README.md

DataLinguistic-70B-4bit-V1.0 Chinese-English Question Answering Model

Model Overview

DataLinguistic-70B-4bit-V1.0 is a Chinese-English question answering model fine-tuned from Huggingface's Llama-2-70b-chat-hf model with 4-bit quantization on DataLinguistic's proprietary datasets.

Model Architecture

DataLinguistic-70B-4bit-V1.0 inherits the encoder-decoder structure from Llama with 70B parameters. On top of that, it is quantized to 4-bit precision before fine-tuning.

Training Datasets

The model was trained on the following open-source datasets:

• Data_OpenSet: Chinese-English question-answering dataset curated from junelee/wizard_vicuna_70k
• Data_OpenSet2: Chinese-English question-answering dataset curated from garage-bAInd/Open-Platypus
• Proprietary Chinese-English question-answering dataset collected internally by DataLinguistic (not open-sourced)

The data is formatted as:

<s>please answer my question in datalynn model: {question} ANS: {answer}</s>

Use Cases

The model can be used for a wide range of Chinese-English question answering and chatbot applications.

Model Advantages

• Based on huge model Llama-70b with 70B parameters
• 4-bit quantization reduces compute
• Fine-tuned on large-scale Chinese-English QA datasets for high quality

Usage

1. Install model from Huggingface
2. Import and initialize model
3. Input question, generate answer

Version

Current version: DataLinguistic-70B-4bit-V1.0

Author

Tang Zhengzheng

Contributors

DataLinguistic team

DataLinguistic-70B-4bit-V1.0 中英文问答模型

模型简介

DataLinguistic-70B-4bit-V1.0是一个基于Huggingface的Llama-2-70b-chat-hf模型进行4bit量化并在DataLinguistic自建数据集上微调的中文英文问答模型。

模型结构

DataLinguistic-70B-4bit-V1.0 inherits the encoder-decoder structure from Llama with 70B parameters. On top of that, it is quantized to 4-bit precision before fine-tuning.

模型训练数据集

模型使用了以下开源数据集进行了训练:

• Data_OpenSet: 基于junelee/wizard_vicuna_70k整理的中英文问答数据集
• Data_OpenSet2: 基于garage-bAInd/Open-Platypus整理的中英文问答数据集
• DataLinguistic内部收集的专属中英文问答数据集(未开源)

数据集采用如下格式:

<s>please answer my question in datalynn model: {问题} ANS: {回答}</s>

应用场景

该模型可广泛应用于中英文问答、聊天机器人等场景。

模型优势

• 基于大模型Llama-70b,参数量达70亿
• 采用4bit量化,降低运算量
• 在大规模中英文问答数据集上进行微调,质量较高

使用步骤

1. 在Huggingface安装模型
2. 导入并初始化模型
3. 输入问题,生成回答

版本信息

当前版本:DataLinguistic-70B-4bit-V1.0

作者

唐正正

贡献者

DataLinguistic团队