wvnvwn/Mistral-7B-Instruct-v0.3-flora-v0

Hugging Face
TEXT GENERATIONConcurrency Cost:1Model Size:7BQuant:FP8Ctx Length:4kPublished:May 21, 2026Architecture:Transformer Warm

The wvnvwn/Mistral-7B-Instruct-v0.3-flora-v0 is a 7 billion parameter instruction-tuned causal language model, based on the Mistral-7B-Instruct-v0.3 architecture. This model was created by wvnvwn through federated LoRA fine-tuning and adapter aggregation, resulting in a full merged model for reproducible evaluation. It is specifically fine-tuned using the 'flora' algorithm on a heterogeneous dataset across 8 clients, making it suitable for tasks benefiting from federated learning approaches.

Loading preview...

Model Overview

This model, wvnvwn/Mistral-7B-Instruct-v0.3-flora-v0, is a 7 billion parameter instruction-tuned causal language model. It is derived from the mistralai/Mistral-7B-Instruct-v0.3 base model, enhanced through a unique federated learning approach.

Key Characteristics

  • Federated LoRA Fine-tuning: The model was fine-tuned using the flora algorithm, involving 8 clients over 3 communication rounds, each with 3 local epochs. This method aggregates adapters from distributed training, then merges them into the base model.
  • Merged Full Model: Unlike adapter-only checkpoints, this repository provides a full merged model, simplifying deployment and ensuring reproducible evaluation without needing separate adapter loading.
  • Base Model Compatibility: It retains the core capabilities of the Mistral-7B-Instruct-v0.3 architecture, making it suitable for general instruction-following tasks.

Potential Use Cases

  • Research in Federated Learning: Ideal for researchers and developers exploring the practical application and evaluation of models trained via federated learning techniques.
  • Instruction Following: Suitable for various instruction-based natural language processing tasks, leveraging its Mistral-7B-Instruct foundation.
  • Reproducible Evaluation: Designed for straightforward and consistent evaluation due to its fully merged nature, eliminating complexities associated with adapter loading.