Dataset · interstellarninja
Hermes: tool-calling training data
interstellarninja/hermes_reasoning_tool_use
51,004 ShareGPT dialogues on when and how LLMs should call tools.
What's inside
The dataset contains 51,004 ShareGPT dialogues that teach LLMs when, how, and whether to call tools. It is gathered using the Nous Research Atropos RL stack and a custom MultiTurnToolCallingEnv, aligned with BFCL v3 evaluation scenarios. Rows are OpenAI-style conversations; Apache-2.0 license, size 392 MB.
What you can do with it
- Training tool-calling LoRA
- RL and RLHF for tool-using agents
- Evaluation on BFCL v3 scenarios
- Tasks
- Question answering
- Languages
- English
- Size
- 10K–100K rows
- Downloads this month
- 4.2K
- Likes
- 210
Models trained on it
- ManniX-ITA/gemma-4-26B-A4B-agentic-GGUFText generation 15.8K
- mradermacher/gemma-4-26B-A4B-agentic-i1-GGUF 9.2K
- cybertruck32489/LFM3-1.7B-RDText generation 3.9K
- ManniX-ITA/gemma-4-26B-A4B-agenticImage text → text 2.3K
- mradermacher/gemma-4-26B-A4B-agentic-GGUF 2.2K
- cybertruck32489/LFM3-1.7B-RD-v2Text generation 1.8K
- mradermacher/TinyBalls-110M-V1-GGUFText generation 445
- br1-pist/Qwen3-4B-AgentCoderText generation 444
How to load it
from datasets import load_dataset
ds = load_dataset("interstellarninja/hermes_reasoning_tool_use")