Model Card for qwen3-4b-instruct-stat-qlora-v2

This model is a QLoRA fine-tuned variant of Qwen3-4B-Instruct, specialized for explaining statistical test pipeline outputs in a clear, structured, and technically correct way. It is designed to transform structured tool_json outputs (e.g. T-Test, ANOVA, correlation, clustering, chi-square results) into high-quality natural language explanations following a strict five-section analytical format. Its' goal is to serve my personal project app available on spaces. The fine-tuning process focused on reducing hallucinations, improving methodological correctness, and strengthening interpretability, while preserving the general instruction-following abilities of the base model.

Model Details

Model Description

• Developed by: João Vaz, Independent research project
• Shared by: Ozymandias2
• Model type: Instruction-tuned causal language model with LoRA adapters
• Language(s) (NLP): English
• License: Same as base model (Qwen/Qwen3-4B-Instruct-2507)
• Finetuned from model: Qwen/Qwen3-4B-Instruct-2507

This model was trained to generate structured statistical explanations using the following fixed template: -Missing Data Analysis -Pre-Test Diagnostics -Test Selection Rationale -Test Results -Interpretation

The model explicitly avoids: -causal language for observational analyses, -hallucinated preprocessing steps, -incorrect test naming or directionality.

Model Sources

• Repository: https://github.com/JoaoLAVaz/data-chat-assistant/tree/v1/result_explainer_study
• Paper : https://github.com/JoaoLAVaz/data-chat-assistant/blob/v1/result_explainer_study/README.md
• Demo (spaces, feel free to restart and test it): https://huggingface.co/spaces/Ozymandias2/data-chat-assistant

Framework versions

• PEFT 0.18.0
• Transformers ≥ 4.40