jerryyan/TraceML-Action-Labeler

TEXT GENERATIONPricing:Input $0.32 / Cached $0.064 / Output $1.6Concurrent Unit Cost:1Model Size:2BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Sep 28, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

jerryyan/TraceML-Action-Labeler is a 1.7 billion parameter Qwen3-based model developed by Jiarui Yan and his collaborators, specifically fine-tuned for analyzing transitions between machine learning solution versions. It functions as an action labeler within the TraceML framework, identifying and categorizing edits to ML code and their impact. The model excels at generating structured JSON outputs detailing coarse and fine actions, intents, edit magnitude, and score effects from code diffs and state labels, making it ideal for automated analysis of ML development lifecycles.

Loading preview...

TraceML Action Labeler (Qwen3-1.7B)

This model, developed by Jiarui Yan and his team as part of the TraceML project (NeurIPS 2026), is a Qwen3-1.7B variant specifically fine-tuned to label actions and intents within transitions between different versions of a machine learning solution. It processes code differences, state labels of both versions, and score changes to output a structured JSON analysis.

Key Capabilities

  • Detailed Action Labeling: Identifies 10 coarse actions (e.g., model, training, data), 85 fine actions, and 6 intents (e.g., optimization, debugging).
  • Impact Analysis: Quantifies edit magnitude (micro, minor, major, overhaul) and score_effect (improving, plateau, regressing, unknown).
  • Natural Language Summaries: Provides short natural-language summaries for the goal and diff.
  • Schema-Constrained Output: Generates JSON output with predefined vocabularies for consistent analysis.
  • Integration with TraceML Toolkit: Designed to work seamlessly with the TraceML toolkit for automated analysis of ML runs, handling prompt building and output parsing.

Good For

  • Automated ML Development Analysis: Ideal for researchers and developers looking to automatically understand and categorize changes in ML codebases.
  • Tracing ML Evolution: Useful for tracking how ML solutions evolve, identifying the purpose and impact of code modifications.
  • Research on ML Development: Supports studies on ML engineering practices by providing structured data on code changes and their effects.
  • Generating Structured Labels: Provides a programmatic way to extract detailed, categorized information about ML code transitions, which can be used for further analysis or dataset creation.

Finetunes
1 models