shibing624 / chinese-alpaca-plus-7b-hf

huggingface.co
Total runs: 1.5K
24-hour runs: 39
7-day runs: 262
30-day runs: 381
Model's Last Updated: December 15 2023
text-generation

Introduction of chinese-alpaca-plus-7b-hf

Model Details of chinese-alpaca-plus-7b-hf

Chinese Alpaca Plus 7B Model

发布中文LLaMA, Alpaca Plus版(7B)模型

推出中文LLaMA, Alpaca Plus版(7B),相比基础版本的改进点如下:

  • 进一步扩充了训练数据,其中LLaMA扩充至120G文本(通用领域),Alpaca扩充至4M指令数据(重点增加了STEM相关数据)
  • Alpaca训练时采用了更大的rank,相比原版具有更低的验证集损失
  • 评测结果显示,Alpaca-Plus-7B相比基础版Alpaca-7B效果更优,部分任务接近或超过13B版本
  • 这一轮比拼:7B获得65.3分,13B获得70.9分,Plus-7B效果75.3分,具体评测结果请参考 效果评测

本模型是 原生LLaMA-7B 合并 中文LLaMA LoRA 中文Alpaca LoRA 后的模型权重 chinese-alpaca-plus-7b-hf ,并转化为HuggingFace版本权重(.bin文件),可以直接使用或者继续训练。

13b-hf权重链接: https://huggingface.co/shibing624/chinese-alpaca-plus-13b-hf

test case:

input_text predict
为什么天空是蓝色的? 天空是蓝色的,是因为大气层中的气体分子会散射太阳光中的蓝色光,使得我们看到的天空是蓝色的。
release model weight
Usage

本项目开源在textgen项目: textgen ,可支持llama模型,通过如下命令调用:

Install package:

pip install -U textgen
from textgen import GptModel
model = GptModel("llama", "shibing624/chinese-alpaca-plus-7b-hf")
r = model.predict(["用一句话描述地球为什么是独一无二的。"])
print(r) # ['地球是独一无二的,因为它拥有独特的大气层、水循环、生物多样性以及其他自然资源,这些都使它成为一个独特的生命支持系统。']
Usage (HuggingFace Transformers)

Without textgen , you can use the model like this:

First, you pass your input through the transformer model, then you get the generated sentence.

Install package:

pip install sentencepiece
pip install transformers>=4.28.0
import torch
import transformers
from transformers import LlamaTokenizer, LlamaForCausalLM

def generate_prompt(text):
    return f"""Below is an instruction that describes a task. Write a response that appropriately completes the request.

### Instruction:
{text}

### Response:"""


tokenizer = LlamaTokenizer.from_pretrained('shibing624/chinese-alpaca-plus-7b-hf')
model = LlamaForCausalLM.from_pretrained('shibing624/chinese-alpaca-plus-7b-hf').half().cuda()
model.eval()

text = '为什么天空是蓝色的?'
prompt = generate_prompt(text)
input_ids = tokenizer.encode(prompt, return_tensors='pt').to('cuda')


with torch.no_grad():
    output_ids = model.generate(
        input_ids=input_ids,
        max_new_tokens=128,
        temperature=1,
        top_k=40,
        top_p=0.9,
        repetition_penalty=1.15
    ).cuda()
output = tokenizer.decode(output_ids[0], skip_special_tokens=True)
print(output.replace(text, '').strip())

output:

为什么天空是蓝色的?
天空是蓝色的,是因为大气层中的气体分子会散射太阳光中的蓝色光,使得我们看到的天空是蓝色的。
模型来源

release合并后的模型权重,一步到位直接使用,省电、减少碳排放。

基于 多LoRA权重合并(适用于Chinese-Alpaca-Plus ) 方法手动合并而成,具体是使用 decapoda-research/llama-7b-hf 底座模型 合并 Chinese-LLaMA-Plus-LoRA和Chinese-Alpaca-Plus-LoRA 两个LoRA权重 得到,并转化为HuggingFace版本权重(.bin文件)。

HuggingFace版本权重(.bin文件)可用于:

  • 使用Transformers进行训练和推理
  • 使用text-generation-webui搭建界面

PyTorch版本权重(.pth文件)可用于:

  • 使用llama.cpp工具进行量化和部署

PyTorch版本权重(.pth文件)链接,8-bit量化版的Alpaca-Plus-7B: Billsfriend/chinese-Alpaca-7b-plus-ggml-q8_0

模型文件组成:

chinese-alpaca-plus-7b-hf
    config.json
    generation_config.json
    pytorch_model-00001-of-00002.bin
    pytorch_model-00002-of-00002.bin
    pytorch_model.bin.index.json
    special_tokens_map.json
    tokenizer.json
    tokenizer.model
    tokenizer_config.json

硬件要求:14G显存

微调数据集

我整理部分公开微调数据集:

  1. 50万条中文ChatGPT指令Belle数据集: BelleGroup/train_0.5M_CN
  2. 100万条中文ChatGPT指令Belle数据集: BelleGroup/train_1M_CN
  3. 5万条英文ChatGPT指令Alpaca数据集: 50k English Stanford Alpaca dataset
  4. 5万条中文GPT4指令Alpaca数据集: shibing624/alpaca-zh
  5. 69万条中文指令Guanaco数据集(Belle50万条+Guanaco19万条): Chinese-Vicuna/guanaco_belle_merge_v1.0

如果需要训练LLaMA模型,请参考 https://github.com/shibing624/textgen

Citation
@software{textgen,
  author = {Xu Ming},
  title = {textgen: Implementation of language model finetune},
  year = {2023},
  url = {https://github.com/shibing624/textgen},
}
Reference

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric Value
Avg. 42.46
ARC (25-shot) 49.23
HellaSwag (10-shot) 70.48
MMLU (5-shot) 38.39
TruthfulQA (0-shot) 39.72
Winogrande (5-shot) 70.09
GSM8K (5-shot) 0.68
DROP (3-shot) 28.61

Runs of shibing624 chinese-alpaca-plus-7b-hf on huggingface.co

1.5K
Total runs
39
24-hour runs
88
3-day runs
262
7-day runs
381
30-day runs

More Information About chinese-alpaca-plus-7b-hf huggingface.co Model

More chinese-alpaca-plus-7b-hf license Visit here:

https://choosealicense.com/licenses/other

chinese-alpaca-plus-7b-hf huggingface.co

chinese-alpaca-plus-7b-hf huggingface.co is an AI model on huggingface.co that provides chinese-alpaca-plus-7b-hf's model effect (), which can be used instantly with this shibing624 chinese-alpaca-plus-7b-hf model. huggingface.co supports a free trial of the chinese-alpaca-plus-7b-hf model, and also provides paid use of the chinese-alpaca-plus-7b-hf. Support call chinese-alpaca-plus-7b-hf model through api, including Node.js, Python, http.

chinese-alpaca-plus-7b-hf huggingface.co Url

https://huggingface.co/shibing624/chinese-alpaca-plus-7b-hf

shibing624 chinese-alpaca-plus-7b-hf online free

chinese-alpaca-plus-7b-hf huggingface.co is an online trial and call api platform, which integrates chinese-alpaca-plus-7b-hf's modeling effects, including api services, and provides a free online trial of chinese-alpaca-plus-7b-hf, you can try chinese-alpaca-plus-7b-hf online for free by clicking the link below.

shibing624 chinese-alpaca-plus-7b-hf online free url in huggingface.co:

https://huggingface.co/shibing624/chinese-alpaca-plus-7b-hf

chinese-alpaca-plus-7b-hf install

chinese-alpaca-plus-7b-hf is an open source model from GitHub that offers a free installation service, and any user can find chinese-alpaca-plus-7b-hf on GitHub to install. At the same time, huggingface.co provides the effect of chinese-alpaca-plus-7b-hf install, users can directly use chinese-alpaca-plus-7b-hf installed effect in huggingface.co for debugging and trial. It also supports api for free installation.

chinese-alpaca-plus-7b-hf install url in huggingface.co:

https://huggingface.co/shibing624/chinese-alpaca-plus-7b-hf

Url of chinese-alpaca-plus-7b-hf

chinese-alpaca-plus-7b-hf huggingface.co Url

Provider of chinese-alpaca-plus-7b-hf huggingface.co

shibing624
ORGANIZATIONS

Other API from shibing624

huggingface.co

Total runs: 68
Run Growth: -16
Growth Rate: -22.86%
Updated: February 19 2024