A collection of a cleaned collection of tool-calling and reasoning conversations derived from the hermes_reasoning_tool_use source. It is structured specifically for Axolotl fine-tuning using a chat-based format that organizes interactions into system, user, and assistant roles.
Use Cases
- Fine-tune large language models for function calling by training on the 'conversations' sequence.
- Train models to follow system-level constraints defined in the 'from': 'system' message blocks.
- Develop reasoning capabilities by processing multi-turn dialogues within the 'value' fields.
Strengths
- Uses a 'chat_template' structure compatible with Axolotl fine-tuning workflows.
- Features a 'conversations' field containing nested message objects for multi-turn dialogue.
- Maps message roles and content using 'from' and 'value' keys respectively within the message_property_mappings.
- Derived from the interstellarninja/hermes_reasoning_tool_use dataset with specific cleaning for training stability.