All posts

Name: RobotNurse
Price range: $899-$18,000+

2026-09-25	Our Model Conversion Ran Five Times at Once	Quantization
2026-09-25	We Sent a V4 Model a V3.1 Tool Prompt	Inference Engines
2026-09-25	The Tag Was 128 Tokens Back: An Open Bug	Inference Engines
2026-09-25	fp16 Residual Streams and a Two-Million Activation	Inference Engines
2026-09-19	Every Sub-4-Bit Number We Had Was Uncalibrated	Quantization
2026-09-19	We Published a 36% Speedup and Withdrew It the Next Day	Hardware Notes
2026-09-18	Our Two Machines Ran Different Models Under the Same Name	Hardware Notes
2026-09-18	Speculative Decoding Won Alone and Lost Batched	Inference Engines
2026-09-18	Your Headline tok/s Is a Short-Prompt Number	Running It Daily
2026-09-17	Is Local AI Right for Your Small Business? An Honest Decision Guide	Buyer's Guide
2026-09-17	Local AI on a Compact Device: What a Spark-Style Box Can (and Can't) Do	Buyer's Guide
2026-09-17	Ollama for Beginners: Private, Local LLMs on Your Own Machine	Buyer's Guide
2026-09-17	Run Local AI on a Mac: A Realistic Guide (What Works, What Doesn't)	Buyer's Guide
2026-09-17	What a Private AI Setup Actually Is (and How to Get One)	Buyer's Guide
2026-09-14	Running a private LLM for a small law firm on a Mac Studio — what actually works
2026-09-14	We indexed 45 local-inference benchmark results — the 4.7x gap nobody looks for
2026-09-12	Every GPU Out-of-Memory Error Was One sysctl	Hardware Notes
2026-09-12	17 GB of Attention Buffers Nobody Read	Inference Engines
2026-09-12	Can SSD Offload Fit a Model Twice Your RAM? The Arithmetic.	Hardware Notes
2026-09-12	Our Benchmark Queues Deadlocked on Each Other's pgrep	Running It Daily
2026-09-12	Five Kernel Ideas That Measured Neutral or Worse	Inference Engines
2026-09-12	We Pinned a Hash of Floating-Point Output	Quantization
2026-09-12	HIPAA-Safe AI in a Small Medical or Dental Practice: What's Actually Allowed	Buyer's Guide
2026-09-12	Client-Confidential AI for Law Firms and CPA Practices	Buyer's Guide
2026-09-12	On-Prem vs Cloud AI: The Real Three-Year Cost, Including When Cloud Is Cheaper	Buyer's Guide
2026-09-12	What a Local AI Setup Actually Costs, Start to Finish	Buyer's Guide
2026-09-12	Replacing ChatGPT with a Local Model: What You Keep and What You Lose	Buyer's Guide
2026-09-12	Sizing a Local-AI Box for a 5-to-20-Person Office	Buyer's Guide
2026-09-11	We Put the Neural Engine to Work on Prefill. It Added 19%.	Inference Engines
2026-09-07	Our Chunked-Prefill Fix Was 72x Slower. Every Test Passed.	Inference Engines
2026-09-06	We Wrote Our Own Inference Engine. It Started at 0.9 tok/s.	Inference Engines
2026-09-06	We Panicked the Same Machine Twice in One Day	Running It Daily
2026-09-06	Five Parallel Agents, Zero Commits	Local Stack
2026-09-05	Our Kernel Benchmark Was Measuring the System Cache	Hardware Notes
2026-09-03	We Evaluated a Faster Inference Engine and Rejected It	Inference Engines
2026-09-03	2x Decode From Block-Diffusion Drafting	Inference Engines
2026-09-02	Benchmark the Hardware Before You Benchmark the Software	Hardware Notes
2026-09-02	Your SSH-Launched Benchmark Is Running at nice 5	Hardware Notes
2026-09-02	Our Prefix Cache Died at Token 1,067	Inference Engines
2026-09-02	Our Baseline Was the Three Worst Samples in the Log	Hardware Notes
2026-09-01	The Health Checker Was Manufacturing the Stalls It Detects	Running It Daily
2026-09-01	Unfusing a Kernel Made It 1.63x Faster	Inference Engines
2026-08-30	The 429 That Was Really a 400	Local Stack
2026-08-29	Our Compression Logged Success and Compressed Nothing	Local Stack
2026-08-29	Four Reasons Your Config Edit Didn't Take	Local Stack
2026-08-27	A 192 GB Model on a 256 GB Box	Quantization
2026-08-27	Four Wrong Diagnoses of a Locked-Out Mac	Running It Daily
2026-08-25	We Built a Voice Clone and Never Marked It as Synthetic	Running It Daily
2026-08-24	One Environment Variable Cost Us 60x on Cache Hits	Inference Engines
2026-08-23	One Abandoned Request Wedged the Whole Server	Running It Daily
2026-08-23	We Chased an Audio Bug for Months. The Instrument Was Broken.	Local Stack
2026-08-20	The Crash We Couldn't Reproduce on Purpose	Running It Daily
2026-08-19	Every Failure Silently Scored 0.5	The Model Zoo
2026-08-19	The Fine-Tune Was Dead Weight. The Reference Clip Was Doing the Work.	Local Stack
2026-08-19	max_tokens=900 Yielded Zero Pairs. 3000 Yielded 59.	Local Stack
2026-08-18	64% of Our TTS Compute Synthesized Audio We Threw Away	Local Stack
2026-08-18	22% of Our Training Set Was Invented by Another Model	Local Stack
2026-08-18	A 9.5x Pessimization That Looked Like a Hardware Limit	Hardware Notes
2026-08-17	One Slow Request Locks Out Everything Else	Inference Engines
2026-08-15	Every Surprising Eval Result Was Our Infrastructure	The Model Zoo
2026-08-15	A Safety Hook Truncated Every Code Generation	The Model Zoo
2026-08-11	You Can't Prompt Your Way to Both Commitment and Abstention	The Model Zoo
2026-08-11	What Thinking Mode Actually Costs	The Model Zoo
2026-08-11	We Spent a Week Tuning Prefill. The Bottleneck Was Decode.	Inference Engines
2026-08-11	Your Engine Isn't Hung. It's Saturated.	Running It Daily
2026-08-10	The Benchmark Said 18%. Production Said 0.07%.	Running It Daily
2026-08-10	The Machine Was Up. Everything On It Was Dead.	Running It Daily
2026-08-09	Our GPU Cluster Is Idle 97.4% of the Time	Running It Daily
2026-08-07	JSON Mode Was Causing the Malformed JSON	Inference Engines
2026-08-06	Zero of Twenty-Five Skills Were Loading	Local Stack
2026-08-05	Our Model Stopped Saying No	The Model Zoo