pankajmathur/model_007
pankajmathur/model_007 is a 69 billion parameter Llama2-70b based language model developed by Pankaj Mathur, fine-tuned for both explanatory and instructional tasks. It leverages a diverse dataset including Open-Platypus, Alpaca, WizardLM, and Orca samples to achieve a hybrid response style. This model is designed for general-purpose instruction following and explanation generation, demonstrating competitive performance across various benchmarks.
Loading preview...
Model_007: A Hybrid Llama2-70b for Explain and Instruct Tasks
pankajmathur/model_007 is a 69 billion parameter language model built upon the Llama2-70b architecture, developed by Pankaj Mathur. This model is uniquely designed to handle both explanatory and instructional prompts, offering a versatile response style. It was fine-tuned using a comprehensive collection of datasets, including Open-Platypus, Alpaca, WizardLM, Dolly-V2, Dolphin Samples, Orca_minis_v1, Alpaca_orca, and WizardLM_orca, to achieve its hybrid capabilities.
Key Capabilities & Performance
Model_007 demonstrates solid performance across a range of tasks, as evaluated using the EleutherAI Language Model Evaluation Harness and reported on the HuggingFaceH4 Open LLM Leaderboard. Key benchmark results include:
- ARC: 0.7108
- HellaSwag: 0.8765
- MMLU: 0.6904
- TruthfulQA: 0.6312
- Winogrande: 0.8335
- Total Average: 0.6320
Prompt Formats
The model supports both Orca and Alpaca prompt formats, allowing flexibility in how users interact with it. Examples for both formats are provided, along with code instructions for integration using the transformers library.
Usage Considerations
This model requires significant GPU VRAM (up to 45GB for 4-bit loading) and is compatible with OobaBooga Web UI. Users should be aware of the license and usage restrictions inherited from the original Llama-2 model. As with all large language models, it may occasionally produce inaccurate, biased, or offensive content, and users are advised to exercise caution and cross-check information.