saucam/mistral-orpo-beta-NeuralBeagle14-7B-dare-ties

TEXT GENERATIONConcurrency Cost:1Model Size:7BQuant:FP8Ctx Length:4kPublished:Mar 16, 2024License:apache-2.0Architecture:Transformer Open Weights Cold

saucam/mistral-orpo-beta-NeuralBeagle14-7B-dare-ties is a 7 billion parameter language model created by saucam, formed by merging kaist-ai/mistral-orpo-beta and mlabonne/NeuralBeagle14-7B using the DARE TIES method. This model combines the strengths of its base components, leveraging the ORPO fine-tuning approach from kaist-ai/mistral-orpo-beta and the general capabilities of NeuralBeagle14-7B. It is designed for general text generation tasks, offering a balanced performance profile derived from its merged architecture.

Loading preview...

Model Overview

saucam/mistral-orpo-beta-NeuralBeagle14-7B-dare-ties is a 7 billion parameter language model developed by saucam. It is a merged model, combining two distinct base models: kaist-ai/mistral-orpo-beta and mlabonne/NeuralBeagle14-7B. The merge was performed using the DARE TIES method, a technique for combining multiple models to leverage their individual strengths.

Key Characteristics

  • Merged Architecture: Integrates the capabilities of kaist-ai/mistral-orpo-beta, which likely incorporates ORPO (Odds Ratio Preference Optimization) fine-tuning, and mlabonne/NeuralBeagle14-7B.
  • DARE TIES Method: Utilizes a specific merging algorithm to combine the weights of the constituent models, with kaist-ai/mistral-orpo-beta contributing 60% and mlabonne/NeuralBeagle14-7B contributing 40% of the density and weight.
  • Configuration: The merge configuration specifies int8_mask: true and dtype: bfloat16, indicating potential optimizations for efficiency and performance.

Usage

This model can be used for various text generation tasks. The provided usage example demonstrates how to load the model and tokenizer using the transformers library and generate text based on a user prompt. It is suitable for applications requiring a capable 7B parameter model that benefits from the combined training of its base components.