A ready-to-train dataset of 18,487 multi-turn tool-calling conversations in ChatML messages format, combining Hermes Agent reasoning traces with NousResearch function-calling data.
Built for SFT training of tool-calling / agentic LLMs with TRL's SFTTrainer.
Quick Start
from datasets import load_dataset
from trl import SFTTrainer