基于 StarCoder2 7B 的编码能力卓越的 Dolphin 模型家族的无审查版本。

7B

227 拉取 更新于4个月前

README

DolphinCoder StarCoder2 7b 🐬

latitude.sh 赞助。

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

此模型基于 StarCoder2-7b,并符合 bigcode-openrail-m 许可。

这只海豚在编码方面非常出色,我用大量的编码数据进行了训练。

此模型未经过审查。我已经过滤了数据集以去除调准和偏见。这使得模型更合规。建议您在将模型作为服务公开之前实施自己的调准层。它将非常符合任何请求,甚至是不道德的请求。请阅读我关于未审查模型的博客文章。 https://erichartford.com/uncensored-models
您应对使用此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 8x L40S 上训练 3 个周期花费了 2 天。

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

感恩

  • 此模型得以实现得益于 latitude.sh 的慷慨赞助。
  • 欢迎微软加入开源 AI!感谢 Orca-Math 数据集!
  • 感谢 BigCode 为训练和发布 StarCoder2 权重。
  • 向数据集的作者致以巨大的谢意:@ise-uiuc, @teknium, @m-a-p
  • 特别感谢@winglian和Axolotl的贡献者,他们打造了最优秀的训练框架!
  • Built with Axolotl
  • 感谢开源AI社区中所有帮助我的其他人。

示例输出

如果您想 financially support我的工作

周边产品