H200 NVL · SM90A · RUNTIME STATUS

Governor inference node

Current process state, model configuration, measured throughput, speculative decoding counters, Spectral compatibility, ports, and ingress.

CONNECTINGExplore Spectrals →
Resident model
Architecture
Context
QuantizationW4A16 · compressed tensors
Compatible Spectrals
Resident width
State collection
CompatibilityGUARDED
Network paths
gov.apiary.vision:443→Caddy→127.0.0.1:9100 status
captain.apiary.vision:443→127.0.0.1:8000 gateway→127.0.0.1:9000 vLLM
Direct warm throughput
276.87 tokens/s
Speculative decoding
live process acceptance
Draftergoogle/gemma-4-31B-it-assistant
Method / depthMTP / 3 tokens
Controlled 3×512 benchmark96.95%
GPU telemetry
GiB
Device
Compute
Temperature
Gateway speed
88.72 tokens/s
Gateway
Engine
Qwen Spectral profile
5120D
Compatible States
ArchitectureQwen3_5ForConditionalGeneration
Targetqwen38
Model profile QWEN
··
Quantization
Drafter
Speed
Latest testNot run
R2 model
R2 compiled artifact
Switch command