--additional_target not supported with unsloth?

[kno10]

Reminder

• I have read the README and searched the existing issues.

Reproduction

It appears the additional_target parameter is not applied when unsloth is used.
The number of trainable parameters only includes the enabled lora layers, not the embedding layer.
For all I can tell, unsloth does support optimizing the embedding layers, too.

LLaMA-Factory/src/llmtuner/model/adapter.py

Lines 144 to 157 in 7f6c248

	if model_args.use_unsloth:
	from unsloth import FastLanguageModel # type: ignore
	unsloth_peft_kwargs = {"model": model, "max_seq_length": model_args.model_max_length}
	model = FastLanguageModel.get_peft_model(**peft_kwargs, **unsloth_peft_kwargs)
	else:
	lora_config = LoraConfig(
	task_type=TaskType.CAUSAL_LM,
	inference_mode=False,
	modules_to_save=finetuning_args.additional_target,
	use_dora=finetuning_args.use_dora,
	**peft_kwargs,
	)
	model = get_peft_model(model, lora_config)

Maybe this is sufficient?

--- a/src/llmtuner/model/adapter.py
+++ b/src/llmtuner/model/adapter.py
@@ -145,6 +145,8 @@ def init_adapter(
                 from unsloth import FastLanguageModel  # type: ignore
 
                 unsloth_peft_kwargs = {"model": model, "max_seq_length": model_args.model_max_length}
+                if finetuning_args.additional_target:
+                    unsloth_peft_kwargs["modules_to_save"] = finetuning_args.additional_target
                 model = FastLanguageModel.get_peft_model(**peft_kwargs, **unsloth_peft_kwargs)
             else:
                 lora_config = LoraConfig(

Expected behavior

No response

System Info

No response

Others

No response