xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2025-12-10 06:35:00 +08:00
Code Issues Packages Projects Releases Wiki Activity

[Kernel] Turn off CUTLASS scaled_mm for Ada Lovelace (#6384)

Browse Source
This commit is contained in:
Tyler Michael Smith 2024-07-14 09:37:19 -04:00 committed by GitHub
parent 6ef3bf912c
commit 9dad5cc859
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
3 changed files with 15 additions and 9 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

8
.buildkite/lm-eval-harness/configs/Meta-Llama-3-8B-Instruct-FP8-compressed-tensors.yaml
View File

@ -1,11 +1,11 @@
# bash .buildkite/lm-eval-harness/run-lm-eval-gsm-vllm-baseline.sh -m nm-testing/Meta-Llama-3-8B-FP8-compressed-tensors-test -b 32 -l 250 -f 5 -t 1
# bash .buildkite/lm-eval-harness/run-lm-eval-gsm-vllm-baseline.sh -m nm-testing/Meta-Llama-3-8B-FP8-compressed-tensors-test -b 32 -l 1000 -f 5 -t 1
model_name: "nm-testing/Meta-Llama-3-8B-FP8-compressed-tensors-test"
tasks:
- name: "gsm8k"
metrics:
- name: "exact_match,strict-match"
value: 0.752
value: 0.755
- name: "exact_match,flexible-extract"
value: 0.752
limit: 250
value: 0.755
limit: 1000
num_fewshot: 5

6
.buildkite/lm-eval-harness/configs/Meta-Llama-3-8B-Instruct-FP8.yaml
View File

@ -4,8 +4,8 @@ tasks:
- name: "gsm8k"
metrics:
- name: "exact_match,strict-match"
value: 0.756
value: 0.753
- name: "exact_match,flexible-extract"
value: 0.752
limit: 250
value: 0.753
limit: 1000
num_fewshot: 5

8
csrc/quantization/cutlass_w8a8/scaled_mm_entry.cu
View File

@ -38,7 +38,13 @@ bool cutlass_scaled_mm_supports_fp8(int64_t cuda_device_capability) {
if (cuda_device_capability >= 90) {
return CUDA_VERSION >= 12000;
} else if (cuda_device_capability >= 89) {
return CUDA_VERSION >= 12040;
// CUTLASS Kernels have not been tuned for Ada Lovelace systems
// and are slower than torch.mm. Return false unconditionally in this case.
return false;
// Once the CUTLASS kernels have been optimized for Lovelace systems,
// use the following check:
// return CUDA_VERSION >= 12040;
}
#endif