pankajmathur/model_101

Model_101: A Hybrid Llama2-70B for Explain & Instruct Tasks

pankajmathur/model_101 is a 69 billion parameter model built upon the Llama2 architecture, developed by Pankaj Mathur. It is specifically designed as a "hybrid" model, capable of both explaining concepts and following instructions effectively. This dual capability is achieved through fine-tuning on a comprehensive collection of datasets, including Open-Platypus, Alpaca, WizardLM, Dolly-V2, Dolphin Samples, Orca_minis_v1, and Alpaca_orca, among others.

Key Capabilities & Performance

This model demonstrates strong performance across various benchmarks, as evaluated using the EleutherAI Language Model Evaluation Harness. Key results from the HuggingFaceH4 Open LLM Leaderboard include:

• ARC (25-shot): 68.69
• HellaSwag (10-shot): 86.42
• MMLU (5-shot): 69.92
• TruthfulQA (0-shot): 58.85
• Winogrande (5-shot): 82.08
• GSM8K (5-shot): 44.81

The model supports both Orca and Alpaca prompt formats, providing flexibility for different instruction-following paradigms. It requires significant GPU VRAM (up to 45GB in 4-bit quantization) for optimal performance.

Ideal Use Cases

• General-purpose instruction following: Excels at understanding and executing a wide array of user commands.
• Explanatory tasks: Capable of generating detailed and coherent explanations on various topics.
• Conversational AI: Suitable for applications requiring nuanced dialogue and response generation.
• Research and development: Provides a robust base for further fine-tuning or experimentation with Llama2-based models.