xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2026-03-23 11:49:09 +08:00
Code Issues Packages Projects Releases Wiki Activity

fix attention free models

Browse Source 
Signed-off-by: Xingyu Liu <charlotteliu12x@gmail.com>
This commit is contained in:
Xingyu Liu 2025-12-08 16:13:31 -08:00
parent 0a4f4724ef
commit 1cf506d89e
1 changed files with 19 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

19
vllm/transformers_utils/model_arch_config_convertor.py
View File

@ -269,6 +269,22 @@ class ModelArchConfigConvertorBase:
return model_arch_config
class MambaModelArchConfigConvertor(ModelArchConfigConvertorBase):
def get_head_size(self) -> int:
return 0
def get_total_num_kv_heads(self) -> int:
return 0
class TerratorchModelArchConfigConvertor(ModelArchConfigConvertorBase):
def get_head_size(self) -> int:
return 0
def get_total_num_kv_heads(self) -> int:
return 0
class Zamba2ModelArchConfigConvertor(ModelArchConfigConvertorBase):
def get_head_size(self) -> int:
return getattr(self.hf_text_config, "attention_head_dim", 0)
@ -357,6 +373,9 @@ class LongCatFlashMTPModelArchConfigConvertor(ModelArchConfigConvertorBase):
# hf_config.model_type -> convertor class
MODEL_ARCH_CONFIG_CONVERTORS = {
"mamba": MambaModelArchConfigConvertor,
"mamba2": MambaModelArchConfigConvertor,
"terratorch": TerratorchModelArchConfigConvertor,
"zamba2": Zamba2ModelArchConfigConvertor,
"mpt": MPTModelArchConfigConvertor,
"dbrx": DbrxModelArchConfigConvertor,