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

Remove gpt-oss hybrid swa gate for trtllm_mha (#19079)

Browse Source
This commit is contained in:
Mohammad Miadh Angkad
2026-02-27 10:30:00 -08:00
committed by GitHub
parent 1b79934d34
commit 35ef38c61b
1 changed files with 0 additions and 10 deletions
Download Patch File Download Diff File Expand all files Collapse all files
python/sglang/srt/server_args.py
-10
View File
@@ -1356,16 +1356,6 @@ class ServerArgs:
f"- Decode: {decode_attn_backend}\n" f"- Decode: {decode_attn_backend}\n"
) )
if (
prefill_attn_backend == "trtllm_mha"
or decode_attn_backend == "trtllm_mha"
):
# TODO: support swa kv indices translation for trtllm_mha attention backend
self.disable_hybrid_swa_memory = True
logger.warning(
"Disable hybrid SWA memory for GPT-OSS model with trtllm_mha attention backend."
)
quant_method = get_quantization_config(hf_config) quant_method = get_quantization_config(hf_config)
is_mxfp4_quant_format = quant_method == "mxfp4" is_mxfp4_quant_format = quant_method == "mxfp4"
if is_blackwell_supported(): if is_blackwell_supported():