GitHub - salesforce/CodeGen: CodeGen is a family of open-source model for program synthesis. Trained on TPU-v4. Competitive with OpenAI Codex.
🚀 Exciting AI Tool Alert! 🤖 Introducing CodeGen by Salesforce AI Research! 🌟 Trained on TPU-v4, this open-source model family excels at program synthesis, rivaling OpenAI Codex. 🎯 Accessible on the Hugging Face Hub for seamless implementation. 🔥 #AI #CodeGen #ProgramSynthesis
- The CodeGen repository by Salesforce AI Research contains models such as CodeGen1, CodeGen2, and CodeGen2.5 for Program Synthesis.
- CodeGen2.5 has shown superior performance compared to 16B models while having only 7B parameters.
- CodeGen2.0 boasts strong infill sampling capability, while CodeGen1.0 was released on par with OpenAI Codex in its time.
- These models are accessible on the Hugging Face Hub for implementation.
- Training of CodeGen models involves using the Jaxformer library for data preprocessing, training, and fine-tuning.
- Citations for the CodeGen papers are provided for any usage.
- CodeGen is an open-source model family focused on program synthesis, competitive with OpenAI Codex.
- The models are trained on TPU-v4 and cover topics such as codex, language model, TPU acceleration, generative model, and program synthesis.