Hermes Function Calling v1
The structured-output and function-calling compilation used to train the Hermes 2 Pro series: ~12k ShareGPT-format samples across five task types — single-turn and multi-turn function calls, agentic JSON mode, single-turn JSON mode, and structured extraction. Includes a cleaned 5k Glaive subset. The go-to set when you need reliable JSON output as well as tool calls.
from datasets import load_dataset
ds = load_dataset("NousResearch/hermes-function-calling-v1")Preview a sample row
{
"conversations": [
{ "from": "system", "value": "You are a function calling AI model. Here are the available tools: <tools>[{\"name\": \"create_event\", ...}]</tools>" },
{ "from": "human", "value": "Schedule a team sync tomorrow at 10am." },
{ "from": "gpt", "value": "<tool_call>{\"name\": \"create_event\", \"arguments\": {\"title\": \"Team sync\", \"time\": \"10:00\"}}</tool_call>" }
]
}Fine-tune with this dataset
Estimated VRAM to fine-tune with QLoRA (4-bit base model + LoRA adapters), using conservative defaults:
Frequently asked questions
Can I use Hermes Function Calling v1 commercially?
Yes — Hermes Function Calling v1 is released under Apache 2.0, a permissive license that allows commercial use, including training models you ship in a product. Check the dataset card for attribution requirements before release.
How much data does Hermes Function Calling v1 contain, and do I need all of it?
Hermes Function Calling v1 contains 12k Samples. You rarely need all of it: for style and format fine-tuning, a few hundred to a few thousand examples are enough — load a slice (e.g. split="train[:1000]") and scale up only if quality plateaus.
What is Hermes Function Calling v1 best used for?
Models that must return valid structured JSON — for agents, extraction, and tool pipelines. It belongs to the Agentic & Function Calling section of our dataset hub, where you'll find alternatives and complementary sets.