Blossom-v3.1-mistral-7b Overview

Blossom-v3.1-mistral-7b is a 7 billion parameter conversational large language model developed by Azure99, built upon the Mistral-7B-v0.1 pre-trained model. It is specifically fine-tuned for robust general capabilities and strong context comprehension in conversational AI.

Key Capabilities

• Conversational AI: Excels in multi-turn dialogue and instruction-following, trained on a diverse dataset including Orca, Wizard, and Chat data.
• Mathematical Reasoning: Incorporates a dedicated 2K Blossom math reasoning dataset in its second training stage, enhancing its ability to handle mathematical queries.
• Multilingual Support: Benefits from high-quality Chinese and English datasets, making it effective for both languages, though a Baichuan2-based variant is recommended for purely Chinese scenarios.
• Efficient Training: Underwent a two-stage training process, starting with 100K single-turn instruction datasets (Wizard, Orca) for 1 epoch, followed by 3 epochs on a mix of math reasoning, multi-turn chat, and sampled data.

Good For

• Developing chatbots and conversational agents requiring strong general understanding.
• Applications needing robust context comprehension in dialogue.
• Tasks involving mathematical reasoning and problem-solving.
• Use cases requiring a balance of English and Chinese language capabilities in a conversational setting.