frozenleaves/vllm
1
0
Fork 0
mirror of https://github.com/vllm-project/vllm.git synced 2025-10-20 23:03:52 +08:00
Code Issues Packages Projects Releases Wiki Activity

[V1] Fix yapf (#11538)

Browse Source 
Signed-off-by: Woosuk Kwon <woosuk.kwon@berkeley.edu>
This commit is contained in:
Woosuk Kwon
2024-12-27 09:47:10 +09:00
committed by GitHub
parent 371d04d39b
commit 81b979f2a8
2 changed files with 21 additions and 19 deletions
Download Patch File Download Diff File Expand all files Collapse all files

24
vllm/v1/sample/ops/penalties.py
View File

@ -2,8 +2,7 @@ from typing import List, Set, Tuple
import torch
from vllm.model_executor.layers.utils import (
apply_penalties as _apply_penalties)
from vllm.model_executor.layers.utils import apply_penalties
from vllm.utils import is_pin_memory_available, make_tensor_with_pad
@ -17,27 +16,30 @@ def apply_min_token_penalties(logits: torch.Tensor,
"""
min_tokens_logits_to_penalize: List[Tuple[int, int]] = []
for index, min_token in enumerate(min_tokens):
if (len(output_token_ids[index]) < min_token):
if len(output_token_ids[index]) < min_token:
for stop_token_id in stop_token_ids[index]:
min_tokens_logits_to_penalize.append((index, stop_token_id))
if min_tokens_logits_to_penalize:
logits[tuple(zip(*min_tokens_logits_to_penalize))] = -float("inf")
def apply_penalties(logits: torch.Tensor, prompt_token_ids: torch.Tensor,
presence_penalties: torch.Tensor,
frequency_penalties: torch.Tensor,
repetition_penalties: torch.Tensor,
output_token_ids: List[List[int]]) -> torch.Tensor:
def apply_all_penalties(
logits: torch.Tensor,
prompt_token_ids: torch.Tensor,
presence_penalties: torch.Tensor,
frequency_penalties: torch.Tensor,
repetition_penalties: torch.Tensor,
output_token_ids: List[List[int]],
) -> torch.Tensor:
"""
Applies presence, frequency and repetition penalties to the logits.
"""
_, vocab_size = logits.shape
output_tokens_t = _convert_to_tensors(output_token_ids, vocab_size,
logits.device)
return _apply_penalties(logits, prompt_token_ids, output_tokens_t,
presence_penalties, frequency_penalties,
repetition_penalties)
return apply_penalties(logits, prompt_token_ids, output_tokens_t,
presence_penalties, frequency_penalties,
repetition_penalties)
def _convert_to_tensors(output_token_ids: List[List[int]], vocab_size: int,

16
vllm/v1/sample/sampler.py
View File

@ -6,8 +6,8 @@ import torch.nn as nn
from vllm.v1.outputs import SamplerOutput
from vllm.v1.sample.metadata import SamplingMetadata
from vllm.v1.sample.ops.penalties import (apply_min_token_penalties,
apply_penalties)
from vllm.v1.sample.ops.penalties import (apply_all_penalties,
apply_min_token_penalties)
from vllm.v1.sample.ops.topk_topp_sampler import TopKTopPSampler
_SAMPLING_EPS = 1e-5
@ -127,10 +127,10 @@ class Sampler(nn.Module):
sampling_metadata.min_tokens)
if not sampling_metadata.no_penalties:
assert sampling_metadata.prompt_token_ids is not None
logits = apply_penalties(logits,
sampling_metadata.prompt_token_ids,
sampling_metadata.presence_penalties,
sampling_metadata.frequency_penalties,
sampling_metadata.repetition_penalties,
sampling_metadata.output_token_ids)
logits = apply_all_penalties(
logits, sampling_metadata.prompt_token_ids,
sampling_metadata.presence_penalties,
sampling_metadata.frequency_penalties,
sampling_metadata.repetition_penalties,
sampling_metadata.output_token_ids)
return logits