Fine-tuning Qwen3:0.6B small model to extract complaint information
If the task is simple and the concurrent calls to the large model API are high, I suggest fine-tuning a small model for a specific task and deploying it locally. This can meet high concurrency and reduce capital consumption. (Local deployment, default hardware environment single card 4090)