gauthierpiarrette/nl2jq-qwen3-0.6b

TEXT GENERATIONConcurrent Unit Cost:1Model Size:0.8BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Jul 10, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

The gauthierpiarrette/nl2jq-qwen3-0.6b is a 0.8 billion parameter LoRA fine-tune of Qwen3-0.6B, developed by gauthierpiarrette. This model specializes in translating natural language requests and JSON samples into jq programs, excelling at copying novel field names from JSON inputs. It serves as a practical baseline for the nl2jq project, offering a balance of latency and accuracy for local command-line interface use.

Loading preview...

nl2jq-qwen3-0.6b: Natural Language to jq Translator

This model, nl2jq-qwen3-0.6b, is a LoRA fine-tune of the Qwen3-0.6B base model, specifically designed to translate natural language requests and JSON samples into jq programs. Developed by gauthierpiarrette as part of the nl2jq project, it addresses the common need for accurately copying novel field names from input JSON, a critical feature for real-world CLI usage.

Key Capabilities & Features

  • Natural Language to jq Translation: Converts user requests and JSON examples into executable jq scripts.
  • Enhanced Field Name Copying: Markedly more reliable at incorporating novel field names from JSON samples compared to smaller, from-scratch models.
  • Optimized for Local Use: Recommended as the default backend for the jqgen CLI, offering a strong balance of latency (1-3 seconds on laptop CPU) and accuracy.
  • Performance: Achieves a pass@1 score of 0.40 on the nl2jq-bench v1.0.0 dataset, significantly outperforming the base Qwen3-0.6B model's 0.01 zero-shot score.
  • Training: Fine-tuned using LoRA (r=32, α=64) over 150k execution-verified synthetic examples, with a focus on per-example-unique field names to ensure robust copying behavior.

Good for

  • CLI Tools: Ideal for integrating into command-line interfaces where users need to quickly generate jq programs from natural language.
  • Data Transformation: Users who frequently manipulate JSON data and prefer natural language queries over manual jq script writing.
  • Local Deployment: Its efficient performance makes it suitable for local execution on standard hardware, providing quick responses without relying on cloud APIs.

Limitations

  • Supports jq 1.7.1 features only (no streaming, modules, or user-defined functions).
  • May produce valid but unintended programs for ambiguous requests.
  • Users should always inspect the generated jq program before execution.

Finetunes
1 models