Update README.md
Browse files
README.md
CHANGED
|
@@ -1,13 +1,51 @@
|
|
| 1 |
-
|
| 2 |
-
|
| 3 |
-
|
| 4 |
-
|
| 5 |
-
|
| 6 |
-
sdk: gradio
|
| 7 |
-
sdk_version: 6.1.0
|
| 8 |
-
app_file: app.py
|
| 9 |
pinned: false
|
| 10 |
-
license:
|
| 11 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 12 |
|
| 13 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
title: Ministral WebGPU
|
| 2 |
+
emoji: ⚡️
|
| 3 |
+
colorFrom: red
|
| 4 |
+
colorTo: yellow
|
| 5 |
+
sdk: static
|
|
|
|
|
|
|
|
|
|
| 6 |
pinned: false
|
| 7 |
+
license: apache-2.0
|
| 8 |
+
short_description: Frontier multimodal AI, running entirely in your browser.
|
| 9 |
+
app_build_command: npm run build
|
| 10 |
+
app_file: dist/index.html
|
| 11 |
+
models:
|
| 12 |
+
- mistralai/Ministral-3-3B-Instruct-2512-ONNX
|
| 13 |
+
- mistralai/Ministral-3-3B-Instruct-2512
|
| 14 |
|
| 15 |
+
|
| 16 |
+
AI Multimodal WebGPU Assistant
|
| 17 |
+
Developer: Muhammad Abdullah Rasheed Research Assistant @ Cambridge | MSc Data Science & AI '25 | Google WTM Scholar
|
| 18 |
+
|
| 19 |
+
Overview
|
| 20 |
+
This project demonstrates cutting-edge browser-based AI by running a complete 3B parameter multimodal language model entirely client-side using WebGPU acceleration. No servers, no API calls, no data sent anywhere - complete privacy and instant inference.
|
| 21 |
+
|
| 22 |
+
Key Features
|
| 23 |
+
Privacy-First Architecture: The entire Ministral-3B model runs locally in your browser using WebGPU - your video feed never leaves your device
|
| 24 |
+
Real-Time Multimodal AI: Live camera feed processing with visual question answering capabilities
|
| 25 |
+
WebGPU Acceleration: Leveraging the latest browser GPU APIs for near-native performance
|
| 26 |
+
Zero Backend Dependencies: No API keys, no server calls, no external services required
|
| 27 |
+
Cross-Platform: Works seamlessly across modern browsers with WebGPU support
|
| 28 |
+
Technical Stack
|
| 29 |
+
Model: Ministral-3-3B-Instruct (quantized for browser deployment)
|
| 30 |
+
Runtime: Transformers.js for in-browser inference
|
| 31 |
+
Compute: WebGPU API for GPU acceleration
|
| 32 |
+
Frontend: Modern JavaScript with WebAssembly integration
|
| 33 |
+
Use Cases
|
| 34 |
+
Visual question answering from live camera feed
|
| 35 |
+
Real-time scene understanding and description
|
| 36 |
+
Privacy-sensitive AI applications
|
| 37 |
+
Edge computing demonstrations
|
| 38 |
+
Educational tool for AI and browser technologies
|
| 39 |
+
Why This Matters
|
| 40 |
+
This project showcases the future of AI deployment - moving powerful language models from cloud servers to the edge, where they can provide instant, private, and accessible intelligence without compromising user privacy or requiring expensive infrastructure.
|
| 41 |
+
|
| 42 |
+
Author
|
| 43 |
+
Muhammad Abdullah Rasheed
|
| 44 |
+
Research Assistant | AI & Machine Learning Researcher
|
| 45 |
+
|
| 46 |
+
🎓 MSc Data Science & AI '25, Google WTM Scholar
|
| 47 |
+
🔬 Research areas: Computer Vision, NLP, Climate AI
|
| 48 |
+
💼 Experience: Gesture Recognition, Backend Development, ML Engineering
|
| 49 |
+
🔗 LinkedIn | GitHub | HuggingFace
|
| 50 |
+
License
|
| 51 |
+
Apache-2.0
|