chrissoria/catllm-json-formatter

TEXT GENERATIONConcurrency Cost:1Model Size:0.5BQuant:BF16Ctx Length:32kTool Calling:SupportedPublished:Mar 6, 2026License:apache-2.0Architecture:Transformer Open Weights Cold

chrissoria/catllm-json-formatter is a fine-tuned Qwen2.5-0.5B-Instruct model designed to convert messy LLM classification outputs into valid JSON format. This 0.5 billion parameter model specializes in cleaning and structuring raw classification results from other LLMs into a consistent {"1": "0", "2": "1", ...} JSON object. Its primary use case is to ensure reliable and parseable output for downstream applications, particularly within the cat-llm framework, by handling various malformed inputs.

Loading preview...

CatLLM JSON Formatter Overview

chrissoria/catllm-json-formatter is a specialized Qwen2.5-0.5B-Instruct model, fine-tuned to address the common problem of inconsistent or malformed JSON output from large language models, specifically for classification tasks. It acts as a robust post-processor, taking raw, potentially messy LLM classification results and converting them into a clean, standardized JSON format like {"1": "0", "2": "1", ...}.

Key Capabilities

  • Standardized JSON Output: Ensures consistent and parseable JSON from varied, malformed LLM classification responses.
  • Handles Diverse Inputs: Trained on 8,000 synthetic examples covering over 26 messy output formats.
  • Scalable Category Support: Reliably processes category counts from N=2 up to N=50, producing large JSON objects with many keys.
  • Integrated with cat-llm: Designed to be used automatically as a fallback formatter within the cat-llm library when json_formatter=True is enabled.

Good for

  • Improving LLM Reliability: Essential for applications requiring strict JSON adherence from LLM classification outputs.
  • Automated Data Processing: Streamlining workflows where classification results need to be programmatically consumed.
  • Robust LLM Integrations: Mitigating issues caused by LLMs occasionally failing to produce perfectly formatted JSON.
  • Developers using cat-llm: Provides an out-of-the-box solution for ensuring clean classification results.