Overview
The thetmon/c1 model is a LoRA adapter designed to enhance the structured output capabilities of the Qwen/Qwen3-4B-Instruct-2507 base model. It was fine-tuned using QLoRA (4-bit, Unsloth), making it an efficient addition for specific tasks. This repository provides only the adapter weights, requiring the base model to be loaded separately.
Key Capabilities
- Improved Structured Output: The primary objective of this adapter is to significantly boost the accuracy of structured data generation.
- Multi-format Support: Excels at producing outputs in various structured formats, including JSON, YAML, XML, TOML, and CSV.
- Efficient Fine-tuning: Utilizes QLoRA (4-bit) for efficient adaptation, making it practical for deployment.
- Focused Training: Loss is applied exclusively to the final assistant output, ensuring that the model's learning is concentrated on generating correct structured data, while intermediate Chain-of-Thought reasoning is masked.
Training Details
The adapter was trained on the u-10bei/structured_data_with_cot_dataset with a maximum sequence length of 2048 over 3 epochs. It used a learning rate of 2e-06 and LoRA parameters of r=64, alpha=128.
When to Use This Model
This adapter is particularly well-suited for applications where precise and reliable structured data generation is critical. If your use case involves extracting information into specific formats or generating configuration files, this model can provide a significant advantage by reducing errors in output structure.