tahamajs/RR / Docs /visualizer /README.md
tahamajs's picture
|
download
raw
979 Bytes
metadata
title: 3D Neural Architecture Visualizer
emoji: 🧠
colorFrom: indigo
colorTo: blue
sdk: static
pinned: false
license: apache-2.0

3D Neural Architecture Visualizer & Compression Atlas

Interactive WebGL 3D dataflow and multi-model architectural comparison across DistilBERT, RoBERTa, GPT-2, and Qwen-2.5.

Key Capabilities

  • Interactive 3D Computational Graphs: Explore exact projection geometries, attention heads, SVD bottlenecks, GPT-2 fused Conv1D, and Qwen-2.5 GQA.
  • Side-by-Side Architectural Comparison: Baseline uncompressed transformer backbone vs. SVD / Low-Rank factorized topology.
  • 240 Research Modules: Verified drop-in attention and compression mechanisms with live LaTeX formulas, tensor shapes, and code snippets.
  • Real-Time Token Physics: Trace tokens through residual skip highways, Q/K/V projections, and MLP feedforward blocks.
  • Zero-Install Static App: Pure client-side Three.js & Vanilla ES modules.

Xet Storage Details

Size:
979 Bytes
·
Xet hash:
7c1874f0cb7cd364c5238b6765a654ad6a7dff02a063f0c2be59ac7f7dd6f10b

Xet efficiently stores files, intelligently splitting them into unique chunks and accelerating uploads and downloads. More info.