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

12
tests/models/utils.py
View File

@@ -428,9 +428,8 @@ def dummy_hf_overrides(
num_hidden_layers = (3 if model_arch
== "Gemma3nForConditionalGeneration" else 1)
text_config.update({
update_dict = {
"num_layers": num_layers,
"num_hidden_layers": num_hidden_layers,
"num_experts": num_experts,
"num_experts_per_tok": 2,
"num_local_experts": num_experts,
@@ -440,7 +439,14 @@ def dummy_hf_overrides(
"n_routed_experts": num_experts,
# For Gemma-3n
"num_kv_shared_layers": 1,
})
}
# Update num_hidden_layers for non-Longcat architectures
if model_arch != "LongcatFlashForCausalLM" \
and model_arch != "LongCatFlashMTPModel":
update_dict["num_hidden_layers"] = num_hidden_layers
text_config.update(update_dict)
if hasattr(hf_config, "vision_config"):
hf_config.vision_config.update({