GenerativeMagic/Llama-Engineer-Evol-7b

TEXT GENERATIONConcurrency Cost:1Model Size:7BQuant:FP8Ctx Length:4kLicense:cc-by-sa-4.0Architecture:Transformer0.0K Open Weights Cold

GenerativeMagic/Llama-Engineer-Evol-7b is a 7 billion parameter Llama 2 chat instruction-tuned model, further fine-tuned on over 80,000 coding samples. This model specializes in code generation and programming assistance, leveraging datasets like Evol-Instruct-Code-80k-v1 and GPTeacher. It is optimized for direct and helpful responses to coding prompts, with a context length of 4096 tokens.

Loading preview...

Llama-Engineer-Evol-7B: Code-Optimized Llama 2 Variant

GenerativeMagic's Llama-Engineer-Evol-7B is a 7 billion parameter model built upon Meta's Llama 2 chat instruction-tuned architecture. Its primary differentiation lies in extensive fine-tuning on a large corpus of over 80,000 coding samples, specifically designed to enhance its programming capabilities. The training data includes the Evol-Instruct-Code-80k-v1 dataset, a replication of the Evol-Instruct-Code methodology described in the WizardCoder paper, and Teknium's GPTeacher Codegen instructions.

Key Capabilities

  • Enhanced Code Generation: Specialized fine-tuning makes it proficient in generating code and assisting with programming tasks.
  • Direct and Concise Responses: The recommended prompt format emphasizes direct, helpful, and short responses, ideal for quick programming queries.
  • Llama 2 Base: Benefits from the robust foundation of the Llama 2 architecture.

Good For

  • Programming Assistance: Ideal for developers seeking quick code snippets, debugging help, or general programming guidance.
  • Integration into Developer Tools: Suitable for applications requiring an LLM focused on code-related interactions.

While formal benchmarks are planned for future iterations, the model's development focused on learning and refining the fine-tuning process for code-centric applications. GGML weights are also available for broader deployment.