allenai/open-instruct-unnatural-instructions-7b
The allenai/open-instruct-unnatural-instructions-7b model is a 7 billion parameter LLaMa-based language model developed by AllenAI. It has been fine-tuned on the Unnatural Instructions dataset, specializing in instruction-following tasks. This model is distributed as a model diff, requiring recovery with an original LLaMa model, and is suitable for research into instruction tuning on open resources.
Loading preview...
Open-Instruct Unnatural Instructions 7B: Instruction-Tuned LLaMa
This model, developed by AllenAI, is a 7 billion parameter LLaMa model fine-tuned specifically on the Unnatural Instructions dataset. It was created as part of the research detailed in the paper "How Far Can Camels Go? Exploring the State of Instruction Tuning on Open Resources" (arXiv:2306.04751).
Key Capabilities & Features
- Instruction Following: Optimized for understanding and executing instructions based on the Unnatural Instructions dataset.
- LLaMa Architecture: Built upon the LLaMa foundation model, leveraging its robust language understanding capabilities.
- Model Diff Distribution: Distributed as a weight difference, requiring users to recover the full model using an existing LLaMa base model and a provided script.
- Specific Input Format: Designed to work optimally with a defined input format:
<|user|> Your message here! <|assistant|>.
Performance Highlights
Performance metrics across various benchmarks are provided, including:
- MMLU (0-shot/5-shot): 42.9 / 38.1
- GSM (Direct/CoT): 3.5 / 5.0
- BBH (Direct/CoT): 31.4 / 30.0
- TydiQA (Gold-Passage/Closed-book): 36.3 / 6.5
- Codex-Eval (Pass@1/Pass@10): 10.3 / 19.8
- AlpacaFarm vs Davinci-003: 8.2
Good for
- Researchers exploring instruction tuning techniques and their effectiveness on open resources.
- Developers interested in deploying a LLaMa-based model with enhanced instruction-following abilities.
- Experimentation with models trained on the Unnatural Instructions dataset.