swift
swift copied to clipboard
Support unsloth for accelerating lora finetuning and decreasing memory consumption
Describe the feature Unsloth currently supports the LLaMA, Mistral and Yi models, and can achieve 170% speed boosting (result from LLaMA-Factory)
Paste any useful information Unsloth: https://github.com/unslothai/unsloth
Additional context Add any other context or information here(其他信息可以写在这里)