Spaces:
Sleeping
Sleeping
| title: TableMind AI | |
| emoji: 📊 | |
| colorFrom: blue | |
| colorTo: indigo | |
| sdk: gradio | |
| sdk_version: 6.20.0 | |
| app_file: app.py | |
| pinned: false | |
| license: mit | |
| short_description: Fine-tuned table Q&A chatbot with RAG and auto-charts | |
| # TableMind AI | |
| Upload an Excel, CSV, PDF, or Word document (any size) and ask questions in plain English. | |
| Numeric/aggregate questions are answered with exact computed results via a pandas execution | |
| layer, not guesses — everything else is answered via retrieval-augmented generation over the | |
| document. Toggle the chart option for a visual on comparison-style questions. | |
| Built with a QLoRA fine-tuned `Qwen/Qwen2.5-1.5B-Instruct`, `sentence-transformers` + `FAISS` | |
| for retrieval, and a text-to-pandas router for numeric accuracy. | |
| > **Hardware: CPU basic.** This version runs entirely on CPU — no ZeroGPU dependency, no | |
| > daily GPU quota, no PRO subscription required. Generation is slower than GPU hardware | |
| > (expect several seconds to tens of seconds per answer depending on question length), but | |
| > it's free and avoids ZeroGPU's worker-allocation/quota issues entirely. | |