From e08bca2840ca8442c4047e31a659b4329ab943f6 Mon Sep 17 00:00:00 2001 From: isaac-vidas <80056737+isaac-vidas@users.noreply.github.com> Date: Mon, 22 Jan 2024 21:15:48 -0500 Subject: [PATCH] Support load fine-tuned LLaVA model (#80) --- python/sglang/srt/models/llama2.py | 4 ++++ 1 file changed, 4 insertions(+) diff --git a/python/sglang/srt/models/llama2.py b/python/sglang/srt/models/llama2.py index 5f728c5a8..c5690838b 100644 --- a/python/sglang/srt/models/llama2.py +++ b/python/sglang/srt/models/llama2.py @@ -303,6 +303,8 @@ class LlamaForCausalLM(nn.Module): # Skip loading extra bias for GPTQ models. if name.endswith(".bias") and name not in params_dict: continue + if name.startswith("model.vision_tower") and name not in params_dict: + continue param = params_dict[name] weight_loader = param.weight_loader weight_loader(param, loaded_weight, shard_id) @@ -311,6 +313,8 @@ class LlamaForCausalLM(nn.Module): # Skip loading extra bias for GPTQ models. if name.endswith(".bias") and name not in params_dict: continue + if name.startswith("model.vision_tower") and name not in params_dict: + continue param = params_dict[name] weight_loader = getattr(param, "weight_loader", default_weight_loader) weight_loader(param, loaded_weight)