Mustafaege/qwen3.5-functioncalling-v1

---
language:
- en
license: apache-2.0
pretty_name: Qwen3.5 Function Calling Dataset v1
size_categories:
- 100K<n<1M
task_categories:
- text-generation
tags:
- function-calling
- tool-use
- sft
- chat
- qwen3
- qwen3.5
- instruction-following
- structured-output
- json
- fine-tuning
- open-source
annotations_creators:
- machine-generated
language_creators:
- found
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
- split: test
path: data/test-*
---

# Qwen3.5 Function Calling Dataset v1

A curated function-calling SFT dataset built from **glaiveai/glaive-function-calling-v2**, converted and standardized into Qwen3 messages format for fine-tuning Qwen3.5 series models.

## Dataset Summary

| Property | Value |
|----------|-------|
| **Total Samples** | 112,960 |
| **Train Split** | 101,664 |
| **Test Split** | 11,296 |
| **Source** | glaiveai/glaive-function-calling-v2 |
| **Format** | Qwen3 messages |
| **Language** | English |
| **License** | Apache 2.0 |

## What is Function Calling?

Function calling allows language models to produce structured JSON outputs to invoke external APIs, tools, or services. A model learns to:
- Parse a user request and identify the appropriate function
- Extract the correct arguments from the conversation
- Return a valid JSON function call

## Dataset Structure

### Data Fields

| Field | Type | Description |
|-------|------|-------------|
| `messages` | `list[dict]` | Conversation turns with `role` and `content` |

### Role Types

| Role | Description |
|------|-------------|
| `system` | Function definitions and assistant instructions |
| `user` | User requests triggering function calls |
| `assistant` | Model responses with function call JSON or natural language |

## Format

All samples are standardized to Qwen3's native messages format:

```json
{
"messages": [
{
"role": "system",
"content": "You are a helpful assistant with access to the following functions. Use them if required -\n{\n \"name\": \"get_exchange_rate\",\n \"description\": \"Get the exchange rate between two currencies\",\n \"parameters\": {\n \"type\": \"object\",\n \"properties\": {\n \"base_currency\": {\"type\": \"string\"},\n \"target_currency\": {\"type\": \"string\"}\n },\n \"required\": [\"base_currency\", \"target_currency\"]\n }\n}"
},
{
"role": "user",
"content": "Can you tell me the current exchange rate from US dollars to Euros?"
},
{
"role": "assistant",
"content": "<functioncall> {\"name\": \"get_exchange_rate\", \"arguments\": {\"base_currency\": \"USD\", \"target_currency\": \"EUR\"}}"
},
{
"role": "user",
"content": "FUNCTION RESPONSE: {\"exchange_rate\": 0.85}"
},
{
"role": "assistant",
"content": "The current exchange rate from US dollars to Euros is 0.85. This means 1 USD equals 0.85 EUR."
}
]
}
```

## Source Conversion

| Source Dataset | Original Format | Conversion |
|----------------|----------------|------------|
| [glaiveai/glaive-function-calling-v2](https://huggingface.co/datasets/glaiveai/glaive-function-calling-v2) | `system` + `chat` (USER:/ASSISTANT: pattern) | Split on role markers → messages list |

### Conversion Logic

```python
# system field -> {"role": "system", "content": ...}
# chat field:
# "USER: ..." -> {"role": "user", "content": ...}
# "ASSISTANT: ..." -> {"role": "assistant", "content": ...}
```

## Usage

```python
from datasets import load_dataset

dataset = load_dataset("Mustafaege/qwen3.5-functioncalling-v1")

print(dataset)
# DatasetDict({
# train: Dataset({features: ['messages'], num_rows: 101664}),
# test: Dataset({features: ['messages'], num_rows: 11296})
# })

# Iterate samples
for example in dataset['train']:
messages = example['messages']
system_prompt = messages[0]['content']
user_msg = messages[1]['content']
print(f"Functions defined: {len(system_prompt.split('name'))-1}")
break
```

## Training with Unsloth

```python
from unsloth import FastLanguageModel
from unsloth.chat_templates import get_chat_template

model, tokenizer = FastLanguageModel.from_pretrained(
model_name = "unsloth/Qwen3-0.6B",
max_seq_length = 2048,
load_in_4bit = True,
)

# Apply chat template
tokenizer = get_chat_template(tokenizer, chat_template="qwen-3")

from trl import SFTTrainer
trainer = SFTTrainer(
model = model,
tokenizer = tokenizer,
train_dataset = dataset['train'],
dataset_text_field = "messages",
)
```

## Related Datasets

| Version | Samples | New Sources | Link |
|---------|---------|-------------|------|
| **v1** (this) | 112,960 | glaive-function-calling-v2 | [Mustafaege/qwen3.5-functioncalling-v1](https://huggingface.co/datasets/Mustafaege/qwen3.5-functioncalling-v1) |
| **v2** | ~225K | + alpaca_function_calling | [Mustafaege/qwen3.5-functioncalling-v2](https://huggingface.co/datasets/Mustafaege/qwen3.5-functioncalling-v2) |

## Citation

If you use this dataset, please cite the original source:

```
@dataset{glaive-function-calling-v2,
author = {GlaiveAI},
title = {Glaive Function Calling v2},
year = {2024},
url = {https://huggingface.co/datasets/glaiveai/glaive-function-calling-v2}
}
```

## License

Apache 2.0 — see [LICENSE](https://www.apache.org/licenses/LICENSE-2.0) for details.

---

Built for Qwen3.5 fine-tuning. Part of the [Mustafaege](https://huggingface.co/Mustafaege) model series.