EngineX-Hygon/sglang
5
0
Fork 0
Code Issues Pull Requests Actions 7 Projects Releases Wiki Activity

Support load fine-tuned LLaVA model (#80)

Browse Source
This commit is contained in:
isaac-vidas
2024-01-22 21:15:48 -05:00
committed by GitHub
parent cd3ccb2ed7
commit e08bca2840
1 changed files with 4 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
python/sglang/srt/models/llama2.py
View File

@@ -303,6 +303,8 @@ class LlamaForCausalLM(nn.Module):
# Skip loading extra bias for GPTQ models. # Skip loading extra bias for GPTQ models.
if name.endswith(".bias") and name not in params_dict: if name.endswith(".bias") and name not in params_dict:
continue continue
if name.startswith("model.vision_tower") and name not in params_dict:
continue
param = params_dict[name] param = params_dict[name]
weight_loader = param.weight_loader weight_loader = param.weight_loader
weight_loader(param, loaded_weight, shard_id) weight_loader(param, loaded_weight, shard_id)
@@ -311,6 +313,8 @@ class LlamaForCausalLM(nn.Module):
# Skip loading extra bias for GPTQ models. # Skip loading extra bias for GPTQ models.
if name.endswith(".bias") and name not in params_dict: if name.endswith(".bias") and name not in params_dict:
continue continue
if name.startswith("model.vision_tower") and name not in params_dict:
continue
param = params_dict[name] param = params_dict[name]
weight_loader = getattr(param, "weight_loader", default_weight_loader) weight_loader = getattr(param, "weight_loader", default_weight_loader)
weight_loader(param, loaded_weight) weight_loader(param, loaded_weight)