423A35C7/LLaMA-Factory
1
0
Fork 0
mirror of https://github.com/hiyouga/LLaMA-Factory.git synced 2025-12-16 20:00:36 +08:00
Code Issues Packages Projects Releases Wiki Activity

initial-commit

Browse Source 
Former-commit-id: aeb85f200b
This commit is contained in:
simonJJJ
2024-08-28 16:51:35 +08:00
parent efd60f0306
commit 8a09b1e732
10 changed files with 337 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
src/llamafactory/model/adapter.py
View File

@@ -212,7 +212,7 @@ def _setup_lora_tuning(
target_modules = find_expanded_modules(model, target_modules, finetuning_args.freeze_trainable_layers)
if model_args.visual_inputs and finetuning_args.freeze_vision_tower:
target_modules = "^(?!.*vision_tower).*(?:{}).*".format("|".join(target_modules))
target_modules = "^(?!.*(?:vision_tower|visual)).*(?:{}).*".format("|".join(target_modules))
if (
finetuning_args.use_dora

2
src/llamafactory/model/model_utils/misc.py
View File

@@ -36,6 +36,8 @@ def find_all_linear_modules(model: "PreTrainedModel", freeze_vision_tower: bool)
forbidden_modules.add("output")
elif model.config.model_type in ["llava", "paligemma"]:
forbidden_modules.add("multi_modal_projector")
elif model.config.model_type in ["qwen2_vl"]:
forbidden_modules.add("merger")
if freeze_vision_tower:
forbidden_modules.add("vision_tower")