70 lines
2.2 KiB
Markdown
70 lines
2.2 KiB
Markdown
|
---
|
|||
|
|
base_model: Qwen/Qwen3-4B-Instruct-2507
|
|||
|
|
datasets:
|
|||
|
|
- u-10bei/dpo-dataset-qwen-cot
|
|||
|
|
language:
|
|||
|
|
- en
|
|||
|
|
license: apache-2.0
|
|||
|
|
library_name: transformers
|
|||
|
|
pipeline_tag: text-generation
|
|||
|
|
tags:
|
|||
|
|
- dpo
|
|||
|
|
- unsloth
|
|||
|
|
- qwen
|
|||
|
|
- alignment
|
|||
|
|
- structured-output
|
|||
|
|
---
|
|||
|
|
|
|||
|
|
# ＜qwen3-4b-nako13-dpo-qwen-cot-merged＞
|
|||
|
|
|
|||
|
|
This model is a high-performance variant of **Qwen/Qwen3-4B-Instruct-2507**, optimized for precise structured data generation.
|
|||
|
|
It was developed through a **two-stage fine-tuning process** to ensure both high knowledge density and strict output formatting.
|
|||
|
|
|
|||
|
|
## Training Process
|
|||
|
|
1. **Stage 1: SFT (Supervised Fine-Tuning)**
|
|||
|
|
- **Base Model**: Qwen/Qwen3-4B-Instruct-2507
|
|||
|
|
- **Adapter**: [nakotsuko13/qwen3-4b-nako13-structured-output-lora](https://huggingface.co/nakotsuko13/qwen3-4b-nako13-structured-output-lora)
|
|||
|
|
- **Focus**: Trained on 16,500+ samples to master JSON, XML, CSV, and YAML structures.
|
|||
|
|
|
|||
|
|
2. **Stage 2: DPO (Direct Preference Optimization)**
|
|||
|
|
- **Dataset**: u-10bei/dpo-dataset-qwen-cot
|
|||
|
|
- **Focus**: Optimized to eliminate conversational filler and provide direct, raw structured outputs.
|
|||
|
|
|
|||
|
|
## Training Configuration (DPO)
|
|||
|
|
- **Method**: DPO (Direct Preference Optimization)
|
|||
|
|
- **Epochs**: 1
|
|||
|
|
- **Learning rate**: 5e-07
|
|||
|
|
- **Beta**: 0.01
|
|||
|
|
- **Max sequence length**: 1024
|
|||
|
|
- **LoRA Config**: r=64, alpha=128 (Merged into final weights)
|
|||
|
|
|
|||
|
|
## Usage
|
|||
|
|
This is a **full-merged 16-bit model**. It can be used directly with standard `transformers` or `vLLM`.
|
|||
|
|
|
|||
|
|
```python
|
|||
|
|
from transformers import AutoModelForCausalLM, AutoTokenizer
|
|||
|
|
import torch
|
|||
|
|
|
|||
|
|
model_id = nakotsuko13/qwen3-4b-nako13-dpo-qwen-cot-merged
|
|||
|
|
|
|||
|
|
tokenizer = AutoTokenizer.from_pretrained(model_id)
|
|||
|
|
model = AutoModelForCausalLM.from_pretrained(
|
|||
|
|
model_id,
|
|||
|
|
torch_dtype=torch.bfloat16,
|
|||
|
|
device_map="auto"
|
|||
|
|
)
|
|||
|
|
|
|||
|
|
# Test inference
|
|||
|
|
prompt = "Your question here"
|
|||
|
|
inputs = tokenizer.apply_chat_template([{"role": "user", "content": prompt}], tokenize=True, add_generation_prompt=True, return_tensors="pt").to("cuda")
|
|||
|
|
outputs = model.generate(**inputs, max_new_tokens=512)
|
|||
|
|
print(tokenizer.decode(outputs[0]))
|
|||
|
|
|
|||
|
|
```
|
|||
|
|
|
|||
|
|
## Sources & License (IMPORTANT)
|
|||
|
|
|
|||
|
|
* **Training Data**: [u-10bei/dpo-dataset-qwen-cot]
|
|||
|
|
* **License**: MIT License. (As per dataset terms).
|
|||
|
|
* **Compliance**: Users must follow the original base model's license terms.
|