README.md

---
license: apache-2.0
base_model: PrimeIntellect/Qwen3-0.6B
tags:
  - text-generation
  - chinese
  - sft
  - qwen3
datasets:
  - ivanleomk/reverse-chinese-poems
language:
  - zh
pipeline_tag: text-generation
---

# Reverse Chinese Text (SFT)

This model is a fine-tuned version of [PrimeIntellect/Qwen3-0.6B](https://huggingface.co/PrimeIntellect/Qwen3-0.6B) trained on the task of reversing Chinese text character-by-character.

## Training

- **Base Model:** PrimeIntellect/Qwen3-0.6B
- **Method:** Supervised Fine-Tuning (SFT)
- **Dataset:** [ivanleomk/reverse-chinese-poems](https://huggingface.co/datasets/ivanleomk/reverse-chinese-poems)
- **Training Steps:** 200
- **Learning Rate:** 2e-5
- **Batch Size:** 16
- **Framework:** [Prime-RL](https://github.com/PrimeIntellect-ai/prime-rl)

## Benchmark Results

Evaluated on 1,000 samples from the test set:

| Model | Character Accuracy | Exact Match Rate |
|-------|-------------------|------------------|
| PrimeIntellect/Qwen3-0.6B (base) | 0.10% | 0.00% |
| **ivanleomk/reverse-chinese-text (SFT)** | **63.55%** | **9.60%** |

## Usage

```python
from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained("ivanleomk/reverse-chinese-text")
tokenizer = AutoTokenizer.from_pretrained("ivanleomk/reverse-chinese-text")

messages = [
    {"role": "system", "content": "You are a text reversal assistant. Given Chinese text, reverse it character by character."},
    {"role": "user", "content": "请反转以下文字：床前明月光"}
]

input_ids = tokenizer.apply_chat_template(messages, return_tensors="pt")
output = model.generate(input_ids, max_new_tokens=100)
print(tokenizer.decode(output[0], skip_special_tokens=True))
# Expected: 光月明前床
```

## License

Apache 2.0