minke.yu/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 13 Packages Projects Releases Wiki Activity

[BUGFIX] fix bug in handle mamba radix cache in server_args (#18723)

Browse Source
This commit is contained in:
Yi Zhang
2026-02-12 21:33:32 +08:00
committed by GitHub
parent 92c5749f41
commit b168723424
1 changed files with 6 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files
python/sglang/srt/server_args.py
+6 -1
View File
@@ -1538,6 +1538,7 @@ class ServerArgs:
self._handle_mamba_radix_cache( self._handle_mamba_radix_cache(
model_arch=model_arch, model_arch=model_arch,
support_mamba_cache=True,
support_mamba_cache_extra_buffer=False, support_mamba_cache_extra_buffer=False,
sm100_default_attention_backend="flashinfer", sm100_default_attention_backend="flashinfer",
) )
@@ -1589,6 +1590,7 @@ class ServerArgs:
) )
self._handle_mamba_radix_cache( self._handle_mamba_radix_cache(
model_arch=model_arch, model_arch=model_arch,
support_mamba_cache=True,
support_mamba_cache_extra_buffer=True, support_mamba_cache_extra_buffer=True,
sm100_default_attention_backend="triton", sm100_default_attention_backend="triton",
) )
@@ -1622,6 +1624,7 @@ class ServerArgs:
]: ]:
self._handle_mamba_radix_cache( self._handle_mamba_radix_cache(
model_arch=model_arch, model_arch=model_arch,
support_mamba_cache=True,
support_mamba_cache_extra_buffer=False, support_mamba_cache_extra_buffer=False,
sm100_default_attention_backend="triton", sm100_default_attention_backend="triton",
) )
@@ -1629,6 +1632,7 @@ class ServerArgs:
elif model_arch in ["Lfm2ForCausalLM"]: elif model_arch in ["Lfm2ForCausalLM"]:
self._handle_mamba_radix_cache( self._handle_mamba_radix_cache(
model_arch=model_arch, model_arch=model_arch,
support_mamba_cache=True,
support_mamba_cache_extra_buffer=False, support_mamba_cache_extra_buffer=False,
sm100_default_attention_backend="flashinfer", sm100_default_attention_backend="flashinfer",
) )
@@ -1698,7 +1702,8 @@ class ServerArgs:
assert ( assert (
not self.enable_mamba_extra_buffer() not self.enable_mamba_extra_buffer()
), f"mamba extra_buffer is not supported for {model_arch} model" ), f"mamba extra_buffer is not supported for {model_arch} model"
elif self.enable_mamba_extra_buffer(): # extra_buffer
if self.enable_mamba_extra_buffer(): # extra_buffer
assert ( assert (
is_cuda() is_cuda()
), "Mamba extra_buffer is only supported on CUDA devices with FLA backend" ), "Mamba extra_buffer is only supported on CUDA devices with FLA backend"