Skip to content

Commit 35ef38c

Browse files
authored
Remove gpt-oss hybrid swa gate for trtllm_mha (#19079)
1 parent 1b79934 commit 35ef38c

File tree

1 file changed

+0
-10
lines changed

1 file changed

+0
-10
lines changed

python/sglang/srt/server_args.py

Lines changed: 0 additions & 10 deletions
Original file line numberDiff line numberDiff line change
@@ -1356,16 +1356,6 @@ def _handle_model_specific_adjustments(self):
13561356
f"- Decode: {decode_attn_backend}\n"
13571357
)
13581358

1359-
if (
1360-
prefill_attn_backend == "trtllm_mha"
1361-
or decode_attn_backend == "trtllm_mha"
1362-
):
1363-
# TODO: support swa kv indices translation for trtllm_mha attention backend
1364-
self.disable_hybrid_swa_memory = True
1365-
logger.warning(
1366-
"Disable hybrid SWA memory for GPT-OSS model with trtllm_mha attention backend."
1367-
)
1368-
13691359
quant_method = get_quantization_config(hf_config)
13701360
is_mxfp4_quant_format = quant_method == "mxfp4"
13711361
if is_blackwell_supported():

0 commit comments

Comments
 (0)