n4/Qwen3-4B-Instruct-2507-sft_166

Hugging Face
TEXT GENERATIONConcurrency Cost:1Model Size:4BQuant:BF16Ctx Length:32kPublished:Feb 23, 2026License:apache-2.0Architecture:Transformer Open Weights Warm

The n4/Qwen3-4B-Instruct-2507-sft_166 is a 4 billion parameter instruction-tuned causal language model, fine-tuned from Qwen/Qwen3-4B-Instruct-2507. This model specializes in generating accurate structured outputs, including JSON, YAML, XML, TOML, and CSV formats. It was developed by n4 through a multi-stage SFT process, focusing on improving reliability for data serialization tasks.

Loading preview...

Model Overview

This model, n4/Qwen3-4B-Instruct-2507-sft_166, is a 4 billion parameter instruction-tuned language model derived from the Qwen/Qwen3-4B-Instruct-2507 base. It has been specifically fine-tuned using a multi-stage Supervised Fine-Tuning (SFT) process to significantly enhance its structured output accuracy across various formats.

Key Capabilities

  • Specialized Structured Output: Excels at generating data in JSON, YAML, XML, TOML, and CSV formats.
  • Multi-stage SFT: Training involved distinct stages, with dedicated focus on YAML and XML generation, using a maximum sequence length of 512 tokens.
  • Merged LoRA Weights: The model incorporates merged LoRA adapters, providing a full, ready-to-use fine-tuned model.

Good For

  • API Development: Generating structured responses for API endpoints.
  • Configuration Management: Creating configuration files in YAML or TOML.
  • Data Serialization: Tasks requiring reliable output in common structured data formats.
  • Automated Data Processing: Use cases where precise data formatting is critical for downstream systems.

Limitations

  • May encounter difficulties with extremely long, deeply nested, or ambiguous schemas.
  • Users should implement validation (e.g., JSON parsing, XML validation) for generated outputs in production environments.