1 min read
Tuning Henry's Output Limits
A one-commit day adjusting token limits on the local model personas behind the churner.
Local LLMAIConfiguration
Read more
2 min read
Making the Venue Churn Pipeline Deterministic
A late night stabilizing a fleet of model profiles, capping inference hangs, and promoting a faster quantized fleet.
PythonLocal LLMReliabilityAIInfrastructure
Read more
2 min read
26 Iterations: Porting Cleanroom Kernels to Gemma 4
A single day of rapid iteration took the Gemma 4 port from a baseline to validated QKV fusion, invalidating several assumptions along the way.
SYCLKernelsLocal LLMPerformanceIteration
Read more
2 min read
All-Local 29 Tokens Per Second and Fused Norm Kernels
Fusing QKV, inlining norm-weight application, and pushing block loads through the Q4_K path.
SYCLKernelsPerformanceLocal LLMOptimization
Read more
2 min read
Standalone Inference Working: Shared Memory and a CPU Fallback
The standalone backend finally produced inference after fixing ABI layout, shared-memory buffers, and a simple CPU fallback.
SYCLKernelsABILocal LLMPerformance
Read more
2 min read
Enabling Local Dispatch Op by Op Until the Graph Ran Entirely on My Kernel
47 commits and a progressive enablement series that ended with full local decode dispatch at 7.0 t/s and 100% correctness.
SYCLKernelsPerformanceDebuggingLocal LLM
Read more
2 min read
Rebranding to ArcLab and Patching a Modern Jinja Parser Into llama-server
Rebrand, hybrid llama-server build, IPEX reverse-engineering notes, and a sandbox with a parallel job queue.
Local LLMSYCLIntelPythonTooling
Read more
2 min read
Hunting the Fused Aggregation Kernel Bug in Expert Parallelism
A 12-hour debugging session documented commit by commit until I isolated the EP bug to a fused aggregation kernel.
Local LLMDebuggingKernelsPythonAI
Read more
2 min read
A TUI for OpenClaw Sessions and a Deep-Dive Into Expert Padding
A day split between shipping a terminal viewer for agent sessions and chasing an expert-alignment bug in local model builds.
PythonTUILocal LLMDebuggingTooling
Read more
2 min read
Turning a Hacky Local Server Into a Documented Project
A long documentation-and-research push that took ArcLLM from working code to a three-phase, gated roadmap.
AIPythonLocal LLMDocumentationArchitecture
Read more
2 min read
A Venue-Expert CTA and Wiring a Temps Indicator Into ArcLLM
Turning on the Whitney venue-expert call to action, plus folding a temperature/utilization indicator into the ArcLLM stack.
LaravelFrontendAILocal LLMUX
Read more
2 min read
Building ArcLLM, a Local OpenAI-Compatible API Stack
One session took ArcLLM from an empty repo to streaming chat, tool calls, the responses API, persistence, and worker queues.
AIPythonLocal LLMAPIInfrastructure
Read more