elifepathways/jats-agentic-annotation-qwen3.5-9b-teacher-distill-v1

VISIONPricing:Input $0.431 / Cached $0.0862 / Output $1.12Concurrent Unit Cost:1Model Size:9BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Sep 2, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

The elifepathways/jats-agentic-annotation-qwen3.5-9b-teacher-distill-v1 is a 9 billion parameter Qwen3.5-based language model, fine-tuned with a teacher-distilled LoRA adapter. It is specifically designed for multi-turn tool-call agentic annotation, generating JATS XML for scientific paper body sections. This model excels at structured data extraction and formatting within a scratchpad environment, achieving a plain F-score of 0.452 on JATS body annotation tasks.

Loading preview...

Model Overview

This model, elifepathways/jats-agentic-annotation-qwen3.5-9b-teacher-distill-v1, is a 9 billion parameter Qwen3.5-based language model. It incorporates a teacher-distilled LoRA adapter, specifically optimized for generating JATS XML annotations for scientific paper body sections. The model operates within a multi-turn tool-call agentic framework, utilizing a scratchpad environment for structured output.

Key Capabilities

  • JATS XML Generation: Specializes in producing JATS XML for scientific paper body content.
  • Agentic Annotation: Designed for multi-turn interactions with a 17-tool JatsEnv, enabling complex annotation workflows.
  • Teacher-Distilled Performance: Fine-tuned using trajectories from a DeepSeek-v4-pro teacher model, enhancing its ability to handle detailed annotation tasks.
  • Improved F-score: Achieves a plain F-score of 0.452 on a 77-document evaluation set, significantly outperforming its base model (0.336).
  • Context Length: Supports a maximum context length of 32768 tokens, suitable for processing substantial document sections.

Use Cases

  • Automated Scientific Document Processing: Ideal for systems requiring automated extraction and structuring of information from scientific papers into JATS XML format.
  • Research Data Management: Can be integrated into pipelines for standardizing research article content for databases or archives.
  • Content Conversion: Useful for converting raw text or markdown from scientific documents into a machine-readable, structured XML format.

Finetunes
1 models