shibing624 / chinese-llama-plus-13b-hf

huggingface.co
Total runs: 1.1K
24-hour runs: 0
7-day runs: 247
30-day runs: -52
Model's Last Updated: December 15 2023
text-generation

Introduction of chinese-llama-plus-13b-hf

Model Details of chinese-llama-plus-13b-hf

Chinese LLaMA Plus 13B Model

发布中文LLaMA-Plus, Alpaca-Plus 13B版本模型

发布中文LLaMA-Plus, Alpaca-Plus 13B版本,改进点如下:

  • 相比基础版进一步扩充了训练数据,其中LLaMA扩充至120G文本,Alpaca扩充至4.3M指令数据,重点增加了科学领域数据,涵盖:物理、化学、生物、医学、地球科学等
  • Alpaca训练时采用了更大的rank,相比基础版具有更低的验证集损失
  • Alpaca评测结果:13B获得74.3分,Plus-7B获得78.2分,Plus-13B获得80.8分,具体评测结果请参考 效果评测
  • 多轮回复长度相比旧模型提升明显(可适当增大温度系数)
  • 知识问答、写作、翻译等方面效果显著提升

本模型是 decapoda-research/llama-13b-hf 底座模型 合并 ziqingyang/chinese-llama-plus-lora-13b LoRA权重, 并转化为HuggingFace版本权重(.bin文件),可以在此中文LLaMA模型上继续指令微调训练,LLaMA模型为底座模型,直接调用会效果不佳。

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中的气体分子散射了太阳光中的短波长蓝光,使得我们看到的天空呈现出蓝色。
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-llama-plus-13b-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-llama-plus-13b-hf')
model = LlamaForCausalLM.from_pretrained('shibing624/chinese-llama-plus-13b-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中的气体分子散射了太阳光中的短波长蓝光,使得我们看到的天空呈现出蓝色。
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-13b-hf 底座模型 合并 ziqingyang/chinese-llama-plus-lora-13b 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: shibing624/chinese-alpaca-plus-13b-pth

模型文件组成:

chinese-alpaca-plus-13b-hf
|-- config.json
|-- generation_config.json
|-- LICENSE
|-- pytorch_model-00001-of-00003.bin
|-- pytorch_model-00002-of-00003.bin
|-- pytorch_model-00003-of-00003.bin
|-- pytorch_model.bin.index.json
|-- README.md
|-- special_tokens_map.json
|-- tokenizer_config.json
`-- tokenizer.model

硬件要求:25G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. 41.06
ARC (25-shot) 46.25
HellaSwag (10-shot) 71.88
MMLU (5-shot) 40.74
TruthfulQA (0-shot) 39.89
Winogrande (5-shot) 73.09
GSM8K (5-shot) 0.53
DROP (3-shot) 15.08

Runs of shibing624 chinese-llama-plus-13b-hf on huggingface.co

1.1K
Total runs
0
24-hour runs
112
3-day runs
247
7-day runs
-52
30-day runs

More Information About chinese-llama-plus-13b-hf huggingface.co Model

More chinese-llama-plus-13b-hf license Visit here:

https://choosealicense.com/licenses/other

chinese-llama-plus-13b-hf huggingface.co

chinese-llama-plus-13b-hf huggingface.co is an AI model on huggingface.co that provides chinese-llama-plus-13b-hf's model effect (), which can be used instantly with this shibing624 chinese-llama-plus-13b-hf model. huggingface.co supports a free trial of the chinese-llama-plus-13b-hf model, and also provides paid use of the chinese-llama-plus-13b-hf. Support call chinese-llama-plus-13b-hf model through api, including Node.js, Python, http.

chinese-llama-plus-13b-hf huggingface.co Url

https://huggingface.co/shibing624/chinese-llama-plus-13b-hf

shibing624 chinese-llama-plus-13b-hf online free

chinese-llama-plus-13b-hf huggingface.co is an online trial and call api platform, which integrates chinese-llama-plus-13b-hf's modeling effects, including api services, and provides a free online trial of chinese-llama-plus-13b-hf, you can try chinese-llama-plus-13b-hf online for free by clicking the link below.

shibing624 chinese-llama-plus-13b-hf online free url in huggingface.co:

https://huggingface.co/shibing624/chinese-llama-plus-13b-hf

chinese-llama-plus-13b-hf install

chinese-llama-plus-13b-hf is an open source model from GitHub that offers a free installation service, and any user can find chinese-llama-plus-13b-hf on GitHub to install. At the same time, huggingface.co provides the effect of chinese-llama-plus-13b-hf install, users can directly use chinese-llama-plus-13b-hf installed effect in huggingface.co for debugging and trial. It also supports api for free installation.

chinese-llama-plus-13b-hf install url in huggingface.co:

https://huggingface.co/shibing624/chinese-llama-plus-13b-hf

Url of chinese-llama-plus-13b-hf

chinese-llama-plus-13b-hf huggingface.co Url

Provider of chinese-llama-plus-13b-hf huggingface.co

shibing624
ORGANIZATIONS

Other API from shibing624