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

[LFM2] Wire Lfm2MoeForCausalLM into the LFM2 serving override tables (#30780)

Browse Source
This commit is contained in:
Piotr Mazurek
2026-07-29 17:02:14 +00:00
committed by GitHub
parent f69af7b7ad
commit d19999b755
2 changed files with 3 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files
python/sglang/srt/arg_groups/overrides.py
+2 -1
View File
@@ -774,7 +774,7 @@ def _granite_moe_hybrid_overrides(server_args: Any, hf_config: Any) -> dict:
return {} return {}
@_register_for("Lfm2ForCausalLM") @_register_for("Lfm2ForCausalLM", "Lfm2MoeForCausalLM")
def _lfm2_overrides(server_args: Any, hf_config: Any) -> dict: def _lfm2_overrides(server_args: Any, hf_config: Any) -> dict:
if is_sm100_supported() and server_args.attention_backend is None: if is_sm100_supported() and server_args.attention_backend is None:
return {"attention_backend": "flashinfer"} return {"attention_backend": "flashinfer"}
@@ -1123,6 +1123,7 @@ _MAMBA_RADIX_CACHE_ARCHS = frozenset(
"JetNemotronForCausalLM", "JetNemotronForCausalLM",
"JetVLMForConditionalGeneration", "JetVLMForConditionalGeneration",
"Lfm2ForCausalLM", "Lfm2ForCausalLM",
"Lfm2MoeForCausalLM",
"ZayaForCausalLM", "ZayaForCausalLM",
} }
) )
python/sglang/srt/server_args.py
+1 -1
View File
@@ -5395,7 +5395,7 @@ class ServerArgs:
# _glm4_moe_overrides). # _glm4_moe_overrides).
pass pass
elif model_arch in ["Lfm2ForCausalLM"]: elif model_arch in ["Lfm2ForCausalLM", "Lfm2MoeForCausalLM"]:
# Attention backend selection moved to the override registry # Attention backend selection moved to the override registry
# (arg_groups/overrides.py: _lfm2_overrides). # (arg_groups/overrides.py: _lfm2_overrides).
assert resolved_view(self).attention_backend != "triton", ( assert resolved_view(self).attention_backend != "triton", (