minke.yu/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 12 Packages Projects Releases Wiki Activity

Use Flashinfer for target_verify in GDN model for SM120 (#20604)

Browse Source
This commit is contained in:
Brayden Zhong
2026-03-17 22:40:56 -07:00
committed by GitHub
parent 97d5386a21
commit 88c40ec16d
1 changed files with 2 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files
python/sglang/srt/layers/attention/attention_registry.py
+2 -1
View File
@@ -210,7 +210,8 @@ def attn_backend_wrapper(runner: "ModelRunner", full_attn_backend: "AttentionBac
runner.server_args.attention_backend == "triton" runner.server_args.attention_backend == "triton"
or runner.server_args.attention_backend == "trtllm_mha" or runner.server_args.attention_backend == "trtllm_mha"
or runner.server_args.attention_backend == "fa4" or runner.server_args.attention_backend == "fa4"
), "triton or trtllm_mha or fa4 backend are the only supported backends on Blackwell GPUs for hybrid GDN models, use --attention-backend triton or --attention-backend trtllm_mha to specify the backend." or runner.server_args.attention_backend == "flashinfer"
), "triton, trtllm_mha, fa4, or flashinfer backend are the only supported backends on Blackwell GPUs for hybrid GDN models, use --attention-backend to specify the backend."
if is_npu(): if is_npu():
assert ( assert (
runner.server_args.attention_backend == "ascend" runner.server_args.attention_backend == "ascend"