Fix Online Softmax: move subplot titles inside chart area (y=0.95)

Fix Online Softmax charts: increase height (350px), margins, and y-axis range for labels

Fix Memory Hierarchy: move labels further right (x=0.65), shorten arrow

Fix Memory Hierarchy diagram: move transfer labels to right of arrow

Fix GQA scaling chart x-axis: use log scale to space tick labels properly

Fix GQA tab chart overlaps

Add separate KV Cache dtype selector (FP16/BF16/FP8/INT8)

Add .gitignore, remove pycache from tracking

Fix legend overlap in context scaling chart, increase batch size limit to 256

Fix text overlap in Memory Budget charts

Add Memory Budget tab with real model and GPU detection

Add GQA/MQA tab with real model configuration

Refactor benchmarks to use real model.config values

Fix NoneType comparison error in prefill/decode charts

fix: Add fallback SDPA benchmark when attention layer fails

feat: Use real HuggingFace model attention layers for benchmarks

Use scatter markers for attention pattern visualization

Fix attention pattern heatmaps with standard colorscales

Fix Prefill vs Decode chart visibility issues

Implement Prefill vs Decode tab (Phase 4)

Use compute capability + SM count for better GPU spec estimation

Add H200 GPU support and improve roofline chart visibility

Auto-detect GPU instead of dropdown selection

Make Roofline chart dynamic with measured benchmark results

Add Phase 3: Benchmark tab with GPU performance measurement

Fix Visualizer tab bugs: softmax data, reset button, causal toggle

Phase 2: Add Visualizer tab with tiling animation, online softmax, and memory hierarchy

Phase 1: Core structure with model configs and placeholder tabs