frozenleaves/vllm
1
0
Fork 0
mirror of https://github.com/vllm-project/vllm.git synced 2025-10-20 14:53:52 +08:00
Code Issues Packages Projects Releases Wiki Activity

[Bugfix] fix an illegal memory access was encountered of marlin kernel + act_order (#18245)

Browse Source
This commit is contained in:
Jinzhen Lin
2025-05-17 07:02:44 +08:00
committed by GitHub
parent 7fdfa01530
commit e73b7dfd69
3 changed files with 29 additions and 22 deletions
Download Patch File Download Diff File Expand all files Collapse all files

25
csrc/moe/marlin_moe_wna16/marlin_template.h
View File

@ -1767,17 +1767,20 @@ __global__ void Marlin(
if constexpr (has_act_order) {
slice_k_start += tb_k * stages;
slice_k_start_shared_fetch += tb_k * stages;
int first_group_id = g_idx[slice_k_start];
int last_g_idx = slice_k_start + stages * tb_k * 2;
if (last_g_idx >= prob_k) {
last_g_idx = prob_k - 1;
}
int last_group_id = g_idx[last_g_idx];
if (last_group_id >= sh_first_group_id + sh_num_groups) {
fetch_act_order_scales_to_shared(false, first_group_id,
last_group_id);
__syncthreads();
if (slice_k_start < prob_k) {
slice_k_start_shared_fetch += tb_k * stages;
int first_group_id = g_idx[slice_k_start];
int last_g_idx = slice_k_start + stages * tb_k * 2;
if (last_g_idx >= prob_k) {
last_g_idx = prob_k - 1;
}
int last_group_id = g_idx[last_g_idx];
if (last_group_id >= sh_first_group_id + sh_num_groups) {
fetch_act_order_scales_to_shared(false, first_group_id,
last_group_id);
__syncthreads();
}
}
}
if (slice_iters == 0) {

24
csrc/quantization/gptq_marlin/marlin_template.h
View File

@ -1588,16 +1588,20 @@ __global__ void Marlin(
if constexpr (has_act_order) {
slice_k_start += tb_k * stages;
slice_k_start_shared_fetch += tb_k * stages;
int first_group_id = g_idx[slice_k_start];
int last_g_idx = slice_k_start + stages * tb_k * 2;
if (last_g_idx >= prob_k) {
last_g_idx = prob_k - 1;
}
int last_group_id = g_idx[last_g_idx];
if (last_group_id >= sh_first_group_id + sh_num_groups) {
fetch_act_order_scales_to_shared(false, first_group_id, last_group_id);
__syncthreads();
if (slice_k_start < prob_k) {
slice_k_start_shared_fetch += tb_k * stages;
int first_group_id = g_idx[slice_k_start];
int last_g_idx = slice_k_start + stages * tb_k * 2;
if (last_g_idx >= prob_k) {
last_g_idx = prob_k - 1;
}
int last_group_id = g_idx[last_g_idx];
if (last_group_id >= sh_first_group_id + sh_num_groups) {
fetch_act_order_scales_to_shared(false, first_group_id,
last_group_id);
__syncthreads();
}
}
}

2
tests/weight_loading/models.txt
View File

@ -2,7 +2,7 @@ gptq_marlin, robertgshaw2/zephyr-7b-beta-channelwise-gptq, main
gptq_marlin, TheBloke/Llama-2-7B-GPTQ, main
gptq_marlin, TheBloke/TinyLlama-1.1B-Chat-v1.0-GPTQ, main
gptq_marlin, TheBloke/TinyLlama-1.1B-Chat-v1.0-GPTQ, gptq-8bit--1g-actorder_True
#gptq_marlin, TheBloke/TinyLlama-1.1B-Chat-v1.0-GPTQ, gptq-8bit-32g-actorder_True
gptq_marlin, TheBloke/TinyLlama-1.1B-Chat-v1.0-GPTQ, gptq-8bit-32g-actorder_True
gptq_marlin, TechxGenus/gemma-1.1-2b-it-GPTQ, main
gptq, robertgshaw2/zephyr-7b-beta-channelwise-gptq, main
gptq, TheBloke/Llama-2-7B-GPTQ, main