chenchenghao/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 11 Packages Projects Releases Wiki Activity

Deprecate disable-mla (#5481)

Browse Source
This commit is contained in:
Baizhou Zhang
2025-04-17 01:43:14 -07:00
committed by GitHub
parent 81c891111f
commit 4fb05583ef
9 changed files with 188 additions and 575 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
python/sglang/srt/model_executor/model_runner.py
View File

@@ -127,10 +127,7 @@ class ModelRunner:
self.page_size = server_args.page_size
self.req_to_token_pool = req_to_token_pool
self.token_to_kv_pool_allocator = token_to_kv_pool_allocator
self.use_mla_backend = (
self.model_config.attention_arch == AttentionArch.MLA
and not server_args.disable_mla
)
self.use_mla_backend = self.model_config.attention_arch == AttentionArch.MLA
self.attention_chunk_size = model_config.attention_chunk_size
# Model-specific adjustment
@@ -150,7 +147,6 @@ class ModelRunner:
"attention_backend": server_args.attention_backend,
"sampling_backend": server_args.sampling_backend,
"triton_attention_reduce_in_fp32": server_args.triton_attention_reduce_in_fp32,
"disable_mla": server_args.disable_mla,
"torchao_config": server_args.torchao_config,
"enable_nan_detection": server_args.enable_nan_detection,
"enable_dp_attention": server_args.enable_dp_attention,