Browse Models
The simplest way to self-host Mistral 7B OpenOrca. Launch a dedicated cloud GPU server running Lab Station OS to download and serve the model using any compatible app or framework.
Download model weights for local inference. Must be used with a compatible app, notebook, or codebase. May run slowly, or not work at all, depending on your system resources, particularly GPU(s) and available VRAM.
Mistral-7B-OpenOrca is a fine-tuned version of Mistral-7B using the OpenOrca dataset. It shows notable improvements over the base model, with 129% better AGI Eval scores and 119% better BigBench-Hard performance. Uses ChatML format and is available in multiple quantized versions.
Mistral-7B-OpenOrca represents a significant advancement in open-source language models, built upon the foundation of the Mistral 7B base model and fine-tuned using the OpenOrca dataset. This dataset was created to replicate the methodology described in Microsoft Research's Orca paper, which demonstrated the effectiveness of learning from complex explanation traces and step-by-step thought processes.
The model utilizes OpenChat packing and was trained using the Axolotl framework. The training process involved 8x A6000 GPUs over 62 hours, with an approximate commodity cost of $400. The model implements OpenAI's Chat Markup Language (ChatML) format, using <|im_start|>
and <|im_end|>
tokens, ensuring compatibility with various tools including oobabooga and Hugging Face's Transformers chat template.
At the time of its release, Mistral-7B-OpenOrca achieved remarkable performance metrics, leading the Hugging Face Leaderboard among models smaller than 30B parameters. The model demonstrated significant improvements over its base version:
The model's efficiency allows it to run on consumer-grade GPUs, making it particularly accessible for researchers and developers. Several quantized versions are available through TheBloke on Hugging Face: