samandar1105/tablemind-qwen2.5-1.5b

TableMind — Table Q&A Assistant (fine-tuned Qwen/Qwen2.5-1.5B-Instruct)

Fine-tuned with QLoRA on 6956 examples derived from WikiTableQuestions.

Intended use

This model is designed to be paired with a retrieval layer (RAG) that supplies the actual table/document content at inference time. On its own it knows the format and behavior of grounded table answering; it does not have memorized knowledge of any particular user's data.

Training

• Base model: Qwen/Qwen2.5-1.5B-Instruct
• Method: QLoRA (4-bit NF4), r=16, alpha=32
• Epochs: 2, LR: 2e-4, max seq len: 1024
• Dataset: WikiTableQuestions, tables truncated to <= 40 rows

Limitations

• English only. Small model (3B) - pair with the RAG + code-execution router described in the TableMind project guide for numeric accuracy on large tables.
• Not intended for open-domain trivia unrelated to the provided table/document.