jsoul/geo-perturbation-grouping-qwen3-4b

TEXT GENERATIONPricing:Input $0.4 / Cached $0.08 / Output $0.8Concurrent Unit Cost:1Model Size:4BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Jul 8, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

The jsoul/geo-perturbation-grouping-qwen3-4b is a 4 billion parameter Qwen3-based causal language model, fine-tuned by jsoul, specifically designed for extracting and organizing genetic-perturbation case/control experimental groups from NCBI GEO gene-expression series metadata. This model excels at identifying specific experimental designs (KD/KO/OE) with single wild-type HGNC target genes and matched time points, providing structured output for curation assistance. Its 32768 token context length supports processing detailed sample metadata for transcriptomic studies.

Loading preview...

Overview

This model, jsoul/geo-perturbation-grouping-qwen3-4b, is a specialized fine-tune of the Qwen/Qwen3-4B-Instruct-2507 base model. Its primary function is to analyze sample metadata from NCBI GEO gene-expression series and identify valid genetic-perturbation case/control experimental groups. It specifically looks for methods like Knockdown (KD), Knockout (KO), or Overexpression (OE), requiring at least two controls and two cases, a single wild-type HGNC target gene, one cell line, and matched time points. If no such groups are found, it reports accordingly.

Key Capabilities

  • Automated Grouping: Organizes samples into structured genetic-perturbation case/control groups.
  • Specific Perturbation Detection: Identifies KD, KO, and OE methods.
  • Structured Output: Provides clear output including cell line, perturbation method, target gene, and lists of control/case GSM IDs.
  • Merged LoRA Adapter: The LoRA adapter has been merged into the base weights, making it a standalone model for direct use with transformers, vLLM, or TGI.
  • Training Data: Trained on the jsoul/geo-perturbation-grouping-train dataset (2,400 examples) with a fixed system prompt and user-provided study metadata.

Intended Use & Limitations

  • Good for: Curation assistance in identifying perturbation case/control groups from GEO transcriptomic studies (RNA-seq / microarray). Outputs require human review.
  • Out of scope: Non-transcriptomic assays, non-human samples, multi-gene or mutant constructs, and drug/stimulus-confounded designs.
  • Limitations: May hallucinate GSM IDs or gene symbols, mis-assign controls, or mishandle very long sample tables. Validation against the source GEO record is always recommended.

Finetunes
1 models