huggingface / alignment-handbook

Robust recipes to align language models with human and AI preferences
https://huggingface.co/HuggingFaceH4
Apache License 2.0
4.45k stars 385 forks source link

Is it possible to visualize the loss graph during fine-tuning from Mistral to Zephyr? #10

Closed sooftware closed 9 months ago

sooftware commented 10 months ago

I really appreciate you sharing this valuable project. Is there a way to view the loss graph when fine-tuning from Mistral to Zephyr? This could be a valuable resource for research.

sebastianschramm commented 10 months ago

When you are running any of the zephyr-7b-beta recipes (and you did not change the default values in the yaml config files for report_to), it will log all metrics to tensorboard. And you can visualize the tensorboard logs during (and after) training by running tensorboard --logdir <RUNS_FOLDER_INSIDE_OF_OTUPTU_DIR>. Just make sure to use the path to the runs directory inside of the output_dir (see the config yaml file) as logdir. For example, if you run everything with the provided default values it should be tensorboard --logir data/zephyr-7b-sft-lora/runs for the LoRA SFT step.

sooftware commented 9 months ago

Thank you.