Zoth AI Model Foundry & Spirit Matrix Pro

A NullAI studio Ā· frontier reasoning, open weights & local companions

āœ¦ Frontier Reasoning & Local Weights

AI Model Foundry & Spirit Matrix

Zoth Studio orchestrates bleeding-edge open-weight LLMs (Nous Hermes, DeepSeek, Qwen-2.5-Coder) alongside frontier clouds (Grok-3, Claude-3.7-Sonnet). Every model embodies a specialized mascot spirit with real-time mathematical telemetry and zero-leak loopback execution.

24+
Indexed Models
1.8k t/s
Peak Cerebras TPS
100%
Offline Fallback
Argon2id
Encrypted Vault
šŸ’  3D Neural Lattice Ā· Core Active

āš” Interactive Model Playground & Benchmarker

Status: Ready Ā· 0ms // Click 'Run Inference' to test live response streaming across local & cloud providers.
šŸ“ šŸ“ˆ šŸŽ² AI Math Pillars Live Observability Trace Open Full Math Pillars Suite āž”
I. Linear Algebra Tensors & Attention
Attn: softmax(QKįµ€ / āˆšd_k) Ā· V | Cosine: 0.89
Dim: d_model=2048, heads=16, RoPE Īø=10000.0, FLOPs=3.0G
II. Multivariable Calculus Gradients & Loss
Loss ā„’_CE: 1.24 nats | PPL: 3.45 | ā€–gā€–ā‚‚: 0.14
AdamW: Ī·=3.0e-4, Ī²ā‚=0.9, Ī²ā‚‚=0.999, Ī»=0.01 (RMSNorm active)
III. Probability & Info Theory Softmax & Entropy
Entropy H(X): 1.32 bits | Top-1: 78.4%
Sampling: Top-p=0.90, Top-k=40, Temp T=0.7 (Confident)
āœ¦ Hardware & Memory Architecture

GPU VRAM & GGUF Quantization Sizer

FP16 Ā· Q8_0 Ā· Q4_K_M Ā· KV Cache Memory
Model Weights
0.84 GB
KV Cache Memory
0.50 GB
Total Recommended VRAM
1.34 GB
Hardware Tier Match
šŸŸ¢ CPU / 4GB iGPU / Laptop