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

7B

227次引用 4个月前更新

自述文件

DolphinCoder StarCoder2 7b 🐬

latitude.sh 赞助。

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

此模型基于StarCoder2-7b,并受bigcode-openrail-m许可的约束。

这只海豚在编程方面真的是很优秀,我使用了很多编程数据来训练。

此模型未经过审查。我已经过滤了数据集,以移除对齐和偏差。这样做使模型更合规。您在将模型作为服务公开之前,建议实现自己的对齐层。它将对任何请求都高度合规,甚至包括不道德的请求。请阅读我关于未审查模型的博客文章。https://erichartford.com/uncensored-models
您对使用此模型创建的任何内容负责。负责任地享受。

训练

使用qLoRA和Axolotl在8x L40S上训练3个周期花了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

  • 该模型得以实现,得益于 latitude.sh 的慷慨赞助。
  • 欢迎Microsoft加入开源AI队伍!感谢你贡献Orca-Math数据集!
  • 非常感谢 BigCode 为培训并发布 StarCoder2 的权重
  • 向数据集作者表示衷心的感谢:@ise-uiuc、@teknium、@m-a-p
  • 并且向 @winglian 和 Axolotl 的贡献者们表示最大的感谢,他们提供了最好的训练框架!
  • Built with Axolotl
  • 感谢开源AI社区中所有在我学习和成长过程中给予我指导和帮助的人们。

示例输出

如果您想资助我的工作

周边产品