allenai/codetulu-2-34b

TEXT GENERATIONConcurrency Cost:2Model Size:34BQuant:FP8Ctx Length:32kPublished:Nov 13, 2023Architecture:Transformer0.0K Cold

allenai/codetulu-2-34b is a 34 billion parameter instruction-tuned language model developed by AllenAI, fine-tuned from CodeLlama-34b-hf. This model is part of the Tulu series, designed as a helpful assistant, and was trained on a diverse mix of publicly available, synthetic, and human datasets. It specializes in understanding and generating responses based on instruction-following tasks, particularly in English.

Loading preview...

Codetulu 2 34B: An Instruction-Tuned Code Assistant

Codetulu 2 34B, developed by AllenAI, is an instruction-tuned language model based on codellama/CodeLlama-34b-hf. It is part of the Tulu series, which focuses on creating helpful assistant models. This 34 billion parameter model was fine-tuned using a curated mix of publicly available, synthetic, and human-created datasets, specifically the Tulu V2 mix dataset.

Key Capabilities

  • Instruction Following: Designed to act as a helpful assistant, responding to diverse instructions.
  • Code-Oriented Foundation: Built upon CodeLlama, suggesting a strong foundation for code-related tasks.
  • English Language Focus: Primarily optimized for English language interactions.
  • Specific Input Format: Requires a precise input format (<|user|> Your message here! <|assistant|> ) for optimal generation quality.

Intended Use Cases

  • General Assistant Tasks: Suitable for a wide range of instruction-based queries.
  • Code-Related Assistance: Leveraging its CodeLlama base, it can be applied to programming support.

Limitations

  • Safety Alignment: The model has not undergone explicit safety alignment (e.g., RLHF for safe completions) and may produce problematic outputs if prompted.
  • License: Distributed under the AI2 ImpACT Low-risk license.