Implementation:Hiyouga LLaMA Factory Glaive Toolcall En Demo Data

Knowledge Sources	Hiyouga_LLaMA_Factory
Domains	NLP, Training_Data
Last Updated	2026-02-06 19:00 GMT

Overview

glaive_toolcall_en_demo.json provides English tool/function calling demo data in ShareGPT format with tool definitions and multi-turn conversations for training models to invoke external functions.

Description

The file contains a JSON array of records, each with a conversations list and a tools string. The conversations follow ShareGPT format using role tags "from" with values "human", "gpt", "function_call", and "observation". The "function_call" turns contain JSON-serialized tool invocations with function name and arguments, while "observation" turns contain the JSON-serialized tool response. The tools field is a JSON string defining available function schemas following the OpenAI function calling format.

Some conversations include tool calls while others are purely conversational, demonstrating that the model should learn when to invoke tools and when to respond directly.

Usage

This demo dataset trains and tests function-calling capabilities in LLaMA Factory. Users reference it by name (glaive_toolcall_en_demo) with --stage sft to fine-tune models for tool use. The tool definitions in the tools field are injected as part of the system context during training.

Code Reference

Source Location

Repository: Hiyouga_LLaMA_Factory
File: data/glaive_toolcall_en_demo.json

Data Format

[
  {
    "conversations": [
      {
        "from": "human",
        "value": "Hi, I have some ingredients and I want to cook something. Can you help me find a recipe?"
      },
      {
        "from": "gpt",
        "value": "Of course! I can help you with that. Please tell me what ingredients you have."
      },
      {
        "from": "human",
        "value": "I have chicken, bell peppers, and rice."
      },
      {
        "from": "function_call",
        "value": "{\"name\": \"search_recipes\", \"arguments\": {\"ingredients\": [\"chicken\", \"bell peppers\", \"rice\"]}}"
      },
      {
        "from": "observation",
        "value": "{\"recipes\": [{\"name\": \"Chicken and Bell Pepper Stir Fry\", ...}]}"
      },
      {
        "from": "gpt",
        "value": "I found two recipes for you..."
      }
    ],
    "tools": "[{\"name\": \"search_recipes\", \"description\": \"Search for recipes based on ingredients\", \"parameters\": {\"type\": \"object\", \"properties\": {\"ingredients\": {\"type\": \"array\", \"items\": {\"type\": \"string\"}, \"description\": \"The ingredients to search for\"}}, \"required\": [\"ingredients\"]}}]"
  }
]

I/O Contract

Schema

Field	Type	Required	Description
conversations	array	Yes	List of conversation turns with `"from"` and `"value"` fields
tools	string	Yes	JSON-serialized array of tool/function definitions following OpenAI function schema

Conversation Roles

Role (from)	Description
human	User message
gpt	Assistant response (text only, no tool call)
function_call	Tool invocation with JSON payload containing `name` and `arguments`
observation	Tool execution result as JSON

Dataset Registry Entry

Property	Value
Key	`glaive_toolcall_en_demo`
file_name	`glaive_toolcall_en_demo.json`
formatting	sharegpt
columns.messages	conversations
columns.tools	tools
Lines	9158

Usage Examples

# Train a model with function-calling capabilities
# llamafactory-cli train \
#     --dataset glaive_toolcall_en_demo \
#     --stage sft \
#     --model_name_or_path meta-llama/Llama-2-7b-hf \
#     --output_dir output/toolcall_demo

# Loading and inspecting the data
import json

with open("data/glaive_toolcall_en_demo.json", "r", encoding="utf-8") as f:
    data = json.load(f)

print(f"Number of conversations: {len(data)}")
sample = data[0]
print(f"Turns in first conversation: {len(sample['conversations'])}")
tools = json.loads(sample['tools'])
print(f"Number of tools defined: {len(tools)}")
for tool in tools:
    print(f"  Tool: {tool['name']} - {tool['description']}")

Related Pages

Hiyouga_LLaMA_Factory_Glaive_Toolcall_Zh_Demo_Data - Chinese version of the tool-calling demo dataset
Hiyouga_LLaMA_Factory_Dataset_Info_Registry - Central dataset registry that indexes this file
Hiyouga_LLaMA_Factory_Alpaca_En_Demo_Data - English SFT demo data (non-tool-calling)

Page Connections

Double-click a node to navigate. Hold to expand connections.

Principle

Implementation

Heuristic

Environment