• Model creator: https://huggingface.co/GeneZC/
• Original model: https://huggingface.co/GeneZC/MiniMA-2-3B/

Original model description:

• en
• zh license: apache-2.0 library_name: transformers datasets:
• EleutherAI/pile
• togethercomputer/RedPajama-Data-1T
• p208p2002/wudao widget:
• text: 4 + 3 = model-index:
• name: MiniMA-2-3B results:
  • task: type: text-generation name: Text Generation dataset: name: AI2 Reasoning Challenge (25-Shot) type: ai2_arc config: ARC-Challenge split: test args: num_few_shot: 25 metrics:
    • type: acc_norm value: 44.71 name: normalized accuracy source: url: https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=GeneZC/MiniMA-2-3B name: Open LLM Leaderboard
  • task: type: text-generation name: Text Generation dataset: name: HellaSwag (10-Shot) type: hellaswag split: validation args: num_few_shot: 10 metrics:
    • type: acc_norm value: 69.33 name: normalized accuracy source: url: https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=GeneZC/MiniMA-2-3B name: Open LLM Leaderboard
  • task: type: text-generation name: Text Generation dataset: name: MMLU (5-Shot) type: cais/mmlu config: all split: test args: num_few_shot: 5 metrics:
    • type: acc value: 41.22 name: accuracy source: url: https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=GeneZC/MiniMA-2-3B name: Open LLM Leaderboard
  • task: type: text-generation name: Text Generation dataset: name: TruthfulQA (0-shot) type: truthful_qa config: multiple_choice split: validation args: num_few_shot: 0 metrics:
    • type: mc2 value: 38.44 source: url: https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=GeneZC/MiniMA-2-3B name: Open LLM Leaderboard
  • task: type: text-generation name: Text Generation dataset: name: Winogrande (5-shot) type: winogrande config: winogrande_xl split: validation args: num_few_shot: 5 metrics:
    • type: acc value: 66.69 name: accuracy source: url: https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=GeneZC/MiniMA-2-3B name: Open LLM Leaderboard
  • task: type: text-generation name: Text Generation dataset: name: GSM8k (5-shot) type: gsm8k config: main split: test args: num_few_shot: 5 metrics:
    • type: acc value: 8.11 name: accuracy source: url: https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=GeneZC/MiniMA-2-3B name: Open LLM Leaderboard

---

MiniMA-2-3B

📑 arXiv | 👻 GitHub | 🤗 HuggingFace-MiniMA | 🤗 HuggingFace-MiniChat | 🤖 ModelScope-MiniMA | 🤖 ModelScope-MiniChat | 🤗 HuggingFace-MiniChat-1.5 | 🤗 HuggingFace-MiniMA-2 | 🤗 HuggingFace-MiniChat-2

🆕 Updates from MiniMA-3B:

• continued from MiniMA-3B without distillation;
• better data mixture;
• more trained tokens.

❗ Must comply with LICENSE of LLaMA-2 since it is derived from LLaMA-2.

A language model continued from MiniMA-3B.

Completing the compute-performance pareto frontier together with MiniMA-3B and other arts.

Standard Benchmarks

Method	TFLOPs	MMLU (5-shot)	CEval (5-shot)	DROP (3-shot)	HumanEval (0-shot)	BBH (3-shot)	GSM8K (8-shot)
Mamba-2.8B	4.6E9	25.58	24.74	15.72	7.32	29.37	3.49
ShearedLLaMA-2.7B	0.8E9	26.97	22.88	19.98	4.88	30.48	3.56
BTLM-3B	11.3E9	27.20	26.00	17.84	10.98	30.87	4.55
StableLM-3B	72.0E9	44.75	31.05	22.35	15.85	32.59	10.99
Qwen-1.8B	23.8E9	44.05	54.75	12.97	14.02	30.80	22.97
Phi-2-2.8B	159.9E9	56.74	34.03	30.74	46.95	44.13	55.42
LLaMA-2-7B	84.0E9	46.00	34.40	31.57	12.80	32.02	14.10
MiniMA-3B	4.0E9	28.51	28.23	22.50	10.98	31.61	8.11
MiniChat-3B	4.0E9	38.40	36.48	22.58	18.29	31.36	29.72
MiniMA-2-3B	13.4E9	40.14	44.65	23.10	14.63	31.43	8.87
MiniChat-2-3B	13.4E9	46.17	43.91	30.26	22.56	34.95	38.13

The following is an example code snippet to use MiniMA-2-3B:

python

1import torch 2 3from transformers import AutoModelForCausalLM, AutoTokenizer 4 5# MiniMA 6tokenizer = AutoTokenizer.from_pretrained("GeneZC/MiniMA-2-3B", use_fast=False) 7# GPU. 8model = AutoModelForCausalLM.from_pretrained("GeneZC/MiniMA-2-3B", use_cache=True, device_map="auto", torch_dtype=torch.float16).eval() 9# CPU. 10# model = AutoModelForCausalLM.from_pretrained("GeneZC/MiniMA-2-3B", use_cache=True, device_map="cpu", torch_dtype=torch.float16).eval() 11 12prompt = "Question: Sherrie tells the truth. Vernell says Sherrie tells the truth. Alexis says Vernell lies. Michaela says Alexis tells the truth. Elanor says Michaela tells the truth. Does Elanor tell the truth?\nAnswer: No\n\nQuestion: Kristian lies. Sherrie says Kristian lies. Delbert says Sherrie lies. Jerry says Delbert tells the truth. Shalonda says Jerry tells the truth. Does Shalonda tell the truth?\nAnswer: No\n\nQuestion: Vina tells the truth. Helene says Vina lies. Kandi says Helene tells the truth. Jamey says Kandi lies. Ka says Jamey lies. Does Ka tell the truth?\nAnswer: No\n\nQuestion: Christie tells the truth. Ka says Christie tells the truth. Delbert says Ka lies. Leda says Delbert tells the truth. Lorine says Leda tells the truth. Does Lorine tell the truth?\nAnswer:" 13input_ids = tokenizer([prompt]).input_ids 14output_ids = model.generate( 15 torch.as_tensor(input_ids).cuda(), 16 do_sample=True, 17 temperature=0.7, 18 max_new_tokens=1024, 19) 20output_ids = output_ids[0][len(input_ids[0]):] 21output = tokenizer.decode(output_ids, skip_special_tokens=True).strip() 22# output: "No"

Bibtex

bibtex

1@article{zhang2023law, 2 title={Towards the Law of Capacity Gap in Distilling Language Models}, 3 author={Zhang, Chen and Song, Dawei and Ye, Zheyu and Gao, Yan}, 4 year={2023}, 5 url={https://arxiv.org/abs/2311.07052} 6}

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric	Value
Avg.	44.75
AI2 Reasoning Challenge (25-Shot)	44.71
HellaSwag (10-Shot)	69.33
MMLU (5-Shot)	41.22
TruthfulQA (0-shot)	38.44
Winogrande (5-shot)	66.69
GSM8k (5-shot)	8.11