chanwit/gemma4-qwen35-reasoning-600

---
license: apache-2.0
language:
- en
tags:
- gemma-4
- reasoning
- thinking
- fine-tuning
- distillation
size_categories:
- n<7K
source_datasets:
- Jackrong/Qwen3.5-reasoning-700x
---

# gemma4-qwen35-reasoning-600

Gemma-4 formatted reasoning distillation dataset.

## Format

This dataset uses the **Gemma-4 thinking format** with channel-based reasoning:

```
<|turn>system
<|think|>You are a helpful assistant...<turn|>
<|turn>user
{problem}<turn|>
<|turn>model
<|channel>thought
{step-by-step reasoning}
<channel|>
{final answer}<turn|>
```

## Source

Converted from Claude Opus/Qwen reasoning traces:
- `Jackrong/Qwen3.5-reasoning-700x`

## Statistics

- **Total examples**: 633
- **Format**: Gemma-4 text-only
- **Purpose**: Fine-tuning for step-by-step reasoning

## Usage

```python
from datasets import load_dataset

dataset = load_dataset("chanwit/gemma4-qwen35-reasoning-600")
```

## License

Apache 2.0 (same as source datasets)