RichardErkhov/MiniLLM_-_MiniLLM-gpt2-120M-gguf

Quantization made by Richard Erkhov.

Github

Discord

Request more models

MiniLLM-gpt2-120M - GGUF

• Model creator: https://huggingface.co/MiniLLM/
• Original model: https://huggingface.co/MiniLLM/MiniLLM-gpt2-120M/

Name	Quant method	Size
MiniLLM-gpt2-120M.Q2_K.gguf	Q2_K	0.08GB
MiniLLM-gpt2-120M.IQ3_XS.gguf	IQ3_XS	0.08GB
MiniLLM-gpt2-120M.IQ3_S.gguf	IQ3_S	0.08GB
MiniLLM-gpt2-120M.Q3_K_S.gguf	Q3_K_S	0.08GB
MiniLLM-gpt2-120M.IQ3_M.gguf	IQ3_M	0.09GB
MiniLLM-gpt2-120M.Q3_K.gguf	Q3_K	0.09GB
MiniLLM-gpt2-120M.Q3_K_M.gguf	Q3_K_M	0.09GB
MiniLLM-gpt2-120M.Q3_K_L.gguf	Q3_K_L	0.1GB
MiniLLM-gpt2-120M.IQ4_XS.gguf	IQ4_XS	0.1GB
MiniLLM-gpt2-120M.Q4_0.gguf	Q4_0	0.1GB
MiniLLM-gpt2-120M.IQ4_NL.gguf	IQ4_NL	0.1GB
MiniLLM-gpt2-120M.Q4_K_S.gguf	Q4_K_S	0.1GB
MiniLLM-gpt2-120M.Q4_K.gguf	Q4_K	0.11GB
MiniLLM-gpt2-120M.Q4_K_M.gguf	Q4_K_M	0.11GB
MiniLLM-gpt2-120M.Q4_1.gguf	Q4_1	0.11GB
MiniLLM-gpt2-120M.Q5_0.gguf	Q5_0	0.11GB
MiniLLM-gpt2-120M.Q5_K_S.gguf	Q5_K_S	0.11GB
MiniLLM-gpt2-120M.Q5_K.gguf	Q5_K	0.12GB
MiniLLM-gpt2-120M.Q5_K_M.gguf	Q5_K_M	0.12GB
MiniLLM-gpt2-120M.Q5_1.gguf	Q5_1	0.12GB
MiniLLM-gpt2-120M.Q6_K.gguf	Q6_K	0.13GB
MiniLLM-gpt2-120M.Q8_0.gguf	Q8_0	0.17GB

Original model description:

license: apache-2.0 datasets: - databricks/databricks-dolly-15k language: - en metrics: - rouge base_model: - openai-community/gpt2 pipeline_tag: text-generation

MiniLLM-gpt2-120M

paper | code

MiniLLM-gpt2-120M is a gpt2-base (120M) model distilled from gpt2-xlarge (1.5B) on databricks-dolly-15k.

Note: MiniLLM requires a SFT model for initilization to perform the PPO optimization.

Evaluation

We ask GPT-4 to give scores for the generated responses of MiniLLM. The prompts are taken from databricks-dolly-15k (test set), self-instruct, and vicuna

Baseline Models

• SFT w/o KD
• KD
• SeqKD

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}
}