(filtered_df[metric].values[0])
return np.array(means)
def get_perf_w_std(df, method, model, metric):
if metric in ["TTFT", "ITL"]:
mean = get_perf(df, method, model, "Mean " + metric + " (ms)")
mean = mean.tolist()
std = get_perf(df, method, model, "Std " + metric + " (ms)")
if std.mean() == 0:
std = None
success = get_perf(df, method, model, "Successful req.")
if std is not None:
std = std / np.sqrt(success)
std = std.tolist()
else:
assert metric == "Tput"
mean = get_perf(df, method, model, "Input Tput (tok/s)") + get_perf(
df, method, model, "Output Tput (tok/s)"
)
mean = mean.tolist()
std = None
return mean, std
def main(args):
results_folder = Path(args.results_folder)
results = []
# collect results
for test_file in results_folder.glob("*_nightly_results.json"):
with open(test_file) as f:
results = results + json.loads(f.read())
# generate markdown table
df = pd.DataFrame.from_dict(results)
md_table = tabulate(df, headers="keys", tablefmt="pipe", showindex=False)
with open(args.description) as f:
description = f.read()
description = description.format(nightly_results_benchmarking_table=md_table)
with open("nightly_results.md", "w") as f:
f.write(description)
if __name__ == "__main__":
args = parse_arguments()
main(args)

View File

@ -0,0 +1,9 @@
# SPDX-License-Identifier: Apache-2.0
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
from lmdeploy.serve.openai.api_client import APIClient
api_client = APIClient("http://localhost:8000")
model_name = api_client.available_models[0]
print(model_name)

View File

@ -181,14 +181,18 @@ launch_vllm_server() {
if echo "$common_params" | jq -e 'has("fp8")' >/dev/null; then
echo "Key 'fp8' exists in common params. Use neuralmagic fp8 model for convenience."
model=$(echo "$common_params" | jq -r '.neuralmagic_quantized_model')
server_command="vllm serve $model \
server_command="python3 \
-m vllm.entrypoints.openai.api_server \
-tp $tp \
--model $model \
--port $port \
$server_args"
else
echo "Key 'fp8' does not exist in common params."
server_command="vllm serve $model \
server_command="python3 \
-m vllm.entrypoints.openai.api_server \
-tp $tp \
--model $model \
--port $port \
$server_args"
fi

View File

@ -0,0 +1,78 @@
#!/bin/bash
set -ex
set -o pipefail
main() {
(which wget && which curl) || (apt-get update && apt-get install -y wget curl)
(which jq) || (apt-get update && apt-get -y install jq)
(which zip) || (apt-get install -y zip)
if [ ! -f /workspace/buildkite-agent ]; then
echo "buildkite-agent binary not found. Skip plotting the results."
exit 0
fi
# initial annotation
#description="$VLLM_SOURCE_CODE_LOC/.buildkite/nightly-benchmarks/nightly-descriptions.md"
# download results
cd "$VLLM_SOURCE_CODE_LOC/benchmarks"
mkdir -p results/
/workspace/buildkite-agent artifact download 'results/*nightly_results.json' results/
ls
ls results/
# upload benchmark results
zip -r results.zip results/
/workspace/buildkite-agent artifact upload "results.zip"
# upload benchmarking scripts
cd "$VLLM_SOURCE_CODE_LOC/"
zip -r nightly-benchmarks.zip .buildkite/ benchmarks/
/workspace/buildkite-agent artifact upload "nightly-benchmarks.zip"
cd "$VLLM_SOURCE_CODE_LOC/.buildkite/nightly-benchmarks/"
# upload benchmarking pipeline
/workspace/buildkite-agent artifact upload "nightly-pipeline.yaml"
cd "$VLLM_SOURCE_CODE_LOC/.buildkite/nightly-benchmarks/"
/workspace/buildkite-agent annotate --style "success" --context "nightly-benchmarks-results" --append < nightly-annotation.md
# The figures should be generated by a separate process outside the CI/CD pipeline
# # generate figures
# python3 -m pip install tabulate pandas matplotlib
# python3 $VLLM_SOURCE_CODE_LOC/.buildkite/nightly-benchmarks/scripts/generate-nightly-markdown.py \
# --description $description \
# --results-folder results/
# python3 $VLLM_SOURCE_CODE_LOC/.buildkite/nightly-benchmarks/scripts/plot-nightly-results.py \
# --description $description \
# --results-folder results/ \
# --dataset sharegpt
# python3 $VLLM_SOURCE_CODE_LOC/.buildkite/nightly-benchmarks/scripts/plot-nightly-results.py \
# --description $description \
# --results-folder results/ \
# --dataset sonnet_2048_128
# python3 $VLLM_SOURCE_CODE_LOC/.buildkite/nightly-benchmarks/scripts/plot-nightly-results.py \
# --description $description \
# --results-folder results/ \
# --dataset sonnet_128_2048
# # upload results and figures
# /workspace/buildkite-agent artifact upload "nightly_results*.png"
# /workspace/buildkite-agent artifact upload $VLLM_SOURCE_CODE_LOC/.buildkite/nightly-benchmarks/nightly-pipeline.yaml
# /workspace/buildkite-agent artifact upload $VLLM_SOURCE_CODE_LOC/.buildkite/nightly-benchmarks/tests/nightly-tests.json
# /workspace/buildkite-agent annotate --style "success" --context "nightly-benchmarks-results" --append < nightly_results.md
}
main "$@"

View File

@ -0,0 +1,464 @@
#!/bin/bash
set -o pipefail
set -x
check_gpus() {
# check the number of GPUs and GPU type.
declare -g gpu_count=$(nvidia-smi --list-gpus | wc -l)
if [[ $gpu_count -gt 0 ]]; then
echo "GPU found."
else
echo "Need at least 1 GPU to run benchmarking."
exit 1
fi
declare -g gpu_type="$(nvidia-smi --query-gpu=name --format=csv,noheader | awk '{print $2}')"
echo "GPU type is $gpu_type"
}
check_hf_token() {
# check if HF_TOKEN is available and valid
if [[ -z "$HF_TOKEN" ]]; then
echo "Error: HF_TOKEN is not set."
exit 1
elif [[ ! "$HF_TOKEN" =~ ^hf_ ]]; then
echo "Error: HF_TOKEN does not start with 'hf_'."
exit 1
else
echo "HF_TOKEN is set and valid."
fi
}
upload_to_buildkite() {
# upload the benchmarking results to buildkite
# if the agent binary is not found, skip uploading the results, exit 0
if [ ! -f /workspace/buildkite-agent ]; then
echo "buildkite-agent binary not found. Skip uploading the results."
return 0
fi
# /workspace/buildkite-agent annotate --style "success" --context "benchmark-results" --append < $RESULTS_FOLDER/${CURRENT_LLM_SERVING_ENGINE}_nightly_results.md
/workspace/buildkite-agent artifact upload "$RESULTS_FOLDER/*"
}
get_current_llm_serving_engine() {
if which lmdeploy >/dev/null; then
echo "Container: lmdeploy"
export CURRENT_LLM_SERVING_ENGINE=lmdeploy
return
fi
if [ -e /tgi-entrypoint.sh ]; then
echo "Container: tgi"
export CURRENT_LLM_SERVING_ENGINE=tgi
return
fi
if which trtllm-build >/dev/null; then
echo "Container: tensorrt-llm"
export CURRENT_LLM_SERVING_ENGINE=trt
return
fi
if [ -e /sgl-workspace ]; then
echo "Container: sglang"
export CURRENT_LLM_SERVING_ENGINE=sglang
return
fi
if [ -e /vllm-workspace ]; then
echo "Container: vllm"
# move to a completely irrelevant directory, to avoid import vllm from current folder
export CURRENT_LLM_SERVING_ENGINE=vllm
return
fi
}
json2args() {
# transforms the JSON string to command line args, and '_' is replaced to '-'
# example:
# input: { "model": "meta-llama/Llama-2-7b-chat-hf", "tensor_parallel_size": 1 }
# output: --model meta-llama/Llama-2-7b-chat-hf --tensor-parallel-size 1
local json_string=$1
local args=$(
echo "$json_string" | jq -r '
to_entries |
map("--" + (.key | gsub("_"; "-")) + " " + (.value | tostring)) |
join(" ")
'
)
echo "$args"
}
kill_gpu_processes() {
pkill -f '[p]ython'
pkill -f '[p]ython3'
pkill -f '[t]ritonserver'
pkill -f '[p]t_main_thread'
pkill -f '[t]ext-generation'
pkill -f '[l]mdeploy'
# vLLM now names the process with VLLM prefix after https://github.com/vllm-project/vllm/pull/21445
pkill -f '[V]LLM'
while [ "$(nvidia-smi --query-gpu=memory.used --format=csv,noheader,nounits | head -n 1)" -ge 1000 ]; do
sleep 1
done
}
wait_for_server() {
# wait for vllm server to start
# return 1 if vllm server crashes
timeout 1200 bash -c '
until curl -s localhost:8000/v1/completions > /dev/null; do
sleep 1
done' && return 0 || return 1
}
ensure_installed() {
# Ensure that the given command is installed by apt-get
local cmd=$1
if ! which "$cmd" >/dev/null; then
apt-get update && apt-get install -y "$cmd"
fi
}
run_serving_tests() {
# run serving tests using `vllm bench serve` command
# $1: a json file specifying serving test cases
local serving_test_file
serving_test_file=$1
# Iterate over serving tests
jq -c '.[]' "$serving_test_file" | while read -r params; do
# get the test name, and append the GPU type back to it.
test_name=$(echo "$params" | jq -r '.test_name')
# if TEST_SELECTOR is set, only run the test cases that match the selector
if [[ -n "$TEST_SELECTOR" ]] && [[ ! "$test_name" =~ $TEST_SELECTOR ]]; then
echo "Skip test case $test_name."
continue
fi
# prepend the current serving engine to the test name
test_name=${CURRENT_LLM_SERVING_ENGINE}_${test_name}
# get common parameters
common_params=$(echo "$params" | jq -r '.common_parameters')
model=$(echo "$common_params" | jq -r '.model')
tp=$(echo "$common_params" | jq -r '.tp')
dataset_name=$(echo "$common_params" | jq -r '.dataset_name')
dataset_path=$(echo "$common_params" | jq -r '.dataset_path')
port=$(echo "$common_params" | jq -r '.port')
num_prompts=$(echo "$common_params" | jq -r '.num_prompts')
reuse_server=$(echo "$common_params" | jq -r '.reuse_server')
# get client and server arguments
server_params=$(echo "$params" | jq -r ".${CURRENT_LLM_SERVING_ENGINE}_server_parameters")
client_params=$(echo "$params" | jq -r ".${CURRENT_LLM_SERVING_ENGINE}_client_parameters")
client_args=$(json2args "$client_params")
qps_list=$(echo "$params" | jq -r '.qps_list')
qps_list=$(echo "$qps_list" | jq -r '.[] | @sh')
echo "Running over qps list $qps_list"
# check if there is enough GPU to run the test
if [[ $gpu_count -lt $tp ]]; then
echo "Required num-shard $tp but only $gpu_count GPU found. Skip testcase $test_name."
continue
fi
if [[ $reuse_server == "true" ]]; then
echo "Reuse previous server for test case $test_name"
else
kill_gpu_processes
bash "$VLLM_SOURCE_CODE_LOC/.buildkite/nightly-benchmarks/scripts/launch-server.sh" \
"$server_params" "$common_params"
fi
if wait_for_server; then
echo ""
echo "$CURRENT_LLM_SERVING_ENGINE server is up and running."
else
echo ""
echo "$CURRENT_LLM_SERVING_ENGINE failed to start within the timeout period."
break
fi
# prepare tokenizer
# this is required for lmdeploy.
cd "$VLLM_SOURCE_CODE_LOC/benchmarks"
rm -rf /tokenizer_cache
mkdir /tokenizer_cache
python3 ../.buildkite/nightly-benchmarks/scripts/download-tokenizer.py \
--model "$model" \
--cachedir /tokenizer_cache
cd "$VLLM_SOURCE_CODE_LOC/benchmarks"
# change model name for lmdeploy (it will not follow standard hf name)
if [[ "$CURRENT_LLM_SERVING_ENGINE" == "lmdeploy" ]]; then
model=$(python ../.buildkite/nightly-benchmarks/scripts/get-lmdeploy-modelname.py)
fi
# iterate over different QPS
for qps in $qps_list; do
# remove the surrounding single quote from qps
if [[ "$qps" == *"inf"* ]]; then
echo "qps was $qps"
qps="inf"
echo "now qps is $qps"
fi
new_test_name=$test_name"_qps_"$qps
backend=$CURRENT_LLM_SERVING_ENGINE
if [[ $backend = "trt" ]]; then
backend="tensorrt-llm"
fi
if [[ "$backend" == *"vllm"* ]]; then
backend="vllm"
fi
if [[ "$dataset_name" = "sharegpt" ]]; then
client_command="vllm bench serve \
--backend $backend \
--tokenizer /tokenizer_cache \
--model $model \
--dataset-name $dataset_name \
--dataset-path $dataset_path \
--num-prompts $num_prompts \
--port $port \
--save-result \
--result-dir $RESULTS_FOLDER \
--result-filename ${new_test_name}.json \
--request-rate $qps \
--ignore-eos \
$client_args"
elif [[ "$dataset_name" = "sonnet" ]]; then
sonnet_input_len=$(echo "$common_params" | jq -r '.sonnet_input_len')
sonnet_output_len=$(echo "$common_params" | jq -r '.sonnet_output_len')
sonnet_prefix_len=$(echo "$common_params" | jq -r '.sonnet_prefix_len')
client_command="vllm bench serve \
--backend $backend \
--tokenizer /tokenizer_cache \
--model $model \
--dataset-name $dataset_name \
--dataset-path $dataset_path \
--num-prompts $num_prompts \
--sonnet-input-len $sonnet_input_len \
--sonnet-output-len $sonnet_output_len \
--sonnet-prefix-len $sonnet_prefix_len \
--port $port \
--save-result \
--result-dir $RESULTS_FOLDER \
--result-filename ${new_test_name}.json \
--request-rate $qps \
--ignore-eos \
$client_args"
else
echo "The dataset name must be either 'sharegpt' or 'sonnet'. Got $dataset_name."
exit 1
fi
echo "Running test case $test_name with qps $qps"
echo "Client command: $client_command"
eval "$client_command"
server_command="None"
# record the benchmarking commands
jq_output=$(jq -n \
--arg server "$server_command" \
--arg client "$client_command" \
--arg gpu "$gpu_type" \
--arg engine "$CURRENT_LLM_SERVING_ENGINE" \
'{
server_command: $server,
client_command: $client,
gpu_type: $gpu,
engine: $engine
}')
echo "$jq_output" >"$RESULTS_FOLDER/${new_test_name}.commands"
done
done
kill_gpu_processes
}
run_genai_perf_tests() {
# run genai-perf tests
# $1: a json file specifying genai-perf test cases
local genai_perf_test_file
genai_perf_test_file=$1
# Iterate over genai-perf tests
jq -c '.[]' "$genai_perf_test_file" | while read -r params; do
# get the test name, and append the GPU type back to it.
test_name=$(echo "$params" | jq -r '.test_name')
# if TEST_SELECTOR is set, only run the test cases that match the selector
if [[ -n "$TEST_SELECTOR" ]] && [[ ! "$test_name" =~ $TEST_SELECTOR ]]; then
echo "Skip test case $test_name."
continue
fi
# prepend the current serving engine to the test name
test_name=${CURRENT_LLM_SERVING_ENGINE}_${test_name}
# get common parameters
common_params=$(echo "$params" | jq -r '.common_parameters')
model=$(echo "$common_params" | jq -r '.model')
tp=$(echo "$common_params" | jq -r '.tp')
dataset_name=$(echo "$common_params" | jq -r '.dataset_name')
dataset_path=$(echo "$common_params" | jq -r '.dataset_path')
port=$(echo "$common_params" | jq -r '.port')
num_prompts=$(echo "$common_params" | jq -r '.num_prompts')
reuse_server=$(echo "$common_params" | jq -r '.reuse_server')
# get client and server arguments
server_params=$(echo "$params" | jq -r ".${CURRENT_LLM_SERVING_ENGINE}_server_parameters")
qps_list=$(echo "$params" | jq -r '.qps_list')
qps_list=$(echo "$qps_list" | jq -r '.[] | @sh')
echo "Running over qps list $qps_list"
# check if there is enough GPU to run the test
if [[ $gpu_count -lt $tp ]]; then
echo "Required num-shard $tp but only $gpu_count GPU found. Skip testcase $test_name."
continue
fi
if [[ $reuse_server == "true" ]]; then
echo "Reuse previous server for test case $test_name"
else
kill_gpu_processes
bash "$VLLM_SOURCE_CODE_LOC/.buildkite/nightly-benchmarks/scripts/launch-server.sh" \
"$server_params" "$common_params"
fi
if wait_for_server; then
echo ""
echo "$CURRENT_LLM_SERVING_ENGINE server is up and running."
else
echo ""
echo "$CURRENT_LLM_SERVING_ENGINE failed to start within the timeout period."
break
fi
# iterate over different QPS
for qps in $qps_list; do
# remove the surrounding single quote from qps
if [[ "$qps" == *"inf"* ]]; then
echo "qps was $qps"
qps=$num_prompts
echo "now qps is $qps"
fi
new_test_name=$test_name"_qps_"$qps
backend=$CURRENT_LLM_SERVING_ENGINE
if [[ "$backend" == *"vllm"* ]]; then
backend="vllm"
fi
#TODO: add output dir.
client_command="genai-perf profile \
-m $model \
--service-kind openai \
--backend vllm \
--endpoint-type chat \
--streaming \
--url localhost:$port \
--request-rate $qps \
--num-prompts $num_prompts \
"
echo "Client command: $client_command"
eval "$client_command"
#TODO: process/record outputs
done
done
kill_gpu_processes
}
prepare_dataset() {
# download sharegpt dataset
cd "$VLLM_SOURCE_CODE_LOC/benchmarks"
wget https://huggingface.co/datasets/anon8231489123/ShareGPT_Vicuna_unfiltered/resolve/main/ShareGPT_V3_unfiltered_cleaned_split.json
# duplicate sonnet by 4x, to allow benchmarking with input length 2048
cd "$VLLM_SOURCE_CODE_LOC/benchmarks"
echo "" > sonnet_4x.txt
for _ in {1..4}
do
cat sonnet.txt >> sonnet_4x.txt
done
}
main() {
# check if the environment variable is successfully injected from yaml
check_gpus
check_hf_token
get_current_llm_serving_engine
pip install -U transformers
pip install -r requirements/dev.txt
which genai-perf
# check storage
df -h
ensure_installed wget
ensure_installed curl
ensure_installed jq
# genai-perf dependency
ensure_installed libb64-0d
prepare_dataset
cd "$VLLM_SOURCE_CODE_LOC/benchmarks"
declare -g RESULTS_FOLDER=results/
mkdir -p $RESULTS_FOLDER
BENCHMARK_ROOT="$VLLM_SOURCE_CODE_LOC/.buildkite/nightly-benchmarks/"
# run the test
run_serving_tests "$BENCHMARK_ROOT/tests/nightly-tests.json"
# run genai-perf tests
run_genai_perf_tests "$BENCHMARK_ROOT/tests/genai-perf-tests.json"
mv artifacts/ $RESULTS_FOLDER/
# upload benchmark results to buildkite
python3 -m pip install tabulate pandas
python3 "$BENCHMARK_ROOT/scripts/summary-nightly-results.py"
upload_to_buildkite
}
main "$@"

View File

@ -15,8 +15,6 @@ check_gpus() {
declare -g gpu_count=$(nvidia-smi --list-gpus | wc -l)
elif command -v amd-smi; then
declare -g gpu_count=$(amd-smi list | grep 'GPU' | wc -l)
elif command -v hl-smi; then
declare -g gpu_count=$(hl-smi --list | grep -i "Module ID" | wc -l)
fi
if [[ $gpu_count -gt 0 ]]; then
@ -25,16 +23,10 @@ check_gpus() {
echo "Need at least 1 GPU to run benchmarking."
exit 1
fi
declare -g arch_suffix=''
if command -v nvidia-smi; then
declare -g gpu_type=$(nvidia-smi --query-gpu=name --format=csv,noheader | awk '{print $2}')
elif command -v amd-smi; then
declare -g gpu_type=$(amd-smi static -g 0 -a | grep 'MARKET_NAME' | awk '{print $2}')
elif command -v hl-smi; then
declare -g gpu_type=$(hl-smi -q | grep "Product Name" | head -n 1 | awk -F ':' '{print $2}' | sed 's/^ *//')
arch_suffix='-hpu'
fi
echo "GPU type is $gpu_type"
}
@ -146,10 +138,6 @@ kill_gpu_processes() {
while [ "$(amd-smi metric -g 0 | grep 'USED_VRAM' | awk '{print $2}')" -ge 1000 ]; do
sleep 1
done
elif command -v hl-smi; then
while [ "$(hl-smi -q | grep "Used" | head -n 1 | awk '{print $3}')" -ge 1000 ]; do
sleep 1
done
fi
# remove vllm config file
@ -206,11 +194,9 @@ run_latency_tests() {
# check if there is enough GPU to run the test
tp=$(echo "$latency_params" | jq -r '.tensor_parallel_size')
if [ "$ON_CPU" == "1" ]; then
pp=$(echo "$latency_params" | jq -r '.pipeline_parallel_size')
world_size=$(($tp*$pp))
if [[ $numa_count -lt $world_size && -z "${REMOTE_HOST}" ]]; then
echo "Required world-size $world_size but only $numa_count NUMA nodes found. Skip testcase $test_name."
if [ "$ON_CPU" == "1" ];then
if [[ $numa_count -lt $tp ]]; then
echo "Required tensor-parallel-size $tp but only $numa_count NUMA nodes found. Skip testcase $test_name."
continue
fi
else
@ -275,11 +261,9 @@ run_throughput_tests() {
# check if there is enough GPU to run the test
tp=$(echo "$throughput_params" | jq -r '.tensor_parallel_size')
if [ "$ON_CPU" == "1" ]; then
pp=$(echo "$throughput_params" | jq -r '.pipeline_parallel_size')
world_size=$(($tp*$pp))
if [[ $numa_count -lt $world_size && -z "${REMOTE_HOST}" ]]; then
echo "Required world-size $world_size but only $numa_count NUMA nodes found. Skip testcase $test_name."
if [ "$ON_CPU" == "1" ];then
if [[ $numa_count -lt $tp ]]; then
echo "Required tensor-parallel-size $tp but only $numa_count NUMA nodes found. Skip testcase $test_name."
continue
fi
else
@ -345,21 +329,12 @@ run_serving_tests() {
qps_list=$(echo "$params" | jq -r '.qps_list')
qps_list=$(echo "$qps_list" | jq -r '.[] | @sh')
echo "Running over qps list $qps_list"
max_concurrency_list=$(echo "$params" | jq -r '.max_concurrency_list')
if [[ -z "$max_concurrency_list" || "$max_concurrency_list" == "null" ]]; then
num_prompts=$(echo "$client_params" | jq -r '.num_prompts')
max_concurrency_list="[$num_prompts]"
fi
max_concurrency_list=$(echo "$max_concurrency_list" | jq -r '.[] | @sh')
echo "Running over max concurrency list $max_concurrency_list"
# check if there is enough resources to run the test
tp=$(echo "$server_params" | jq -r '.tensor_parallel_size')
if [ "$ON_CPU" == "1" ]; then
pp=$(echo "$server_params" | jq -r '.pipeline_parallel_size')
world_size=$(($tp*$pp))
if [[ $numa_count -lt $world_size && -z "${REMOTE_HOST}" ]]; then
echo "Required world-size $world_size but only $numa_count NUMA nodes found. Skip testcase $test_name."
if [ "$ON_CPU" == "1" ];then
if [[ $numa_count -lt $tp ]]; then
echo "Required tensor-parallel-size $tp but only $numa_count NUMA nodes found. Skip testcase $test_name."
continue
fi
else
@ -377,7 +352,8 @@ run_serving_tests() {
continue
fi
server_command="$server_envs vllm serve \
server_command="$server_envs python3 \
-m vllm.entrypoints.openai.api_server \
$server_args"
# run the server
@ -414,39 +390,35 @@ run_serving_tests() {
echo "now qps is $qps"
fi
# iterate over different max_concurrency
for max_concurrency in $max_concurrency_list; do
new_test_name=$test_name"_qps_"$qps"_concurrency_"$max_concurrency
echo " new test name $new_test_name"
# pass the tensor parallel size to the client so that it can be displayed
# on the benchmark dashboard
client_command="vllm bench serve \
--save-result \
--result-dir $RESULTS_FOLDER \
--result-filename ${new_test_name}.json \
--request-rate $qps \
--max-concurrency $max_concurrency \
--metadata "tensor_parallel_size=$tp" \
$client_args $client_remote_args "
new_test_name=$test_name"_qps_"$qps
echo "Running test case $test_name with qps $qps"
echo "Client command: $client_command"
# pass the tensor parallel size to the client so that it can be displayed
# on the benchmark dashboard
client_command="vllm bench serve \
--save-result \
--result-dir $RESULTS_FOLDER \
--result-filename ${new_test_name}.json \
--request-rate $qps \
--metadata "tensor_parallel_size=$tp" \
$client_args $client_remote_args "
bash -c "$client_command"
echo "Running test case $test_name with qps $qps"
echo "Client command: $client_command"
# record the benchmarking commands
jq_output=$(jq -n \
--arg server "$server_command" \
--arg client "$client_command" \
--arg gpu "$gpu_type" \
'{
server_command: $server,
client_command: $client,
gpu_type: $gpu
}')
echo "$jq_output" >"$RESULTS_FOLDER/${new_test_name}.commands"
bash -c "$client_command"
# record the benchmarking commands
jq_output=$(jq -n \
--arg server "$server_command" \
--arg client "$client_command" \
--arg gpu "$gpu_type" \
'{
server_command: $server,
client_command: $client,
gpu_type: $gpu
}')
echo "$jq_output" >"$RESULTS_FOLDER/${new_test_name}.commands"
done
done
# clean up
@ -463,10 +435,14 @@ main() {
ARCH='-cpu'
else
check_gpus
ARCH="$arch_suffix"
fi
check_hf_token
# Set to v1 to run v1 benchmark
if [[ "${ENGINE_VERSION:-v0}" == "v1" ]]; then
export VLLM_USE_V1=1
fi
# dependencies
(which wget && which curl) || (apt-get update && apt-get install -y wget curl)
(which jq) || (apt-get update && apt-get -y install jq)
@ -482,12 +458,7 @@ main() {
ensure_sharegpt_downloaded
declare -g RESULTS_FOLDER=results/
mkdir -p $RESULTS_FOLDER
QUICK_BENCHMARK_ROOT=../.buildkite/performance-benchmarks/
# dump vllm info via vllm collect-env
env_output=$(vllm collect-env)
echo "$env_output" >"$RESULTS_FOLDER/vllm_env.txt"
QUICK_BENCHMARK_ROOT=../.buildkite/nightly-benchmarks/
# benchmarking
run_serving_tests $QUICK_BENCHMARK_ROOT/tests/"${SERVING_JSON:-serving-tests$ARCH.json}"

View File

@ -0,0 +1,82 @@
# SPDX-License-Identifier: Apache-2.0
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
import datetime
import json
import os
from pathlib import Path
import pandas as pd
from tabulate import tabulate
results_folder = Path("results/")
# serving results and the keys that will be printed into markdown
serving_results = []
serving_column_mapping = {
"test_name": "Test name",
"gpu_type": "GPU",
"completed": "Successful req.",
"request_throughput": "Tput (req/s)",
"mean_ttft_ms": "Mean TTFT (ms)",
"std_ttft_ms": "Std TTFT (ms)",
"median_ttft_ms": "Median TTFT (ms)",
"mean_itl_ms": "Mean ITL (ms)",
"std_itl_ms": "Std ITL (ms)",
"median_itl_ms": "Median ITL (ms)",
"mean_tpot_ms": "Mean TPOT (ms)",
"std_tpot_ms": "Std TPOT (ms)",
"median_tpot_ms": "Median TPOT (ms)",
"total_token_throughput": "Total Token Tput (tok/s)",
"output_throughput": "Output Tput (tok/s)",
"total_input_tokens": "Total input tokens",
"total_output_tokens": "Total output tokens",
"engine": "Engine",
}
if __name__ == "__main__":
# collect results
for test_file in results_folder.glob("*.json"):
with open(test_file) as f:
raw_result = json.loads(f.read())
# attach the benchmarking command to raw_result
with open(test_file.with_suffix(".commands")) as f:
command = json.loads(f.read())
raw_result.update(command)
# update the test name of this result
raw_result.update({"test_name": test_file.stem})
# add the result to raw_result
serving_results.append(raw_result)
continue
serving_results = pd.DataFrame.from_dict(serving_results)
if not serving_results.empty:
serving_results = serving_results[list(serving_column_mapping.keys())].rename(
columns=serving_column_mapping
)
serving_md_table_with_headers = tabulate(
serving_results, headers="keys", tablefmt="pipe", showindex=False
)
# remove the first line of header
serving_md_table_lines = serving_md_table_with_headers.split("\n")
serving_md_table_without_header = "\n".join(serving_md_table_lines[2:])
prefix = datetime.datetime.now().strftime("%Y-%m-%d_%H-%M-%S")
prefix = prefix + "_" + os.environ.get("CURRENT_LLM_SERVING_ENGINE")
# document benchmarking results in markdown
with open(results_folder / f"{prefix}_nightly_results.md", "w") as f:
# document results with header.
# for those who wants to reproduce our benchmark.
f.write(serving_md_table_with_headers)
f.write("\n")
# document benchmarking results in json
with open(results_folder / f"{prefix}_nightly_results.json", "w") as f:
results = serving_results.to_dict(orient="records")
f.write(json.dumps(results))

View File

@ -0,0 +1,23 @@
#!/bin/sh
TOKEN=$(curl -s -L "https://public.ecr.aws/token?service=public.ecr.aws&scope=repository:q9t5s3a7/vllm-ci-postmerge-repo:pull" | jq -r .token)
if [[ "$BUILDKITE_BRANCH" == "main" ]]; then
URL="https://public.ecr.aws/v2/q9t5s3a7/vllm-ci-postmerge-repo/manifests/$BUILDKITE_COMMIT"
else
URL="https://public.ecr.aws/v2/q9t5s3a7/vllm-ci-test-repo/manifests/$BUILDKITE_COMMIT"
fi
TIMEOUT_SECONDS=10
retries=0
while [ $retries -lt 1000 ]; do
if [ "$(curl -s --max-time "$TIMEOUT_SECONDS" -L -H "Authorization: Bearer $TOKEN" -o /dev/null -w "%{http_code}" "$URL")" -eq 200 ]; then
exit 0
fi
echo "Waiting for image to be available..."
retries=$((retries + 1))
sleep 5
done
exit 1

View File

@ -0,0 +1,30 @@
[
{
"test_name": "latency_llama8B_tp1",
"environment_variables": {
"VLLM_ALLOW_LONG_MAX_MODEL_LEN": 1,
"VLLM_CPU_KVCACHE_SPACE": 40
},
"parameters": {
"model": "meta-llama/Meta-Llama-3.1-8B-Instruct",
"tensor_parallel_size": 1,
"load_format": "dummy",
"num_iters_warmup": 5,
"num_iters": 15
}
},
{
"test_name": "latency_llama8B_tp4",
"environment_variables": {
"VLLM_ALLOW_LONG_MAX_MODEL_LEN": 1,
"VLLM_CPU_KVCACHE_SPACE": 40
},
"parameters": {
"model": "meta-llama/Meta-Llama-3.1-8B-Instruct",
"tensor_parallel_size": 4,
"load_format": "dummy",
"num_iters_warmup": 5,
"num_iters": 15
}
}
]

View File

@ -2,7 +2,6 @@
{
"test_name": "serving_llama8B_tp1_sharegpt",
"qps_list": [1, 4, 16, "inf"],
"max_concurrency_list": [32],
"server_environment_variables": {
"VLLM_RPC_TIMEOUT": 100000,
"VLLM_ALLOW_LONG_MAX_MODEL_LEN": 1,
@ -11,7 +10,7 @@
"VLLM_CPU_KVCACHE_SPACE": 40
},
"server_parameters": {
"model": "meta-llama/Llama-3.1-8B-Instruct",
"model": "meta-llama/Meta-Llama-3.1-8B-Instruct",
"tensor_parallel_size": 1,
"dtype": "bfloat16",
"distributed_executor_backend": "mp",
@ -24,17 +23,17 @@
"load_format": "dummy"
},
"client_parameters": {
"model": "meta-llama/Llama-3.1-8B-Instruct",
"model": "meta-llama/Meta-Llama-3.1-8B-Instruct",
"backend": "vllm",
"dataset_name": "sharegpt",
"dataset_path": "./ShareGPT_V3_unfiltered_cleaned_split.json",
"num_prompts": 32
"max_concurrency": 60,
"num_prompts": 200
}
},
{
"test_name": "serving_llama8B_tp2_sharegpt",
"qps_list": [1, 4, 16, "inf"],
"max_concurrency_list": [32],
"server_environment_variables": {
"VLLM_RPC_TIMEOUT": 100000,
"VLLM_ALLOW_LONG_MAX_MODEL_LEN": 1,
@ -43,7 +42,7 @@
"VLLM_CPU_KVCACHE_SPACE": 40
},
"server_parameters": {
"model": "meta-llama/Llama-3.1-8B-Instruct",
"model": "meta-llama/Meta-Llama-3.1-8B-Instruct",
"tensor_parallel_size": 2,
"dtype": "bfloat16",
"distributed_executor_backend": "mp",
@ -56,17 +55,49 @@
"load_format": "dummy"
},
"client_parameters": {
"model": "meta-llama/Llama-3.1-8B-Instruct",
"model": "meta-llama/Meta-Llama-3.1-8B-Instruct",
"backend": "vllm",
"dataset_name": "sharegpt",
"dataset_path": "./ShareGPT_V3_unfiltered_cleaned_split.json",
"num_prompts": 32
"max_concurrency": 60,
"num_prompts": 200
}
},
{
"test_name": "serving_llama8B_tp4_sharegpt",
"qps_list": [1, 4, 16, "inf"],
"server_environment_variables": {
"VLLM_RPC_TIMEOUT": 100000,
"VLLM_ALLOW_LONG_MAX_MODEL_LEN": 1,
"VLLM_ENGINE_ITERATION_TIMEOUT_S": 120,
"VLLM_CPU_SGL_KERNEL": 1,
"VLLM_CPU_KVCACHE_SPACE": 40
},
"server_parameters": {
"model": "meta-llama/Meta-Llama-3.1-8B-Instruct",
"tensor_parallel_size": 4,
"dtype": "bfloat16",
"distributed_executor_backend": "mp",
"block_size": 128,
"trust_remote_code": "",
"disable_log_stats": "",
"enforce_eager": "",
"max_num_batched_tokens": 2048,
"max_num_seqs": 256,
"load_format": "dummy"
},
"client_parameters": {
"model": "meta-llama/Meta-Llama-3.1-8B-Instruct",
"backend": "vllm",
"dataset_name": "sharegpt",
"dataset_path": "./ShareGPT_V3_unfiltered_cleaned_split.json",
"max_concurrency": 60,
"num_prompts": 200
}
},
{
"test_name": "serving_llama8B_tp1_random_128_128",
"qps_list": [1, 4, 16, "inf"],
"max_concurrency_list": [32],
"server_environment_variables": {
"VLLM_RPC_TIMEOUT": 100000,
"VLLM_ALLOW_LONG_MAX_MODEL_LEN": 1,
@ -75,7 +106,7 @@
"VLLM_CPU_KVCACHE_SPACE": 40
},
"server_parameters": {
"model": "meta-llama/Llama-3.1-8B-Instruct",
"model": "meta-llama/Meta-Llama-3.1-8B-Instruct",
"tensor_parallel_size": 1,
"dtype": "bfloat16",
"distributed_executor_backend": "mp",
@ -89,19 +120,19 @@
"load_format": "dummy"
},
"client_parameters": {
"model": "meta-llama/Llama-3.1-8B-Instruct",
"model": "meta-llama/Meta-Llama-3.1-8B-Instruct",
"backend": "vllm",
"dataset_name": "random",
"random-input-len": 128,
"random-output-len": 128,
"ignore-eos": "",
"num_prompts": 32
"max_concurrency": 1000,
"num_prompts": 1000
}
},
{
"test_name": "serving_llama8B_tp2_random_128_128",
"qps_list": [1, 4, 16, "inf"],
"max_concurrency_list": [32],
"server_environment_variables": {
"VLLM_RPC_TIMEOUT": 100000,
"VLLM_ALLOW_LONG_MAX_MODEL_LEN": 1,
@ -110,7 +141,7 @@
"VLLM_CPU_KVCACHE_SPACE": 40
},
"server_parameters": {
"model": "meta-llama/Llama-3.1-8B-Instruct",
"model": "meta-llama/Meta-Llama-3.1-8B-Instruct",
"tensor_parallel_size": 2,
"dtype": "bfloat16",
"distributed_executor_backend": "mp",
@ -124,19 +155,19 @@
"load_format": "dummy"
},
"client_parameters": {
"model": "meta-llama/Llama-3.1-8B-Instruct",
"model": "meta-llama/Meta-Llama-3.1-8B-Instruct",
"backend": "vllm",
"dataset_name": "random",
"random-input-len": 128,
"random-output-len": 128,
"ignore-eos": "",
"num_prompts": 32
"max_concurrency": 1000,
"num_prompts": 1000
}
},
{
"test_name": "serving_llama8B_tp1_random_128_2048",
"test_name": "serving_llama8B_tp4_random_128_128",
"qps_list": [1, 4, 16, "inf"],
"max_concurrency_list": [32],
"server_environment_variables": {
"VLLM_RPC_TIMEOUT": 100000,
"VLLM_ALLOW_LONG_MAX_MODEL_LEN": 1,
@ -145,8 +176,8 @@
"VLLM_CPU_KVCACHE_SPACE": 40
},
"server_parameters": {
"model": "meta-llama/Llama-3.1-8B-Instruct",
"tensor_parallel_size": 1,
"model": "meta-llama/Meta-Llama-3.1-8B-Instruct",
"tensor_parallel_size": 4,
"dtype": "bfloat16",
"distributed_executor_backend": "mp",
"block_size": 128,
@ -159,118 +190,14 @@
"load_format": "dummy"
},
"client_parameters": {
"model": "meta-llama/Llama-3.1-8B-Instruct",
"model": "meta-llama/Meta-Llama-3.1-8B-Instruct",
"backend": "vllm",
"dataset_name": "random",
"random-input-len": 128,
"random-output-len": 2048,
"ignore-eos": "",
"num_prompts": 32
}
},
{
"test_name": "serving_llama8B_tp2_random_128_2048",
"qps_list": [1, 4, 16, "inf"],
"max_concurrency_list": [32],
"server_environment_variables": {
"VLLM_RPC_TIMEOUT": 100000,
"VLLM_ALLOW_LONG_MAX_MODEL_LEN": 1,
"VLLM_ENGINE_ITERATION_TIMEOUT_S": 120,
"VLLM_CPU_SGL_KERNEL": 1,
"VLLM_CPU_KVCACHE_SPACE": 40
},
"server_parameters": {
"model": "meta-llama/Llama-3.1-8B-Instruct",
"tensor_parallel_size": 2,
"dtype": "bfloat16",
"distributed_executor_backend": "mp",
"block_size": 128,
"trust_remote_code": "",
"enable_chunked_prefill": "",
"disable_log_stats": "",
"enforce_eager": "",
"max_num_batched_tokens": 2048,
"max_num_seqs": 256,
"load_format": "dummy"
},
"client_parameters": {
"model": "meta-llama/Llama-3.1-8B-Instruct",
"backend": "vllm",
"dataset_name": "random",
"random-input-len": 128,
"random-output-len": 2048,
"ignore-eos": "",
"num_prompts": 32
}
},
{
"test_name": "serving_llama8B_tp1_random_2048_128",
"qps_list": [1, 4, 16, "inf"],
"max_concurrency_list": [32],
"server_environment_variables": {
"VLLM_RPC_TIMEOUT": 100000,
"VLLM_ALLOW_LONG_MAX_MODEL_LEN": 1,
"VLLM_ENGINE_ITERATION_TIMEOUT_S": 120,
"VLLM_CPU_SGL_KERNEL": 1,
"VLLM_CPU_KVCACHE_SPACE": 40
},
"server_parameters": {
"model": "meta-llama/Llama-3.1-8B-Instruct",
"tensor_parallel_size": 1,
"dtype": "bfloat16",
"distributed_executor_backend": "mp",
"block_size": 128,
"trust_remote_code": "",
"enable_chunked_prefill": "",
"disable_log_stats": "",
"enforce_eager": "",
"max_num_batched_tokens": 2048,
"max_num_seqs": 256,
"load_format": "dummy"
},
"client_parameters": {
"model": "meta-llama/Llama-3.1-8B-Instruct",
"backend": "vllm",
"dataset_name": "random",
"random-input-len": 2048,
"random-output-len": 128,
"ignore-eos": "",
"num_prompts": 32
}
},
{
"test_name": "serving_llama8B_tp2_random_2048_128",
"qps_list": [1, 4, 16, "inf"],
"max_concurrency_list": [32],
"server_environment_variables": {
"VLLM_RPC_TIMEOUT": 100000,
"VLLM_ALLOW_LONG_MAX_MODEL_LEN": 1,
"VLLM_ENGINE_ITERATION_TIMEOUT_S": 120,
"VLLM_CPU_SGL_KERNEL": 1,
"VLLM_CPU_KVCACHE_SPACE": 40
},
"server_parameters": {
"model": "meta-llama/Llama-3.1-8B-Instruct",
"tensor_parallel_size": 2,
"dtype": "bfloat16",
"distributed_executor_backend": "mp",
"block_size": 128,
"trust_remote_code": "",
"enable_chunked_prefill": "",
"disable_log_stats": "",
"enforce_eager": "",
"max_num_batched_tokens": 2048,
"max_num_seqs": 256,
"load_format": "dummy"
},
"client_parameters": {
"model": "meta-llama/Llama-3.1-8B-Instruct",
"backend": "vllm",
"dataset_name": "random",
"random-input-len": 2048,
"random-output-len": 128,
"ignore-eos": "",
"num_prompts": 32
"max_concurrency": 1000,
"num_prompts": 1000
}
}
]

View File

@ -0,0 +1,205 @@
[
{
"test_name": "serving_llama8B_pp1_sharegpt",
"qps_list": [1, 4, 16, "inf"],
"server_environment_variables": {
"VLLM_RPC_TIMEOUT": 100000,
"VLLM_ALLOW_LONG_MAX_MODEL_LEN": 1,
"VLLM_ENGINE_ITERATION_TIMEOUT_S": 120,
"VLLM_CPU_SGL_KERNEL": 1,
"VLLM_CPU_KVCACHE_SPACE": 40
},
"server_parameters": {
"model": "meta-llama/Meta-Llama-3.1-8B-Instruct",
"pipeline_parallel_size": 1,
"dtype": "bfloat16",
"distributed_executor_backend": "mp",
"block_size": 128,
"trust_remote_code": "",
"disable_log_stats": "",
"enforce_eager": "",
"max_num_batched_tokens": 2048,
"max_num_seqs": 256,
"load_format": "dummy"
},
"client_parameters": {
"model": "meta-llama/Meta-Llama-3.1-8B-Instruct",
"backend": "vllm",
"dataset_name": "sharegpt",
"dataset_path": "./ShareGPT_V3_unfiltered_cleaned_split.json",
"max_concurrency": 60,
"num_prompts": 200
}
},
{
"test_name": "serving_llama8B_pp3_sharegpt",
"qps_list": [1, 4, 16, "inf"],
"server_environment_variables": {
"VLLM_RPC_TIMEOUT": 100000,
"VLLM_ALLOW_LONG_MAX_MODEL_LEN": 1,
"VLLM_ENGINE_ITERATION_TIMEOUT_S": 120,
"VLLM_CPU_SGL_KERNEL": 1,
"VLLM_CPU_KVCACHE_SPACE": 40
},
"server_parameters": {
"model": "meta-llama/Meta-Llama-3.1-8B-Instruct",
"pipeline_parallel_size": 3,
"dtype": "bfloat16",
"distributed_executor_backend": "mp",
"block_size": 128,
"trust_remote_code": "",
"disable_log_stats": "",
"enforce_eager": "",
"max_num_batched_tokens": 2048,
"max_num_seqs": 256,
"load_format": "dummy"
},
"client_parameters": {
"model": "meta-llama/Meta-Llama-3.1-8B-Instruct",
"backend": "vllm",
"dataset_name": "sharegpt",
"dataset_path": "./ShareGPT_V3_unfiltered_cleaned_split.json",
"max_concurrency": 60,
"num_prompts": 200
}
},
{
"test_name": "serving_llama8B_tp2pp6_sharegpt",
"qps_list": [1, 4, 16, "inf"],
"server_environment_variables": {
"VLLM_RPC_TIMEOUT": 100000,
"VLLM_ALLOW_LONG_MAX_MODEL_LEN": 1,
"VLLM_ENGINE_ITERATION_TIMEOUT_S": 120,
"VLLM_CPU_SGL_KERNEL": 1,
"VLLM_CPU_KVCACHE_SPACE": 40
},
"server_parameters": {
"model": "meta-llama/Meta-Llama-3.1-8B-Instruct",
"tensor_parallel_size": 2,
"pipeline_parallel_size": 3,
"dtype": "bfloat16",
"distributed_executor_backend": "mp",
"block_size": 128,
"trust_remote_code": "",
"disable_log_stats": "",
"enforce_eager": "",
"max_num_batched_tokens": 2048,
"max_num_seqs": 256,
"load_format": "dummy"
},
"client_parameters": {
"model": "meta-llama/Meta-Llama-3.1-8B-Instruct",
"backend": "vllm",
"dataset_name": "sharegpt",
"dataset_path": "./ShareGPT_V3_unfiltered_cleaned_split.json",
"max_concurrency": 60,
"num_prompts": 200
}
},
{
"test_name": "serving_llama8B_pp1_random_128_128",
"qps_list": [1, 4, 16, "inf"],
"server_environment_variables": {
"VLLM_RPC_TIMEOUT": 100000,
"VLLM_ALLOW_LONG_MAX_MODEL_LEN": 1,
"VLLM_ENGINE_ITERATION_TIMEOUT_S": 120,
"VLLM_CPU_SGL_KERNEL": 1,
"VLLM_CPU_KVCACHE_SPACE": 40
},
"server_parameters": {
"model": "meta-llama/Meta-Llama-3.1-8B-Instruct",
"pipeline_parallel_size": 1,
"dtype": "bfloat16",
"distributed_executor_backend": "mp",
"block_size": 128,
"trust_remote_code": "",
"enable_chunked_prefill": "",
"disable_log_stats": "",
"enforce_eager": "",
"max_num_batched_tokens": 2048,
"max_num_seqs": 256,
"load_format": "dummy"
},
"client_parameters": {
"model": "meta-llama/Meta-Llama-3.1-8B-Instruct",
"backend": "vllm",
"dataset_name": "random",
"random-input-len": 128,
"random-output-len": 128,
"ignore-eos": "",
"max_concurrency": 1000,
"num_prompts": 1000
}
},
{
"test_name": "serving_llama8B_pp3_random_128_128",
"qps_list": [1, 4, 16, "inf"],
"server_environment_variables": {
"VLLM_RPC_TIMEOUT": 100000,
"VLLM_ALLOW_LONG_MAX_MODEL_LEN": 1,
"VLLM_ENGINE_ITERATION_TIMEOUT_S": 120,
"VLLM_CPU_SGL_KERNEL:": 1,
"VLLM_CPU_KVCACHE_SPACE": 40
},
"server_parameters": {
"model": "meta-llama/Meta-Llama-3.1-8B-Instruct",
"pipeline_parallel_size": 3,
"dtype": "bfloat16",
"distributed_executor_backend": "mp",
"block_size": 128,
"trust_remote_code": "",
"enable_chunked_prefill": "",
"disable_log_stats": "",
"enforce_eager": "",
"max_num_batched_tokens": 2048,
"max_num_seqs": 256,
"load_format": "dummy"
},
"client_parameters": {
"model": "meta-llama/Meta-Llama-3.1-8B-Instruct",
"backend": "vllm",
"dataset_name": "random",
"random-input-len": 128,
"random-output-len": 128,
"ignore-eos": "",
"max_concurrency": 1000,
"num_prompts": 1000
}
},
{
"test_name": "serving_llama8B_tp2pp3_random_128_128",
"qps_list": [1, 4, 16, "inf"],
"server_environment_variables": {
"VLLM_RPC_TIMEOUT": 100000,
"VLLM_ALLOW_LONG_MAX_MODEL_LEN": 1,
"VLLM_ENGINE_ITERATION_TIMEOUT_S": 120,
"VLLM_CPU_SGL_KERNEL": 1,
"VLLM_CPU_KVCACHE_SPACE": 40
},
"server_parameters": {
"model": "meta-llama/Meta-Llama-3.1-8B-Instruct",
"tensor_parallel_size": 2,
"pipeline_parallel_size": 3,
"dtype": "bfloat16",
"distributed_executor_backend": "mp",
"block_size": 128,
"trust_remote_code": "",
"enable_chunked_prefill": "",
"disable_log_stats": "",
"enforce_eager": "",
"max_num_batched_tokens": 2048,
"max_num_seqs": 256,
"load_format": "dummy"
},
"client_parameters": {
"model": "meta-llama/Meta-Llama-3.1-8B-Instruct",
"backend": "vllm",
"dataset_name": "random",
"random-input-len": 128,
"random-output-len": 128,
"ignore-eos": "",
"max_concurrency": 1000,
"num_prompts": 1000
}
}
]

View File

@ -0,0 +1,168 @@
[
{
"test_name": "serving_llama8B_tp1_sharegpt",
"qps_list": [1, 4, 16, "inf"],
"server_environment_variables": {
"VLLM_RPC_TIMEOUT": 100000,
"VLLM_ALLOW_LONG_MAX_MODEL_LEN": 1,
"VLLM_ENGINE_ITERATION_TIMEOUT_S": 120,
"VLLM_CPU_SGL_KERNEL": 1,
"VLLM_CPU_KVCACHE_SPACE": 40
},
"server_parameters": {
"model": "meta-llama/Meta-Llama-3.1-8B-Instruct",
"tensor_parallel_size": 1,
"dtype": "bfloat16",
"distributed_executor_backend": "mp",
"block_size": 128,
"trust_remote_code": "",
"disable_log_stats": "",
"enforce_eager": "",
"max_num_batched_tokens": 2048,
"max_num_seqs": 256,
"load_format": "dummy"
},
"client_parameters": {
"model": "meta-llama/Meta-Llama-3.1-8B-Instruct",
"backend": "vllm",
"dataset_name": "sharegpt",
"dataset_path": "./ShareGPT_V3_unfiltered_cleaned_split.json",
"max_concurrency": 60,
"num_prompts": 200
}
},
{
"test_name": "serving_llama8B_tp2_sharegpt",
"qps_list": [1, 4, 16, "inf"],
"server_environment_variables": {
"VLLM_RPC_TIMEOUT": 100000,
"VLLM_ALLOW_LONG_MAX_MODEL_LEN": 1,
"VLLM_ENGINE_ITERATION_TIMEOUT_S": 120,
"VLLM_CPU_SGL_KERNEL": 1,
"VLLM_CPU_KVCACHE_SPACE": 40
},
"server_parameters": {
"model": "meta-llama/Meta-Llama-3.1-8B-Instruct",
"tensor_parallel_size": 2,
"dtype": "bfloat16",
"distributed_executor_backend": "mp",
"block_size": 128,
"trust_remote_code": "",
"disable_log_stats": "",
"enforce_eager": "",
"max_num_batched_tokens": 2048,
"max_num_seqs": 256,
"load_format": "dummy"
},
"client_parameters": {
"model": "meta-llama/Meta-Llama-3.1-8B-Instruct",
"backend": "vllm",
"dataset_name": "sharegpt",
"dataset_path": "./ShareGPT_V3_unfiltered_cleaned_split.json",
"max_concurrency": 60,
"num_prompts": 200
}
},
{
"test_name": "serving_llama8B_tp4_sharegpt",
"qps_list": [1, 4, 16, "inf"],
"server_environment_variables": {
"VLLM_RPC_TIMEOUT": 100000,
"VLLM_ALLOW_LONG_MAX_MODEL_LEN": 1,
"VLLM_ENGINE_ITERATION_TIMEOUT_S": 120,
"VLLM_CPU_SGL_KERNEL": 1,
"VLLM_CPU_KVCACHE_SPACE": 40
},
"server_parameters": {
"model": "meta-llama/Meta-Llama-3.1-8B-Instruct",
"tensor_parallel_size": 4,
"dtype": "bfloat16",
"distributed_executor_backend": "mp",
"block_size": 128,
"trust_remote_code": "",
"disable_log_stats": "",
"enforce_eager": "",
"max_num_batched_tokens": 2048,
"max_num_seqs": 256,
"load_format": "dummy"
},
"client_parameters": {
"model": "meta-llama/Meta-Llama-3.1-8B-Instruct",
"backend": "vllm",
"dataset_name": "sharegpt",
"dataset_path": "./ShareGPT_V3_unfiltered_cleaned_split.json",
"max_concurrency": 60,
"num_prompts": 200
}
},
{
"test_name": "serving_llama8B_tp4_random_1024_128",
"qps_list": [1, 4, 16, "inf"],
"server_environment_variables": {
"VLLM_RPC_TIMEOUT": 100000,
"VLLM_ALLOW_LONG_MAX_MODEL_LEN": 1,
"VLLM_ENGINE_ITERATION_TIMEOUT_S": 120,
"VLLM_CPU_SGL_KERNEL": 1,
"VLLM_CPU_KVCACHE_SPACE": 40
},
"server_parameters": {
"model": "meta-llama/Meta-Llama-3.1-8B-Instruct",
"tensor_parallel_size": 4,
"dtype": "bfloat16",
"distributed_executor_backend": "mp",
"block_size": 128,
"trust_remote_code": "",
"enable_chunked_prefill": "",
"disable_log_stats": "",
"enforce_eager": "",
"max_num_batched_tokens": 2048,
"max_num_seqs": 256,
"load_format": "dummy"
},
"client_parameters": {
"model": "meta-llama/Meta-Llama-3.1-8B-Instruct",
"backend": "vllm",
"dataset_name": "random",
"random-input-len": 1024,
"random-output-len": 128,
"ignore-eos": "",
"max_concurrency": 100,
"num_prompts": 100
}
},
{
"test_name": "serving_llama8B_pp6_random_1024_128",
"qps_list": [1, 4, 16, "inf"],
"server_environment_variables": {
"VLLM_RPC_TIMEOUT": 100000,
"VLLM_ALLOW_LONG_MAX_MODEL_LEN": 1,
"VLLM_ENGINE_ITERATION_TIMEOUT_S": 120,
"VLLM_CPU_SGL_KERNEL": 1,
"VLLM_CPU_KVCACHE_SPACE": 40
},
"server_parameters": {
"model": "meta-llama/Meta-Llama-3.1-8B-Instruct",
"pipeline_parallel_size": 6,
"dtype": "bfloat16",
"distributed_executor_backend": "mp",
"block_size": 128,
"trust_remote_code": "",
"enable_chunked_prefill": "",
"disable_log_stats": "",
"enforce_eager": "",
"max_num_batched_tokens": 2048,
"max_num_seqs": 256,
"load_format": "dummy"
},
"client_parameters": {
"model": "meta-llama/Meta-Llama-3.1-8B-Instruct",
"backend": "vllm",
"dataset_name": "random",
"random-input-len": 1024,
"random-output-len": 128,
"ignore-eos": "",
"max_concurrency": 100,
"num_prompts": 100
}
}
]

View File

@ -0,0 +1,32 @@
[
{
"test_name": "throughput_llama8B_tp1",
"environment_variables": {
"VLLM_ALLOW_LONG_MAX_MODEL_LEN": 1,
"VLLM_CPU_KVCACHE_SPACE": 40
},
"parameters": {
"model": "meta-llama/Meta-Llama-3.1-8B-Instruct",
"tensor_parallel_size": 1,
"load_format": "dummy",
"dataset": "./ShareGPT_V3_unfiltered_cleaned_split.json",
"num_prompts": 200,
"backend": "vllm"
}
},
{
"test_name": "throughput_llama8B_tp4",
"environment_variables": {
"VLLM_ALLOW_LONG_MAX_MODEL_LEN": 1,
"VLLM_CPU_KVCACHE_SPACE": 40
},
"parameters": {
"model": "meta-llama/Meta-Llama-3.1-8B-Instruct",
"tensor_parallel_size": 4,
"load_format": "dummy",
"dataset": "./ShareGPT_V3_unfiltered_cleaned_split.json",
"num_prompts": 200,
"backend": "vllm"
}
}
]

View File

@ -1,456 +0,0 @@
# SPDX-License-Identifier: Apache-2.0
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
import argparse
import json
import os
from importlib import util
import pandas as pd
pd.options.display.float_format = "{:.2f}".format
plotly_found = util.find_spec("plotly.express") is not None
def compare_data_columns(
files, name_column, data_column, info_cols, drop_column, debug=False
):
"""
Align concatenation by keys derived from info_cols instead of row order.
- Pick one canonical key list: subset of info_cols present in ALL files.
- For each file: set index to those keys, aggregate duplicates
- (mean for metric, first for names).
- Concat along axis=1 (indexes align), then reset_index so callers can
- group by columns.
- If --debug, add a <file_label>_name column per file.
"""
print("\ncompare_data_column:", data_column)
frames = []
raw_data_cols = []
compare_frames = []
# 1) choose a canonical key list from info_cols that exists in ALL files
cols_per_file = []
for f in files:
try:
df_tmp = pd.read_json(f, orient="records")
except Exception as err:
raise ValueError(f"Failed to read {f}") from err
cols_per_file.append(set(df_tmp.columns))
key_cols = [c for c in info_cols if all(c in cset for cset in cols_per_file)]
if not key_cols:
# soft fallback: use any info_cols present in the first file
key_cols = [c for c in info_cols if c in list(cols_per_file[0])]
if not key_cols:
raise ValueError(
"No common key columns found from info_cols across the input files."
)
# 2) build a single "meta" block (keys as columns) once, aligned by the key index
meta_added = False
for file in files:
df = pd.read_json(file, orient="records")
# Keep rows that actually have the compared metric (same as original behavior)
if drop_column in df.columns:
df = df.dropna(subset=[drop_column], ignore_index=True)
# Stabilize numeric key columns (harmless if missing)
for c in (
"Input Len",
"Output Len",
"TP Size",
"PP Size",
"# of max concurrency.",
"qps",
):
if c in df.columns:
df[c] = pd.to_numeric(df[c], errors="coerce")
# Ensure all key columns exist
for c in key_cols:
if c not in df.columns:
df[c] = pd.NA
# Set index = key_cols and aggregate duplicates → unique MultiIndex
df_idx = df.set_index(key_cols, drop=False)
# meta (key columns), unique per key
meta = df_idx[key_cols]
if not meta.index.is_unique:
meta = meta.groupby(level=key_cols, dropna=False).first()
# metric series for this file, aggregated to one row per key
file_label = "/".join(file.split("/")[:-1]) or os.path.basename(file)
s = df_idx[data_column]
if not s.index.is_unique:
s = s.groupby(level=key_cols, dropna=False).mean()
s.name = file_label # column label like original
# add meta once (from first file) so keys are the leftmost columns
if not meta_added:
frames.append(meta)
meta_added = True
# (NEW) debug: aligned test-name column per file
if debug and name_column in df_idx.columns:
name_s = df_idx[name_column]
if not name_s.index.is_unique:
name_s = name_s.groupby(level=key_cols, dropna=False).first()
name_s.name = f"{file_label}_name"
frames.append(name_s)
frames.append(s)
raw_data_cols.append(file_label)
compare_frames.append(s)
# Generalize ratio: for any file N>=2, add ratio (fileN / file1)
if len(compare_frames) >= 2:
base = compare_frames[0]
current = compare_frames[-1]
if "P99" in data_column or "Median" in data_column:
ratio = base / current # for latency
else:
ratio = current / base
ratio = ratio.mask(base == 0) # avoid inf when baseline is 0
ratio.name = f"Ratio 1 vs {len(compare_frames)}"
frames.append(ratio)
# 4) concat on columns with aligned MultiIndex;
# then reset_index to return keys as columns
concat_df = pd.concat(frames, axis=1)
concat_df = concat_df.reset_index(drop=True).reset_index()
if "index" in concat_df.columns:
concat_df = concat_df.drop(columns=["index"])
# Ensure key/info columns appear first (in your info_cols order)
front = [c for c in info_cols if c in concat_df.columns]
rest = [c for c in concat_df.columns if c not in front]
concat_df = concat_df[front + rest]
print(raw_data_cols)
return concat_df, raw_data_cols
def split_json_by_tp_pp(
input_file: str = "benchmark_results.json", output_root: str = "."
) -> list[str]:
"""
Split a benchmark JSON into separate folders by (TP Size, PP Size).
Creates: <output_root>/tp{TP}_pp{PP}/benchmark_results.json
Returns: list of file paths written.
"""
# Load JSON data into DataFrame
with open(input_file, encoding="utf-8") as f:
data = json.load(f)
# If the JSON is a dict with a list under common keys, use that list
if isinstance(data, dict):
for key in ("results", "serving_results", "benchmarks", "data"):
if isinstance(data.get(key), list):
data = data[key]
break
df = pd.DataFrame(data)
# Keep only "serving" tests
name_col = next(
(c for c in ["Test name", "test_name", "Test Name"] if c in df.columns), None
)
if name_col:
df = df[
df[name_col].astype(str).str.contains(r"serving", case=False, na=False)
].copy()
# Handle alias column names
rename_map = {
"tp_size": "TP Size",
"tensor_parallel_size": "TP Size",
"pp_size": "PP Size",
"pipeline_parallel_size": "PP Size",
}
df.rename(
columns={k: v for k, v in rename_map.items() if k in df.columns}, inplace=True
)
# Ensure TP/PP columns exist (default to 1 if missing)
if "TP Size" not in df.columns:
df["TP Size"] = 1
if "PP Size" not in df.columns:
df["PP Size"] = 1
# make sure TP/PP are numeric ints with no NaN
df["TP Size"] = (
pd.to_numeric(df.get("TP Size", 1), errors="coerce").fillna(1).astype(int)
)
df["PP Size"] = (
pd.to_numeric(df.get("PP Size", 1), errors="coerce").fillna(1).astype(int)
)
# Split into separate folders
saved_paths: list[str] = []
for (tp, pp), group_df in df.groupby(["TP Size", "PP Size"], dropna=False):
folder_name = os.path.join(output_root, f"tp{int(tp)}_pp{int(pp)}")
os.makedirs(folder_name, exist_ok=True)
filepath = os.path.join(folder_name, "benchmark_results.json")
group_df.to_json(filepath, orient="records", indent=2, force_ascii=False)
print(f"Saved: {filepath}")
saved_paths.append(filepath)
return saved_paths
def _add_limit_line(fig, y_value, label):
# Visible dashed line + annotation
fig.add_hline(
y=y_value,
line_dash="dash",
line_color="red" if "ttft" in label.lower() else "blue",
annotation_text=f"{label}: {y_value} ms",
annotation_position="top left",
)
# Optional: add a legend item (as a transparent helper trace)
if plot and plotly_found:
import plotly.graph_objects as go
fig.add_trace(
go.Scatter(
x=[None],
y=[None],
mode="lines",
line=dict(
dash="dash", color="red" if "ttft" in label.lower() else "blue"
),
name=f"{label}",
)
)
def _find_concurrency_col(df: pd.DataFrame) -> str:
for c in [
"# of max concurrency.",
"# of max concurrency",
"Max Concurrency",
"max_concurrency",
"Concurrency",
]:
if c in df.columns:
return c
# Fallback: guess an integer-like column (harmless if unused)
for c in df.columns:
if df[c].dtype.kind in "iu" and df[c].nunique() > 1 and df[c].min() >= 1:
return c
return "# of max concurrency."
def _highlight_threshold(
df: pd.DataFrame, threshold: float
) -> "pd.io.formats.style.Styler":
"""Highlight numeric per-configuration columns with value <= threshold."""
conc_col = _find_concurrency_col(df)
key_cols = [
c
for c in ["Model", "Dataset Name", "Input Len", "Output Len", conc_col]
if c in df.columns
]
conf_cols = [
c for c in df.columns if c not in key_cols and not str(c).startswith("Ratio")
]
conf_cols = [c for c in conf_cols if pd.api.types.is_numeric_dtype(df[c])]
return df.style.map(
lambda v: "background-color:#e6ffe6;font-weight:bold;"
if pd.notna(v) and v <= threshold
else "",
subset=conf_cols,
)
if __name__ == "__main__":
parser = argparse.ArgumentParser()
parser.add_argument(
"-f", "--file", action="append", type=str, help="input file name"
)
parser.add_argument(
"--debug", action="store_true", help="show all information for debugging"
)
parser.add_argument(
"--plot",
action=argparse.BooleanOptionalAction,
default=True,
help="plot perf diagrams or not --no-plot --plot",
)
parser.add_argument(
"-x",
"--xaxis",
type=str,
default="# of max concurrency.",
help="column name to use as X Axis in comparison graph",
)
parser.add_argument(
"-l",
"--latency",
type=str,
default="p99",
help="take median|p99 for latency like TTFT/TPOT",
)
parser.add_argument(
"--ttft-max-ms",
type=float,
default=3000.0,
help="Reference limit for TTFT plots (ms)",
)
parser.add_argument(
"--tpot-max-ms",
type=float,
default=100.0,
help="Reference limit for TPOT plots (ms)",
)
args = parser.parse_args()
drop_column = "P99"
name_column = "Test name"
info_cols = [
"Model",
"Dataset Name",
"Input Len",
"Output Len",
"TP Size",
"PP Size",
"# of max concurrency.",
"qps",
]
if "median" in args.latency:
data_cols_to_compare = ["Output Tput (tok/s)", "Median TTFT (ms)", "Median"]
html_msgs_for_data_cols = [
"Compare Output Tokens /n",
"Median TTFT /n",
"Median TPOT /n",
]
drop_column = "P99"
elif "p99" in args.latency:
data_cols_to_compare = ["Output Tput (tok/s)", "P99 TTFT (ms)", "P99"]
html_msgs_for_data_cols = [
"Compare Output Tokens /n",
"P99 TTFT /n",
"P99 TPOT /n",
]
if len(args.file) == 1:
files = split_json_by_tp_pp(args.file[0], output_root="splits")
info_cols = [c for c in info_cols if c not in ("TP Size", "PP Size")]
else:
files = args.file
print("comparing : " + ", ".join(files))
debug = args.debug
plot = args.plot
# For Plot feature, assign y axis from one of info_cols
y_axis_index = info_cols.index(args.xaxis) if args.xaxis in info_cols else 6
with open("perf_comparison.html", "w") as text_file:
for i in range(len(data_cols_to_compare)):
output_df, raw_data_cols = compare_data_columns(
files,
name_column,
data_cols_to_compare[i],
info_cols,
drop_column,
debug=debug,
)
# For Plot feature, insert y axis from one of info_cols
raw_data_cols.insert(0, info_cols[y_axis_index])
filtered_info_cols = info_cols[:-2]
existing_group_cols = [
c for c in filtered_info_cols if c in output_df.columns
]
if not existing_group_cols:
raise ValueError(
f"No valid group-by columns "
f"Expected subset: {filtered_info_cols}, "
f"but DataFrame has: {list(output_df.columns)}"
)
# output_df_sorted = output_df.sort_values(by=existing_group_cols)
output_df_sorted = output_df.sort_values(by=args.xaxis)
output_groups = output_df_sorted.groupby(existing_group_cols, dropna=False)
for name, group in output_groups:
group_name = (
",".join(map(str, name)).replace(",", "_").replace("/", "-")
)
group_html_name = "perf_comparison_" + group_name + ".html"
metric_name = str(data_cols_to_compare[i]).lower()
if "tok/s" in metric_name:
html = group.to_html()
elif "ttft" in metric_name:
styler = _highlight_threshold(group, args.ttft_max_ms).format(
{c: "{:.2f}" for c in group.select_dtypes("number").columns},
na_rep="",
)
html = styler.to_html(
table_attributes='border="1" class="dataframe"'
)
elif (
"tpot" in metric_name
or "median" in metric_name
or "p99" in metric_name
):
styler = _highlight_threshold(group, args.tpot_max_ms).format(
{c: "{:.2f}" for c in group.select_dtypes("number").columns},
na_rep="",
)
html = styler.to_html(
table_attributes='border="1" class="dataframe"'
)
text_file.write(html_msgs_for_data_cols[i])
text_file.write(html)
with open(group_html_name, "a+") as sub_text_file:
sub_text_file.write(html_msgs_for_data_cols[i])
sub_text_file.write(html)
if plot and plotly_found:
import plotly.express as px
df = group[raw_data_cols]
df_sorted = df.sort_values(by=info_cols[y_axis_index])
# Melt DataFrame for plotting
df_melted = df_sorted.melt(
id_vars=info_cols[y_axis_index],
var_name="Configuration",
value_name=data_cols_to_compare[i],
)
title = (
data_cols_to_compare[i] + " vs " + info_cols[y_axis_index]
)
# Create Plotly line chart
fig = px.line(
df_melted,
x=info_cols[y_axis_index],
y=data_cols_to_compare[i],
color="Configuration",
title=title,
markers=True,
)
# ---- Add threshold lines based on metric name ----
if "ttft" in metric_name:
_add_limit_line(fig, args.ttft_max_ms, "TTFT limit")
elif (
"tpot" in metric_name
or "median" in metric_name
or "p99" in metric_name
):
_add_limit_line(fig, args.tpot_max_ms, "TPOT limit")
# Export to HTML
text_file.write(
fig.to_html(full_html=True, include_plotlyjs="cdn")
)
sub_text_file.write(
fig.to_html(full_html=True, include_plotlyjs="cdn")
)

View File

@ -1,26 +0,0 @@
[
{
"test_name": "latency_llama8B_tp2",
"environment_variables": {
"VLLM_RPC_TIMEOUT": 100000,
"VLLM_ALLOW_LONG_MAX_MODEL_LEN": 1,
"VLLM_ENGINE_ITERATION_TIMEOUT_S": 120,
"VLLM_CPU_SGL_KERNEL": 1,
"VLLM_CPU_KVCACHE_SPACE": 40
},
"parameters": {
"model": "meta-llama/Llama-3.1-8B-Instruct",
"tensor_parallel_size": 2,
"dtype": "bfloat16",
"distributed_executor_backend": "mp",
"block_size": 128,
"trust_remote_code": "",
"disable_log_stats": "",
"enforce_eager": "",
"max_num_batched_tokens": 2048,
"max_num_seqs": 256,
"num_iters_warmup": 5,
"num_iters": 15
}
}
]

View File

@ -1,55 +0,0 @@
[
{
"test_name": "latency_llama8B_tp1",
"environment_variables": {
"PT_HPU_LAZY_MODE": 1,
"VLLM_CONTIGUOUS_PA": 1,
"VLLM_DEFRAG": 1
},
"parameters": {
"model": "meta-llama/Meta-Llama-3.1-8B-Instruct",
"tensor_parallel_size": 1,
"load_format": "dummy",
"num-iters-warmup": 5,
"num-iters": 15,
"max-model-len": 256,
"async-scheduling": ""
}
},
{
"test_name": "latency_llama70B_tp4",
"environment_variables": {
"PT_HPU_LAZY_MODE": 1,
"PT_HPU_ENABLE_LAZY_COLLECTIVES": 1,
"VLLM_CONTIGUOUS_PA": 1,
"VLLM_DEFRAG": 1
},
"parameters": {
"model": "meta-llama/Meta-Llama-3.1-70B-Instruct",
"tensor_parallel_size": 4,
"load_format": "dummy",
"num-iters-warmup": 5,
"num-iters": 15,
"max-model-len": 256,
"async-scheduling": ""
}
},
{
"test_name": "latency_mixtral8x7B_tp2",
"environment_variables": {
"PT_HPU_LAZY_MODE": 1,
"PT_HPU_ENABLE_LAZY_COLLECTIVES": 1,
"VLLM_CONTIGUOUS_PA": 1,
"VLLM_DEFRAG": 1
},
"parameters": {
"model": "mistralai/Mixtral-8x7B-Instruct-v0.1",
"tensor_parallel_size": 2,
"load_format": "dummy",
"num-iters-warmup": 5,
"num-iters": 15,
"max-model-len": 256,
"async-scheduling": ""
}
}
]

View File

@ -1,610 +0,0 @@
[
{
"test_name": "serving_llama8B_bf16_tp1_sharegpt",
"qps_list": ["inf"],
"max_concurrency_list": [12, 16, 24, 32, 64, 128, 200],
"server_environment_variables": {
"VLLM_RPC_TIMEOUT": 100000,
"VLLM_ALLOW_LONG_MAX_MODEL_LEN": 1,
"VLLM_ENGINE_ITERATION_TIMEOUT_S": 120,
"VLLM_CPU_SGL_KERNEL": 1,
"VLLM_CPU_KVCACHE_SPACE": 40
},
"server_parameters": {
"model": "meta-llama/Llama-3.1-8B-Instruct",
"tensor_parallel_size": 1,
"dtype": "bfloat16",
"distributed_executor_backend": "mp",
"block_size": 128,
"trust_remote_code": "",
"disable_log_stats": "",
"enforce_eager": "",
"max_num_batched_tokens": 2048,
"max_num_seqs": 256,
"load_format": "dummy"
},
"client_parameters": {
"model": "meta-llama/Llama-3.1-8B-Instruct",
"backend": "vllm",
"dataset_name": "sharegpt",
"dataset_path": "./ShareGPT_V3_unfiltered_cleaned_split.json",
"num_prompts": 200
}
},
{
"test_name": "serving_llama8B_bf16_tp2_sharegpt",
"qps_list": ["inf"],
"max_concurrency_list": [12, 16, 24, 32, 64, 128, 200],
"server_environment_variables": {
"VLLM_RPC_TIMEOUT": 100000,
"VLLM_ALLOW_LONG_MAX_MODEL_LEN": 1,
"VLLM_ENGINE_ITERATION_TIMEOUT_S": 120,
"VLLM_CPU_SGL_KERNEL": 1,
"VLLM_CPU_KVCACHE_SPACE": 40
},
"server_parameters": {
"model": "meta-llama/Llama-3.1-8B-Instruct",
"tensor_parallel_size": 2,
"dtype": "bfloat16",
"distributed_executor_backend": "mp",
"block_size": 128,
"trust_remote_code": "",
"disable_log_stats": "",
"enforce_eager": "",
"max_num_batched_tokens": 2048,
"max_num_seqs": 256,
"load_format": "dummy"
},
"client_parameters": {
"model": "meta-llama/Llama-3.1-8B-Instruct",
"backend": "vllm",
"dataset_name": "sharegpt",
"dataset_path": "./ShareGPT_V3_unfiltered_cleaned_split.json",
"num_prompts": 200
}
},
{
"test_name": "serving_llama8B_bf16_tp4_sharegpt",
"qps_list": ["inf"],
"max_concurrency_list": [12, 16, 24, 32, 64, 128, 200],
"server_environment_variables": {
"VLLM_RPC_TIMEOUT": 100000,
"VLLM_ALLOW_LONG_MAX_MODEL_LEN": 1,
"VLLM_ENGINE_ITERATION_TIMEOUT_S": 120,
"VLLM_CPU_SGL_KERNEL": 1,
"VLLM_CPU_KVCACHE_SPACE": 40
},
"server_parameters": {
"model": "meta-llama/Llama-3.1-8B-Instruct",
"tensor_parallel_size": 4,
"dtype": "bfloat16",
"distributed_executor_backend": "mp",
"block_size": 128,
"trust_remote_code": "",
"disable_log_stats": "",
"enforce_eager": "",
"max_num_batched_tokens": 2048,
"max_num_seqs": 256,
"load_format": "dummy"
},
"client_parameters": {
"model": "meta-llama/Llama-3.1-8B-Instruct",
"backend": "vllm",
"dataset_name": "sharegpt",
"dataset_path": "./ShareGPT_V3_unfiltered_cleaned_split.json",
"num_prompts": 200
}
},
{
"test_name": "serving_llama8B_bf16_tp1_random_128_128",
"qps_list": ["inf"],
"max_concurrency_list": [12, 16, 24, 32, 64, 128, 200, 1000],
"server_environment_variables": {
"VLLM_RPC_TIMEOUT": 100000,
"VLLM_ALLOW_LONG_MAX_MODEL_LEN": 1,
"VLLM_ENGINE_ITERATION_TIMEOUT_S": 120,
"VLLM_CPU_SGL_KERNEL": 1,
"VLLM_CPU_KVCACHE_SPACE": 40
},
"server_parameters": {
"model": "meta-llama/Llama-3.1-8B-Instruct",
"tensor_parallel_size": 1,
"dtype": "bfloat16",
"distributed_executor_backend": "mp",
"block_size": 128,
"trust_remote_code": "",
"enable_chunked_prefill": "",
"disable_log_stats": "",
"enforce_eager": "",
"max_num_batched_tokens": 2048,
"max_num_seqs": 256,
"load_format": "dummy"
},
"client_parameters": {
"model": "meta-llama/Llama-3.1-8B-Instruct",
"backend": "vllm",
"dataset_name": "random",
"random-input-len": 128,
"random-output-len": 128,
"ignore-eos": "",
"num_prompts": 1000
}
},
{
"test_name": "serving_llama8B_bf16_tp2_random_128_128",
"qps_list": ["inf"],
"max_concurrency_list": [12, 16, 24, 32, 64, 128, 200, 1000],
"server_environment_variables": {
"VLLM_RPC_TIMEOUT": 100000,
"VLLM_ALLOW_LONG_MAX_MODEL_LEN": 1,
"VLLM_ENGINE_ITERATION_TIMEOUT_S": 120,
"VLLM_CPU_SGL_KERNEL": 1,
"VLLM_CPU_KVCACHE_SPACE": 40
},
"server_parameters": {
"model": "meta-llama/Llama-3.1-8B-Instruct",
"tensor_parallel_size": 2,
"dtype": "bfloat16",
"distributed_executor_backend": "mp",
"block_size": 128,
"trust_remote_code": "",
"enable_chunked_prefill": "",
"disable_log_stats": "",
"enforce_eager": "",
"max_num_batched_tokens": 2048,
"max_num_seqs": 256,
"load_format": "dummy"
},
"client_parameters": {
"model": "meta-llama/Llama-3.1-8B-Instruct",
"backend": "vllm",
"dataset_name": "random",
"random-input-len": 128,
"random-output-len": 128,
"ignore-eos": "",
"num_prompts": 1000
}
},
{
"test_name": "serving_llama8B_bf16_tp4_random_128_128",
"qps_list": ["inf"],
"max_concurrency_list": [12, 16, 24, 32, 64, 128, 200, 1000],
"server_environment_variables": {
"VLLM_RPC_TIMEOUT": 100000,
"VLLM_ALLOW_LONG_MAX_MODEL_LEN": 1,
"VLLM_ENGINE_ITERATION_TIMEOUT_S": 120,
"VLLM_CPU_SGL_KERNEL": 1,
"VLLM_CPU_KVCACHE_SPACE": 40
},
"server_parameters": {
"model": "meta-llama/Llama-3.1-8B-Instruct",
"tensor_parallel_size": 4,
"dtype": "bfloat16",
"distributed_executor_backend": "mp",
"block_size": 128,
"trust_remote_code": "",
"enable_chunked_prefill": "",
"disable_log_stats": "",
"enforce_eager": "",
"max_num_batched_tokens": 2048,
"max_num_seqs": 256,
"load_format": "dummy"
},
"client_parameters": {
"model": "meta-llama/Llama-3.1-8B-Instruct",
"backend": "vllm",
"dataset_name": "random",
"random-input-len": 128,
"random-output-len": 128,
"num_prompts": 1000
}
},
{
"test_name": "serving_llama8B_int8_tp1_sharegpt",
"qps_list": ["inf"],
"max_concurrency_list": [12, 16, 24, 32, 64, 128, 200],
"server_environment_variables": {
"VLLM_RPC_TIMEOUT": 100000,
"VLLM_ALLOW_LONG_MAX_MODEL_LEN": 1,
"VLLM_ENGINE_ITERATION_TIMEOUT_S": 120,
"VLLM_CPU_SGL_KERNEL": 1,
"VLLM_CPU_KVCACHE_SPACE": 40
},
"server_parameters": {
"model": "RedHatAI/Meta-Llama-3.1-8B-Instruct-quantized.w8a8",
"tensor_parallel_size": 1,
"dtype": "bfloat16",
"distributed_executor_backend": "mp",
"block_size": 128,
"trust_remote_code": "",
"disable_log_stats": "",
"enforce_eager": "",
"max_num_batched_tokens": 2048,
"max_num_seqs": 256,
"load_format": "dummy"
},
"client_parameters": {
"model": "RedHatAI/Meta-Llama-3.1-8B-Instruct-quantized.w8a8",
"backend": "vllm",
"dataset_name": "sharegpt",
"dataset_path": "./ShareGPT_V3_unfiltered_cleaned_split.json",
"num_prompts": 200
}
},
{
"test_name": "serving_llama8B_int8_tp2_sharegpt",
"qps_list": ["inf"],
"max_concurrency_list": [12, 16, 24, 32, 64, 128, 200],
"server_environment_variables": {
"VLLM_RPC_TIMEOUT": 100000,
"VLLM_ALLOW_LONG_MAX_MODEL_LEN": 1,
"VLLM_ENGINE_ITERATION_TIMEOUT_S": 120,
"VLLM_CPU_SGL_KERNEL": 1,
"VLLM_CPU_KVCACHE_SPACE": 40
},
"server_parameters": {
"model": "RedHatAI/Meta-Llama-3.1-8B-Instruct-quantized.w8a8",
"tensor_parallel_size": 2,
"dtype": "bfloat16",
"distributed_executor_backend": "mp",
"block_size": 128,
"trust_remote_code": "",
"disable_log_stats": "",
"enforce_eager": "",
"max_num_batched_tokens": 2048,
"max_num_seqs": 256,
"load_format": "dummy"
},
"client_parameters": {
"model": "RedHatAI/Meta-Llama-3.1-8B-Instruct-quantized.w8a8",
"backend": "vllm",
"dataset_name": "sharegpt",
"dataset_path": "./ShareGPT_V3_unfiltered_cleaned_split.json",
"num_prompts": 200
}
},
{
"test_name": "serving_llama8B_int8_tp4_sharegpt",
"qps_list": ["inf"],
"max_concurrency_list": [12, 16, 24, 32, 64, 128, 200],
"server_environment_variables": {
"VLLM_RPC_TIMEOUT": 100000,
"VLLM_ALLOW_LONG_MAX_MODEL_LEN": 1,
"VLLM_ENGINE_ITERATION_TIMEOUT_S": 120,
"VLLM_CPU_SGL_KERNEL": 1,
"VLLM_CPU_KVCACHE_SPACE": 40
},
"server_parameters": {
"model": "RedHatAI/Meta-Llama-3.1-8B-Instruct-quantized.w8a8",
"tensor_parallel_size": 4,
"dtype": "bfloat16",
"distributed_executor_backend": "mp",
"block_size": 128,
"trust_remote_code": "",
"disable_log_stats": "",
"enforce_eager": "",
"max_num_batched_tokens": 2048,
"max_num_seqs": 256,
"load_format": "dummy"
},
"client_parameters": {
"model": "RedHatAI/Meta-Llama-3.1-8B-Instruct-quantized.w8a8",
"backend": "vllm",
"dataset_name": "sharegpt",
"dataset_path": "./ShareGPT_V3_unfiltered_cleaned_split.json",
"num_prompts": 200
}
},
{
"test_name": "serving_llama8B_int8_tp1_random_128_128",
"qps_list": ["inf"],
"max_concurrency_list": [12, 16, 24, 32, 64, 128, 200, 1000],
"server_environment_variables": {
"VLLM_RPC_TIMEOUT": 100000,
"VLLM_ALLOW_LONG_MAX_MODEL_LEN": 1,
"VLLM_ENGINE_ITERATION_TIMEOUT_S": 120,
"VLLM_CPU_SGL_KERNEL": 1,
"VLLM_CPU_KVCACHE_SPACE": 40
},
"server_parameters": {
"model": "RedHatAI/Meta-Llama-3.1-8B-Instruct-quantized.w8a8",
"tensor_parallel_size": 1,
"dtype": "bfloat16",
"distributed_executor_backend": "mp",
"block_size": 128,
"trust_remote_code": "",
"enable_chunked_prefill": "",
"disable_log_stats": "",
"enforce_eager": "",
"max_num_batched_tokens": 2048,
"max_num_seqs": 256,
"load_format": "dummy"
},
"client_parameters": {
"model": "RedHatAI/Meta-Llama-3.1-8B-Instruct-quantized.w8a8",
"backend": "vllm",
"dataset_name": "random",
"random-input-len": 128,
"random-output-len": 128,
"ignore-eos": "",
"num_prompts": 1000
}
},
{
"test_name": "serving_llama8B_int8_tp2_random_128_128",
"qps_list": ["inf"],
"max_concurrency_list": [12, 16, 24, 32, 64, 128, 200, 1000],
"server_environment_variables": {
"VLLM_RPC_TIMEOUT": 100000,
"VLLM_ALLOW_LONG_MAX_MODEL_LEN": 1,
"VLLM_ENGINE_ITERATION_TIMEOUT_S": 120,
"VLLM_CPU_SGL_KERNEL": 1,
"VLLM_CPU_KVCACHE_SPACE": 40
},
"server_parameters": {
"model": "RedHatAI/Meta-Llama-3.1-8B-Instruct-quantized.w8a8",
"tensor_parallel_size": 2,
"dtype": "bfloat16",
"distributed_executor_backend": "mp",
"block_size": 128,
"trust_remote_code": "",
"enable_chunked_prefill": "",
"disable_log_stats": "",
"enforce_eager": "",
"max_num_batched_tokens": 2048,
"max_num_seqs": 256,
"load_format": "dummy"
},
"client_parameters": {
"model": "RedHatAI/Meta-Llama-3.1-8B-Instruct-quantized.w8a8",
"backend": "vllm",
"dataset_name": "random",
"random-input-len": 128,
"random-output-len": 128,
"ignore-eos": "",
"num_prompts": 1000
}
},
{
"test_name": "serving_llama8B_int8_tp4_random_128_128",
"qps_list": ["inf"],
"max_concurrency_list": [12, 16, 24, 32, 64, 128, 200, 1000],
"server_environment_variables": {
"VLLM_RPC_TIMEOUT": 100000,
"VLLM_ALLOW_LONG_MAX_MODEL_LEN": 1,
"VLLM_ENGINE_ITERATION_TIMEOUT_S": 120,
"VLLM_CPU_SGL_KERNEL": 1,
"VLLM_CPU_KVCACHE_SPACE": 40
},
"server_parameters": {
"model": "RedHatAI/Meta-Llama-3.1-8B-Instruct-quantized.w8a8",
"tensor_parallel_size": 4,
"dtype": "bfloat16",
"distributed_executor_backend": "mp",
"block_size": 128,
"trust_remote_code": "",
"enable_chunked_prefill": "",
"disable_log_stats": "",
"enforce_eager": "",
"max_num_batched_tokens": 2048,
"max_num_seqs": 256,
"load_format": "dummy"
},
"client_parameters": {
"model": "RedHatAI/Meta-Llama-3.1-8B-Instruct-quantized.w8a8",
"backend": "vllm",
"dataset_name": "random",
"random-input-len": 128,
"random-output-len": 128,
"ignore-eos": "",
"num_prompts": 1000
}
},
{
"test_name": "serving_llama8B_int4_tp1_sharegpt",
"qps_list": ["inf"],
"max_concurrency_list": [12, 16, 24, 32, 64, 128, 200],
"server_environment_variables": {
"VLLM_RPC_TIMEOUT": 100000,
"VLLM_ALLOW_LONG_MAX_MODEL_LEN": 1,
"VLLM_ENGINE_ITERATION_TIMEOUT_S": 120,
"VLLM_CPU_SGL_KERNEL": 1,
"VLLM_CPU_KVCACHE_SPACE": 40
},
"server_parameters": {
"model": "hugging-quants/Meta-Llama-3.1-8B-Instruct-AWQ-INT4",
"quantization": "awq",
"tensor_parallel_size": 1,
"dtype": "bfloat16",
"distributed_executor_backend": "mp",
"block_size": 128,
"trust_remote_code": "",
"disable_log_stats": "",
"enforce_eager": "",
"max_num_batched_tokens": 2048,
"max_num_seqs": 256,
"load_format": "dummy"
},
"client_parameters": {
"model": "hugging-quants/Meta-Llama-3.1-8B-Instruct-AWQ-INT4",
"backend": "vllm",
"dataset_name": "sharegpt",
"dataset_path": "./ShareGPT_V3_unfiltered_cleaned_split.json",
"num_prompts": 200
}
},
{
"test_name": "serving_llama8B_int4_tp2_sharegpt",
"qps_list": ["inf"],
"max_concurrency_list": [12, 16, 24, 32, 64, 128, 200],
"server_environment_variables": {
"VLLM_RPC_TIMEOUT": 100000,
"VLLM_ALLOW_LONG_MAX_MODEL_LEN": 1,
"VLLM_ENGINE_ITERATION_TIMEOUT_S": 120,
"VLLM_CPU_SGL_KERNEL": 1,
"VLLM_CPU_KVCACHE_SPACE": 40
},
"server_parameters": {
"model": "hugging-quants/Meta-Llama-3.1-8B-Instruct-AWQ-INT4",
"quantization": "awq",
"tensor_parallel_size": 2,
"dtype": "bfloat16",
"distributed_executor_backend": "mp",
"block_size": 128,
"trust_remote_code": "",
"disable_log_stats": "",
"enforce_eager": "",
"max_num_batched_tokens": 2048,
"max_num_seqs": 256,
"load_format": "dummy"
},
"client_parameters": {
"model": "hugging-quants/Meta-Llama-3.1-8B-Instruct-AWQ-INT4",
"backend": "vllm",
"dataset_name": "sharegpt",
"dataset_path": "./ShareGPT_V3_unfiltered_cleaned_split.json",
"num_prompts": 200
}
},
{
"test_name": "serving_llama8B_int4_tp4_sharegpt",
"qps_list": ["inf"],
"max_concurrency_list": [12, 16, 24, 32, 64, 128, 200],
"server_environment_variables": {
"VLLM_RPC_TIMEOUT": 100000,
"VLLM_ALLOW_LONG_MAX_MODEL_LEN": 1,
"VLLM_ENGINE_ITERATION_TIMEOUT_S": 120,
"VLLM_CPU_SGL_KERNEL": 1,
"VLLM_CPU_KVCACHE_SPACE": 40
},
"server_parameters": {
"model": "hugging-quants/Meta-Llama-3.1-8B-Instruct-AWQ-INT4",
"quantization": "awq",
"tensor_parallel_size": 4,
"dtype": "bfloat16",
"distributed_executor_backend": "mp",
"block_size": 128,
"trust_remote_code": "",
"disable_log_stats": "",
"enforce_eager": "",
"max_num_batched_tokens": 2048,
"max_num_seqs": 256,
"load_format": "dummy"
},
"client_parameters": {
"model": "hugging-quants/Meta-Llama-3.1-8B-Instruct-AWQ-INT4",
"backend": "vllm",
"dataset_name": "sharegpt",
"dataset_path": "./ShareGPT_V3_unfiltered_cleaned_split.json",
"num_prompts": 200
}
},
{
"test_name": "serving_llama8B_int4_tp1_random_128_128",
"qps_list": ["inf"],
"max_concurrency_list": [12, 16, 24, 32, 64, 128, 200, 1000],
"server_environment_variables": {
"VLLM_RPC_TIMEOUT": 100000,
"VLLM_ALLOW_LONG_MAX_MODEL_LEN": 1,
"VLLM_ENGINE_ITERATION_TIMEOUT_S": 120,
"VLLM_CPU_SGL_KERNEL": 1,
"VLLM_CPU_KVCACHE_SPACE": 40
},
"server_parameters": {
"model": "hugging-quants/Meta-Llama-3.1-8B-Instruct-AWQ-INT4",
"quantization": "awq",
"tensor_parallel_size": 1,
"dtype": "bfloat16",
"distributed_executor_backend": "mp",
"block_size": 128,
"trust_remote_code": "",
"enable_chunked_prefill": "",
"disable_log_stats": "",
"enforce_eager": "",
"max_num_batched_tokens": 2048,
"max_num_seqs": 256,
"load_format": "dummy"
},
"client_parameters": {
"model": "hugging-quants/Meta-Llama-3.1-8B-Instruct-AWQ-INT4",
"backend": "vllm",
"dataset_name": "random",
"random-input-len": 128,
"random-output-len": 128,
"ignore-eos": "",
"num_prompts": 1000
}
},
{
"test_name": "serving_llama8B_int4_tp2_random_128_128",
"qps_list": ["inf"],
"max_concurrency_list": [12, 16, 24, 32, 64, 128, 200, 1000],
"server_environment_variables": {
"VLLM_RPC_TIMEOUT": 100000,
"VLLM_ALLOW_LONG_MAX_MODEL_LEN": 1,
"VLLM_ENGINE_ITERATION_TIMEOUT_S": 120,
"VLLM_CPU_SGL_KERNEL": 1,
"VLLM_CPU_KVCACHE_SPACE": 40
},
"server_parameters": {
"model": "hugging-quants/Meta-Llama-3.1-8B-Instruct-AWQ-INT4",
"quantization": "awq",
"tensor_parallel_size": 2,
"dtype": "bfloat16",
"distributed_executor_backend": "mp",
"block_size": 128,
"trust_remote_code": "",
"enable_chunked_prefill": "",
"disable_log_stats": "",
"enforce_eager": "",
"max_num_batched_tokens": 2048,
"max_num_seqs": 256,
"load_format": "dummy"
},
"client_parameters": {
"model": "hugging-quants/Meta-Llama-3.1-8B-Instruct-AWQ-INT4",
"backend": "vllm",
"dataset_name": "random",
"random-input-len": 128,
"random-output-len": 128,
"ignore-eos": "",
"num_prompts": 1000
}
},
{
"test_name": "serving_llama8B_int4_tp4_random_128_128",
"qps_list": ["inf"],
"max_concurrency_list": [12, 16, 24, 32, 64, 128, 200, 1000],
"server_environment_variables": {
"VLLM_RPC_TIMEOUT": 100000,
"VLLM_ALLOW_LONG_MAX_MODEL_LEN": 1,
"VLLM_ENGINE_ITERATION_TIMEOUT_S": 120,
"VLLM_CPU_SGL_KERNEL": 1,
"VLLM_CPU_KVCACHE_SPACE": 40
},
"server_parameters": {
"model": "hugging-quants/Meta-Llama-3.1-8B-Instruct-AWQ-INT4",
"quantization": "awq",
"tensor_parallel_size": 4,
"dtype": "bfloat16",
"distributed_executor_backend": "mp",
"block_size": 128,
"trust_remote_code": "",
"enable_chunked_prefill": "",
"disable_log_stats": "",
"enforce_eager": "",
"max_num_batched_tokens": 2048,
"max_num_seqs": 256,
"load_format": "dummy"
},
"client_parameters": {
"model": "hugging-quants/Meta-Llama-3.1-8B-Instruct-AWQ-INT4",
"backend": "vllm",
"dataset_name": "random",
"random-input-len": 128,
"random-output-len": 128,
"ignore-eos": "",
"num_prompts": 1000
}
}
]

View File

@ -1,82 +0,0 @@
[
{
"test_name": "serving_llama8B_tp1_sharegpt",
"qps_list": [1, 4, 16, "inf"],
"server_environment_variables": {
"PT_HPU_LAZY_MODE": 1,
"VLLM_CONTIGUOUS_PA": 1,
"VLLM_DEFRAG": 1
},
"server_parameters": {
"model": "meta-llama/Meta-Llama-3.1-8B-Instruct",
"tensor_parallel_size": 1,
"swap_space": 16,
"disable_log_stats": "",
"load_format": "dummy",
"max-model-len": 2048,
"max-num-seqs": 256,
"async-scheduling": ""
},
"client_parameters": {
"model": "meta-llama/Meta-Llama-3.1-8B-Instruct",
"backend": "vllm",
"dataset_name": "sharegpt",
"dataset_path": "./ShareGPT_V3_unfiltered_cleaned_split.json",
"num_prompts": 200
}
},
{
"test_name": "serving_llama70B_tp4_sharegpt",
"qps_list": [1, 4, 16, "inf"],
"server_environment_variables": {
"PT_HPU_LAZY_MODE": 1,
"PT_HPU_ENABLE_LAZY_COLLECTIVES": 1,
"VLLM_CONTIGUOUS_PA": 1,
"VLLM_DEFRAG": 1
},
"server_parameters": {
"model": "meta-llama/Meta-Llama-3.1-70B-Instruct",
"tensor_parallel_size": 4,
"swap_space": 16,
"disable_log_stats": "",
"load_format": "dummy",
"max-model-len": 2048,
"max-num-seqs": 256,
"async-scheduling": ""
},
"client_parameters": {
"model": "meta-llama/Meta-Llama-3.1-70B-Instruct",
"backend": "vllm",
"dataset_name": "sharegpt",
"dataset_path": "./ShareGPT_V3_unfiltered_cleaned_split.json",
"num_prompts": 200
}
},
{
"test_name": "serving_mixtral8x7B_tp2_sharegpt",
"qps_list": [1, 4, 16, "inf"],
"server_environment_variables": {
"PT_HPU_LAZY_MODE": 1,
"PT_HPU_ENABLE_LAZY_COLLECTIVES": 1,
"VLLM_CONTIGUOUS_PA": 1,
"VLLM_DEFRAG": 1
},
"server_parameters": {
"model": "mistralai/Mixtral-8x7B-Instruct-v0.1",
"tensor_parallel_size": 2,
"swap_space": 16,
"disable_log_stats": "",
"load_format": "dummy",
"max-model-len": 2048,
"max-num-seqs": 256,
"async-scheduling": ""
},
"client_parameters": {
"model": "mistralai/Mixtral-8x7B-Instruct-v0.1",
"backend": "vllm",
"dataset_name": "sharegpt",
"dataset_path": "./ShareGPT_V3_unfiltered_cleaned_split.json",
"num_prompts": 200
}
}
]

View File

@ -1,27 +0,0 @@
[
{
"test_name": "throughput_llama8B_tp2",
"environment_variables": {
"VLLM_RPC_TIMEOUT": 100000,
"VLLM_ALLOW_LONG_MAX_MODEL_LEN": 1,
"VLLM_ENGINE_ITERATION_TIMEOUT_S": 120,
"VLLM_CPU_SGL_KERNEL": 1,
"VLLM_CPU_KVCACHE_SPACE": 40
},
"parameters": {
"model": "meta-llama/Llama-3.1-8B-Instruct",
"tensor_parallel_size": 2,
"dtype": "bfloat16",
"distributed_executor_backend": "mp",
"block_size": 128,
"trust_remote_code": "",
"disable_log_stats": "",
"enforce_eager": "",
"max_num_batched_tokens": 2048,
"max_num_seqs": 256,
"dataset": "./ShareGPT_V3_unfiltered_cleaned_split.json",
"num_prompts": 200,
"backend": "vllm"
}
}
]

View File

@ -1,61 +0,0 @@
[
{
"test_name": "throughput_llama8B_tp1",
"environment_variables": {
"PT_HPU_LAZY_MODE": 1,
"VLLM_CONTIGUOUS_PA": 1,
"VLLM_DEFRAG": 1
},
"parameters": {
"model": "meta-llama/Meta-Llama-3.1-8B-Instruct",
"tensor_parallel_size": 1,
"load_format": "dummy",
"dataset_path": "./ShareGPT_V3_unfiltered_cleaned_split.json",
"num_prompts": 1000,
"backend": "vllm",
"max-model-len": 2048,
"max-num-seqs": 512,
"async-scheduling": ""
}
},
{
"test_name": "throughput_llama70B_tp4",
"environment_variables": {
"PT_HPU_LAZY_MODE": 1,
"PT_HPU_ENABLE_LAZY_COLLECTIVES": 1,
"VLLM_CONTIGUOUS_PA": 1,
"VLLM_DEFRAG": 1
},
"parameters": {
"model": "meta-llama/Meta-Llama-3.1-70B-Instruct",
"tensor_parallel_size": 4,
"load_format": "dummy",
"dataset_path": "./ShareGPT_V3_unfiltered_cleaned_split.json",
"num_prompts": 1000,
"backend": "vllm",
"max-model-len": 2048,
"max-num-seqs": 512,
"async-scheduling": ""
}
},
{
"test_name": "throughput_mixtral8x7B_tp2",
"environment_variables": {
"PT_HPU_LAZY_MODE": 1,
"PT_HPU_ENABLE_LAZY_COLLECTIVES": 1,
"VLLM_CONTIGUOUS_PA": 1,
"VLLM_DEFRAG": 1
},
"parameters": {
"model": "mistralai/Mixtral-8x7B-Instruct-v0.1",
"tensor_parallel_size": 2,
"load_format": "dummy",
"dataset_path": "./ShareGPT_V3_unfiltered_cleaned_split.json",
"num_prompts": 1000,
"backend": "vllm",
"max-model-len": 2048,
"max-num-seqs": 512,
"async-scheduling": ""
}
}
]

46
.buildkite/pyproject.toml Normal file
View File

@ -0,0 +1,46 @@
# This local pyproject file is part of the migration from yapf to ruff format.
# It uses the same core rules as the main pyproject.toml file, but with the
# following differences:
# - ruff line length is overridden to 88
# - deprecated typing ignores (UP006, UP035) have been removed
[tool.ruff]
line-length = 88
[tool.ruff.lint.per-file-ignores]
"vllm/third_party/**" = ["ALL"]
"vllm/version.py" = ["F401"]
"vllm/_version.py" = ["ALL"]
[tool.ruff.lint]
select = [
# pycodestyle
"E",
# Pyflakes
"F",
# pyupgrade
"UP",
# flake8-bugbear
"B",
# flake8-simplify
"SIM",
# isort
"I",
# flake8-logging-format
"G",
]
ignore = [
# star imports
"F405", "F403",
# lambda expression assignment
"E731",
# Loop control variable not used within loop body
"B007",
# f-string format
"UP032",
# Can remove once 3.10+ is the minimum Python version
"UP007",
]
[tool.ruff.format]
docstring-code-format = true

View File

@ -1,37 +1,5 @@
steps:
# aarch64 + CUDA builds
- label: "Build arm64 wheel - CUDA 12.9"
depends_on: ~
id: build-wheel-arm64-cuda-12-9
agents:
queue: arm64_cpu_queue_postmerge
commands:
# #NOTE: torch_cuda_arch_list is derived from upstream PyTorch build files here:
# https://github.com/pytorch/pytorch/blob/main/.ci/aarch64_linux/aarch64_ci_build.sh#L7
- "DOCKER_BUILDKIT=1 docker build --build-arg max_jobs=16 --build-arg USE_SCCACHE=1 --build-arg GIT_REPO_CHECK=1 --build-arg CUDA_VERSION=12.9.1 --build-arg VLLM_MAIN_CUDA_VERSION=12.9 --build-arg torch_cuda_arch_list='8.7 8.9 9.0 10.0+PTX 12.0' --tag vllm-ci:build-image --target build --progress plain -f docker/Dockerfile ."
- "mkdir artifacts"
- "docker run --rm -v $(pwd)/artifacts:/artifacts_host vllm-ci:build-image bash -c 'cp -r dist /artifacts_host && chmod -R a+rw /artifacts_host'"
- "bash .buildkite/scripts/upload-wheels.sh"
env:
DOCKER_BUILDKIT: "1"
# aarch64 build
- label: "Build arm64 CPU wheel"
depends_on: ~
id: build-wheel-arm64-cpu
agents:
queue: arm64_cpu_queue_postmerge
commands:
- "DOCKER_BUILDKIT=1 docker build --build-arg max_jobs=16 --build-arg GIT_REPO_CHECK=1 --build-arg VLLM_BUILD_ACL=ON --tag vllm-ci:build-image --target vllm-build --progress plain -f docker/Dockerfile.cpu ."
- "mkdir artifacts"
- "docker run --rm -v $(pwd)/artifacts:/artifacts_host vllm-ci:build-image bash -c 'cp -r dist /artifacts_host && chmod -R a+rw /artifacts_host'"
- "bash .buildkite/scripts/upload-wheels.sh"
env:
DOCKER_BUILDKIT: "1"
# x86 + CUDA builds
- label: "Build wheel - CUDA 12.8"
depends_on: ~
id: build-wheel-cuda-12-8
agents:
queue: cpu_queue_postmerge
@ -43,79 +11,80 @@ steps:
env:
DOCKER_BUILDKIT: "1"
- label: "Build wheel - CUDA 12.9"
depends_on: ~
id: build-wheel-cuda-12-9
- label: "Build wheel - CUDA 12.6"
id: build-wheel-cuda-12-6
agents:
queue: cpu_queue_postmerge
commands:
- "DOCKER_BUILDKIT=1 docker build --build-arg max_jobs=16 --build-arg USE_SCCACHE=1 --build-arg GIT_REPO_CHECK=1 --build-arg CUDA_VERSION=12.9.1 --tag vllm-ci:build-image --target build --progress plain -f docker/Dockerfile ."
- "DOCKER_BUILDKIT=1 docker build --build-arg max_jobs=16 --build-arg USE_SCCACHE=1 --build-arg GIT_REPO_CHECK=1 --build-arg CUDA_VERSION=12.6.3 --build-arg torch_cuda_arch_list='7.0 7.5 8.0 8.9 9.0+PTX' --tag vllm-ci:build-image --target build --progress plain -f docker/Dockerfile ."
- "mkdir artifacts"
- "docker run --rm -v $(pwd)/artifacts:/artifacts_host vllm-ci:build-image bash -c 'cp -r dist /artifacts_host && chmod -R a+rw /artifacts_host'"
- "bash .buildkite/scripts/upload-wheels.sh"
env:
DOCKER_BUILDKIT: "1"
- label: "Build wheel - CUDA 13.0"
depends_on: ~
id: build-wheel-cuda-13-0
# Note(simon): We can always build CUDA 11.8 wheel to ensure the build is working.
# However, this block can be uncommented to save some compute hours.
# - block: "Build CUDA 11.8 wheel"
# key: block-build-cu118-wheel
- label: "Build wheel - CUDA 11.8"
# depends_on: block-build-cu118-wheel
id: build-wheel-cuda-11-8
agents:
queue: cpu_queue_postmerge
commands:
- "DOCKER_BUILDKIT=1 docker build --build-arg max_jobs=16 --build-arg USE_SCCACHE=1 --build-arg GIT_REPO_CHECK=1 --build-arg CUDA_VERSION=13.0.1 --build-arg BUILD_BASE_IMAGE=nvidia/cuda:13.0.1-devel-ubuntu22.04 --tag vllm-ci:build-image --target build --progress plain -f docker/Dockerfile ."
- "DOCKER_BUILDKIT=1 docker build --build-arg max_jobs=16 --build-arg USE_SCCACHE=1 --build-arg GIT_REPO_CHECK=1 --build-arg CUDA_VERSION=11.8.0 --build-arg torch_cuda_arch_list='7.0 7.5 8.0 8.9 9.0+PTX' --tag vllm-ci:build-image --target build --progress plain -f docker/Dockerfile ."
- "mkdir artifacts"
- "docker run --rm -v $(pwd)/artifacts:/artifacts_host vllm-ci:build-image bash -c 'cp -r dist /artifacts_host && chmod -R a+rw /artifacts_host'"
- "bash .buildkite/scripts/upload-wheels.sh"
env:
DOCKER_BUILDKIT: "1"
# Build release images (12.9)
- label: "Build release image (x86)"
- block: "Build release image"
depends_on: ~
id: build-release-image-x86
key: block-release-image-build
- label: "Build release image"
depends_on: block-release-image-build
id: build-release-image
agents:
queue: cpu_queue_postmerge
commands:
- "aws ecr-public get-login-password --region us-east-1 | docker login --username AWS --password-stdin public.ecr.aws/q9t5s3a7"
- "DOCKER_BUILDKIT=1 docker build --build-arg max_jobs=16 --build-arg USE_SCCACHE=1 --build-arg GIT_REPO_CHECK=1 --build-arg CUDA_VERSION=12.9.1 --build-arg FLASHINFER_AOT_COMPILE=true --build-arg INSTALL_KV_CONNECTORS=true --tag public.ecr.aws/q9t5s3a7/vllm-release-repo:$BUILDKITE_COMMIT-$(uname -m) --target vllm-openai --progress plain -f docker/Dockerfile ."
- "docker push public.ecr.aws/q9t5s3a7/vllm-release-repo:$BUILDKITE_COMMIT-$(uname -m)"
# re-tag to default image tag and push, just in case arm64 build fails
- "docker tag public.ecr.aws/q9t5s3a7/vllm-release-repo:$BUILDKITE_COMMIT-$(uname -m) public.ecr.aws/q9t5s3a7/vllm-release-repo:$BUILDKITE_COMMIT"
- "DOCKER_BUILDKIT=1 docker build --build-arg max_jobs=16 --build-arg USE_SCCACHE=1 --build-arg GIT_REPO_CHECK=1 --build-arg CUDA_VERSION=12.8.1 --build-arg INSTALL_KV_CONNECTORS=true --tag public.ecr.aws/q9t5s3a7/vllm-release-repo:$BUILDKITE_COMMIT --target vllm-openai --progress plain -f docker/Dockerfile ."
- "docker push public.ecr.aws/q9t5s3a7/vllm-release-repo:$BUILDKITE_COMMIT"
- label: "Build release image (arm64)"
depends_on: ~
id: build-release-image-arm64
agents:
queue: arm64_cpu_queue_postmerge
commands:
- "aws ecr-public get-login-password --region us-east-1 | docker login --username AWS --password-stdin public.ecr.aws/q9t5s3a7"
- "DOCKER_BUILDKIT=1 docker build --build-arg max_jobs=16 --build-arg USE_SCCACHE=1 --build-arg GIT_REPO_CHECK=1 --build-arg CUDA_VERSION=12.9.1 --build-arg FLASHINFER_AOT_COMPILE=true --build-arg torch_cuda_arch_list='8.7 8.9 9.0 10.0+PTX 12.0' --build-arg INSTALL_KV_CONNECTORS=true --tag public.ecr.aws/q9t5s3a7/vllm-release-repo:$BUILDKITE_COMMIT-$(uname -m) --target vllm-openai --progress plain -f docker/Dockerfile ."
- "docker push public.ecr.aws/q9t5s3a7/vllm-release-repo:$BUILDKITE_COMMIT-$(uname -m)"
# Add job to create multi-arch manifest
- label: "Create multi-arch manifest"
depends_on:
- build-release-image-x86
- build-release-image-arm64
id: create-multi-arch-manifest
agents:
queue: cpu_queue_postmerge
commands:
- "aws ecr-public get-login-password --region us-east-1 | docker login --username AWS --password-stdin public.ecr.aws/q9t5s3a7"
- "docker manifest create public.ecr.aws/q9t5s3a7/vllm-release-repo:$BUILDKITE_COMMIT public.ecr.aws/q9t5s3a7/vllm-release-repo:$BUILDKITE_COMMIT-x86_64 public.ecr.aws/q9t5s3a7/vllm-release-repo:$BUILDKITE_COMMIT-aarch64 --amend"
- "docker manifest push public.ecr.aws/q9t5s3a7/vllm-release-repo:$BUILDKITE_COMMIT"
- label: "Annotate release workflow"
depends_on:
- create-multi-arch-manifest
- build-release-image
- build-wheel-cuda-12-8
- build-wheel-cuda-12-6
- build-wheel-cuda-11-8
id: annotate-release-workflow
agents:
queue: cpu_queue_postmerge
commands:
- "bash .buildkite/scripts/annotate-release.sh"
- label: "Build and publish TPU release image"
depends_on: ~
if: build.env("NIGHTLY") == "1"
agents:
queue: tpu_queue_postmerge
commands:
- "yes | docker system prune -a"
- "git fetch --all"
- "DOCKER_BUILDKIT=1 docker build --build-arg max_jobs=16 --build-arg USE_SCCACHE=1 --build-arg GIT_REPO_CHECK=1 --tag vllm/vllm-tpu:nightly --tag vllm/vllm-tpu:$BUILDKITE_COMMIT --progress plain -f docker/Dockerfile.tpu ."
- "docker push vllm/vllm-tpu:nightly"
- "docker push vllm/vllm-tpu:$BUILDKITE_COMMIT"
plugins:
- docker-login#v3.0.0:
username: vllmbot
password-env: DOCKERHUB_TOKEN
env:
DOCKER_BUILDKIT: "1"
- input: "Provide Release version here"
id: input-release-version
fields:
@ -138,46 +107,18 @@ steps:
env:
DOCKER_BUILDKIT: "1"
- block: "Build arm64 CPU release image"
key: block-arm64-cpu-release-image-build
- block: "Build Neuron release image"
key: block-neuron-release-image-build
depends_on: ~
- label: "Build and publish arm64 CPU release image"
depends_on: block-arm64-cpu-release-image-build
- label: "Build and publish Neuron release image"
depends_on: block-neuron-release-image-build
agents:
queue: arm64_cpu_queue_postmerge
queue: neuron-postmerge
commands:
- "aws ecr-public get-login-password --region us-east-1 | docker login --username AWS --password-stdin public.ecr.aws/q9t5s3a7"
- "DOCKER_BUILDKIT=1 docker build --build-arg max_jobs=16 --build-arg GIT_REPO_CHECK=1 --tag public.ecr.aws/q9t5s3a7/vllm-arm64-cpu-release-repo:$(buildkite-agent meta-data get release-version) --tag public.ecr.aws/q9t5s3a7/vllm-arm64-cpu-release-repo:latest --progress plain --target vllm-openai -f docker/Dockerfile.cpu ."
- "docker push public.ecr.aws/q9t5s3a7/vllm-arm64-cpu-release-repo:latest"
- "docker push public.ecr.aws/q9t5s3a7/vllm-arm64-cpu-release-repo:$(buildkite-agent meta-data get release-version)"
- "DOCKER_BUILDKIT=1 docker build --build-arg max_jobs=16 --build-arg GIT_REPO_CHECK=1 --tag public.ecr.aws/q9t5s3a7/vllm-neuron-release-repo:$(buildkite-agent meta-data get release-version) --tag public.ecr.aws/q9t5s3a7/vllm-neuron-release-repo:latest --progress plain -f docker/Dockerfile.neuron ."
- "docker push public.ecr.aws/q9t5s3a7/vllm-neuron-release-repo:latest"
- "docker push public.ecr.aws/q9t5s3a7/vllm-neuron-release-repo:$(buildkite-agent meta-data get release-version)"
env:
DOCKER_BUILDKIT: "1"
- label: "Build and publish nightly multi-arch image to DockerHub"
depends_on:
- create-multi-arch-manifest
if: build.env("NIGHTLY") == "1"
agents:
queue: cpu_queue_postmerge
commands:
- "aws ecr-public get-login-password --region us-east-1 | docker login --username AWS --password-stdin public.ecr.aws/q9t5s3a7"
- "docker pull public.ecr.aws/q9t5s3a7/vllm-release-repo:$BUILDKITE_COMMIT-x86_64"
- "docker pull public.ecr.aws/q9t5s3a7/vllm-release-repo:$BUILDKITE_COMMIT-aarch64"
- "docker tag public.ecr.aws/q9t5s3a7/vllm-release-repo:$BUILDKITE_COMMIT-x86_64 vllm/vllm-openai:nightly-x86_64"
- "docker tag public.ecr.aws/q9t5s3a7/vllm-release-repo:$BUILDKITE_COMMIT-aarch64 vllm/vllm-openai:nightly-aarch64"
- "docker push vllm/vllm-openai:nightly-x86_64"
- "docker push vllm/vllm-openai:nightly-aarch64"
- "docker manifest create vllm/vllm-openai:nightly vllm/vllm-openai:nightly-x86_64 vllm/vllm-openai:nightly-aarch64 --amend"
- "docker manifest create vllm/vllm-openai:nightly-$BUILDKITE_COMMIT vllm/vllm-openai:nightly-x86_64 vllm/vllm-openai:nightly-aarch64 --amend"
- "docker manifest push vllm/vllm-openai:nightly"
- "docker manifest push vllm/vllm-openai:nightly-$BUILDKITE_COMMIT"
# Clean up old nightly builds (keep only last 14)
- "bash .buildkite/scripts/cleanup-nightly-builds.sh"
plugins:
- docker-login#v3.0.0:
username: vllmbot
password-env: DOCKERHUB_TOKEN
env:
DOCKER_BUILDKIT: "1"
DOCKERHUB_USERNAME: "vllmbot"

View File

@ -14,33 +14,18 @@ buildkite-agent annotate --style 'info' --context 'release-workflow' << EOF
To download the wheel:
\`\`\`
aws s3 cp s3://vllm-wheels/${RELEASE_VERSION}/vllm-${RELEASE_VERSION}-cp38-abi3-manylinux1_x86_64.whl .
aws s3 cp s3://vllm-wheels/${RELEASE_VERSION}/vllm-${RELEASE_VERSION}-cp38-abi3-manylinux2014_aarch64.whl .
aws s3 cp s3://vllm-wheels/${RELEASE_VERSION}+cu126/vllm-${RELEASE_VERSION}+cu126-cp38-abi3-manylinux1_x86_64.whl .
aws s3 cp s3://vllm-wheels/${RELEASE_VERSION}+cu129/vllm-${RELEASE_VERSION}+cu129-cp38-abi3-manylinux1_x86_64.whl .
aws s3 cp s3://vllm-wheels/${RELEASE_VERSION}+cu118/vllm-${RELEASE_VERSION}+cu118-cp38-abi3-manylinux1_x86_64.whl .
\`\`\`
To download and upload the image:
\`\`\`
docker pull public.ecr.aws/q9t5s3a7/vllm-release-repo:${BUILDKITE_COMMIT}-x86_64
docker pull public.ecr.aws/q9t5s3a7/vllm-release-repo:${BUILDKITE_COMMIT}-aarch64
docker tag public.ecr.aws/q9t5s3a7/vllm-release-repo:${BUILDKITE_COMMIT}-x86_64 vllm/vllm-openai:x86_64
docker tag vllm/vllm-openai:x86_64 vllm/vllm-openai:latest-x86_64
docker tag vllm/vllm-openai:x86_64 vllm/vllm-openai:v${RELEASE_VERSION}-x86_64
docker push vllm/vllm-openai:latest-x86_64
docker push vllm/vllm-openai:v${RELEASE_VERSION}-x86_64
docker tag public.ecr.aws/q9t5s3a7/vllm-release-repo:${BUILDKITE_COMMIT}-aarch64 vllm/vllm-openai:aarch64
docker tag vllm/vllm-openai:aarch64 vllm/vllm-openai:latest-aarch64
docker tag vllm/vllm-openai:aarch64 vllm/vllm-openai:v${RELEASE_VERSION}-aarch64
docker push vllm/vllm-openai:latest-aarch64
docker push vllm/vllm-openai:v${RELEASE_VERSION}-aarch64
docker manifest create vllm/vllm-openai:latest vllm/vllm-openai:latest-x86_64 vllm/vllm-openai:latest-aarch64 --amend
docker manifest create vllm/vllm-openai:v${RELEASE_VERSION} vllm/vllm-openai:v${RELEASE_VERSION}-x86_64 vllm/vllm-openai:v${RELEASE_VERSION}-aarch64 --amend
docker manifest push vllm/vllm-openai:latest
docker manifest push vllm/vllm-openai:v${RELEASE_VERSION}
docker pull public.ecr.aws/q9t5s3a7/vllm-release-repo:${BUILDKITE_COMMIT}
docker tag public.ecr.aws/q9t5s3a7/vllm-release-repo:${BUILDKITE_COMMIT} vllm/vllm-openai
docker tag vllm/vllm-openai vllm/vllm-openai:latest
docker tag vllm/vllm-openai vllm/vllm-openai:v${RELEASE_VERSION}
docker push vllm/vllm-openai:latest
docker push vllm/vllm-openai:v${RELEASE_VERSION}
\`\`\`
EOF

View File

@ -1,120 +0,0 @@
#!/bin/bash
set -ex
# Clean up old nightly builds from DockerHub, keeping only the last 14 builds
# This script uses DockerHub API to list and delete old tags with "nightly-" prefix
# DockerHub API endpoint for vllm/vllm-openai repository
REPO_API_URL="https://hub.docker.com/v2/repositories/vllm/vllm-openai/tags"
# Get DockerHub credentials from environment
if [ -z "$DOCKERHUB_TOKEN" ]; then
echo "Error: DOCKERHUB_TOKEN environment variable is not set"
exit 1
fi
if [ -z "$DOCKERHUB_USERNAME" ]; then
echo "Error: DOCKERHUB_USERNAME environment variable is not set"
exit 1
fi
# Get DockerHub bearer token
echo "Getting DockerHub bearer token..."
set +x
BEARER_TOKEN=$(curl -s -X POST \
-H "Content-Type: application/json" \
-d "{\"username\": \"$DOCKERHUB_USERNAME\", \"password\": \"$DOCKERHUB_TOKEN\"}" \
"https://hub.docker.com/v2/users/login" | jq -r '.token')
set -x
if [ -z "$BEARER_TOKEN" ] || [ "$BEARER_TOKEN" = "null" ]; then
echo "Error: Failed to get DockerHub bearer token"
exit 1
fi
# Function to get all tags from DockerHub
get_all_tags() {
local page=1
local all_tags=""
while true; do
set +x
local response=$(curl -s -H "Authorization: Bearer $BEARER_TOKEN" \
"$REPO_API_URL?page=$page&page_size=100")
set -x
# Get both last_updated timestamp and tag name, separated by |
local tags=$(echo "$response" | jq -r '.results[] | select(.name | startswith("nightly-")) | "\(.last_updated)|\(.name)"')
if [ -z "$tags" ]; then
break
fi
all_tags="$all_tags$tags"$'\n'
page=$((page + 1))
done
# Sort by timestamp (newest first) and extract just the tag names
echo "$all_tags" | sort -r | cut -d'|' -f2
}
delete_tag() {
local tag_name="$1"
echo "Deleting tag: $tag_name"
local delete_url="https://hub.docker.com/v2/repositories/vllm/vllm-openai/tags/$tag_name"
set +x
local response=$(curl -s -X DELETE -H "Authorization: Bearer $BEARER_TOKEN" "$delete_url")
set -x
if echo "$response" | jq -e '.detail' > /dev/null 2>&1; then
echo "Warning: Failed to delete tag $tag_name: $(echo "$response" | jq -r '.detail')"
else
echo "Successfully deleted tag: $tag_name"
fi
}
# Get all nightly- prefixed tags, sorted by last_updated timestamp (newest first)
echo "Fetching all tags from DockerHub..."
all_tags=$(get_all_tags)
if [ -z "$all_tags" ]; then
echo "No tags found to clean up"
exit 0
fi
# Count total tags
total_tags=$(echo "$all_tags" | wc -l)
echo "Found $total_tags tags"
# Keep only the last 14 builds (including the current one)
tags_to_keep=14
tags_to_delete=$((total_tags - tags_to_keep))
if [ $tags_to_delete -le 0 ]; then
echo "No tags need to be deleted (only $total_tags tags found, keeping $tags_to_keep)"
exit 0
fi
echo "Will delete $tags_to_delete old tags, keeping the newest $tags_to_keep"
# Get tags to delete (skip the first $tags_to_keep tags)
tags_to_delete_list=$(echo "$all_tags" | tail -n +$((tags_to_keep + 1)))
if [ -z "$tags_to_delete_list" ]; then
echo "No tags to delete"
exit 0
fi
# Delete old tags
echo "Deleting old tags..."
while IFS= read -r tag; do
if [ -n "$tag" ]; then
delete_tag "$tag"
# Add a small delay to avoid rate limiting
sleep 1
fi
done <<< "$tags_to_delete_list"
echo "Cleanup completed successfully"

View File

@ -86,6 +86,10 @@ if [[ $commands == *"pytest -v -s models/test_registry.py"* ]]; then
commands=${commands//"pytest -v -s models/test_registry.py"/"pytest -v -s models/test_registry.py -k 'not BambaForCausalLM and not GritLM and not Mamba2ForCausalLM and not Zamba2ForCausalLM'"}
fi
if [[ $commands == *"VLLM_USE_V1=0 pytest -v -s models/test_initialization.py -k 'not llama4 and not plamo2'"* ]]; then
commands=${commands//"VLLM_USE_V1=0 pytest -v -s models/test_initialization.py -k 'not llama4 and not plamo2'"/"VLLM_USE_V1=0 pytest -v -s models/test_initialization.py -k 'not llama4 and not plamo2 and not BambaForCausalLM and not Gemma2ForCausalLM and not Grok1ModelForCausalLM and not Zamba2ForCausalLM and not Gemma2Model and not GritLM'"}
fi
if [[ $commands == *"pytest -v -s compile/test_basic_correctness.py"* ]]; then
commands=${commands//"pytest -v -s compile/test_basic_correctness.py"/"VLLM_USE_TRITON_FLASH_ATTN=0 pytest -v -s compile/test_basic_correctness.py"}
fi
@ -117,6 +121,7 @@ fi
if [[ $commands == *" kernels/quantization"* ]]; then
commands="${commands} \
--ignore=kernels/quantization/test_int8_quant.py \
--ignore=kernels/quantization/test_aqlm.py \
--ignore=kernels/quantization/test_machete_mm.py \
--ignore=kernels/quantization/test_block_fp8.py \
--ignore=kernels/quantization/test_block_int8.py \
@ -160,9 +165,16 @@ if [[ $commands == *" entrypoints/llm "* ]]; then
--ignore=entrypoints/llm/test_chat.py \
--ignore=entrypoints/llm/test_accuracy.py \
--ignore=entrypoints/llm/test_init.py \
--ignore=entrypoints/llm/test_generate_multiple_loras.py \
--ignore=entrypoints/llm/test_prompt_validation.py "}
fi
#Obsolete currently
##ignore certain Entrypoints/llm tests
#if [[ $commands == *" && pytest -v -s entrypoints/llm/test_guided_generate.py"* ]]; then
# commands=${commands//" && pytest -v -s entrypoints/llm/test_guided_generate.py"/" "}
#fi
# --ignore=entrypoints/openai/test_encoder_decoder.py \
# --ignore=entrypoints/openai/test_embedding.py \
# --ignore=entrypoints/openai/test_oot_registration.py
@ -173,14 +185,6 @@ fi
PARALLEL_JOB_COUNT=8
MYPYTHONPATH=".."
# Test that we're launching on the machine that has
# proper access to GPUs
render_gid=$(getent group render | cut -d: -f3)
if [[ -z "$render_gid" ]]; then
echo "Error: 'render' group not found. This is required for GPU access." >&2
exit 1
fi
# check if the command contains shard flag, we will run all shards in parallel because the host have 8 GPUs.
if [[ $commands == *"--shard-id="* ]]; then
# assign job count as the number of shards used
@ -194,7 +198,6 @@ if [[ $commands == *"--shard-id="* ]]; then
--device /dev/kfd $BUILDKITE_AGENT_META_DATA_RENDER_DEVICES \
--network=host \
--shm-size=16gb \
--group-add "$render_gid" \
--rm \
-e HIP_VISIBLE_DEVICES="${GPU}" \
-e HF_TOKEN \
@ -226,8 +229,8 @@ else
--device /dev/kfd $BUILDKITE_AGENT_META_DATA_RENDER_DEVICES \
--network=host \
--shm-size=16gb \
--group-add "$render_gid" \
--rm \
-e HIP_VISIBLE_DEVICES=0 \
-e HF_TOKEN \
-e AWS_ACCESS_KEY_ID \
-e AWS_SECRET_ACCESS_KEY \

View File

@ -25,28 +25,25 @@ function cpu_tests() {
# offline inference
podman exec -it "$container_id" bash -c "
set -xve
python3 examples/offline_inference/basic/generate.py --model facebook/opt-125m" >> $HOME/test_basic.log
set -e
python3 examples/offline_inference/basic/generate.py --model facebook/opt-125m"
# Run basic model test
podman exec -it "$container_id" bash -c "
set -evx
set -e
pip install pytest pytest-asyncio einops peft Pillow soundfile transformers_stream_generator matplotlib
pip install sentence-transformers datamodel_code_generator
# Note: disable Bart until supports V1
# pytest -v -s tests/models/language/generation/test_bart.py -m cpu_model
pytest -v -s tests/models/language/generation/test_bart.py -m cpu_model
pytest -v -s tests/models/language/generation/test_common.py::test_models[False-5-32-openai-community/gpt2]
pytest -v -s tests/models/language/generation/test_common.py::test_models[False-5-32-facebook/opt-125m]
pytest -v -s tests/models/language/generation/test_common.py::test_models[False-5-32-google/gemma-1.1-2b-it]
pytest -v -s tests/models/language/pooling/test_classification.py::test_models[float-jason9693/Qwen2.5-1.5B-apeach]
# TODO: Below test case tests/models/language/pooling/test_embedding.py::test_models[True-ssmits/Qwen2-7B-Instruct-embed-base] fails on ppc64le. Disabling it for time being.
# pytest -v -s tests/models/language/pooling/test_embedding.py -m cpu_model" >> $HOME/test_rest.log
pytest -v -s tests/models/language/pooling/test_embedding.py -m cpu_model"
}
# All of CPU tests are expected to be finished less than 40 mins.
export container_id
export -f cpu_tests
timeout 120m bash -c cpu_tests
timeout 40m bash -c cpu_tests

View File

@ -25,8 +25,8 @@ numactl -C "$CORE_RANGE" -N "$NUMA_NODE" docker build --tag cpu-test-"$NUMA_NODE
numactl -C "$CORE_RANGE" -N "$NUMA_NODE" docker build --build-arg VLLM_CPU_DISABLE_AVX512="true" --tag cpu-test-"$NUMA_NODE"-avx2 --target vllm-test -f docker/Dockerfile.cpu .
# Run the image, setting --shm-size=4g for tensor parallel.
docker run -itd --cpuset-cpus="$CORE_RANGE" --cpuset-mems="$NUMA_NODE" --entrypoint /bin/bash -v ~/.cache/huggingface:/root/.cache/huggingface --privileged=true -e HF_TOKEN --env VLLM_CPU_KVCACHE_SPACE=16 --env VLLM_CPU_CI_ENV=1 -e E2E_OMP_THREADS="$OMP_CORE_RANGE" --shm-size=4g --name cpu-test-"$NUMA_NODE" cpu-test-"$NUMA_NODE"
docker run -itd --cpuset-cpus="$CORE_RANGE" --cpuset-mems="$NUMA_NODE" --entrypoint /bin/bash -v ~/.cache/huggingface:/root/.cache/huggingface --privileged=true -e HF_TOKEN --env VLLM_CPU_KVCACHE_SPACE=16 --env VLLM_CPU_CI_ENV=1 -e E2E_OMP_THREADS="$OMP_CORE_RANGE" --shm-size=4g --name cpu-test-"$NUMA_NODE"-avx2 cpu-test-"$NUMA_NODE"-avx2
docker run -itd --cpuset-cpus="$CORE_RANGE" --cpuset-mems="$NUMA_NODE" --entrypoint /bin/bash -v ~/.cache/huggingface:/root/.cache/huggingface --privileged=true -e HF_TOKEN --env VLLM_CPU_KVCACHE_SPACE=4 --env VLLM_CPU_CI_ENV=1 -e E2E_OMP_THREADS="$OMP_CORE_RANGE" --shm-size=4g --name cpu-test-"$NUMA_NODE" cpu-test-"$NUMA_NODE"
docker run -itd --cpuset-cpus="$CORE_RANGE" --cpuset-mems="$NUMA_NODE" --entrypoint /bin/bash -v ~/.cache/huggingface:/root/.cache/huggingface --privileged=true -e HF_TOKEN --env VLLM_CPU_KVCACHE_SPACE=4 --env VLLM_CPU_CI_ENV=1 -e E2E_OMP_THREADS="$OMP_CORE_RANGE" --shm-size=4g --name cpu-test-"$NUMA_NODE"-avx2 cpu-test-"$NUMA_NODE"-avx2
function cpu_tests() {
set -e
@ -46,74 +46,57 @@ function cpu_tests() {
set -e
python3 examples/offline_inference/basic/generate.py --model facebook/opt-125m"
# Run kernel tests
docker exec cpu-test-"$NUMA_NODE" bash -c "
set -e
pytest -x -v -s tests/kernels/test_onednn.py"
# Run basic model test
docker exec cpu-test-"$NUMA_NODE" bash -c "
set -e
# Note: disable until supports V1
# pytest -x -v -s tests/kernels/attention/test_cache.py -m cpu_model
# pytest -x -v -s tests/kernels/attention/test_mla_decode_cpu.py -m cpu_model
# pytest -v -s tests/kernels/attention/test_cache.py -m cpu_model
# pytest -v -s tests/kernels/attention/test_mla_decode_cpu.py -m cpu_model
pytest -x -v -s tests/models/language/generation -m cpu_model
VLLM_CPU_SGL_KERNEL=1 pytest -x -v -s tests/models/language/generation -m cpu_model
# Note: disable Bart until supports V1
pytest -v -s tests/models/language/generation -m cpu_model \
--ignore=tests/models/language/generation/test_bart.py
VLLM_CPU_SGL_KERNEL=1 pytest -v -s tests/models/language/generation -m cpu_model \
--ignore=tests/models/language/generation/test_bart.py
pytest -x -v -s tests/models/language/pooling -m cpu_model
pytest -x -v -s tests/models/multimodal/generation \
pytest -v -s tests/models/language/pooling -m cpu_model
pytest -v -s tests/models/multimodal/generation \
--ignore=tests/models/multimodal/generation/test_mllama.py \
--ignore=tests/models/multimodal/generation/test_pixtral.py \
-m cpu_model"
# Run compressed-tensor test
docker exec cpu-test-"$NUMA_NODE" bash -c "
set -e
pytest -x -s -v \
tests/quantization/test_compressed_tensors.py::test_compressed_tensors_w8a8_logprobs"
pytest -s -v \
tests/quantization/test_compressed_tensors.py::test_compressed_tensors_w8a8_logprobs[False-10-32-neuralmagic/Llama-3.2-1B-quantized.w8a8]"
# Note: disable it until supports V1
# Run AWQ test
# docker exec cpu-test-"$NUMA_NODE" bash -c "
# set -e
# VLLM_USE_V1=0 pytest -x -s -v \
# VLLM_USE_V1=0 pytest -s -v \
# tests/quantization/test_ipex_quant.py"
# Run multi-lora tests
docker exec cpu-test-"$NUMA_NODE" bash -c "
set -e
pytest -x -s -v \
pytest -s -v \
tests/lora/test_qwen2vl.py"
# online serving: tp+pp
# online serving
docker exec cpu-test-"$NUMA_NODE" bash -c '
set -e
VLLM_CPU_OMP_THREADS_BIND=$E2E_OMP_THREADS VLLM_CPU_SGL_KERNEL=1 vllm serve meta-llama/Llama-3.2-3B-Instruct -tp=2 -pp=2 &
server_pid=$!
timeout 600 bash -c "until curl localhost:8000/v1/models; do sleep 1; done" || exit 1
vllm bench serve \
--backend vllm \
--dataset-name random \
--model meta-llama/Llama-3.2-3B-Instruct \
--num-prompts 20 \
--endpoint /v1/completions
kill -s SIGTERM $server_pid &'
# online serving: tp+dp
docker exec cpu-test-"$NUMA_NODE" bash -c '
set -e
VLLM_CPU_OMP_THREADS_BIND=$E2E_OMP_THREADS VLLM_CPU_SGL_KERNEL=1 vllm serve meta-llama/Llama-3.2-3B-Instruct -tp=2 -dp=2 &
server_pid=$!
timeout 600 bash -c "until curl localhost:8000/v1/models; do sleep 1; done" || exit 1
vllm bench serve \
--backend vllm \
--dataset-name random \
--model meta-llama/Llama-3.2-3B-Instruct \
--num-prompts 20 \
--endpoint /v1/completions
kill -s SIGTERM $server_pid &'
--endpoint /v1/completions'
}
# All of CPU tests are expected to be finished less than 40 mins.
export -f cpu_tests
timeout 2h bash -c "cpu_tests $CORE_RANGE $NUMA_NODE"
timeout 1.5h bash -c "cpu_tests $CORE_RANGE $NUMA_NODE"

View File

@ -0,0 +1,64 @@
#!/bin/bash
# This script build the Neuron docker image and run the API server inside the container.
# It serves a sanity check for compilation and basic model usage.
set -e
set -v
image_name="neuron/vllm-ci"
container_name="neuron_$(tr -dc A-Za-z0-9 < /dev/urandom | head -c 10; echo)"
HF_CACHE="$(realpath ~)/huggingface"
mkdir -p "${HF_CACHE}"
HF_MOUNT="/root/.cache/huggingface"
HF_TOKEN=$(aws secretsmanager get-secret-value --secret-id "ci/vllm-neuron/hf-token" --region us-west-2 --query 'SecretString' --output text | jq -r .VLLM_NEURON_CI_HF_TOKEN)
NEURON_COMPILE_CACHE_URL="$(realpath ~)/neuron_compile_cache"
mkdir -p "${NEURON_COMPILE_CACHE_URL}"
NEURON_COMPILE_CACHE_MOUNT="/root/.cache/neuron_compile_cache"
# Try building the docker image
aws ecr-public get-login-password --region us-east-1 | docker login --username AWS --password-stdin public.ecr.aws
# prune old image and containers to save disk space, and only once a day
# by using a timestamp file in tmp.
if [ -f /tmp/neuron-docker-build-timestamp ]; then
last_build=$(cat /tmp/neuron-docker-build-timestamp)
current_time=$(date +%s)
if [ $((current_time - last_build)) -gt 86400 ]; then
# Remove dangling images (those that are not tagged and not used by any container)
docker image prune -f
# Remove unused volumes / force the system prune for old images as well.
docker volume prune -f && docker system prune -f
echo "$current_time" > /tmp/neuron-docker-build-timestamp
fi
else
date "+%s" > /tmp/neuron-docker-build-timestamp
fi
docker build -t "${image_name}" -f docker/Dockerfile.neuron .
# Setup cleanup
remove_docker_container() {
docker image rm -f "${image_name}" || true;
}
trap remove_docker_container EXIT
# Run the image
docker run --rm -it --device=/dev/neuron0 --network bridge \
-v "${HF_CACHE}:${HF_MOUNT}" \
-e "HF_HOME=${HF_MOUNT}" \
-e "HF_TOKEN=${HF_TOKEN}" \
-v "${NEURON_COMPILE_CACHE_URL}:${NEURON_COMPILE_CACHE_MOUNT}" \
-e "NEURON_COMPILE_CACHE_URL=${NEURON_COMPILE_CACHE_MOUNT}" \
--name "${container_name}" \
${image_name} \
/bin/bash -c "
set -e; # Exit on first error
python3 /workspace/vllm/examples/offline_inference/neuron.py;
python3 -m pytest /workspace/vllm/tests/neuron/1_core/ -v --capture=tee-sys;
for f in /workspace/vllm/tests/neuron/2_core/*.py; do
echo \"Running test file: \$f\";
python3 -m pytest \$f -v --capture=tee-sys;
done
"

View File

@ -1,191 +0,0 @@
#!/bin/bash
# This script build the Ascend NPU docker image and run the offline inference inside the container.
# It serves a sanity check for compilation and basic model usage.
set -ex
# Base ubuntu image with basic ascend development libraries and python installed
VLLM_ASCEND_REPO="https://github.com/vllm-project/vllm-ascend.git"
CONFIG_FILE_REMOTE_PATH="tests/e2e/vllm_interface/vllm_test.cfg"
TEST_RUN_CONFIG_FILE="vllm_test.cfg"
VLLM_ASCEND_TMP_DIR=
# Get the test run configuration file from the vllm-ascend repository
fetch_vllm_test_cfg() {
VLLM_ASCEND_TMP_DIR=$(mktemp -d)
# Ensure that the temporary directory is cleaned up when an exception occurs during configuration file retrieval
cleanup() {
rm -rf "${VLLM_ASCEND_TMP_DIR}"
}
trap cleanup EXIT
GIT_TRACE=1 git clone -v --depth 1 "${VLLM_ASCEND_REPO}" "${VLLM_ASCEND_TMP_DIR}"
if [ ! -f "${VLLM_ASCEND_TMP_DIR}/${CONFIG_FILE_REMOTE_PATH}" ]; then
echo "Error: file '${CONFIG_FILE_REMOTE_PATH}' does not exist in the warehouse" >&2
exit 1
fi
# If the file already exists locally, just overwrite it
cp "${VLLM_ASCEND_TMP_DIR}/${CONFIG_FILE_REMOTE_PATH}" "${TEST_RUN_CONFIG_FILE}"
echo "Copied ${CONFIG_FILE_REMOTE_PATH} to ${TEST_RUN_CONFIG_FILE}"
# Since the trap will be overwritten later, and when it is executed here, the task of cleaning up resources
# when the trap is abnormal has been completed, so the temporary resources are manually deleted here.
rm -rf "${VLLM_ASCEND_TMP_DIR}"
trap - EXIT
}
# Downloads test run configuration file from a remote URL.
# Loads the configuration into the current script environment.
get_config() {
if [ ! -f "${TEST_RUN_CONFIG_FILE}" ]; then
echo "Error: file '${TEST_RUN_CONFIG_FILE}' does not exist in the warehouse" >&2
exit 1
fi
source "${TEST_RUN_CONFIG_FILE}"
echo "Base docker image name that get from configuration: ${BASE_IMAGE_NAME}"
return 0
}
# get test running configuration.
fetch_vllm_test_cfg
get_config
# Check if the function call was successful. If not, exit the script.
if [ $? -ne 0 ]; then
exit 1
fi
image_name="npu/vllm-ci:${BUILDKITE_COMMIT}_${EPOCHSECONDS}"
container_name="npu_${BUILDKITE_COMMIT}_$(tr -dc A-Za-z0-9 < /dev/urandom | head -c 10; echo)"
# BUILDKITE_AGENT_NAME format is {hostname}-{agent_idx}-{npu_card_num}cards
agent_idx=$(echo "${BUILDKITE_AGENT_NAME}" | awk -F'-' '{print $(NF-1)}')
echo "agent_idx: ${agent_idx}"
builder_name="cachebuilder${agent_idx}"
builder_cache_dir="/mnt/docker-cache${agent_idx}"
mkdir -p ${builder_cache_dir}
# Try building the docker image
cat <<EOF | DOCKER_BUILDKIT=1 docker build \
--add-host cache-service-vllm.nginx-pypi-cache.svc.cluster.local:${PYPI_CACHE_HOST} \
--builder ${builder_name} --cache-from type=local,src=${builder_cache_dir} \
--cache-to type=local,dest=${builder_cache_dir},mode=max \
--progress=plain --load -t ${image_name} -f - .
FROM ${BASE_IMAGE_NAME}
# Define environments
ENV DEBIAN_FRONTEND=noninteractive
RUN pip config set global.index-url http://cache-service-vllm.nginx-pypi-cache.svc.cluster.local:${PYPI_CACHE_PORT}/pypi/simple && \
pip config set global.trusted-host cache-service-vllm.nginx-pypi-cache.svc.cluster.local && \
apt-get update -y && \
apt-get install -y python3-pip git vim wget net-tools gcc g++ cmake libnuma-dev && \
rm -rf /var/cache/apt/* && \
rm -rf /var/lib/apt/lists/*
# Install for pytest to make the docker build cache layer always valid
RUN --mount=type=cache,target=/root/.cache/pip \
pip install pytest>=6.0 modelscope
WORKDIR /workspace/vllm
# Install vLLM dependencies in advance. Effect: As long as common.txt remains unchanged, the docker cache layer will be valid.
COPY requirements/common.txt /workspace/vllm/requirements/common.txt
RUN --mount=type=cache,target=/root/.cache/pip \
pip install -r requirements/common.txt
COPY . .
# Install vLLM
RUN --mount=type=cache,target=/root/.cache/pip \
VLLM_TARGET_DEVICE="empty" python3 -m pip install -v -e /workspace/vllm/ --extra-index https://download.pytorch.org/whl/cpu/ && \
python3 -m pip uninstall -y triton
# Install vllm-ascend
WORKDIR /workspace
ARG VLLM_ASCEND_REPO=https://github.com/vllm-project/vllm-ascend.git
ARG VLLM_ASCEND_TAG=main
RUN git config --global url."https://gh-proxy.test.osinfra.cn/https://github.com/".insteadOf "https://github.com/" && \
git clone --depth 1 \$VLLM_ASCEND_REPO --branch \$VLLM_ASCEND_TAG /workspace/vllm-ascend
# Install vllm dependencies in advance. Effect: As long as common.txt remains unchanged, the docker cache layer will be valid.
RUN --mount=type=cache,target=/root/.cache/pip \
pip install -r /workspace/vllm-ascend/requirements.txt
RUN --mount=type=cache,target=/root/.cache/pip \
export PIP_EXTRA_INDEX_URL=https://mirrors.huaweicloud.com/ascend/repos/pypi && \
source /usr/local/Ascend/ascend-toolkit/set_env.sh && \
source /usr/local/Ascend/nnal/atb/set_env.sh && \
export LD_LIBRARY_PATH=\$LD_LIBRARY_PATH:/usr/local/Ascend/ascend-toolkit/latest/`uname -i`-linux/devlib && \
python3 -m pip install -v -e /workspace/vllm-ascend/ --extra-index https://download.pytorch.org/whl/cpu/
ENV VLLM_WORKER_MULTIPROC_METHOD=spawn
ENV VLLM_USE_MODELSCOPE=True
WORKDIR /workspace/vllm-ascend
CMD ["/bin/bash"]
EOF
# Setup cleanup
remove_docker_container() {
docker rm -f "${container_name}" || true;
docker image rm -f "${image_name}" || true;
docker system prune -f || true;
}
trap remove_docker_container EXIT
# Generate corresponding --device args based on BUILDKITE_AGENT_NAME
# Ascend NPU BUILDKITE_AGENT_NAME format is {hostname}-{agent_idx}-{npu_card_num}cards, and agent_idx starts from 1.
# e.g. atlas-a2-001-1-2cards means this is the 1-th agent on atlas-a2-001 host, and it has 2 NPU cards.
# returns --device /dev/davinci0 --device /dev/davinci1
parse_and_gen_devices() {
local input="$1"
local index cards_num
if [[ "$input" =~ ([0-9]+)-([0-9]+)cards$ ]]; then
index="${BASH_REMATCH[1]}"
cards_num="${BASH_REMATCH[2]}"
else
echo "parse error" >&2
return 1
fi
local devices=""
local i=0
while (( i < cards_num )); do
local dev_idx=$(((index - 1)*cards_num + i ))
devices="$devices --device /dev/davinci${dev_idx}"
((i++))
done
# trim leading space
devices="${devices#"${devices%%[![:space:]]*}"}"
# Output devices: assigned to the caller variable
printf '%s' "$devices"
}
devices=$(parse_and_gen_devices "${BUILDKITE_AGENT_NAME}") || exit 1
# Run the image and execute the Out-Of-Tree (OOT) platform interface test case on Ascend NPU hardware.
# This test checks whether the OOT platform interface is functioning properly in conjunction with
# the hardware plugin vllm-ascend.
model_cache_dir=/mnt/modelscope${agent_idx}
mkdir -p ${model_cache_dir}
docker run \
${devices} \
--device /dev/davinci_manager \
--device /dev/devmm_svm \
--device /dev/hisi_hdc \
-v /usr/local/dcmi:/usr/local/dcmi \
-v /usr/local/bin/npu-smi:/usr/local/bin/npu-smi \
-v /usr/local/Ascend/driver/lib64/:/usr/local/Ascend/driver/lib64/ \
-v /usr/local/Ascend/driver/version.info:/usr/local/Ascend/driver/version.info \
-v /etc/ascend_install.info:/etc/ascend_install.info \
-v ${model_cache_dir}:/root/.cache/modelscope \
--entrypoint="" \
--name "${container_name}" \
"${image_name}" \
bash -c '
set -e
pytest -v -s tests/e2e/vllm_interface/
'

View File

@ -61,12 +61,13 @@ echo "Results will be stored in: $RESULTS_DIR"
echo "--- Installing Python dependencies ---"
python3 -m pip install --progress-bar off git+https://github.com/thuml/depyf.git \
&& python3 -m pip install --progress-bar off pytest pytest-asyncio tpu-info \
&& python3 -m pip install --progress-bar off "lm-eval @ git+https://github.com/EleutherAI/lm-evaluation-harness.git@206b7722158f58c35b7ffcd53b035fdbdda5126d" \
&& python3 -m pip install --progress-bar off hf-transfer tblib==3.1.0
&& python3 -m pip install --progress-bar off lm_eval[api]==0.4.4 \
&& python3 -m pip install --progress-bar off hf-transfer
echo "--- Python dependencies installed ---"
export VLLM_USE_V1=1
export VLLM_XLA_CHECK_RECOMPILATION=1
export VLLM_XLA_CACHE_PATH=
echo "Using VLLM V1"
echo "--- Hardware Information ---"
# tpu-info

View File

@ -61,12 +61,13 @@ echo "Results will be stored in: $RESULTS_DIR"
echo "--- Installing Python dependencies ---"
python3 -m pip install --progress-bar off git+https://github.com/thuml/depyf.git \
&& python3 -m pip install --progress-bar off pytest pytest-asyncio tpu-info \
&& python3 -m pip install --progress-bar off "lm-eval @ git+https://github.com/EleutherAI/lm-evaluation-harness.git@206b7722158f58c35b7ffcd53b035fdbdda5126d" \
&& python3 -m pip install --progress-bar off hf-transfer tblib==3.1.0
&& python3 -m pip install --progress-bar off lm_eval[api]==0.4.4 \
&& python3 -m pip install --progress-bar off hf-transfer
echo "--- Python dependencies installed ---"
export VLLM_USE_V1=1
export VLLM_XLA_CHECK_RECOMPILATION=1
export VLLM_XLA_CACHE_PATH=
echo "Using VLLM V1"
echo "--- Hardware Information ---"
# tpu-info

View File

@ -20,32 +20,24 @@ trap remove_docker_container EXIT
# Run the image and test offline inference/tensor parallel
docker run \
--device /dev/dri:/dev/dri \
--net=host \
--ipc=host \
--privileged \
--device /dev/dri \
-v /dev/dri/by-path:/dev/dri/by-path \
--entrypoint="" \
-e "HF_TOKEN=${HF_TOKEN}" \
-e "ZE_AFFINITY_MASK=${ZE_AFFINITY_MASK}" \
--name "${container_name}" \
"${image_name}" \
bash -c '
set -e
echo $ZE_AFFINITY_MASK
pip install tblib==3.1.0
python3 examples/offline_inference/basic/generate.py --model facebook/opt-125m --block-size 64 --enforce-eager
python3 examples/offline_inference/basic/generate.py --model facebook/opt-125m --block-size 64 -O3 -O.cudagraph_mode=NONE
python3 examples/offline_inference/basic/generate.py --model facebook/opt-125m --block-size 64 --enforce-eager -tp 2 --distributed-executor-backend ray
python3 examples/offline_inference/basic/generate.py --model facebook/opt-125m --block-size 64 --enforce-eager -tp 2 --distributed-executor-backend mp
VLLM_ATTENTION_BACKEND=TRITON_ATTN python3 examples/offline_inference/basic/generate.py --model facebook/opt-125m --block-size 64 --enforce-eager
sh -c '
VLLM_USE_V1=1 python3 examples/offline_inference/basic/generate.py --model facebook/opt-125m --block-size 64 --enforce-eager
VLLM_USE_V1=1 python3 examples/offline_inference/basic/generate.py --model facebook/opt-125m --block-size 64 --enforce-eager -tp 2 --distributed-executor-backend ray
VLLM_USE_V1=1 python3 examples/offline_inference/basic/generate.py --model facebook/opt-125m --block-size 64 --enforce-eager -tp 2 --distributed-executor-backend mp
cd tests
pytest -v -s v1/core
pytest -v -s v1/engine
pytest -v -s v1/sample --ignore=v1/sample/test_logprobs.py --ignore=v1/sample/test_logprobs_e2e.py
pytest -v -s v1/worker --ignore=v1/worker/test_gpu_model_runner.py
pytest -v -s v1/structured_output
pytest -v -s v1/spec_decode --ignore=v1/spec_decode/test_max_len.py --ignore=v1/spec_decode/test_tree_attention.py --ignore=v1/spec_decode/test_speculators_eagle3.py
pytest -v -s v1/kv_connector/unit --ignore=v1/kv_connector/unit/test_multi_connector.py --ignore=v1/kv_connector/unit/test_nixl_connector.py --ignore=v1/kv_connector/unit/test_shared_storage_connector.py
pytest -v -s v1/spec_decode --ignore=v1/spec_decode/test_max_len.py --ignore=v1/spec_decode/test_eagle.py
pytest -v -s v1/kv_connector/unit --ignore=v1/kv_connector/unit/test_multi_connector.py --ignore=v1/kv_connector/unit/test_nixl_connector.py
pytest -v -s v1/test_serial_utils.py
pytest -v -s v1/test_utils.py
pytest -v -s v1/test_metrics_reader.py
'

View File

@ -18,7 +18,7 @@ vllm bench throughput --input-len 256 --output-len 256 --output-json throughput_
bench_throughput_exit_code=$?
# run server-based benchmarks and upload the result to buildkite
vllm serve meta-llama/Llama-2-7b-chat-hf &
python3 -m vllm.entrypoints.openai.api_server --model meta-llama/Llama-2-7b-chat-hf &
server_pid=$!
wget https://huggingface.co/datasets/anon8231489123/ShareGPT_Vicuna_unfiltered/resolve/main/ShareGPT_V3_unfiltered_cleaned_split.json

View File

@ -1,59 +0,0 @@
#!/bin/bash
# SPDX-License-Identifier: Apache-2.0
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
# Setup script for Prime-RL integration tests
# This script prepares the environment for running Prime-RL tests with nightly vLLM
set -euo pipefail
SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
REPO_ROOT="$(cd "${SCRIPT_DIR}/../.." && pwd)"
PRIME_RL_REPO="https://github.com/PrimeIntellect-ai/prime-rl.git"
PRIME_RL_DIR="${REPO_ROOT}/prime-rl"
echo "Setting up Prime-RL integration test environment..."
# Clean up any existing Prime-RL directory
if [ -d "${PRIME_RL_DIR}" ]; then
echo "Removing existing Prime-RL directory..."
rm -rf "${PRIME_RL_DIR}"
fi
# Install UV if not available
if ! command -v uv &> /dev/null; then
echo "Installing UV package manager..."
curl -LsSf https://astral.sh/uv/install.sh | sh
source $HOME/.local/bin/env
fi
# Clone Prime-RL repository at specific branch for reproducible tests
PRIME_RL_BRANCH="integ-vllm-main"
echo "Cloning Prime-RL repository at branch: ${PRIME_RL_BRANCH}..."
git clone --branch "${PRIME_RL_BRANCH}" --single-branch "${PRIME_RL_REPO}" "${PRIME_RL_DIR}"
cd "${PRIME_RL_DIR}"
echo "Setting up UV project environment..."
export UV_PROJECT_ENVIRONMENT=/usr/local
ln -s /usr/bin/python3 /usr/local/bin/python
# Remove vllm pin from pyproject.toml
echo "Removing vllm pin from pyproject.toml..."
sed -i '/vllm==/d' pyproject.toml
# Sync Prime-RL dependencies
echo "Installing Prime-RL dependencies..."
uv sync --inexact && uv sync --inexact --all-extras
# Verify installation
echo "Verifying installations..."
uv run python -c "import vllm; print(f'vLLM version: {vllm.__version__}')"
uv run python -c "import prime_rl; print('Prime-RL imported successfully')"
echo "Prime-RL integration test environment setup complete!"
echo "Running Prime-RL integration tests..."
export WANDB_MODE=offline # this makes this test not require a WANDB_API_KEY
uv run pytest -vs tests/integration/test_rl.py -m gpu
echo "Prime-RL integration tests completed!"

View File

@ -1,62 +0,0 @@
#!/usr/bin/env bash
set -euxo pipefail
# args: [THRESHOLD] [NUM_QUESTIONS] [START_PORT]
THRESHOLD=${1:-0.25}
NUM_Q=${2:-1319}
PORT=${3:-8010}
OUT_DIR=${OUT_DIR:-/tmp/vllm-scheduled}
mkdir -p "${OUT_DIR}"
wait_for_server() {
local port=$1
timeout 600 bash -c '
until curl -sf "http://127.0.0.1:'"$port"'/health" > /dev/null; do
sleep 1
done'
}
MODEL="deepseek-ai/DeepSeek-V2-lite"
BACKENDS=("deepep_high_throughput" "deepep_low_latency")
cleanup() {
if [[ -n "${SERVER_PID:-}" ]] && kill -0 "${SERVER_PID}" 2>/dev/null; then
kill "${SERVER_PID}" 2>/dev/null || true
for _ in {1..20}; do
kill -0 "${SERVER_PID}" 2>/dev/null || break
sleep 0.5
done
kill -9 "${SERVER_PID}" 2>/dev/null || true
fi
}
trap cleanup EXIT
for BACK in "${BACKENDS[@]}"; do
VLLM_DEEP_GEMM_WARMUP=skip \
VLLM_ALL2ALL_BACKEND=$BACK \
vllm serve "$MODEL" \
--enforce-eager \
--tensor-parallel-size 2 \
--data-parallel-size 2 \
--enable-expert-parallel \
--enable-eplb \
--trust-remote-code \
--max-model-len 2048 \
--port $PORT &
SERVER_PID=$!
wait_for_server $PORT
TAG=$(echo "$MODEL" | tr '/: \\n' '_____')
OUT="${OUT_DIR}/${TAG}_${BACK}.json"
python3 tests/evals/gsm8k/gsm8k_eval.py --host http://127.0.0.1 --port $PORT --num-questions ${NUM_Q} --save-results ${OUT}
python3 - <<PY
import json; acc=json.load(open('${OUT}'))['accuracy']
print(f"${MODEL} ${BACK}: accuracy {acc:.3f}")
assert acc >= ${THRESHOLD}, f"${MODEL} ${BACK} accuracy {acc}"
PY
cleanup
SERVER_PID=
sleep 1
PORT=$((PORT+1))
done

View File

@ -1,61 +0,0 @@
#!/usr/bin/env bash
set -euxo pipefail
# args: [THRESHOLD] [NUM_QUESTIONS] [START_PORT]
THRESHOLD=${1:-0.8}
NUM_Q=${2:-1319}
PORT=${3:-8020}
OUT_DIR=${OUT_DIR:-/tmp/vllm-scheduled}
mkdir -p "${OUT_DIR}"
wait_for_server() {
local port=$1
timeout 600 bash -c '
until curl -sf "http://127.0.0.1:'"$port"'/health" > /dev/null; do
sleep 1
done'
}
MODEL="QWen/Qwen3-30B-A3B-FP8"
BACKENDS=("deepep_high_throughput" "deepep_low_latency")
cleanup() {
if [[ -n "${SERVER_PID:-}" ]] && kill -0 "${SERVER_PID}" 2>/dev/null; then
kill "${SERVER_PID}" 2>/dev/null || true
for _ in {1..20}; do
kill -0 "${SERVER_PID}" 2>/dev/null || break
sleep 0.5
done
kill -9 "${SERVER_PID}" 2>/dev/null || true
fi
}
trap cleanup EXIT
for BACK in "${BACKENDS[@]}"; do
VLLM_DEEP_GEMM_WARMUP=skip \
VLLM_ALL2ALL_BACKEND=$BACK \
vllm serve "$MODEL" \
--enforce-eager \
--tensor-parallel-size 2 \
--data-parallel-size 2 \
--enable-expert-parallel \
--trust-remote-code \
--max-model-len 2048 \
--port $PORT &
SERVER_PID=$!
wait_for_server $PORT
TAG=$(echo "$MODEL" | tr '/: \\n' '_____')
OUT="${OUT_DIR}/${TAG}_${BACK}.json"
python3 tests/evals/gsm8k/gsm8k_eval.py --host http://127.0.0.1 --port $PORT --num-questions ${NUM_Q} --save-results ${OUT}
python3 - <<PY
import json; acc=json.load(open('${OUT}'))['accuracy']
print(f"${MODEL} ${BACK}: accuracy {acc:.3f}")
assert acc >= ${THRESHOLD}, f"${MODEL} ${BACK} accuracy {acc}"
PY
cleanup
SERVER_PID=
sleep 1
PORT=$((PORT+1))
done

View File

@ -17,7 +17,7 @@ if [ "$disk_usage" -gt "$threshold" ]; then
# Remove dangling images (those that are not tagged and not used by any container)
docker image prune -f
# Remove unused volumes / force the system prune for old images as well.
docker volume prune -f && docker system prune --force --filter "until=24h" --all
docker volume prune -f && docker system prune --force --filter "until=72h" --all
echo "Docker images and volumes cleanup completed."
else
echo "Disk usage is below $threshold%. No cleanup needed."

View File

@ -9,6 +9,6 @@ MAX_NUM_BATCHED_TOKENS=1024
TENSOR_PARALLEL_SIZE=1
MAX_MODEL_LEN=2048
DOWNLOAD_DIR=/mnt/disks/persist
EXPECTED_THROUGHPUT=8.7
EXPECTED_THROUGHPUT=10.0
INPUT_LEN=1800
OUTPUT_LEN=128

View File

@ -42,7 +42,7 @@ echo "lanching vllm..."
echo "logging to $VLLM_LOG"
echo
vllm serve $MODEL \
VLLM_USE_V1=1 vllm serve $MODEL \
--seed 42 \
--max-num-seqs $MAX_NUM_SEQS \
--max-num-batched-tokens $MAX_NUM_BATCHED_TOKENS \

View File

@ -14,19 +14,8 @@ fi
# Get the single wheel file
wheel="${wheel_files[0]}"
# Detect architecture and rename 'linux' to appropriate manylinux version
arch=$(uname -m)
if [[ $arch == "x86_64" ]]; then
manylinux_version="manylinux1"
elif [[ $arch == "aarch64" ]]; then
manylinux_version="manylinux2014"
else
echo "Warning: Unknown architecture $arch, using manylinux1 as default"
manylinux_version="manylinux1"
fi
# Rename 'linux' to the appropriate manylinux version in the wheel filename
new_wheel="${wheel/linux/$manylinux_version}"
# Rename 'linux' to 'manylinux1' in the wheel filename
new_wheel="${wheel/linux/manylinux1}"
mv -- "$wheel" "$new_wheel"
wheel="$new_wheel"
@ -58,25 +47,31 @@ python3 .buildkite/generate_index.py --wheel "$normal_wheel"
aws s3 cp "$wheel" "s3://vllm-wheels/$BUILDKITE_COMMIT/"
aws s3 cp "$normal_wheel" "s3://vllm-wheels/$BUILDKITE_COMMIT/"
if [[ $normal_wheel == *"cu129"* ]]; then
# only upload index.html for cu129 wheels (default wheels) as it
# is available on both x86 and arm64
if [[ $normal_wheel == *"cu118"* ]]; then
# if $normal_wheel matches cu118, do not upload the index.html
echo "Skipping index files for cu118 wheels"
elif [[ $normal_wheel == *"cu126"* ]]; then
# if $normal_wheel matches cu126, do not upload the index.html
echo "Skipping index files for cu126 wheels"
else
# only upload index.html for cu128 wheels (default wheels)
aws s3 cp index.html "s3://vllm-wheels/$BUILDKITE_COMMIT/vllm/index.html"
aws s3 cp "s3://vllm-wheels/nightly/index.html" "s3://vllm-wheels/$BUILDKITE_COMMIT/index.html"
else
echo "Skipping index files for non-cu129 wheels"
fi
# generate index for nightly
aws s3 cp "$wheel" "s3://vllm-wheels/nightly/"
aws s3 cp "$normal_wheel" "s3://vllm-wheels/nightly/"
if [[ $normal_wheel == *"cu129"* ]]; then
# only upload index.html for cu129 wheels (default wheels) as it
# is available on both x86 and arm64
aws s3 cp index.html "s3://vllm-wheels/nightly/vllm/index.html"
if [[ $normal_wheel == *"cu118"* ]]; then
# if $normal_wheel matches cu118, do not upload the index.html
echo "Skipping index files for cu118 wheels"
elif [[ $normal_wheel == *"cu126"* ]]; then
# if $normal_wheel matches cu126, do not upload the index.html
echo "Skipping index files for cu126 wheels"
else
echo "Skipping index files for non-cu129 wheels"
# only upload index.html for cu128 wheels (default wheels)
aws s3 cp index.html "s3://vllm-wheels/nightly/vllm/index.html"
fi
aws s3 cp "$wheel" "s3://vllm-wheels/$version/"

File diff suppressed because it is too large Load Diff

File diff suppressed because it is too large Load Diff

View File

@ -1,47 +0,0 @@
[run]
# Track the installed vllm package (this is what actually gets imported during tests)
# Use wildcard pattern to match the installed location
source =
vllm
*/dist-packages/vllm
*/site-packages/vllm
omit =
*/tests/*
*/test_*
*/__pycache__/*
*/build/*
*/dist/*
*/vllm.egg-info/*
*/third_party/*
*/examples/*
*/benchmarks/*
*/docs/*
[paths]
# Map all possible vllm locations to a canonical "vllm" path
# This ensures coverage.combine properly merges data from different test runs
source =
vllm
/vllm-workspace/src/vllm
/vllm-workspace/vllm
*/site-packages/vllm
*/dist-packages/vllm
[report]
exclude_lines =
pragma: no cover
def __repr__
if self.debug:
if settings.DEBUG
raise AssertionError
raise NotImplementedError
if 0:
if __name__ == .__main__.:
class .*\bProtocol\):
@(abc\.)?abstractmethod
[html]
directory = htmlcov
[xml]
output = coverage.xml

View File

@ -1,4 +0,0 @@
# Migrate from `yapf` & `isort` to `ruff`
d6953beb91da4e9c99be4c0a1304a2d24189535c
# Convert `Optional[x]` to `x | None` and `Union[x, y]` to `x | y`
8fcaaf6a165e661f63fc51be906bc05b0767332f

View File

@ -1,24 +0,0 @@
# doc: https://github.com/pytorch/test-infra/blob/main/tools/stronghold/docs/bc_linter_config.md
version: 1
paths:
# We temporarily disable globally, and will only enable with `annotations.include`
# include:
# - "vllm/v1/attetion/*.py"
# - "vllm/v1/core/*.py"
exclude:
- "**/*.py"
scan:
functions: true # check free functions and methods
classes: true # check classes/dataclasses
public_only: true # ignore names starting with "_" at any level
annotations:
include: # decorators that forceinclude a symbol
- name: "bc_linter_include" # matched by simple name or dotted suffix
propagate_to_members: false # for classes, include methods/inner classes
exclude: # decorators that forceexclude a symbol
- name: "bc_linter_skip" # matched by simple name or dotted suffix
propagate_to_members: true # for classes, exclude methods/inner classes
excluded_violations: [] # e.g. ["ParameterRenamed", "FieldTypeChanged"]

99
.github/CODEOWNERS vendored
View File

@ -2,86 +2,62 @@
# for more info about CODEOWNERS file
# This lists cover the "core" components of vLLM that require careful review
/vllm/attention @LucasWilkinson
/vllm/attention/backends/abstract.py @WoosukKwon @zhuohan123 @youkaichao @alexm-redhat @comaniac @njhill
/vllm/executor/executor_base.py @zhuohan123 @youkaichao @alexm-redhat @comaniac @njhill @22quinn
/vllm/model_executor/layers/fused_moe @mgoin @pavanimajety
/vllm/model_executor/layers/quantization @mgoin @robertgshaw2-redhat @tlrmchlsmth @yewentao256 @pavanimajety
/vllm/model_executor/layers/mamba @tdoublep
/vllm/model_executor/model_loader @22quinn
/vllm/multimodal @DarkLight1337 @ywang96 @NickLucche
/vllm/core @zhuohan123 @youkaichao @alexm-redhat @comaniac @njhill
/vllm/engine/llm_engine.py @zhuohan123 @youkaichao @alexm-redhat @comaniac @njhill
/vllm/executor/executor_base.py @zhuohan123 @youkaichao @alexm-redhat @comaniac @njhill
/vllm/worker/worker_base.py @zhuohan123 @youkaichao @alexm-redhat @comaniac @njhill
/vllm/worker/worker.py @zhuohan123 @youkaichao @alexm-redhat @comaniac @njhill
/vllm/model_executor/layers/sampler.py @zhuohan123 @youkaichao @alexm-redhat @comaniac @njhill
/vllm/model_executor/layers/quantization @mgoin @robertgshaw2-redhat @tlrmchlsmth @yewentao256
/vllm/multimodal @DarkLight1337 @ywang96
/vllm/vllm_flash_attn @LucasWilkinson
/vllm/lora @jeejeelee
/vllm/reasoning @aarnphm @chaunceyjiang
/vllm/entrypoints @aarnphm @chaunceyjiang
/vllm/reasoning @aarnphm
/vllm/entrypoints @aarnphm
/vllm/compilation @zou3519 @youkaichao @ProExpertProg
/vllm/distributed/kv_transfer @NickLucche @ApostaC
CMakeLists.txt @tlrmchlsmth @LucasWilkinson
# Any change to the VllmConfig changes can have a large user-facing impact,
# so spam a lot of people
/vllm/config @simon-mo @WoosukKwon @youkaichao @robertgshaw2-redhat @mgoin @tlrmchlsmth @houseroad @hmellor @yewentao256 @ProExpertProg
/vllm/config/cache.py @simon-mo @WoosukKwon @youkaichao @robertgshaw2-redhat @mgoin @tlrmchlsmth @houseroad @hmellor @yewentao256 @ProExpertProg @heheda12345
# vLLM V1
/vllm/v1/attention @LucasWilkinson
/vllm/v1/attention/backends/mla @pavanimajety
/vllm/v1/attention/backends/flashinfer.py @mgoin @pavanimajety
/vllm/v1/attention/backends/triton_attn.py @tdoublep
/vllm/v1/core @WoosukKwon @robertgshaw2-redhat @njhill @ywang96 @comaniac @alexm-redhat @heheda12345 @ApostaC
/vllm/v1/sample @22quinn @houseroad @njhill
/vllm/v1/spec_decode @benchislett @luccafong
/vllm/v1/structured_output @mgoin @russellb @aarnphm @benchislett
/vllm/v1/kv_cache_interface.py @heheda12345
/vllm/v1/offloading @ApostaC
/vllm/v1 @WoosukKwon @robertgshaw2-redhat @njhill @ywang96 @comaniac @alexm-redhat
/vllm/v1/structured_output @mgoin @russellb @aarnphm
# Test ownership
/.buildkite/lm-eval-harness @mgoin @simon-mo
/tests/async_engine @njhill @robertgshaw2-redhat @simon-mo
/tests/basic_correctness/test_chunked_prefill @rkooo567 @comaniac
/tests/distributed/test_multi_node_assignment.py @youkaichao
/tests/distributed/test_pipeline_parallel.py @youkaichao
/tests/distributed/test_same_node.py @youkaichao
/tests/entrypoints @DarkLight1337 @robertgshaw2-redhat @simon-mo @aarnphm @NickLucche
/tests/evals @mgoin
/tests/kernels @mgoin @tlrmchlsmth @WoosukKwon @yewentao256
/tests/entrypoints @DarkLight1337 @robertgshaw2-redhat @simon-mo @aarnphm
/tests/kernels @tlrmchlsmth @WoosukKwon @yewentao256
/tests/models @DarkLight1337 @ywang96
/tests/multimodal @DarkLight1337 @ywang96 @NickLucche
/tests/quantization @mgoin @robertgshaw2-redhat @yewentao256 @pavanimajety
/tests/multimodal @DarkLight1337 @ywang96
/tests/prefix_caching @comaniac @KuntaiDu
/tests/quantization @mgoin @robertgshaw2-redhat @yewentao256
/tests/test_inputs.py @DarkLight1337 @ywang96
/tests/v1/entrypoints/llm/test_struct_output_generate.py @mgoin @russellb @aarnphm
/tests/v1/structured_output @mgoin @russellb @aarnphm
/tests/v1/core @WoosukKwon @robertgshaw2-redhat @njhill @ywang96 @comaniac @alexm-redhat @heheda12345 @ApostaC
/tests/weight_loading @mgoin @youkaichao @yewentao256
/tests/lora @jeejeelee
/tests/models/language/generation/test_hybrid.py @tdoublep
/tests/v1/kv_connector/nixl_integration @NickLucche
/tests/v1/kv_connector @ApostaC
/tests/v1/offloading @ApostaC
# Transformers backend
/vllm/model_executor/models/transformers @hmellor
/tests/models/test_transformers.py @hmellor
# Docs
/docs/mkdocs @hmellor
/docs/**/*.yml @hmellor
/requirements/docs.txt @hmellor
.readthedocs.yaml @hmellor
/docs @hmellor
mkdocs.yaml @hmellor
# Linting
.markdownlint.yaml @hmellor
.pre-commit-config.yaml @hmellor
/tools/pre_commit @hmellor
# CPU
/vllm/v1/worker/cpu* @bigPYJ1151
/vllm/v1/worker/^cpu @bigPYJ1151
/csrc/cpu @bigPYJ1151
/vllm/platforms/cpu.py @bigPYJ1151
/cmake/cpu_extension.cmake @bigPYJ1151
/docker/Dockerfile.cpu @bigPYJ1151
# Intel GPU
/vllm/v1/worker/xpu* @jikunshang
/vllm/v1/worker/^xpu @jikunshang
/vllm/platforms/xpu.py @jikunshang
/docker/Dockerfile.xpu @jikunshang
@ -89,9 +65,6 @@ mkdocs.yaml @hmellor
/vllm/attention/backends/dual_chunk_flash_attn.py @sighingnow
/vllm/model_executor/models/qwen* @sighingnow
# MTP-specific files
/vllm/model_executor/models/deepseek_mtp.py @luccafong
# Mistral-specific files
/vllm/model_executor/models/mistral*.py @patrickvonplaten
/vllm/model_executor/models/mixtral*.py @patrickvonplaten
@ -99,31 +72,3 @@ mkdocs.yaml @hmellor
/vllm/model_executor/models/pixtral*.py @patrickvonplaten
/vllm/transformers_utils/configs/mistral.py @patrickvonplaten
/vllm/transformers_utils/tokenizers/mistral.py @patrickvonplaten
# Kernels
/vllm/attention/ops/chunked_prefill_paged_decode.py @tdoublep
/vllm/attention/ops/triton_unified_attention.py @tdoublep
# ROCm related: specify owner with write access to notify AMD folks for careful code review
/docker/Dockerfile.rocm* @gshtras
/vllm/v1/attention/backends/rocm*.py @gshtras
/vllm/v1/attention/backends/mla/rocm*.py @gshtras
/vllm/attention/ops/rocm*.py @gshtras
/vllm/model_executor/layers/fused_moe/rocm*.py @gshtras
# TPU
/vllm/v1/worker/tpu* @NickLucche
/vllm/platforms/tpu.py @NickLucche
/vllm/v1/sample/tpu @NickLucche
/vllm/tests/v1/tpu @NickLucche
# KVConnector installation files
/requirements/kv_connectors.txt @NickLucche
# Pooling models
/examples/*/pooling/ @noooop
/tests/models/*/pooling* @noooop
/tests/entrypoints/pooling @noooop
/vllm/config/pooler.py @noooop
/vllm/pooling_params.py @noooop
/vllm/model_executor/layers/pooler.py @noooop

View File

@ -43,6 +43,10 @@ body:
Any other things you would like to mention.
validations:
required: false
- type: markdown
attributes:
value: >
Thanks for contributing 🎉! The vLLM core team hosts a biweekly RFC review session at 9:30AM Pacific Time, while most RFCs can be discussed online, you can optionally sign up for a slot to discuss your RFC online [here](https://docs.google.com/document/d/1CiLVBZeIVfR7_PNAKVSusxpceywkoOOB78qoWqHvSZc/edit).
- type: checkboxes
id: askllm
attributes:

View File

@ -7,6 +7,8 @@ PLEASE FILL IN THE PR DESCRIPTION HERE ENSURING ALL CHECKLIST ITEMS (AT THE BOTT
## Test Result
## (Optional) Documentation Update
---
<details>
<summary> Essential Elements of an Effective PR Description Checklist </summary>
@ -15,7 +17,6 @@ PLEASE FILL IN THE PR DESCRIPTION HERE ENSURING ALL CHECKLIST ITEMS (AT THE BOTT
- [ ] The test plan, such as providing test command.
- [ ] The test results, such as pasting the results comparison before and after, or e2e results
- [ ] (Optional) The necessary documentation update, such as updating `supported_models.md` and `examples` for a new model.
- [ ] (Optional) Release notes update. If your change is user facing, please update the release notes draft in the [Google Doc](https://docs.google.com/document/d/1YyVqrgX4gHTtrstbq8oWUImOyPCKSGnJ7xtTpmXzlRs/edit?tab=t.0).
</details>
**BEFORE SUBMITTING, PLEASE READ <https://docs.vllm.ai/en/latest/contributing>** (anything written below this line will be removed by GitHub Actions)

77
.github/mergify.yml vendored
View File

@ -2,7 +2,6 @@ pull_request_rules:
- name: label-documentation
description: Automatically apply documentation label
conditions:
- label != stale
- or:
- files~=^[^/]+\.md$
- files~=^docs/
@ -11,13 +10,10 @@ pull_request_rules:
label:
add:
- documentation
comment:
message: "Documentation preview: https://vllm--{{number}}.org.readthedocs.build/en/{{number}}/"
- name: label-ci-build
description: Automatically apply ci/build label
conditions:
- label != stale
- or:
- files~=^\.github/
- files~=\.buildkite/
@ -34,7 +30,6 @@ pull_request_rules:
- name: label-deepseek
description: Automatically apply deepseek label
conditions:
- label != stale
- or:
- files~=^examples/.*deepseek.*\.py
- files~=^tests/.*deepseek.*\.py
@ -51,7 +46,6 @@ pull_request_rules:
- name: label-frontend
description: Automatically apply frontend label
conditions:
- label != stale
- files~=^vllm/entrypoints/
actions:
label:
@ -61,7 +55,6 @@ pull_request_rules:
- name: label-llama
description: Automatically apply llama label
conditions:
- label != stale
- or:
- files~=^examples/.*llama.*\.py
- files~=^tests/.*llama.*\.py
@ -77,7 +70,6 @@ pull_request_rules:
- name: label-multi-modality
description: Automatically apply multi-modality label
conditions:
- label != stale
- or:
- files~=^vllm/multimodal/
- files~=^tests/multimodal/
@ -91,7 +83,6 @@ pull_request_rules:
- name: label-new-model
description: Automatically apply new-model label
conditions:
- label != stale
- and:
- files~=^vllm/model_executor/models/
- files=vllm/model_executor/models/registry.py
@ -103,12 +94,11 @@ pull_request_rules:
- name: label-performance
description: Automatically apply performance label
conditions:
- label != stale
- or:
- files~=^benchmarks/
- files~=^vllm/benchmarks/
- files~=^tests/benchmarks/
- files~=^\.buildkite/performance-benchmarks/
- files~=^\.buildkite/nightly-benchmarks/
actions:
label:
add:
@ -117,7 +107,6 @@ pull_request_rules:
- name: label-qwen
description: Automatically apply qwen label
conditions:
- label != stale
- or:
- files~=^examples/.*qwen.*\.py
- files~=^tests/.*qwen.*\.py
@ -132,20 +121,12 @@ pull_request_rules:
- name: label-gpt-oss
description: Automatically apply gpt-oss label
conditions:
- label != stale
- or:
- files~=^examples/.*gpt[-_]?oss.*\.py
- files~=^tests/.*gpt[-_]?oss.*\.py
- files~=^tests/entrypoints/openai/test_response_api_with_harmony.py
- files~=^tests/entrypoints/test_context.py
- files~=^vllm/model_executor/models/.*gpt[-_]?oss.*\.py
- files~=^vllm/model_executor/layers/.*gpt[-_]?oss.*\.py
- files~=^vllm/entrypoints/harmony_utils.py
- files~=^vllm/entrypoints/tool_server.py
- files~=^vllm/entrypoints/tool.py
- files~=^vllm/entrypoints/context.py
- title~=(?i)gpt[-_]?oss
- title~=(?i)harmony
actions:
label:
add:
@ -154,7 +135,6 @@ pull_request_rules:
- name: label-rocm
description: Automatically apply rocm label
conditions:
- label != stale
- or:
- files~=^csrc/rocm/
- files~=^docker/Dockerfile.rocm
@ -175,7 +155,6 @@ pull_request_rules:
- name: label-structured-output
description: Automatically apply structured-output label
conditions:
- label != stale
- or:
- files~=^benchmarks/structured_schemas/
- files=benchmarks/benchmark_serving_structured_output.py
@ -185,7 +164,7 @@ pull_request_rules:
- files=examples/online_serving/openai_chat_completion_structured_outputs.py
- files=examples/online_serving/openai_chat_completion_structured_outputs_with_reasoning.py
- files~=^tests/v1/structured_output/
- files=tests/v1/entrypoints/llm/test_struct_output_generate.py
- files=tests/v1/entrypoints/llm/test_guided_generate.py
- files~=^vllm/v1/structured_output/
actions:
label:
@ -195,7 +174,6 @@ pull_request_rules:
- name: label-speculative-decoding
description: Automatically apply speculative-decoding label
conditions:
- label != stale
- or:
- files~=^vllm/v1/spec_decode/
- files~=^tests/v1/spec_decode/
@ -211,7 +189,6 @@ pull_request_rules:
- name: label-v1
description: Automatically apply v1 label
conditions:
- label != stale
- or:
- files~=^vllm/v1/
- files~=^tests/v1/
@ -224,7 +201,6 @@ pull_request_rules:
description: Automatically apply tpu label
# Keep this list in sync with `label-tpu-remove` conditions
conditions:
- label != stale
- or:
- files~=tpu.py
- files~=_tpu
@ -240,7 +216,6 @@ pull_request_rules:
description: Automatically remove tpu label
# Keep this list in sync with `label-tpu` conditions
conditions:
- label != stale
- and:
- -files~=tpu.py
- -files~=_tpu
@ -255,9 +230,9 @@ pull_request_rules:
- name: label-tool-calling
description: Automatically add tool-calling label
conditions:
- label != stale
- or:
- files~=^tests/tool_use/
- files~=^tests/mistral_tool_use/
- files~=^tests/entrypoints/openai/tool_parsers/
- files=tests/entrypoints/openai/test_chat_with_tool_reasoning.py
- files~=^vllm/entrypoints/openai/tool_parsers/
@ -274,9 +249,8 @@ pull_request_rules:
- name: ping author on conflicts and add 'needs-rebase' label
conditions:
- label != stale
- conflict
- -closed
- conflict
- -closed
actions:
label:
add:
@ -290,55 +264,20 @@ pull_request_rules:
- name: assign reviewer for tensorizer changes
conditions:
- label != stale
- or:
- files~=^vllm/model_executor/model_loader/tensorizer.py
- files~=^vllm/model_executor/model_loader/tensorizer_loader.py
- files~=^tests/entrypoints/openai/test_tensorizer_entrypoint.py
- files~=^tests/model_executor/model_loader/tensorizer_loader/
- files~=^tests/tensorizer_loader/
actions:
assign:
users:
- "sangstar"
- name: assign reviewer for modelopt changes
conditions:
- label != stale
- or:
- files~=^vllm/model_executor/layers/quantization/modelopt\.py$
- files~=^vllm/model_executor/layers/quantization/__init__\.py$
- files~=^tests/models/quantization/test_modelopt\.py$
- files~=^tests/quantization/test_modelopt\.py$
- files~=^tests/models/quantization/test_nvfp4\.py$
- files~=^docs/features/quantization/modelopt\.md$
actions:
assign:
users:
- "Edwardf0t1"
- name: remove 'needs-rebase' label when conflict is resolved
conditions:
- -conflict
- -closed
- -conflict
- -closed
actions:
label:
remove:
- needs-rebase
- name: label-kv-connector
description: Automatically apply kv-connector label
conditions:
- label != stale
- or:
- files~=^examples/online_serving/disaggregated[^/]*/.*
- files~=^examples/offline_inference/disaggregated[^/]*/.*
- files~=^examples/others/lmcache/
- files~=^tests/v1/kv_connector/
- files~=^vllm/distributed/kv_transfer/
- title~=(?i)\bP/?D\b
- title~=(?i)NIXL
- title~=(?i)LMCache
actions:
label:
add:
- kv-connector

View File

@ -1,21 +0,0 @@
# scale-config.yml:
# Powers what instance types are available for GHA auto-scaled
# runners. Runners listed here will be available as self hosted
# runners, configuration is directly pulled from the main branch.
# runner_types:
# runner_label:
# instance_type: m4.large
# os: linux
# # min_available defaults to the global cfg in the ALI Terraform
# min_available: undefined
# # when max_available value is not defined, no max runners is enforced
# max_available: undefined
# disk_size: 50
# is_ephemeral: true
runner_types:
linux.2xlarge:
disk_size: 150
instance_type: c5.2xlarge
is_ephemeral: true
os: linux

View File

@ -10,7 +10,7 @@ jobs:
runs-on: ubuntu-latest
steps:
- name: Add label
uses: actions/github-script@ed597411d8f924073f98dfc5c65a23a2325f34cd # v8.0.0
uses: actions/github-script@60a0d83039c74a4aee543508d2ffcb1c3799cdea # v7.0.1
with:
script: |
github.rest.issues.addLabels({

View File

@ -1,29 +0,0 @@
name: BC Lint
on:
pull_request:
types:
- opened
- synchronize
- reopened
- labeled
- unlabeled
jobs:
bc_lint:
if: github.repository_owner == 'vllm-project'
runs-on: ubuntu-latest
steps:
- name: Run BC Lint Action
uses: pytorch/test-infra/.github/actions/bc-lint@main
with:
repo: ${{ github.event.pull_request.head.repo.full_name }}
base_sha: ${{ github.event.pull_request.base.sha }}
head_sha: ${{ github.event.pull_request.head.sha }}
suppression: ${{ contains(github.event.pull_request.labels.*.name, 'suppress-bc-linter') }}
docs_link: 'https://github.com/pytorch/test-infra/wiki/BC-Linter'
config_dir: .github
concurrency:
group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.sha }}
cancel-in-progress: true

View File

@ -16,7 +16,7 @@ jobs:
uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # v4.2.2
- name: Set up Python
uses: actions/setup-python@e797f83bcb11b83ae66e0230d6156d7c80228e7c # v6.0.0
uses: actions/setup-python@42375524e23c412d93fb67b49958b491fce71c38 # v5.4.0
with:
python-version: '3.12'

View File

@ -1,361 +0,0 @@
name: Label issues based on keywords
on:
issues:
types: [opened, edited, reopened]
permissions:
issues: write # needed so the workflow can add labels
contents: read
concurrency:
group: issue-labeler-${{ github.event.issue.number }}
cancel-in-progress: true
jobs:
add-labels:
runs-on: ubuntu-latest
steps:
- name: Label issues based on keywords
id: label-step
uses: actions/github-script@ed597411d8f924073f98dfc5c65a23a2325f34cd # v8.0.0
with:
script: |
// Configuration: Add new labels and keywords here
const labelConfig = {
rocm: {
// Keyword search - matches whole words only (with word boundaries)
keywords: [
{
term: "composable kernel",
searchIn: "both"
},
{
term: "rccl",
searchIn: "body" // only search in body
},
{
term: "migraphx",
searchIn: "title" // only search in title
},
{
term: "hipgraph",
searchIn: "both"
},
{
term: "ROCm System Management Interface",
searchIn: "body"
},
],
// Substring search - matches anywhere in text (partial matches)
substrings: [
{
term: "VLLM_ROCM_",
searchIn: "both"
},
{
term: "aiter",
searchIn: "title"
},
{
term: "rocm",
searchIn: "title"
},
{
term: "amd",
searchIn: "title"
},
{
term: "hip-",
searchIn: "both"
},
{
term: "gfx",
searchIn: "both"
},
{
term: "cdna",
searchIn: "both"
},
{
term: "rdna",
searchIn: "both"
},
{
term: "torch_hip",
searchIn: "body" // only in body
},
{
term: "_hip",
searchIn: "both"
},
{
term: "hip_",
searchIn: "both"
},
// ROCm tools and libraries
{
term: "hipify",
searchIn: "both"
},
],
// Regex patterns - for complex pattern matching
regexPatterns: [
{
pattern: "\\bmi\\d{3}[a-z]*\\b",
description: "AMD GPU names (mi + 3 digits + optional letters)",
flags: "gi",
searchIn: "both" // "title", "body", or "both"
}
],
},
// Add more label configurations here as needed
// example: {
// keywords: [...],
// substrings: [...],
// regexPatterns: [...]
// },
};
// Helper function to create regex based on search type
function createSearchRegex(term, type) {
// Escape special regex characters in the term
const escapedTerm = term.replace(/[.*+?^${}()|[\]\\]/g, '\\$&');
switch (type) {
case 'keyword':
// Word boundary search - matches whole words only
return new RegExp(`\\b${escapedTerm}\\b`, "gi");
case 'substring':
// Substring search - matches anywhere in the text
return new RegExp(escapedTerm, "gi");
default:
throw new Error(`Unknown search type: ${type}`);
}
}
// Helper function to find matching terms in text with line information
function findMatchingTermsWithLines(text, searchTerms = [], searchType = 'keyword', searchLocation = '') {
const matches = [];
const lines = text.split('\n');
for (const termConfig of searchTerms) {
let regex;
let term, searchIn, pattern, description, flags;
// Handle different input formats (string or object)
if (typeof termConfig === 'string') {
term = termConfig;
searchIn = 'both'; // default
} else {
term = termConfig.term;
searchIn = termConfig.searchIn || 'both';
pattern = termConfig.pattern;
description = termConfig.description;
flags = termConfig.flags;
}
// Skip if this term shouldn't be searched in the current location
if (searchIn !== 'both' && searchIn !== searchLocation) {
continue;
}
// Create appropriate regex
if (searchType === 'regex') {
regex = new RegExp(pattern, flags || "gi");
} else {
regex = createSearchRegex(term, searchType);
}
const termMatches = [];
// Check each line for matches
lines.forEach((line, lineIndex) => {
const lineMatches = line.match(regex);
if (lineMatches) {
lineMatches.forEach(match => {
termMatches.push({
match: match,
lineNumber: lineIndex + 1,
lineContent: line.trim(),
searchType: searchType,
searchLocation: searchLocation,
originalTerm: term || pattern,
description: description,
// Show context around the match in the line
context: line.length > 100 ?
line.substring(Math.max(0, line.toLowerCase().indexOf(match.toLowerCase()) - 30),
line.toLowerCase().indexOf(match.toLowerCase()) + match.length + 30) + '...'
: line.trim()
});
});
}
});
if (termMatches.length > 0) {
matches.push({
term: term || (description || pattern),
searchType: searchType,
searchLocation: searchLocation,
searchIn: searchIn,
pattern: pattern,
matches: termMatches,
count: termMatches.length
});
}
}
return matches;
}
// Helper function to check if label should be added
async function processLabel(labelName, config) {
const body = context.payload.issue.body || "";
const title = context.payload.issue.title || "";
core.notice(`Processing label: ${labelName}`);
core.notice(`Issue Title: "${title}"`);
core.notice(`Issue Body length: ${body.length} characters`);
let shouldAddLabel = false;
let allMatches = [];
let reason = '';
const keywords = config.keywords || [];
const substrings = config.substrings || [];
const regexPatterns = config.regexPatterns || [];
core.notice(`Searching with ${keywords.length} keywords, ${substrings.length} substrings, and ${regexPatterns.length} regex patterns`);
// Search in title
if (title.trim()) {
core.notice(`Searching in title: "${title}"`);
const titleKeywordMatches = findMatchingTermsWithLines(title, keywords, 'keyword', 'title');
const titleSubstringMatches = findMatchingTermsWithLines(title, substrings, 'substring', 'title');
const titleRegexMatches = findMatchingTermsWithLines(title, regexPatterns, 'regex', 'title');
allMatches.push(...titleKeywordMatches, ...titleSubstringMatches, ...titleRegexMatches);
}
// Search in body
if (body.trim()) {
core.notice(`Searching in body (${body.length} characters)`);
const bodyKeywordMatches = findMatchingTermsWithLines(body, keywords, 'keyword', 'body');
const bodySubstringMatches = findMatchingTermsWithLines(body, substrings, 'substring', 'body');
const bodyRegexMatches = findMatchingTermsWithLines(body, regexPatterns, 'regex', 'body');
allMatches.push(...bodyKeywordMatches, ...bodySubstringMatches, ...bodyRegexMatches);
}
if (allMatches.length > 0) {
core.notice(`Found ${allMatches.length} matching term(s):`);
for (const termMatch of allMatches) {
const locationText = termMatch.searchLocation === 'title' ? 'title' : 'body';
const searchInText = termMatch.searchIn === 'both' ? 'both' : termMatch.searchIn;
if (termMatch.searchType === 'regex') {
core.notice(` 📍 Regex: "${termMatch.term}" (pattern: ${termMatch.pattern}) found ${termMatch.count} time(s) in ${locationText} (configured to search in: ${searchInText}):`);
} else {
core.notice(` 📍 Term: "${termMatch.term}" (${termMatch.searchType} search) found ${termMatch.count} time(s) in ${locationText} (configured to search in: ${searchInText}):`);
}
// Show details for each match
termMatch.matches.forEach((match, index) => {
core.notice(` ${index + 1}. Line ${match.lineNumber} in ${match.searchLocation}: "${match.match}" [${match.searchType}]`);
if (match.description) {
core.notice(` Description: ${match.description}`);
}
core.notice(` Context: ${match.context}`);
if (match.lineContent !== match.context) {
core.notice(` Full line: ${match.lineContent}`);
}
});
}
shouldAddLabel = true;
const totalMatches = allMatches.reduce((sum, t) => sum + t.count, 0);
const titleMatches = allMatches.filter(t => t.searchLocation === 'title').reduce((sum, t) => sum + t.count, 0);
const bodyMatches = allMatches.filter(t => t.searchLocation === 'body').reduce((sum, t) => sum + t.count, 0);
const keywordMatches = allMatches.filter(t => t.searchType === 'keyword').reduce((sum, t) => sum + t.count, 0);
const substringMatches = allMatches.filter(t => t.searchType === 'substring').reduce((sum, t) => sum + t.count, 0);
const regexMatches = allMatches.filter(t => t.searchType === 'regex').reduce((sum, t) => sum + t.count, 0);
reason = `Found ${totalMatches} total matches (${titleMatches} in title, ${bodyMatches} in body) - ${keywordMatches} keyword matches, ${substringMatches} substring matches, ${regexMatches} regex matches`;
}
core.notice(`Final decision: ${shouldAddLabel ? 'ADD LABEL' : 'DO NOT ADD LABEL'}`);
core.notice(`Reason: ${reason || 'No matching terms found'}`);
if (shouldAddLabel) {
const existingLabels = context.payload.issue.labels.map(l => l.name);
if (!existingLabels.includes(labelName)) {
await github.rest.issues.addLabels({
owner: context.repo.owner,
repo: context.repo.repo,
issue_number: context.issue.number,
labels: [labelName],
});
core.notice(`Label "${labelName}" added. ${reason}`);
return true;
}
core.notice(`Label "${labelName}" already present.`);
return false;
}
core.notice(`No matching terms found for label "${labelName}".`);
return false;
}
// Process all configured labels
const labelsAddedResults = await Promise.all(
Object.entries(labelConfig).map(([labelName, config]) =>
processLabel(labelName, config).then(added => ({ labelName, added }))
)
);
const numLabelsAdded = labelsAddedResults.filter(r => r.added).length;
core.notice(`Processing complete. ${numLabelsAdded} label(s) added.`);
// Return which labels were added for the next step
const addedLabels = labelsAddedResults.filter(r => r.added).map(r => r.labelName);
core.setOutput('labels_added', JSON.stringify(addedLabels));
return addedLabels;
- name: CC users for labeled issues
if: steps.label-step.outputs.labels_added != '[]'
uses: actions/github-script@ed597411d8f924073f98dfc5c65a23a2325f34cd # v8.0.0
with:
script: |
// Configuration: Map labels to GitHub users to CC
// You can add multiple users per label, and multiple label configurations
const ccConfig = {
rocm: {
users: ['hongxiayang', 'tjtanaa', 'vllmellm'], // Add more users as needed: ['user1', 'user2', 'user3']
message: 'CC {users} for ROCm-related issue' // {users} will be replaced with @mentions
},
// Add more label -> user mappings here
// Example:
// cuda: {
// users: ['user1', 'user2'],
// message: 'CC {users} for CUDA-related issue'
// },
// performance: {
// users: ['perfexpert'],
// message: 'CC {users} for performance issue'
// },
};
const labelsAdded = JSON.parse('${{ steps.label-step.outputs.labels_added }}');
core.notice(`Labels added: ${labelsAdded.join(', ')}`);
// Get existing comments to check for already mentioned users
const comments = await github.rest.issues.listComments({
owner: context.repo.owner,
repo: context.repo.repo,
issue_number: context.issue.number,
});
const issueBody = context.payload.issue.body || '';
const allExistingText = issueBody + '\n' + comments.data.map(c => c.body).join('\n');
// Process each label that was added
for (const label of labelsAdded) {
if (ccConfig[label]) {
const config = ccConfig[label];
const usersToMention = [];
// Check which users haven't been mentioned yet
for (const user of config.users) {
const mentionPattern = new RegExp(`@${user}\\b`, 'i');
if (!mentionPattern.test(allExistingText)) {
usersToMention.push(user);
} else {
core.notice(`@${user} already mentioned for label "${label}", skipping`);
}
}
// Post comment if there are users to mention
if (usersToMention.length > 0) {
const mentions = usersToMention.map(u => `@${u}`).join(' ');
const message = config.message.replace('{users}', mentions);
await github.rest.issues.createComment({
owner: context.repo.owner,
repo: context.repo.repo,
issue_number: context.issue.number,
body: message
});
core.notice(`CC comment added for label "${label}": ${mentions}`);
} else {
core.notice(`All users for label "${label}" already mentioned, skipping comment`);
}
}
}

89
.github/workflows/lint-and-deploy.yaml vendored Normal file
View File

@ -0,0 +1,89 @@
name: Lint and Deploy Charts
on: pull_request
concurrency:
group: ${{ github.workflow }}-${{ github.ref }}
cancel-in-progress: true
permissions:
contents: read
jobs:
lint-and-deploy:
runs-on: ubuntu-latest
steps:
- name: Checkout
uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # v4.2.2
with:
fetch-depth: 0
- name: Set up Helm
uses: azure/setup-helm@b9e51907a09c216f16ebe8536097933489208112 # v4.3.0
with:
version: v3.14.4
#Python is required because ct lint runs Yamale and yamllint which require Python.
- uses: actions/setup-python@42375524e23c412d93fb67b49958b491fce71c38 # v5.4.0
with:
python-version: '3.13'
- name: Set up chart-testing
uses: helm/chart-testing-action@0d28d3144d3a25ea2cc349d6e59901c4ff469b3b # v2.7.0
with:
version: v3.10.1
- name: Run chart-testing (lint)
run: ct lint --target-branch ${{ github.event.repository.default_branch }} --chart-dirs examples/online_serving/chart-helm --charts examples/online_serving/chart-helm
- name: Setup minio
run: |
docker network create vllm-net
docker run -d -p 9000:9000 --name minio --net vllm-net \
-e "MINIO_ACCESS_KEY=minioadmin" \
-e "MINIO_SECRET_KEY=minioadmin" \
-v /tmp/data:/data \
-v /tmp/config:/root/.minio \
minio/minio server /data
export AWS_ACCESS_KEY_ID=minioadmin
export AWS_SECRET_ACCESS_KEY=minioadmin
export AWS_EC2_METADATA_DISABLED=true
mkdir opt-125m
cd opt-125m && curl -O -Ls "https://huggingface.co/facebook/opt-125m/resolve/main/{pytorch_model.bin,config.json,generation_config.json,merges.txt,special_tokens_map.json,tokenizer_config.json,vocab.json}" && cd ..
aws --endpoint-url http://127.0.0.1:9000/ s3 mb s3://testbucket
aws --endpoint-url http://127.0.0.1:9000/ s3 cp opt-125m/ s3://testbucket/opt-125m --recursive
- name: Create kind cluster
uses: helm/kind-action@a1b0e391336a6ee6713a0583f8c6240d70863de3 # v1.12.0
- name: Build the Docker image vllm cpu
run: docker buildx build -f docker/Dockerfile.cpu -t vllm-cpu-env .
- name: Configuration of docker images, network and namespace for the kind cluster
run: |
docker pull amazon/aws-cli:2.6.4
kind load docker-image amazon/aws-cli:2.6.4 --name chart-testing
kind load docker-image vllm-cpu-env:latest --name chart-testing
docker network connect vllm-net "$(docker ps -aqf "name=chart-testing-control-plane")"
kubectl create ns ns-vllm
- name: Run chart-testing (install)
run: |
export AWS_ACCESS_KEY_ID=minioadmin
export AWS_SECRET_ACCESS_KEY=minioadmin
sleep 30 && kubectl -n ns-vllm logs -f "$(kubectl -n ns-vllm get pods | awk '/deployment/ {print $1;exit}')" &
helm install --wait --wait-for-jobs --timeout 5m0s --debug --create-namespace --namespace=ns-vllm test-vllm examples/online_serving/chart-helm -f examples/online_serving/chart-helm/values.yaml --set secrets.s3endpoint=http://minio:9000 --set secrets.s3bucketname=testbucket --set secrets.s3accesskeyid=$AWS_ACCESS_KEY_ID --set secrets.s3accesskey=$AWS_SECRET_ACCESS_KEY --set resources.requests.cpu=1 --set resources.requests.memory=4Gi --set resources.limits.cpu=2 --set resources.limits.memory=5Gi --set image.env[0].name=VLLM_CPU_KVCACHE_SPACE --set image.env[1].name=VLLM_LOGGING_LEVEL --set image.env[2].name=VLLM_CPU_CI_ENV --set-string image.env[0].value="1" --set-string image.env[1].value="DEBUG" --set-string image.env[2].value="1" --set-string extraInit.s3modelpath="opt-125m/" --set-string 'resources.limits.nvidia\.com/gpu=0' --set-string 'resources.requests.nvidia\.com/gpu=0' --set-string image.repository="vllm-cpu-env"
- name: curl test
run: |
kubectl -n ns-vllm port-forward service/test-vllm-service 8001:80 &
sleep 10
CODE="$(curl -v -f --location http://localhost:8001/v1/completions \
--header "Content-Type: application/json" \
--data '{
"model": "opt-125m",
"prompt": "San Francisco is a",
"max_tokens": 7,
"temperature": 0
}'):$CODE"
echo "$CODE"

View File

@ -17,7 +17,7 @@ jobs:
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # v4.2.2
- uses: actions/setup-python@e797f83bcb11b83ae66e0230d6156d7c80228e7c # v6.0.0
- uses: actions/setup-python@42375524e23c412d93fb67b49958b491fce71c38 # v5.4.0
with:
python-version: "3.12"
- run: echo "::add-matcher::.github/workflows/matchers/actionlint.json"

111
.github/workflows/publish.yml vendored Normal file
View File

@ -0,0 +1,111 @@
# This workflow will upload a Python Package to Release asset
# For more information see: https://help.github.com/en/actions/language-and-framework-guides/using-python-with-github-actions
name: Create Release
on:
push:
tags:
- v*
# Needed to create release and upload assets
permissions:
contents: write
jobs:
release:
# Retrieve tag and create release
name: Create Release
runs-on: ubuntu-latest
outputs:
upload_url: ${{ steps.create_release.outputs.upload_url }}
steps:
- name: Checkout
uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # v4.2.2
- name: Extract branch info
shell: bash
run: |
echo "release_tag=${GITHUB_REF#refs/*/}" >> "$GITHUB_ENV"
- name: Create Release
id: create_release
uses: actions/github-script@60a0d83039c74a4aee543508d2ffcb1c3799cdea # v7.0.1
env:
RELEASE_TAG: ${{ env.release_tag }}
with:
github-token: "${{ secrets.GITHUB_TOKEN }}"
script: |
const script = require('.github/workflows/scripts/create_release.js')
await script(github, context, core)
# NOTE(simon): No longer build wheel using GitHub Actions. See buildkite's release workflow.
# wheel:
# name: Build Wheel
# runs-on: ${{ matrix.os }}
# needs: release
# strategy:
# fail-fast: false
# matrix:
# os: ['ubuntu-20.04']
# python-version: ['3.9', '3.10', '3.11', '3.12']
# pytorch-version: ['2.4.0'] # Must be the most recent version that meets requirements/cuda.txt.
# cuda-version: ['11.8', '12.1']
# steps:
# - name: Checkout
# uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # v4.2.2
# - name: Setup ccache
# uses: hendrikmuhs/ccache-action@ed74d11c0b343532753ecead8a951bb09bb34bc9 # v1.2.14
# with:
# create-symlink: true
# key: ${{ github.job }}-${{ matrix.python-version }}-${{ matrix.cuda-version }}
# - name: Set up Linux Env
# if: ${{ runner.os == 'Linux' }}
# run: |
# bash -x .github/workflows/scripts/env.sh
# - name: Set up Python
# uses: actions/setup-python@0b93645e9fea7318ecaed2b359559ac225c90a2b # v5.3.0
# with:
# python-version: ${{ matrix.python-version }}
# - name: Install CUDA ${{ matrix.cuda-version }}
# run: |
# bash -x .github/workflows/scripts/cuda-install.sh ${{ matrix.cuda-version }} ${{ matrix.os }}
# - name: Install PyTorch ${{ matrix.pytorch-version }} with CUDA ${{ matrix.cuda-version }}
# run: |
# bash -x .github/workflows/scripts/pytorch-install.sh ${{ matrix.python-version }} ${{ matrix.pytorch-version }} ${{ matrix.cuda-version }}
# - name: Build wheel
# shell: bash
# env:
# CMAKE_BUILD_TYPE: Release # do not compile with debug symbol to reduce wheel size
# run: |
# bash -x .github/workflows/scripts/build.sh ${{ matrix.python-version }} ${{ matrix.cuda-version }}
# wheel_name=$(find dist -name "*whl" -print0 | xargs -0 -n 1 basename)
# asset_name=${wheel_name//"linux"/"manylinux1"}
# echo "wheel_name=${wheel_name}" >> "$GITHUB_ENV"
# echo "asset_name=${asset_name}" >> "$GITHUB_ENV"
# - name: Upload Release Asset
# uses: actions/upload-release-asset@e8f9f06c4b078e705bd2ea027f0926603fc9b4d5 # v1.0.2
# env:
# GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
# with:
# upload_url: ${{ needs.release.outputs.upload_url }}
# asset_path: ./dist/${{ env.wheel_name }}
# asset_name: ${{ env.asset_name }}
# asset_content_type: application/*
# (Danielkinz): This last step will publish the .whl to pypi. Warning: untested
# - name: Publish package
# uses: pypa/gh-action-pypi-publish@release/v1.8
# with:
# repository-url: https://test.pypi.org/legacy/
# password: ${{ secrets.PYPI_API_TOKEN }}
# skip-existing: true

View File

@ -9,46 +9,19 @@ jobs:
runs-on: ubuntu-latest
steps:
- name: Remind to run full CI on PR
uses: actions/github-script@ed597411d8f924073f98dfc5c65a23a2325f34cd # v8.0.0
uses: actions/github-script@60a0d83039c74a4aee543508d2ffcb1c3799cdea # v7.0.1
with:
script: |
try {
// Get the PR author
const prAuthor = context.payload.pull_request.user.login;
// Check if this is the author's first PR in this repository
// Use GitHub's search API to find all PRs by this author
const { data: searchResults } = await github.rest.search.issuesAndPullRequests({
q: `repo:${context.repo.owner}/${context.repo.repo} type:pr author:${prAuthor}`,
per_page: 100
});
const authorPRCount = searchResults.total_count;
console.log(`Found ${authorPRCount} PRs by ${prAuthor}`);
// Only post comment if this is the first PR (only one PR by this author)
if (authorPRCount === 1) {
console.log(`Posting welcome comment for first-time contributor: ${prAuthor}`);
await github.rest.issues.createComment({
owner: context.repo.owner,
repo: context.repo.repo,
issue_number: context.issue.number,
body: '👋 Hi! Thank you for contributing to the vLLM project.\n\n' +
'💬 Join our developer Slack at https://slack.vllm.ai to discuss your PR in #pr-reviews, coordinate on features in #feat- channels, or join special interest groups in #sig- channels.\n\n' +
'Just a reminder: PRs would not trigger full CI run by default. Instead, it would only run `fastcheck` CI which starts running only a small and essential subset of CI tests to quickly catch errors. \n\n' +
'You ask your reviewers to trigger select CI tests on top of `fastcheck` CI. \n\n' +
'Once the PR is approved and ready to go, your PR reviewer(s) can run CI to test the changes comprehensively before merging.\n\n' +
'To run CI, PR reviewers can either: Add `ready` label to the PR or enable auto-merge.\n\n' +
'If you have any questions, please reach out to us on Slack at https://slack.vllm.ai.\n\n' +
'🚀'
});
} else {
console.log(`Skipping comment for ${prAuthor} - not their first PR (${authorPRCount} PRs found)`);
}
} catch (error) {
console.error('Error checking PR history or posting comment:', error);
// Don't fail the workflow, just log the error
}
github.rest.issues.createComment({
owner: context.repo.owner,
repo: context.repo.repo,
issue_number: context.issue.number,
body: '👋 Hi! Thank you for contributing to the vLLM project.\n\n' +
'💬 Join our developer Slack at https://slack.vllm.ai to discuss your PR in #pr-reviews, coordinate on features in #feat- channels, or join special interest groups in #sig- channels.\n\n' +
'Just a reminder: PRs would not trigger full CI run by default. Instead, it would only run `fastcheck` CI which starts running only a small and essential subset of CI tests to quickly catch errors. You can run other CI tests on top of those by going to your `fastcheck` build on Buildkite UI (linked in the PR checks section) and unblock them. If you do not have permission to unblock, ping `simon-mo` or `khluu` to add you in our Buildkite org.\n\n' +
'Once the PR is approved and ready to go, your PR reviewer(s) can run CI to test the changes comprehensively before merging.\n\n' +
'To run CI, PR reviewers can either: Add `ready` label to the PR or enable auto-merge.\n\n' +
'🚀'
})
env:
GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}

View File

@ -13,7 +13,7 @@ jobs:
actions: write
runs-on: ubuntu-latest
steps:
- uses: actions/stale@5f858e3efba33a5ca4407a664cc011ad407f2008 # v10.1.0
- uses: actions/stale@5bef64f19d7facfb25b37b414482c7164d639639 # v9.1.0
with:
# Increasing this value ensures that changes to this workflow
# propagate to all issues and PRs in days rather than months

19
.gitignore vendored
View File

@ -4,7 +4,7 @@
# vllm-flash-attn built from source
vllm/vllm_flash_attn/*
# triton jit
# triton jit
.triton
# Byte-compiled / optimized / DLL files
@ -94,9 +94,6 @@ ipython_config.py
# generated files
**/generated/**
# uv
uv.lock
# pyenv
# For a library or package, you might want to ignore these files since the code is
# intended to run in multiple environments; otherwise, check them in:
@ -180,14 +177,6 @@ cython_debug/
# VSCode
.vscode/
# Claude
CLAUDE.md
.claude/
# Codex
AGENTS.md
.codex/
# DS Store
.DS_Store
@ -218,9 +207,3 @@ shellcheck*/
# Ignore moe/marlin_moe gen code
csrc/moe/marlin_moe_wna16/kernel_*
# Ignore ep_kernels_workspace folder
ep_kernels_workspace/
# Allow tracked library source folders under submodules (e.g., benchmarks/lib)
!vllm/benchmarks/lib/

View File

@ -4,6 +4,7 @@ MD013: false
MD024:
siblings_only: true
MD033: false
MD042: false
MD045: false
MD046: false
MD051: false

View File

@ -6,19 +6,30 @@ default_stages:
- manual # Run in CI
exclude: 'vllm/third_party/.*'
repos:
- repo: https://github.com/astral-sh/ruff-pre-commit
rev: v0.14.0
- repo: https://github.com/google/yapf
rev: v0.43.0
hooks:
- id: ruff-check
- id: yapf
args: [--in-place, --verbose]
# Keep the same list from yapfignore here to avoid yapf failing without any inputs
exclude: '(.buildkite|benchmarks|build|examples)/.*'
- repo: https://github.com/astral-sh/ruff-pre-commit
rev: v0.11.7
hooks:
- id: ruff
args: [--output-format, github, --fix]
- id: ruff-format
files: ^(.buildkite|benchmarks|examples)/.*
- repo: https://github.com/crate-ci/typos
rev: v1.38.1
rev: v1.34.0
hooks:
- id: typos
args: [--force-exclude]
- repo: https://github.com/PyCQA/isort
rev: 6.0.1
hooks:
- id: isort
- repo: https://github.com/pre-commit/mirrors-clang-format
rev: v21.1.2
rev: v20.1.3
hooks:
- id: clang-format
exclude: 'csrc/(moe/topk_softmax_kernels.cu|quantization/gguf/(ggml-common.h|dequantize.cuh|vecdotq.cuh|mmq.cuh|mmvq.cuh))|vllm/third_party/.*'
@ -35,55 +46,61 @@ repos:
hooks:
- id: actionlint
- repo: https://github.com/astral-sh/uv-pre-commit
rev: 0.9.1
rev: 0.6.17
hooks:
- id: pip-compile
args: [requirements/test.in, -o, requirements/test.txt, --index-strategy, unsafe-best-match, --torch-backend, cu129, --python-platform, x86_64-manylinux_2_28]
args: [requirements/test.in, -o, requirements/test.txt, --index-strategy, unsafe-best-match, --torch-backend, cu128]
files: ^requirements/test\.(in|txt)$
- repo: local
hooks:
- id: format-torch-nightly-test
name: reformat nightly_torch_test.txt to be in sync with test.in
language: python
entry: python tools/pre_commit/generate_nightly_torch_test.py
entry: python tools/generate_nightly_torch_test.py
files: ^requirements/test\.(in|txt)$
- id: mypy-local
name: Run mypy locally for lowest supported Python version
entry: python tools/pre_commit/mypy.py 0 "3.10"
name: Run mypy for local Python installation
entry: tools/mypy.sh 0 "local"
language: python
types: [python]
additional_dependencies: &mypy_deps [mypy==1.11.1, types-cachetools, types-setuptools, types-PyYAML, types-requests, pydantic]
stages: [pre-commit] # Don't run in CI
<<: &mypy_common
language: python
types_or: [python, pyi]
require_serial: true
additional_dependencies: [mypy==1.11.1, regex, types-cachetools, types-setuptools, types-PyYAML, types-requests, types-torch, pydantic]
- id: mypy-3.9 # TODO: Use https://github.com/pre-commit/mirrors-mypy when mypy setup is less awkward
name: Run mypy for Python 3.9
entry: tools/mypy.sh 1 "3.9"
language: python
types: [python]
additional_dependencies: *mypy_deps
stages: [manual] # Only run in CI
- id: mypy-3.10 # TODO: Use https://github.com/pre-commit/mirrors-mypy when mypy setup is less awkward
name: Run mypy for Python 3.10
entry: python tools/pre_commit/mypy.py 1 "3.10"
<<: *mypy_common
entry: tools/mypy.sh 1 "3.10"
language: python
types: [python]
additional_dependencies: *mypy_deps
stages: [manual] # Only run in CI
- id: mypy-3.11 # TODO: Use https://github.com/pre-commit/mirrors-mypy when mypy setup is less awkward
name: Run mypy for Python 3.11
entry: python tools/pre_commit/mypy.py 1 "3.11"
<<: *mypy_common
entry: tools/mypy.sh 1 "3.11"
language: python
types: [python]
additional_dependencies: *mypy_deps
stages: [manual] # Only run in CI
- id: mypy-3.12 # TODO: Use https://github.com/pre-commit/mirrors-mypy when mypy setup is less awkward
name: Run mypy for Python 3.12
entry: python tools/pre_commit/mypy.py 1 "3.12"
<<: *mypy_common
stages: [manual] # Only run in CI
- id: mypy-3.13 # TODO: Use https://github.com/pre-commit/mirrors-mypy when mypy setup is less awkward
name: Run mypy for Python 3.13
entry: python tools/pre_commit/mypy.py 1 "3.13"
<<: *mypy_common
entry: tools/mypy.sh 1 "3.12"
language: python
types: [python]
additional_dependencies: *mypy_deps
stages: [manual] # Only run in CI
- id: shellcheck
name: Lint shell scripts
entry: tools/pre_commit/shellcheck.sh
entry: tools/shellcheck.sh
language: script
types: [shell]
- id: png-lint
name: Lint PNG exports from excalidraw
entry: tools/pre_commit/png-lint.sh
entry: tools/png-lint.sh
language: script
types: [png]
- id: signoff-commit
@ -100,12 +117,12 @@ repos:
stages: [commit-msg]
- id: check-spdx-header
name: Check SPDX headers
entry: python tools/pre_commit/check_spdx_header.py
entry: python tools/check_spdx_header.py
language: python
types: [python]
- id: check-root-lazy-imports
name: Check root lazy imports
entry: python tools/pre_commit/check_init_lazy_imports.py
entry: python tools/check_init_lazy_imports.py
language: python
types: [python]
- id: check-filenames
@ -119,11 +136,11 @@ repos:
pass_filenames: false
- id: update-dockerfile-graph
name: Update Dockerfile dependency graph
entry: tools/pre_commit/update-dockerfile-graph.sh
entry: tools/update-dockerfile-graph.sh
language: script
- id: enforce-import-regex-instead-of-re
name: Enforce import regex as re
entry: python tools/pre_commit/enforce_regex_import.py
entry: python tools/enforce_regex_import.py
language: python
types: [python]
pass_filenames: false
@ -131,22 +148,25 @@ repos:
# forbid directly import triton
- id: forbid-direct-triton-import
name: "Forbid direct 'import triton'"
entry: python tools/pre_commit/check_triton_import.py
entry: python tools/check_triton_import.py
language: python
types: [python]
pass_filenames: false
additional_dependencies: [regex]
- id: check-pickle-imports
name: Prevent new pickle/cloudpickle imports
entry: python tools/pre_commit/check_pickle_imports.py
entry: python tools/check_pickle_imports.py
language: python
types: [python]
additional_dependencies: [regex]
pass_filenames: false
additional_dependencies: [pathspec, regex]
- id: validate-config
name: Validate configuration has default values and that each field has a docstring
entry: python tools/pre_commit/validate_config.py
entry: python tools/validate_config.py
language: python
additional_dependencies: [regex]
types: [python]
pass_filenames: true
files: vllm/config.py|tests/test_config.py|vllm/entrypoints/openai/cli_args.py
# Keep `suggestion` last
- id: suggestion
name: Suggestion

View File

@ -13,7 +13,6 @@ build:
mkdocs:
configuration: mkdocs.yaml
fail_on_warning: true
# Optionally declare the Python requirements required to build your docs
python:

View File

@ -1,2 +1 @@
collect_env.py
vllm/model_executor/layers/fla/ops/*.py

View File

@ -13,10 +13,6 @@ cmake_minimum_required(VERSION 3.26)
# cmake --install . --component _C
project(vllm_extensions LANGUAGES CXX)
set(CMAKE_CXX_STANDARD 17)
set(CMAKE_CXX_STANDARD_REQUIRED ON)
# CUDA by default, can be overridden by using -DVLLM_TARGET_DEVICE=... (used by setup.py)
set(VLLM_TARGET_DEVICE "cuda" CACHE STRING "Target device backend for vLLM")
message(STATUS "Build type: ${CMAKE_BUILD_TYPE}")
@ -34,10 +30,10 @@ install(CODE "set(CMAKE_INSTALL_LOCAL_ONLY TRUE)" ALL_COMPONENTS)
# Supported python versions. These versions will be searched in order, the
# first match will be selected. These should be kept in sync with setup.py.
#
set(PYTHON_SUPPORTED_VERSIONS "3.10" "3.11" "3.12" "3.13")
set(PYTHON_SUPPORTED_VERSIONS "3.9" "3.10" "3.11" "3.12")
# Supported AMD GPU architectures.
set(HIP_SUPPORTED_ARCHS "gfx906;gfx908;gfx90a;gfx942;gfx950;gfx1030;gfx1100;gfx1101;gfx1200;gfx1201;gfx1150;gfx1151")
set(HIP_SUPPORTED_ARCHS "gfx906;gfx908;gfx90a;gfx942;gfx950;gfx1030;gfx1100;gfx1101;gfx1200;gfx1201")
#
# Supported/expected torch versions for CUDA/ROCm.
@ -49,8 +45,8 @@ set(HIP_SUPPORTED_ARCHS "gfx906;gfx908;gfx90a;gfx942;gfx950;gfx1030;gfx1100;gfx1
# requirements.txt files and should be kept consistent. The ROCm torch
# versions are derived from docker/Dockerfile.rocm
#
set(TORCH_SUPPORTED_VERSION_CUDA "2.9.0")
set(TORCH_SUPPORTED_VERSION_ROCM "2.9.0")
set(TORCH_SUPPORTED_VERSION_CUDA "2.7.1")
set(TORCH_SUPPORTED_VERSION_ROCM "2.7.0")
#
# Try to find python package with an executable that exactly matches
@ -86,9 +82,6 @@ find_package(Torch REQUIRED)
# Supported NVIDIA architectures.
# This check must happen after find_package(Torch) because that's when CMAKE_CUDA_COMPILER_VERSION gets defined
if(DEFINED CMAKE_CUDA_COMPILER_VERSION AND
CMAKE_CUDA_COMPILER_VERSION VERSION_GREATER_EQUAL 13.0)
set(CUDA_SUPPORTED_ARCHS "7.5;8.0;8.6;8.7;8.9;9.0;10.0;11.0;12.0")
elseif(DEFINED CMAKE_CUDA_COMPILER_VERSION AND
CMAKE_CUDA_COMPILER_VERSION VERSION_GREATER_EQUAL 12.8)
set(CUDA_SUPPORTED_ARCHS "7.0;7.2;7.5;8.0;8.6;8.7;8.9;9.0;10.0;10.1;12.0")
else()
@ -178,25 +171,6 @@ if(NVCC_THREADS AND VLLM_GPU_LANG STREQUAL "CUDA")
list(APPEND VLLM_GPU_FLAGS "--threads=${NVCC_THREADS}")
endif()
#
# Set compression mode for CUDA >=13.x.
#
if(VLLM_GPU_LANG STREQUAL "CUDA" AND
DEFINED CMAKE_CUDA_COMPILER_VERSION AND
CMAKE_CUDA_COMPILER_VERSION VERSION_GREATER_EQUAL 13.0)
list(APPEND VLLM_GPU_FLAGS "--compress-mode=size")
endif()
#
# Set CUDA include flags for CXX compiler.
#
if(VLLM_GPU_LANG STREQUAL "CUDA")
set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -I${CUDA_TOOLKIT_ROOT_DIR}/include")
if(CUDA_VERSION VERSION_GREATER_EQUAL 13.0)
set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -I${CUDA_TOOLKIT_ROOT_DIR}/include/cccl")
endif()
endif()
#
# Use FetchContent for C++ dependencies that are compiled as part of vLLM's build process.
# setup.py will override FETCHCONTENT_BASE_DIR to play nicely with sccache.
@ -269,12 +243,13 @@ set(VLLM_EXT_SRC
"csrc/sampler.cu"
"csrc/cuda_view.cu"
"csrc/quantization/gptq/q_gemm.cu"
"csrc/quantization/w8a8/int8/scaled_quant.cu"
"csrc/quantization/w8a8/fp8/common.cu"
"csrc/quantization/compressed_tensors/int8_quant_kernels.cu"
"csrc/quantization/fp8/common.cu"
"csrc/quantization/fused_kernels/fused_layernorm_dynamic_per_token_quant.cu"
"csrc/quantization/gguf/gguf_kernel.cu"
"csrc/quantization/activation_kernels.cu"
"csrc/cuda_utils_kernels.cu"
"csrc/prepare_inputs/advance_step.cu"
"csrc/custom_all_reduce.cu"
"csrc/torch_bindings.cpp")
@ -282,7 +257,7 @@ if(VLLM_GPU_LANG STREQUAL "CUDA")
SET(CUTLASS_ENABLE_HEADERS_ONLY ON CACHE BOOL "Enable only the header library")
# Set CUTLASS_REVISION. Used for FetchContent. Also fixes some bogus messages when building.
set(CUTLASS_REVISION "v4.2.1" CACHE STRING "CUTLASS revision to use")
set(CUTLASS_REVISION "v4.0.0" CACHE STRING "CUTLASS revision to use")
# Use the specified CUTLASS source directory for compilation if VLLM_CUTLASS_SRC_DIR is provided
if (DEFINED ENV{VLLM_CUTLASS_SRC_DIR})
@ -312,15 +287,17 @@ if(VLLM_GPU_LANG STREQUAL "CUDA")
FetchContent_MakeAvailable(cutlass)
list(APPEND VLLM_EXT_SRC
"csrc/quantization/aqlm/gemm_kernels.cu"
"csrc/quantization/awq/gemm_kernels.cu"
"csrc/permute_cols.cu"
"csrc/quantization/w8a8/cutlass/scaled_mm_entry.cu"
"csrc/quantization/cutlass_w8a8/scaled_mm_entry.cu"
"csrc/quantization/fp4/nvfp4_quant_entry.cu"
"csrc/quantization/fp4/nvfp4_scaled_mm_entry.cu"
"csrc/quantization/fp4/nvfp4_blockwise_moe_kernel.cu"
"csrc/sparse/cutlass/sparse_scaled_mm_entry.cu"
"csrc/cutlass_extensions/common.cpp"
"csrc/quantization/w8a8/fp8/per_token_group_quant.cu"
"csrc/quantization/w8a8/int8/per_token_group_quant.cu")
"csrc/attention/mla/cutlass_mla_entry.cu"
"csrc/quantization/fp8/per_token_group_quant.cu")
set_gencode_flags_for_srcs(
SRCS "${VLLM_EXT_SRC}"
@ -374,27 +351,20 @@ if(VLLM_GPU_LANG STREQUAL "CUDA")
set_gencode_flags_for_srcs(
SRCS "${MARLIN_TEMPLATE_KERNEL_SRC}"
CUDA_ARCHS "${MARLIN_ARCHS}")
if(${CMAKE_CUDA_COMPILER_VERSION} VERSION_GREATER_EQUAL 12.8)
set_source_files_properties(${MARLIN_TEMPLATE_KERNEL_SRC}
PROPERTIES COMPILE_FLAGS "-static-global-template-stub=false")
endif()
list(APPEND VLLM_EXT_SRC ${MARLIN_TEMPLATE_KERNEL_SRC})
set(MARLIN_SRCS
"csrc/quantization/marlin/dense/marlin_cuda_kernel.cu"
"csrc/quantization/marlin/sparse/marlin_24_cuda_kernel.cu"
"csrc/quantization/marlin/qqq/marlin_qqq_gemm_kernel.cu"
"csrc/quantization/gptq_marlin/gptq_marlin.cu"
"csrc/quantization/gptq_marlin/gptq_marlin_repack.cu"
"csrc/quantization/gptq_marlin/awq_marlin_repack.cu")
set_gencode_flags_for_srcs(
SRCS "${MARLIN_SRCS}"
CUDA_ARCHS "${MARLIN_ARCHS}")
if(${CMAKE_CUDA_COMPILER_VERSION} VERSION_GREATER_EQUAL 12.8)
set_source_files_properties("csrc/quantization/gptq_marlin/gptq_marlin.cu"
PROPERTIES COMPILE_FLAGS "-static-global-template-stub=false")
endif()
list(APPEND VLLM_EXT_SRC "${MARLIN_SRCS}")
message(STATUS "Building Marlin kernels for archs: ${MARLIN_ARCHS}")
else()
message(STATUS "Not building Marlin kernels as no compatible archs found"
@ -424,11 +394,11 @@ if(VLLM_GPU_LANG STREQUAL "CUDA")
cuda_archs_loose_intersection(SCALED_MM_ARCHS "9.0a;" "${CUDA_ARCHS}")
if(${CMAKE_CUDA_COMPILER_VERSION} VERSION_GREATER_EQUAL 12.0 AND SCALED_MM_ARCHS)
set(SRCS
"csrc/quantization/w8a8/cutlass/scaled_mm_c3x_sm90.cu"
"csrc/quantization/w8a8/cutlass/c3x/scaled_mm_sm90_fp8.cu"
"csrc/quantization/w8a8/cutlass/c3x/scaled_mm_sm90_int8.cu"
"csrc/quantization/w8a8/cutlass/c3x/scaled_mm_azp_sm90_int8.cu"
"csrc/quantization/w8a8/cutlass/c3x/scaled_mm_blockwise_sm90_fp8.cu")
"csrc/quantization/cutlass_w8a8/scaled_mm_c3x_sm90.cu"
"csrc/quantization/cutlass_w8a8/c3x/scaled_mm_sm90_fp8.cu"
"csrc/quantization/cutlass_w8a8/c3x/scaled_mm_sm90_int8.cu"
"csrc/quantization/cutlass_w8a8/c3x/scaled_mm_azp_sm90_int8.cu"
"csrc/quantization/cutlass_w8a8/c3x/scaled_mm_blockwise_sm90_fp8.cu")
set_gencode_flags_for_srcs(
SRCS "${SRCS}"
CUDA_ARCHS "${SCALED_MM_ARCHS}")
@ -452,16 +422,12 @@ if(VLLM_GPU_LANG STREQUAL "CUDA")
# The cutlass_scaled_mm kernels for Geforce Blackwell SM120 (c3x, i.e. CUTLASS 3.x) require
# CUDA 12.8 or later
if(${CMAKE_CUDA_COMPILER_VERSION} VERSION_GREATER_EQUAL 13.0)
cuda_archs_loose_intersection(SCALED_MM_ARCHS "12.0f" "${CUDA_ARCHS}")
else()
cuda_archs_loose_intersection(SCALED_MM_ARCHS "12.0a" "${CUDA_ARCHS}")
endif()
cuda_archs_loose_intersection(SCALED_MM_ARCHS "12.0;12.0a" "${CUDA_ARCHS}")
if(${CMAKE_CUDA_COMPILER_VERSION} VERSION_GREATER_EQUAL 12.8 AND SCALED_MM_ARCHS)
set(SRCS
"csrc/quantization/w8a8/cutlass/scaled_mm_c3x_sm120.cu"
"csrc/quantization/w8a8/cutlass/c3x/scaled_mm_sm120_fp8.cu"
"csrc/quantization/w8a8/cutlass/c3x/scaled_mm_blockwise_sm120_fp8.cu"
"csrc/quantization/cutlass_w8a8/scaled_mm_c3x_sm120.cu"
"csrc/quantization/cutlass_w8a8/c3x/scaled_mm_sm120_fp8.cu"
"csrc/quantization/cutlass_w8a8/c3x/scaled_mm_blockwise_sm120_fp8.cu"
)
set_gencode_flags_for_srcs(
SRCS "${SRCS}"
@ -486,16 +452,12 @@ if(VLLM_GPU_LANG STREQUAL "CUDA")
# The cutlass_scaled_mm kernels for Blackwell SM100 (c3x, i.e. CUTLASS 3.x)
# require CUDA 12.8 or later
if(${CMAKE_CUDA_COMPILER_VERSION} VERSION_GREATER_EQUAL 13.0)
cuda_archs_loose_intersection(SCALED_MM_ARCHS "10.0f;11.0f;12.0f" "${CUDA_ARCHS}")
else()
cuda_archs_loose_intersection(SCALED_MM_ARCHS "10.0a;10.1a;10.3a;12.0a;12.1a" "${CUDA_ARCHS}")
endif()
cuda_archs_loose_intersection(SCALED_MM_ARCHS "10.0a;10.1a" "${CUDA_ARCHS}")
if(${CMAKE_CUDA_COMPILER_VERSION} VERSION_GREATER_EQUAL 12.8 AND SCALED_MM_ARCHS)
set(SRCS
"csrc/quantization/w8a8/cutlass/scaled_mm_c3x_sm100.cu"
"csrc/quantization/w8a8/cutlass/c3x/scaled_mm_sm100_fp8.cu"
"csrc/quantization/w8a8/cutlass/c3x/scaled_mm_blockwise_sm100_fp8.cu"
"csrc/quantization/cutlass_w8a8/scaled_mm_c3x_sm100.cu"
"csrc/quantization/cutlass_w8a8/c3x/scaled_mm_sm100_fp8.cu"
"csrc/quantization/cutlass_w8a8/c3x/scaled_mm_blockwise_sm100_fp8.cu"
)
set_gencode_flags_for_srcs(
SRCS "${SRCS}"
@ -526,7 +488,7 @@ if(VLLM_GPU_LANG STREQUAL "CUDA")
# subtract out the archs that are already built for 3x
list(REMOVE_ITEM SCALED_MM_2X_ARCHS ${SCALED_MM_3X_ARCHS})
if (SCALED_MM_2X_ARCHS)
set(SRCS "csrc/quantization/w8a8/cutlass/scaled_mm_c2x.cu")
set(SRCS "csrc/quantization/cutlass_w8a8/scaled_mm_c2x.cu")
set_gencode_flags_for_srcs(
SRCS "${SRCS}"
CUDA_ARCHS "${SCALED_MM_2X_ARCHS}")
@ -570,15 +532,10 @@ if(VLLM_GPU_LANG STREQUAL "CUDA")
# The nvfp4_scaled_mm_sm120 kernels for Geforce Blackwell SM120 require
# CUDA 12.8 or later
if(${CMAKE_CUDA_COMPILER_VERSION} VERSION_GREATER_EQUAL 13.0)
cuda_archs_loose_intersection(FP4_ARCHS "12.0f" "${CUDA_ARCHS}")
else()
cuda_archs_loose_intersection(FP4_ARCHS "12.0a" "${CUDA_ARCHS}")
endif()
cuda_archs_loose_intersection(FP4_ARCHS "12.0;12.0a" "${CUDA_ARCHS}")
if(${CMAKE_CUDA_COMPILER_VERSION} VERSION_GREATER_EQUAL 12.8 AND FP4_ARCHS)
set(SRCS
"csrc/quantization/fp4/nvfp4_quant_kernels.cu"
"csrc/quantization/fp4/activation_nvfp4_quant_fusion_kernels.cu"
"csrc/quantization/fp4/nvfp4_scaled_mm_sm120_kernels.cu")
set_gencode_flags_for_srcs(
SRCS "${SRCS}"
@ -593,15 +550,10 @@ if(VLLM_GPU_LANG STREQUAL "CUDA")
endif()
# FP4 Archs and flags
if(${CMAKE_CUDA_COMPILER_VERSION} VERSION_GREATER_EQUAL 13.0)
cuda_archs_loose_intersection(FP4_ARCHS "10.0f;11.0f;12.0f" "${CUDA_ARCHS}")
else()
cuda_archs_loose_intersection(FP4_ARCHS "10.0a;10.1a;12.0a;12.1a" "${CUDA_ARCHS}")
endif()
cuda_archs_loose_intersection(FP4_ARCHS "10.0a" "${CUDA_ARCHS}")
if(${CMAKE_CUDA_COMPILER_VERSION} VERSION_GREATER_EQUAL 12.8 AND FP4_ARCHS)
set(SRCS
"csrc/quantization/fp4/nvfp4_quant_kernels.cu"
"csrc/quantization/fp4/activation_nvfp4_quant_fusion_kernels.cu"
"csrc/quantization/fp4/nvfp4_experts_quant.cu"
"csrc/quantization/fp4/nvfp4_scaled_mm_kernels.cu"
"csrc/quantization/fp4/nvfp4_blockwise_moe_kernel.cu")
@ -619,13 +571,10 @@ if(VLLM_GPU_LANG STREQUAL "CUDA")
endif()
# CUTLASS MLA Archs and flags
if(${CMAKE_CUDA_COMPILER_VERSION} VERSION_GREATER_EQUAL 13.0)
cuda_archs_loose_intersection(MLA_ARCHS "10.0f;11.0f;12.0f" "${CUDA_ARCHS}")
else()
cuda_archs_loose_intersection(MLA_ARCHS "10.0a;10.1a;10.3a;12.0a;12.1a" "${CUDA_ARCHS}")
endif()
cuda_archs_loose_intersection(MLA_ARCHS "10.0a" "${CUDA_ARCHS}")
if(${CMAKE_CUDA_COMPILER_VERSION} VERSION_GREATER_EQUAL 12.8 AND MLA_ARCHS)
set(SRCS
"csrc/attention/mla/cutlass_mla_kernels.cu"
"csrc/attention/mla/sm100_cutlass_mla_kernel.cu")
set_gencode_flags_for_srcs(
SRCS "${SRCS}"
@ -649,7 +598,7 @@ if(VLLM_GPU_LANG STREQUAL "CUDA")
# if it's possible to compile MoE kernels that use its output.
cuda_archs_loose_intersection(SCALED_MM_ARCHS "9.0a" "${CUDA_ARCHS}")
if(${CMAKE_CUDA_COMPILER_VERSION} VERSION_GREATER_EQUAL 12.3 AND SCALED_MM_ARCHS)
set(SRCS "csrc/quantization/w8a8/cutlass/moe/grouped_mm_c3x_sm90.cu")
set(SRCS "csrc/quantization/cutlass_w8a8/moe/grouped_mm_c3x_sm90.cu")
set_gencode_flags_for_srcs(
SRCS "${SRCS}"
CUDA_ARCHS "${SCALED_MM_ARCHS}")
@ -667,13 +616,9 @@ if(VLLM_GPU_LANG STREQUAL "CUDA")
endif()
endif()
if(${CMAKE_CUDA_COMPILER_VERSION} VERSION_GREATER_EQUAL 13.0)
cuda_archs_loose_intersection(SCALED_MM_ARCHS "10.0f;11.0f" "${CUDA_ARCHS}")
else()
cuda_archs_loose_intersection(SCALED_MM_ARCHS "10.0a" "${CUDA_ARCHS}")
endif()
cuda_archs_loose_intersection(SCALED_MM_ARCHS "10.0a" "${CUDA_ARCHS}")
if(${CMAKE_CUDA_COMPILER_VERSION} VERSION_GREATER_EQUAL 12.8 AND SCALED_MM_ARCHS)
set(SRCS "csrc/quantization/w8a8/cutlass/moe/grouped_mm_c3x_sm100.cu")
set(SRCS "csrc/quantization/cutlass_w8a8/moe/grouped_mm_c3x_sm100.cu")
set_gencode_flags_for_srcs(
SRCS "${SRCS}"
CUDA_ARCHS "${SCALED_MM_ARCHS}")
@ -692,13 +637,9 @@ if(VLLM_GPU_LANG STREQUAL "CUDA")
endif()
# moe_data.cu is used by all CUTLASS MoE kernels.
if(${CMAKE_CUDA_COMPILER_VERSION} VERSION_GREATER_EQUAL 13.0)
cuda_archs_loose_intersection(CUTLASS_MOE_DATA_ARCHS "9.0a;10.0f;11.0f;12.0f" "${CUDA_ARCHS}")
else()
cuda_archs_loose_intersection(CUTLASS_MOE_DATA_ARCHS "9.0a;10.0a;10.1a;10.3a;12.0a;12.1a" "${CUDA_ARCHS}")
endif()
cuda_archs_loose_intersection(CUTLASS_MOE_DATA_ARCHS "9.0a;10.0a" "${CUDA_ARCHS}")
if(${CMAKE_CUDA_COMPILER_VERSION} VERSION_GREATER_EQUAL 12.3 AND CUTLASS_MOE_DATA_ARCHS)
set(SRCS "csrc/quantization/w8a8/cutlass/moe/moe_data.cu")
set(SRCS "csrc/quantization/cutlass_w8a8/moe/moe_data.cu")
set_gencode_flags_for_srcs(
SRCS "${SRCS}"
CUDA_ARCHS "${CUTLASS_MOE_DATA_ARCHS}")
@ -715,13 +656,9 @@ if(VLLM_GPU_LANG STREQUAL "CUDA")
endif()
endif()
if(${CMAKE_CUDA_COMPILER_VERSION} VERSION_GREATER_EQUAL 13.0)
cuda_archs_loose_intersection(SCALED_MM_ARCHS "10.0f;11.0f;12.0f" "${CUDA_ARCHS}")
else()
cuda_archs_loose_intersection(SCALED_MM_ARCHS "10.0a;10.1a;10.3a;12.0a;12.1a" "${CUDA_ARCHS}")
endif()
cuda_archs_loose_intersection(SCALED_MM_ARCHS "10.0a" "${CUDA_ARCHS}")
if(${CMAKE_CUDA_COMPILER_VERSION} VERSION_GREATER_EQUAL 12.8 AND SCALED_MM_ARCHS)
set(SRCS "csrc/quantization/w8a8/cutlass/moe/blockwise_scaled_group_mm_sm100.cu")
set(SRCS "csrc/quantization/cutlass_w8a8/moe/blockwise_scaled_group_mm_sm100.cu")
set_gencode_flags_for_srcs(
SRCS "${SRCS}"
CUDA_ARCHS "${SCALED_MM_ARCHS}")
@ -808,44 +745,6 @@ if(VLLM_GPU_LANG STREQUAL "CUDA")
"found in CUDA target architectures")
endif()
endif()
# Only build W4A8 kernels if we are building for something compatible with sm90a
cuda_archs_loose_intersection(W4A8_ARCHS "9.0a" "${CUDA_ARCHS}")
if(${CMAKE_CUDA_COMPILER_VERSION} VERSION_GREATER_EQUAL 12.0 AND W4A8_ARCHS)
set(SRCS
"csrc/quantization/cutlass_w4a8/w4a8_mm_entry.cu")
set_gencode_flags_for_srcs(
SRCS "${SRCS}"
CUDA_ARCHS "${W4A8_ARCHS}")
list(APPEND VLLM_EXT_SRC "${SRCS}")
message(STATUS "Building W4A8 kernels for archs: ${W4A8_ARCHS}")
else()
if (NOT ${CMAKE_CUDA_COMPILER_VERSION} VERSION_GREATER_EQUAL 12.0
AND W4A8_ARCHS)
message(STATUS "Not building W4A8 kernels as CUDA Compiler version is "
"not >= 12.0, we recommend upgrading to CUDA 12.0 or "
"later if you intend on running w4a16 quantized models on "
"Hopper.")
else()
message(STATUS "Not building W4A8 kernels as no compatible archs "
"found in CUDA target architectures")
endif()
endif()
# Hadacore kernels
cuda_archs_loose_intersection(HADACORE_ARCHS "8.0;8.9;9.0" "${CUDA_ARCHS}")
if(HADACORE_ARCHS)
set(SRCS "csrc/quantization/hadamard/hadacore/hadamard_transform_cuda.cu")
set_gencode_flags_for_srcs(
SRCS "${SRCS}"
CUDA_ARCHS "${HADACORE_ARCHS}")
list(APPEND VLLM_EXT_SRC "${SRCS}")
message(STATUS "Building hadacore")
endif()
# if CUDA endif
endif()
@ -883,13 +782,10 @@ target_compile_definitions(_C PRIVATE CUTLASS_ENABLE_DIRECT_CUDA_DRIVER_CALL=1)
set(VLLM_MOE_EXT_SRC
"csrc/moe/torch_bindings.cpp"
"csrc/moe/moe_align_sum_kernels.cu"
"csrc/moe/moe_lora_align_sum_kernels.cu"
"csrc/moe/topk_softmax_kernels.cu")
if(VLLM_GPU_LANG STREQUAL "CUDA")
list(APPEND VLLM_MOE_EXT_SRC
"csrc/moe/moe_wna16.cu"
"csrc/moe/grouped_topk_kernels.cu")
list(APPEND VLLM_MOE_EXT_SRC "csrc/moe/moe_wna16.cu")
endif()
if(VLLM_GPU_LANG STREQUAL "CUDA")
@ -958,10 +854,6 @@ if(VLLM_GPU_LANG STREQUAL "CUDA")
set_gencode_flags_for_srcs(
SRCS "${MOE_WNAA16_MARLIN_SRC}"
CUDA_ARCHS "${MARLIN_MOE_ARCHS}")
if(${CMAKE_CUDA_COMPILER_VERSION} VERSION_GREATER_EQUAL 12.8)
set_source_files_properties(${MOE_WNAA16_MARLIN_SRC}
PROPERTIES COMPILE_FLAGS "-static-global-template-stub=false")
endif()
list(APPEND VLLM_MOE_EXT_SRC ${MOE_WNAA16_MARLIN_SRC})
@ -1008,7 +900,6 @@ endif()
# For CUDA we also build and ship some external projects.
if (VLLM_GPU_LANG STREQUAL "CUDA")
include(cmake/external_projects/flashmla.cmake)
include(cmake/external_projects/qutlass.cmake)
# vllm-flash-attn should be last as it overwrites some CMake functions
include(cmake/external_projects/vllm_flash_attn.cmake)

View File

@ -2,6 +2,7 @@ include LICENSE
include requirements/common.txt
include requirements/cuda.txt
include requirements/rocm.txt
include requirements/neuron.txt
include requirements/cpu.txt
include CMakeLists.txt

View File

@ -14,28 +14,18 @@ Easy, fast, and cheap LLM serving for everyone
| <a href="https://docs.vllm.ai"><b>Documentation</b></a> | <a href="https://blog.vllm.ai/"><b>Blog</b></a> | <a href="https://arxiv.org/abs/2309.06180"><b>Paper</b></a> | <a href="https://x.com/vllm_project"><b>Twitter/X</b></a> | <a href="https://discuss.vllm.ai"><b>User Forum</b></a> | <a href="https://slack.vllm.ai"><b>Developer Slack</b></a> |
</p>
---
Join us at the [PyTorch Conference, October 22-23](https://events.linuxfoundation.org/pytorch-conference/) and [Ray Summit, November 3-5](https://www.anyscale.com/ray-summit/2025) in San Francisco for our latest updates on vLLM and to meet the vLLM team! Register now for the largest vLLM community events of the year!
---
*Latest News* 🔥
- [2025/11] We hosted [vLLM Beijing Meetup](https://mp.weixin.qq.com/s/xSrYXjNgr1HbCP4ExYNG1w) focusing on distributed inference and diverse accelerator support with vLLM! Please find the meetup slides [here](https://drive.google.com/drive/folders/1nQJ8ZkLSjKxvu36sSHaceVXtttbLvvu-?usp=drive_link).
- [2025/10] We hosted [vLLM Shanghai Meetup](https://mp.weixin.qq.com/s/__xb4OyOsImz-9eAVrdlcg) focused on hands-on vLLM inference optimization! Please find the meetup slides [here](https://drive.google.com/drive/folders/1KqwjsFJLfEsC8wlDugnrR61zsWHt94Q6).
- [2025/09] We hosted [vLLM Toronto Meetup](https://luma.com/e80e0ymm) focused on tackling inference at scale and speculative decoding with speakers from NVIDIA and Red Hat! Please find the meetup slides [here](https://docs.google.com/presentation/d/1IYJYmJcu9fLpID5N5RbW_vO0XLo0CGOR14IXOjB61V8/edit?usp=sharing).
- [2025/08] We hosted [vLLM Shenzhen Meetup](https://mp.weixin.qq.com/s/k8ZBO1u2_2odgiKWH_GVTQ) focusing on the ecosystem around vLLM! Please find the meetup slides [here](https://drive.google.com/drive/folders/1Ua2SVKVSu-wp5vou_6ElraDt2bnKhiEA).
- [2025/08] We hosted [vLLM Singapore Meetup](https://www.sginnovate.com/event/vllm-sg-meet). We shared V1 updates, disaggregated serving and MLLM speedups with speakers from Embedded LLM, AMD, WekaIO, and A*STAR. Please find the meetup slides [here](https://drive.google.com/drive/folders/1ncf3GyqLdqFaB6IeB834E5TZJPLAOiXZ?usp=sharing).
- [2025/08] We hosted [vLLM Shanghai Meetup](https://mp.weixin.qq.com/s/pDmAXHcN7Iqc8sUKgJgGtg) focusing on building, developing, and integrating with vLLM! Please find the meetup slides [here](https://drive.google.com/drive/folders/1OvLx39wnCGy_WKq8SiVKf7YcxxYI3WCH).
- [2025/08] We hosted [vLLM Beijing Meetup](https://mp.weixin.qq.com/s/dgkWg1WFpWGO2jCdTqQHxA) focusing on large-scale LLM deployment! Please find the meetup slides [here](https://drive.google.com/drive/folders/1Pid6NSFLU43DZRi0EaTcPgXsAzDvbBqF) and the recording [here](https://www.chaspark.com/#/live/1166916873711665152).
- [2025/05] We hosted [NYC vLLM Meetup](https://lu.ma/c1rqyf1f)! Please find the meetup slides [here](https://docs.google.com/presentation/d/1_q_aW_ioMJWUImf1s1YM-ZhjXz8cUeL0IJvaquOYBeA/edit?usp=sharing).
- [2025/05] vLLM is now a hosted project under PyTorch Foundation! Please find the announcement [here](https://pytorch.org/blog/pytorch-foundation-welcomes-vllm/).
- [2025/01] We are excited to announce the alpha release of vLLM V1: A major architectural upgrade with 1.7x speedup! Clean code, optimized execution loop, zero-overhead prefix caching, enhanced multimodal support, and more. Please check out our blog post [here](https://blog.vllm.ai/2025/01/27/v1-alpha-release.html).
<details>
<summary>Previous News</summary>
- [2025/08] We hosted [vLLM Korea Meetup](https://luma.com/cgcgprmh) with Red Hat and Rebellions! We shared the latest advancements in vLLM along with project spotlights from the vLLM Korea community. Please find the meetup slides [here](https://drive.google.com/file/d/1bcrrAE1rxUgx0mjIeOWT6hNe2RefC5Hm/view).
- [2025/08] We hosted [vLLM Beijing Meetup](https://mp.weixin.qq.com/s/dgkWg1WFpWGO2jCdTqQHxA) focusing on large-scale LLM deployment! Please find the meetup slides [here](https://drive.google.com/drive/folders/1Pid6NSFLU43DZRi0EaTcPgXsAzDvbBqF) and the recording [here](https://www.chaspark.com/#/live/1166916873711665152).
- [2025/05] We hosted [NYC vLLM Meetup](https://lu.ma/c1rqyf1f)! Please find the meetup slides [here](https://docs.google.com/presentation/d/1_q_aW_ioMJWUImf1s1YM-ZhjXz8cUeL0IJvaquOYBeA/edit?usp=sharing).
- [2025/04] We hosted [Asia Developer Day](https://www.sginnovate.com/event/limited-availability-morning-evening-slots-remaining-inaugural-vllm-asia-developer-day)! Please find the meetup slides from the vLLM team [here](https://docs.google.com/presentation/d/19cp6Qu8u48ihB91A064XfaXruNYiBOUKrBxAmDOllOo/edit?usp=sharing).
- [2025/03] We hosted [vLLM x Ollama Inference Night](https://lu.ma/vllm-ollama)! Please find the meetup slides from the vLLM team [here](https://docs.google.com/presentation/d/16T2PDD1YwRnZ4Tu8Q5r6n53c5Lr5c73UV9Vd2_eBo4U/edit?usp=sharing).
- [2025/03] We hosted [the first vLLM China Meetup](https://mp.weixin.qq.com/s/n77GibL2corAtQHtVEAzfg)! Please find the meetup slides from vLLM team [here](https://docs.google.com/presentation/d/1REHvfQMKGnvz6p3Fd23HhSO4c8j5WPGZV0bKYLwnHyQ/edit?usp=sharing).
@ -84,7 +74,7 @@ vLLM is flexible and easy to use with:
- Tensor, pipeline, data and expert parallelism support for distributed inference
- Streaming outputs
- OpenAI-compatible API server
- Support for NVIDIA GPUs, AMD CPUs and GPUs, Intel CPUs and GPUs, PowerPC CPUs, and TPU. Additionally, support for diverse hardware plugins such as Intel Gaudi, IBM Spyre and Huawei Ascend.
- Support NVIDIA GPUs, AMD CPUs and GPUs, Intel CPUs and GPUs, PowerPC CPUs, TPU, and AWS Neuron
- Prefix caching support
- Multi-LoRA support
@ -151,7 +141,6 @@ Compute Resources:
- Trainy
- UC Berkeley
- UC San Diego
- Volcengine
Slack Sponsor: Anyscale

View File

@ -42,9 +42,4 @@ For certain security issues of CRITICAL, HIGH, or MODERATE severity level, we ma
* If you wish to be added to the prenotification group, please send an email copying all the members of the [vulnerability management team](https://docs.vllm.ai/en/latest/contributing/vulnerability_management.html). Each vendor contact will be analyzed on a case-by-case basis.
* Organizations and vendors who either ship or use vLLM, are eligible to join the prenotification group if they meet at least one of the following qualifications
* Substantial internal deployment leveraging the upstream vLLM project.
* Established internal security teams and comprehensive compliance measures.
* Active and consistent contributions to the upstream vLLM project.
* We may withdraw organizations from receiving future prenotifications if they release fixes or any other information about issues before they are public. Group membership may also change based on policy refinements for who may be included.

Some files were not shown because too many files have changed in this diff Show More