---
title: TenderHub WebAI Verification Worker
emoji: "document"
colorFrom: "blue"
colorTo: "purple"
sdk: gradio
sdk_version: "4.44.0"
python_version: "3.11"
app_file: app.py
pinned: false
license: mit
tags:
  - document-processing
  - tender-analysis
  - verification
  - multimodal-ai
short_description: Secondary verification layer for tender document analysis using webAI-ColVec1-4b
---

# TenderHub WebAI Verification Worker

This HF Space provides a secondary verification layer for tender document processing using the webAI-ColVec1-4b multimodal model with ZeroGPU acceleration.

## Features

- **Multimodal Analysis**: Direct document understanding with vision-language model
- **ZeroGPU Support**: On-demand GPU processing for efficient inference
- **Memory Optimization**: 8-bit quantization + FlashAttention-2 to prevent OOM on free tier
- **Cross-Validation**: Comparison with primary worker results
- **Structured Output**: JSON-formatted analysis and comparison metrics

## Memory Strategy

- **8-bit Quantization**: Reduces 4B model memory from ~8GB to ~4GB with better quality
- **FlashAttention-2**: Optimized attention mechanism with ~40% memory reduction
- **Aggressive Cleanup**: Manual garbage collection to prevent 4GB+ ghost memory
- **CPU Loading**: Model loads on CPU, moves to GPU only during inference
- **Batch Size 1**: Processes one document at a time
- **Automatic Cleanup**: GPU cache clearing after each inference

## Usage

1. Upload a tender document (PDF or image)
2. Click "Verify Document" for analysis
3. Review structured output and comparison metrics

For automated processing, the worker also runs a background job queue processor.