README.md

metadata

base_model:
  - facebook/opt-2.7B
datasets:
  - databricks/databricks-dolly-15k
language:
  - en
license: apache-2.0
metrics:
  - rouge
pipeline_tag: text-generation
library_name: transformers

SeqKD-OPT-2.7B

paper | code

SeqKD-OPT-2.7B is an OPT-2.7B model distilled from OPT-13B on databricks-dolly-15k with sequence-level forward KLD.

It is used as a baseline for MiniLLM.

Other Baselines

• SFT w/o KD
• KD

Citation

@inproceedings{minillm,
  title={MiniLLM: Knowledge Distillation of Large Language Models},
  author={Gu, Yuxian and Dong, Li and Wei, Furu and Huang, Minlie},
  booktitle={Proceedings of ICLR},
  year={2024}
}