biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[Model] Add LongCat-Flash (#23991)

Browse Source 
Signed-off-by: yangxurui <yangxurui@meituan.com>
Co-authored-by: yangxurui <yangxurui@meituan.com>
This commit is contained in:
XuruiYang
2025-09-25 12:53:40 +08:00
committed by GitHub
parent 90b139cfff
commit 845adb3ec6
31 changed files with 1357 additions and 66 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
vllm/config/model.py
View File

@@ -1131,7 +1131,8 @@ class ModelConfig:
if not hasattr(self.hf_text_config, "model_type"):
return False
elif self.hf_text_config.model_type in \
('deepseek_v2', 'deepseek_v3', 'deepseek_mtp', 'kimi_k2'):
('deepseek_v2', 'deepseek_v3', 'deepseek_mtp',
'kimi_k2', 'longcat_flash'):
return self.hf_text_config.kv_lora_rank is not None
elif self.hf_text_config.model_type == 'eagle':
# if the model is an EAGLE module, check for the
@@ -1257,6 +1258,9 @@ class ModelConfig:
or self.hf_config.model_type == "qwen3_next_mtp"):
total_num_hidden_layers = getattr(self.hf_text_config,
"num_nextn_predict_layers", 0)
elif (self.hf_config.model_type == "longcat_flash_mtp"):
total_num_hidden_layers = getattr(self.hf_text_config,
"num_nextn_predict_layers", 1)
else:
total_num_hidden_layers = getattr(self.hf_text_config,
"num_hidden_layers", 0)