ljh728/Qwen3-0.6B-JSON-SFT-GRPO

TEXT GENERATIONPricing:Input $0.04 / Cached $0.008 / Output $0.08Concurrent Unit Cost:1Model Size:0.8BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Sep 4, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

The ljh728/Qwen3-0.6B-JSON-SFT-GRPO is a 0.8 billion parameter Qwen3-based language model, fine-tuned from NotoriousH2/Qwen3-0.6B-JSON-SFT. Developed by ljh728, this model was trained using Unsloth and Huggingface's TRL library, achieving a 2x speed improvement during fine-tuning. It is specifically optimized for tasks requiring JSON-formatted output, making it suitable for structured data generation and API interactions.

Loading preview...

Model Overview

The ljh728/Qwen3-0.6B-JSON-SFT-GRPO is a 0.8 billion parameter Qwen3-based language model, fine-tuned by ljh728. It builds upon the NotoriousH2/Qwen3-0.6B-JSON-SFT model and features a context length of 32768 tokens. A key aspect of its development is the utilization of Unsloth and Huggingface's TRL library, which enabled a 2x faster fine-tuning process.

Key Capabilities

  • JSON-Specific Fine-tuning: This model is specifically fine-tuned for generating JSON-formatted output, making it highly effective for tasks requiring structured data.
  • Efficient Training: Leverages Unsloth for accelerated fine-tuning, indicating potential for rapid adaptation to new JSON-centric tasks.
  • Qwen3 Architecture: Benefits from the underlying Qwen3 architecture, providing a solid foundation for language understanding and generation.

Good For

  • Structured Data Generation: Ideal for applications that need to output information in a consistent JSON format.
  • API Interaction: Can be used to generate JSON payloads or parse JSON responses for API calls.
  • Rapid Prototyping: The efficient fine-tuning process suggests it could be quickly adapted for specific JSON schema requirements.