Vigogne V2.2

• Refactor training code
• Introduce the Vigostral-Chat model

Vigogne V2.1

• Integrate QLoRA support for improved training efficiency
• Integrate FlashAttention support and implement training example packing
• Implement generation script for Orca-style data
• Add support for serving using FastChat and vLLM
• Introduce Vigogne-Chat-V2.0 models (blog)
• Include LangChain integration

Vigogne V2.0

• Integrate data generation scripts for self-instruct, self-chatting, and translation
• Expand the training scripts to include chat-optimized models, and incorporate DeepSpeed support for more efficient training
• Release Vigogne-Instruct v2 models trained on a more extensive dataset
• Release Vigogne-Chat models and introduce the Gradio Demo for them

Vigogne V1.1

• Enhance the translation quality of the Stanford Alpaca dataset
• Update the Gradio demo for Vigogne-Instruct models to include streaming mode

Vigogne V1.0

• Share a French translation of the Stanford Alpaca dataset (the quality may not be perfect)
• Share fine-tuned LLaMA and BLOOM models
• Add instructions for deploying using llama.cpp