Start with LLMGauge
LLMGauge is the WumboLabs evidence tool: a local-first CLI for validating practical LLM testing workflows on real consumer hardware.
WUMBOLABS / LOCAL LLM TESTING
Real hardware. Real testing. No hype.
WumboLabs builds and documents practical local AI testing workflows on consumer hardware. LLMGauge is the flagship tool: a local-first CLI for running, validating, scoring, and reporting real GGUF model tests through llama.cpp.
LLMGauge is the WumboLabs evidence tool: a local-first CLI for validating practical LLM testing workflows on real consumer hardware.
Read bounded reports, baselines, fit tests, and lab notes from local testing and practical lab work.
Monolith is the local AI workbench layer for tracking models and future LLMGauge import workflows.
Validated guided setup, clean-clone onboarding, doctor/smoke checks, dry-run planning, real llama.cpp runs, artifact validation, manual scoring, report regeneration, and export-index generation.
Current LLMGauge work is focused on proving that a fresh user can configure, run, validate, score, report, and export real local model results without hidden assumptions.
Published a WumboJetsII RTX 5070 12GB practical-use test comparing Gemmable 4 12B, Gemma 4 12B QAT Q4, and Gemma 4 12B UD-Q5.
Tested Mellum2 Instruct and Thinking Q4_K_M through LLMGauge across 8k, ladder, fake-tool, synthetic preload, and 64k agent-backend runs.
Moving WumboLabs away from generic AI landing-page structure and toward a public lab-console layout.
Website now consumes public-safe Monolith metadata and roadmap content directly from the Monolith repo.
Monolith roadmap sync now reads like a roadmap: current phase, next milestone, upcoming phases, and future direction.