README.md

---
language:
- en
- zh
license: apache-2.0
library_name: transformers
widget:
- text: <s> [|User|] Hi 👋  </s>[|Assistant|]
model-index:
- name: MiniChat-1.5-3B
  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: 46.5
      name: normalized accuracy
    source:
      url: https://huggingface.co./spaces/HuggingFaceH4/open_llm_leaderboard?query=GeneZC/MiniChat-1.5-3B
      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: 68.28
      name: normalized accuracy
    source:
      url: https://huggingface.co./spaces/HuggingFaceH4/open_llm_leaderboard?query=GeneZC/MiniChat-1.5-3B
      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: 46.67
      name: accuracy
    source:
      url: https://huggingface.co./spaces/HuggingFaceH4/open_llm_leaderboard?query=GeneZC/MiniChat-1.5-3B
      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: 50.71
    source:
      url: https://huggingface.co./spaces/HuggingFaceH4/open_llm_leaderboard?query=GeneZC/MiniChat-1.5-3B
      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: 65.04
      name: accuracy
    source:
      url: https://huggingface.co./spaces/HuggingFaceH4/open_llm_leaderboard?query=GeneZC/MiniChat-1.5-3B
      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: 24.18
      name: accuracy
    source:
      url: https://huggingface.co./spaces/HuggingFaceH4/open_llm_leaderboard?query=GeneZC/MiniChat-1.5-3B
      name: Open LLM Leaderboard
---

## MiniChat-1.5-3B

📑 [arXiv](https://arxiv.org/abs/2311.07052) | 👻 [GitHub](https://github.com/GeneZC/MiniMA) | 🤗 [HuggingFace-MiniMA](https://huggingface.co./GeneZC/MiniMA-3B) | 🤗 [HuggingFace-MiniChat](https://huggingface.co./GeneZC/MiniChat-3B) | 🤗 [HuggingFace-MiniChat-1.5](https://huggingface.co./GeneZC/MiniChat-1.5-3B) | 🤖 [ModelScope-MiniMA](https://modelscope.cn/models/GeneZC/MiniMA-3B) | 🤖 [ModelScope-MiniChat](https://modelscope.cn/models/GeneZC/MiniChat-3B)

🆕 **Updates from MiniChat-3B**: 
- better data mixture;
- use of [NEFTune](https://arxiv.org/abs/2310.05914);
- use of [DPO](https://arxiv.org/abs/2305.18290).

❗ Must comply with LICENSE of LLaMA2 since it is derived from LLaMA2.

A language model distilled and finetuned from an adapted version of LLaMA2-7B following "Towards the Law of Capacity Gap in Distilling Language Models".

Outperforming a wide range of 3B competitors in GPT4 evaluation and even competing with several 7B chat models.

<img src="./teaser_b.jpg" alt="teaser_b" width="687" />

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

```python
import torch

from transformers import AutoModelForCausalLM, AutoTokenizer

from conversation import get_default_conv_template

# MiniChat
tokenizer = AutoTokenizer.from_pretrained("GeneZC/MiniChat-3B", use_fast=False)
# GPU.
model = AutoModelForCausalLM.from_pretrained("GeneZC/MiniChat-3B", use_cache=True, device_map="auto", torch_dtype=torch.float16).eval()
# CPU.
# model = AutoModelForCausalLM.from_pretrained("GeneZC/MiniChat-3B", use_cache=True, device_map="cpu", torch_dtype=torch.float16).eval()

conv = get_default_conv_template("minichat")

question = "Implement a program to find the common elements in two arrays without using any extra data structures."
conv.append_message(conv.roles[0], question)
conv.append_message(conv.roles[1], None)
prompt = conv.get_prompt()
input_ids = tokenizer([prompt]).input_ids
output_ids = model.generate(
    torch.as_tensor(input_ids).cuda(),
    do_sample=True,
    temperature=0.7,
    max_new_tokens=1024,
)
output_ids = output_ids[0][len(input_ids[0]):]
output = tokenizer.decode(output_ids, skip_special_tokens=True).strip()
# output: "def common_elements(arr1, arr2):\n    if len(arr1) == 0:\n        return []\n    if len(arr2) == 0:\n        return arr1\n\n    common_elements = []\n    for element in arr1:\n        if element in arr2:\n            common_elements.append(element)\n\n    return common_elements"
# Multiturn conversation could be realized by continuously appending questions to `conv`.
```

## Bibtex

```bibtex
@article{zhang2023law,
    title={Towards the Law of Capacity Gap in Distilling Language Models},
    author={Zhang, Chen and Song, Dawei and Ye, Zheyu and Gao, Yan},
    year={2023},
    url={https://arxiv.org/abs/2311.07052}
}
```
# [Open LLM Leaderboard Evaluation Results](https://huggingface.co./spaces/HuggingFaceH4/open_llm_leaderboard)
Detailed results can be found [here](https://huggingface.co./datasets/open-llm-leaderboard/details_GeneZC__MiniChat-1.5-3B)

|             Metric              |Value|
|---------------------------------|----:|
|Avg.                             |50.23|
|AI2 Reasoning Challenge (25-Shot)|46.50|
|HellaSwag (10-Shot)              |68.28|
|MMLU (5-Shot)                    |46.67|
|TruthfulQA (0-shot)              |50.71|
|Winogrande (5-shot)              |65.04|
|GSM8k (5-shot)                   |24.18|