Llama Monitor

Stopped

Server Control

Model Preset Context KV Key KV Val Tensor Split Batch Slots Port no-mmap ngram-spec

gfx906: Q4_0/Q4_1/Q8_0 use optimized MMVQ. Q4_K_M uses generic path (slower). turbo3 KV = 4.6x compression.

Server Logs

Inference

Prompt Speed

—

Generation Speed

—

Context (KV Cache)

—

Slots

—

Server Status

—

GPUs

GPU	Temp	Load	VRAM	Power	SCLK	MCLK

Server Logs