Wilson Wu

Fixes https://github.com/unslothai/unsloth/issues/1948

This causes from_pretrained to error out when model_name points to a peft adapter and revision is specified. I think you can fix this by deleting this one line: https://github.com/unslothai/unsloth/blob/8d7662e49e0102559ba63ecb06e47325148cb33e/unsloth/models/loader.py#L205

Trainer.training_step incorrectly normalizes mean token loss when n_gpu > 1

5

### System Info ``` - `transformers` version: 4.46.0 - Platform: Linux-5.15.0-136-generic-x86_64-with-glibc2.35 - Python version: 3.10.12 - Huggingface_hub version: 0.29.2 - Safetensors version: 0.5.3 - Accelerate version: 1.4.0 - Accelerate config:...