基于StarCoder2 7B的Dolphin模型家族的无审查版本,擅长编码。

7B

227 Pulls 更新于4个月前

说明文档

DolphinCoder StarCoder2 7b 🐬

latitude.sh 赞助。

加入我们的Discord! https://discord.gg/cognitivecomputations

本模型基于StarCoder2-7b,并遵循bigcode-openrail-m许可证。

DolphinCoder 真的很擅长编码,我使用了大量编程数据来训练它。

此模型未经审查。我已经过滤数据集以去除对齐和偏差。这使得模型更加合规。建议在将模型公开作为服务之前自行实现对齐层。它将对任何请求高度合规,甚至包括不道德的请求。请阅读有关非审查模型的博客文章。https://erichartford.com/uncensored-models
您应对使用本模型创建的任何内容负责。请负责任地享受。

训练

使用qLoRA和Axolotl在8个L40S上训练3个epoch需要2天时间

提示格式
此模型使用ChatML提示格式。

<|im_start|>system
You are DolphinCoder, a helpful AI programming assistant.<|im_end|>
<|im_start|>user
{prompt}<|im_end|>
<|im_start|>assistant

示例

<|im_start|>system
You are DolphinCoder, a master at software engineering and coding in any programming language.
<|im_start|>user
Please write me a program in golang that parses all the lines in a file, and reverses them character-wise, and saves it to a new file.
<|im_start|>assistant

量化模型

感谢

  • 本模型得以实现,多亏了 latitude.sh 的慷慨赞助。
  • 欢迎Microsoft加入开源AI!感谢Orca-Math数据集!
  • BigCode 致以最衷心的感谢,它训练并发布了StarCoder2的权重。
  • 向数据集作者:@ise-uiuc, @teknium, @m-a-p 表示极大感谢。
  • 还要向 @winglian 和 Axolotl 贡献者表示极大感谢,他们创造了最佳训练框架!
  • Built with Axolotl
  • 感谢开源AI社区中所有教导和帮助过我的其他人。

示例输出

如果您想经济上支持我的工作

周边产品