423A35C7/LLaMA-Factory
1
0
Fork 0
mirror of https://github.com/hiyouga/LLaMA-Factory.git synced 2025-07-31 10:42:50 +08:00
Code Issues Packages Projects Releases Wiki Activity

[model] unsloth resume from checkpoint bug (#8423)

Browse Source 
Co-authored-by: viyer <vivek_iyer2@apple.com>
This commit is contained in:
Vivek Iyer 2025-06-23 04:43:54 -04:00 committed by GitHub
parent 8a3bddc7fa
commit 1221533542
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
2 changed files with 3 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

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

@ -188,7 +188,7 @@ def _setup_lora_tuning(
if adapter_to_resume is not None: # resume lora training
if model_args.use_unsloth:
model = load_unsloth_peft_model(config, model_args, is_trainable=is_trainable)
model = load_unsloth_peft_model(config, model_args, finetuning_args, is_trainable=is_trainable)
else:
model = PeftModel.from_pretrained(model, adapter_to_resume, is_trainable=is_trainable, **init_kwargs)

4
src/llamafactory/model/model_utils/unsloth.py
View File

@ -80,12 +80,12 @@ def get_unsloth_peft_model(
def load_unsloth_peft_model(
config: "PretrainedConfig", model_args: "ModelArguments", is_trainable: bool
config: "PretrainedConfig", model_args: "ModelArguments", finetuning_args: "FinetuningArguments", is_trainable: bool
) -> "PreTrainedModel":
r"""Load peft model with unsloth. Used in both training and inference."""
from unsloth import FastLanguageModel # type: ignore
unsloth_kwargs = _get_unsloth_kwargs(config, model_args.adapter_name_or_path[0], model_args)
unsloth_kwargs = _get_unsloth_kwargs(config, model_args.adapter_name_or_path[0], model_args, finetuning_args)
try:
if not is_trainable:
unsloth_kwargs["use_gradient_checkpointing"] = False