pankajmathur/orca_mini_v3_7b

Hugging Face
TEXT GENERATIONConcurrency Cost:1Model Size:7BQuant:FP8Ctx Length:4kPublished:Aug 7, 2023License:otherArchitecture:Transformer0.0K Warm

pankajmathur/orca_mini_v3_7b is a 7 billion parameter Llama 2-based causal language model fine-tuned by Pankaj Mathur on Orca-style datasets. This model is designed to follow instructions effectively, leveraging its training on complex explanation traces. It demonstrates competitive performance on various benchmarks, including ARC, HellaSwag, and MMLU, making it suitable for general instruction-following tasks.

Loading preview...

Model Overview

pankajmathur/orca_mini_v3_7b is a 7 billion parameter language model built upon the Llama 2 architecture, fine-tuned by Pankaj Mathur. Its training incorporates Orca-style datasets, which are known for emphasizing complex explanation traces, aiming to enhance the model's instruction-following capabilities.

Key Capabilities & Performance

This model is designed to be an effective AI assistant, excelling at following instructions. Evaluation results from the HuggingFace Open LLM Leaderboard indicate its performance across several benchmarks:

  • ARC Challenge: 57.17% (acc_norm)
  • HellaSwag: 79.66% (acc_norm)
  • MMLU: 52.34% (acc_norm)
  • TruthfulQA: 50.29% (mc2)

These metrics suggest a balanced capability across common reasoning, common sense, and knowledge-based tasks for its size class.

Use Cases

This model is particularly well-suited for:

  • Instruction Following: Generating responses based on explicit user instructions.
  • General Chatbot Applications: Serving as an interactive AI assistant.
  • Exploration of Open-Source LLMs: Providing a performant 7B parameter option for developers interested in Llama 2 derivatives with enhanced instruction-following.

Limitations

As with many LLMs, orca_mini_v3_7b may occasionally produce inaccurate or misleading information. There is also a possibility of generating inappropriate, biased, or offensive content, necessitating careful usage and cross-verification of outputs.