README.md

metadata

license: other
license_name: yi-license
license_link: LICENSE
datasets:
  - garage-bAInd/Open-Platypus

Instruction tune of Yi-34b with Open-Platypus (fp16)

Overview

This is chargoddard/Yi-34B-Llama, with instruction tuning performed with the garage-bAInd/Open-Platypus dataset. That base model is 01-ai/Yi-34B, but using llama2 model definitions and tokenizer to remove any remote code requirements.

This is a (merged) QLoRA fine-tune (rank 64).

The finetune was performed with 1x RTX 6000 Ada (~18 hours to this checkpoint). It is possible this is rather undertrained, as this checkpoint is at 1 epoch. I began to see some performance degradation after that; more hyperparameter tuning is probably warranted.

How to Use

Use as you would any llama-2 model.

Prompting:

Model was trained with legacy airoboros <2.0 system prompt. See bhenrym14/airoboros-33b-gpt4-1.4.1-lxctx-PI-16384-fp16 model card for details.