Improve model performance for your use case by fine-tuning Qwen models on QwenCloud with SFT and LoRA.
Fine-tuning adapts a pre-trained model to your specific use case using your own data. Compared to prompt engineering alone, fine-tuning can:
Fine-tuning jobs and the models they produce belong to the region where you create them and cannot be used in another region.
Before creating a fine-tuning job, you need:
Fine-tuning on QwenCloud uses SFT (Supervised Fine-Tuning) -- you train the model on input/output pairs to teach it desired responses. Training data is formatted as ChatML messages with user/assistant turns.
All fine-tuning jobs use LoRA (Efficient Training), which trains low-rank adapter weights on top of the frozen base model. This approach is faster and lower cost than full parameter training.
You can also select a previously fine-tuned custom model as the base model to continue training iteratively. When creating a fine-tuning job, switch to the Custom Models tab in the base model selector to choose one of your own models.
Each line of your JSONL file contains a multi-turn conversation in ChatML format:
- Improve accuracy on domain-specific tasks
- Reduce latency by replacing few-shot prompts with learned behavior
- Align outputs with human preferences or brand voice
- Replace larger models with fine-tuned smaller models at lower cost
Fine-tuning is available only in Singapore and China North 2 (Beijing). It is not available in China (Hong Kong), Germany (Frankfurt), Japan (Tokyo), or US (Virginia), and its menu is hidden in the console in those regions. To use it, switch to Singapore or China North 2 (Beijing) with the region selector in the upper-right corner of the console. See Feature availability by region.
Prerequisites
Before creating a fine-tuning job, you need:
- A QwenCloud account. Log in to the console.
- A published dataset in JSONL format. See Create a dataset and Manage datasets to publish it.
Training method
Fine-tuning on QwenCloud uses SFT (Supervised Fine-Tuning) -- you train the model on input/output pairs to teach it desired responses. Training data is formatted as ChatML messages with user/assistant turns.
Training mode
All fine-tuning jobs use LoRA (Efficient Training), which trains low-rank adapter weights on top of the frozen base model. This approach is faster and lower cost than full parameter training.
Supported models
| Model | Training method | Training mode |
|---|---|---|
| Qwen3.6-Flash | SFT | LoRA |
Data format
Each line of your JSONL file contains a multi-turn conversation in ChatML format:
Dataset tips
- Scale: Fine-tuning requires at least several hundred high-quality examples. Thousands of examples typically yield better results.
- Diversity: Cover the range of scenarios your model will encounter in production. Balance data proportions to match actual usage patterns.
- Quality: Ensure assistant responses reflect the exact behavior you want. Low-quality outputs in training data lead to low-quality model behavior.
Next steps
- Create a dataset -- Prepare training data for fine-tuning
- Create a fine-tuning job in the console
- Manage fine-tuning jobs -- Monitor progress and publish checkpoints to create custom models
- Deployment overview -- Deploy your custom model to serve inference requests