ML INTELLIGENCE

AI/ML Dashboard

LLM benchmarks, latest research papers, and model performance analytics.

MODEL ANALYTICS

Benchmark Visualizations

Compare language models on intelligence, speed, and cost dimensions.

SCATTER PLOTS

Loading benchmarks...

Loading benchmarks...

MODEL COMPARISON

Multi-Benchmark Radar

Compare top models across all benchmark dimensions simultaneously.

TOP 5 MODEL RADAR

Loading benchmarks...

INFERENCE

Throughput Simulator

Estimate theoretical inference throughput from model parameters, quantization, and hardware bandwidth.

INFERENCE SIMULATOR

MODEL MEMORY

140.0 GB

70B × 2 bytes

FLOPs / TOKEN

140.00 GFLOP

2 × params

PEAK tok/s

3.21

bandwidth / (2·p·bpp)

SEQUENCE COMPUTE

286.72 TFLOP

× 2048 seq · 1 batch

Roofline estimate | memory-bound decode | bars scale with params & bandwidth

HARDWARE

GPU Requirement Calculator

Input model parameters, quantization, and serving engine to find compatible GPUs with estimated throughput.

GPU REQUIREMENT CALCULATOR

Model VRAM

14.0 GB

KV Cache (4096 ctx x 1 batch)

18.7 GB

Total Required

34.1 GB

DATACENTER

H100 SXM5BEST
VRAM: 80GBBW: 3350GB/sFP16: 1979TFTDP: 700W~102 tok/s$30.0K
Headroom: 45.9 GB free
H100 PCIe
VRAM: 80GBBW: 2039GB/sFP16: 1513TFTDP: 350W~62 tok/s$25.0K
Headroom: 45.9 GB free
A100 80GB SXM4
VRAM: 80GBBW: 2039GB/sFP16: 624TFTDP: 400W~62 tok/s$15.0K
Headroom: 45.9 GB free
A100 40GB SXM4
VRAM: 40GBBW: 1555GB/sFP16: 624TFTDP: 400W~47 tok/s$10.0K
Headroom: 5.9 GB free
L40S
VRAM: 48GBBW: 864GB/sFP16: 366TFTDP: 350W~26 tok/s$8.0K
Headroom: 13.9 GB free

WORKSTATION

RTX 6000 Ada
VRAM: 48GBBW: 960GB/sFP16: 362TFTDP: 300W~29 tok/s$6.8K
Headroom: 13.9 GB free
RTX A6000
VRAM: 48GBBW: 768GB/sFP16: 155TFTDP: 300W~23 tok/s$4.5K
Headroom: 13.9 GB free

VRAM = weights + KV cache (4096 tokens x 1 batch) + 10% overhead | tok/s = min(bandwidth-bound, compute-bound) x vLLM efficiency (85%) | Estimates exclude CUDA graphs and continuous batching

NEURAL NETWORKS

Interactive Neural Network

Watch a neural network learn in real-time. Forward pass, backpropagation, and gradient flow visualized across layers.

Loading neural_net...

DIMENSIONALITY REDUCTION

t-SNE Visualization

High-dimensional data projected to 2D in real-time. The t-SNE algorithm clusters similar data points in WASM.

Loading tsne...

OPTIMIZATION

Gradient Descent Visualization

3D loss landscape with optimizer trajectory. Watch the gradient descent ball navigate valleys and saddle points.

Loading gradient_descent...

LEADERBOARD

LLM Benchmark Table

Sortable comparison across MMLU, HumanEval, GSM8K, HellaSwag, ARC, and TruthfulQA benchmarks.

LLM BENCHMARK LEADERBOARD

Loading benchmarks...

RESEARCH

Latest ArXiv Papers

Newest submissions in AI, Machine Learning, and Computation & Language. Updated hourly.

LATEST ARXIV PAPERS

Fetching latest AI/ML papers from ArXiv...