kanshan/TensorRT-LLMs
1
0
Fork 0
mirror of https://github.com/NVIDIA/TensorRT-LLM.git synced 2026-01-13 22:18:36 +08:00
Code Issues Actions 1 Packages Projects Releases Wiki Activity

[TRTLLM-8777][feat] Update DeepGEMM to the latest commit to include optimizations for DeepSeek-v3.2 (#9380)

Browse Source 
Signed-off-by: Fanrong Li <23290157+lfr-0531@users.noreply.github.com>
This commit is contained in:
Fanrong Li 2025-11-25 08:58:08 +08:00 committed by GitHub
parent 786d308b88
commit 5a99c9734d
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
2 changed files with 6 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
3rdparty/CMakeLists.txt vendored
View File

@ -39,7 +39,7 @@ FetchContent_Declare(
FetchContent_Declare(
deepgemm
GIT_REPOSITORY https://github.com/ruoqianguo/DeepGEMM
GIT_TAG 9fa5965e265e27995f539e0dd73a06351a8a9eaf
GIT_TAG 6cb8161516302550785d9af924d2778afef1f3f6 # swapab_sm100 branch
GIT_SUBMODULES_RECURSE
ON
SOURCE_SUBDIR

10
tests/unittest/_torch/attention/sparse/test_dsa_indexer.py
View File

@ -308,9 +308,9 @@ def test_deepgemm_fp8_mqa_logits_basic():
"""
torch.manual_seed(0)
num_heads, head_dim = 32, 128
seq_len = 512
seq_len_kv = 1024
num_heads, head_dim = 64, 128
seq_len = 2048
seq_len_kv = 4096
#[seq_len, num_heads, head_dim]
q = torch.randn(
seq_len,
@ -335,8 +335,8 @@ def test_deepgemm_fp8_mqa_logits_basic():
)
# ks[i] -> ke[i] for each q[i]
ks = torch.zeros(seq_len, dtype=torch.int, device="cuda")
ke = torch.arange(seq_len, dtype=torch.int, device="cuda") + (
seq_len_kv - seq_len) + 1 # +1 for exclusive end
ke = torch.arange(seq_len, dtype=torch.int,
device="cuda") + (seq_len_kv - seq_len)
# Convert to FP8
q_fp8 = q.to(torch.float8_e4m3fn)