11 long-form essays. Deep dives, opinions, tutorials.
Ref PG-ART-011 Date 2026-08-17 Category LLM Read 8 min
A local inference server answers four requests at the same time with no flag. The --parallel flag sets a slot count, and a named count cuts the KV cache into fixed parts.
Read article > Ref PG-ART-010 Date 2026-08-16 Category LLM Read 9 min
Five temperatures, five samples each, on a local 1.2B: the invented answer survived every setting, including 0. What temperature actually controls, and the knob that fixed it.
Read article > Ref PG-ART-008 Date 2026-07-27 Category LLM Read 7 min
Apache, MIT, Llama Community, Gemma terms and the new OpenMDW-1.1 - what each one obliges you to do, which model families use it, and the three traps that are not in the licence name.
Read article > Ref PG-ART-009 Date 2026-07-27 Category Ocr Read 10 min
Open OCR in July 2026: sizes, VRAM, licences and decode settings - from a 4 MB recogniser that cannot hallucinate to a 0.9B model that outscores Gemini 3 Pro.
Read article > Ref PG-ART-007 Date 2026-07-21 Category Coding Agent Read 7 min
A coding agent spends most of its money reading, not writing. Prewalk lets an expensive model do the reading, then hands the whole context to a cheap model at the exact right moment: the first edit.
Read article > Ref PG-ART-006 Date 2026-07-03 Category LLM Read 7 min
Claude Code hid a watermark that flagged Chinese users. Chinese models write weaker code for Americans. If you cannot diff your stack, it can decide who you are.
Read article > Ref PG-ART-005 Date 2026-05-01 Category Context Read 7 min
Bigger is not better. The advertised number lies. Pick the smallest window that fits the work and you save money, time, and accuracy.
Read article > Ref PG-ART-004 Date 2026-04-30 Category Llama Cpp Read 4 min
43 words a second. 0.29 second wait. 20 of 20 coding answers correct. 19.5 GB used out of 24. The card runs this AI very well - and it can hold a 128k window with 4 chats at the same time.
Read article > Ref PG-ART-003 Date 2026-04-29 Category Llama Cpp Read 4 min
44 words a second. 0.26 second wait. 19 of 20 coding answers worked. 18 GB used out of 24. The card runs this AI well - and it can hold a 128k window with 4 chats at the same time.
Read article > Ref PG-ART-002 Date 2026-04-22 Category VLLM Read 5 min
105 words a second solo. 160 words a second with 2 chats. 0.56 second wait. 32 GB card almost full. Tested on vLLM.
Read article > Ref PG-ART-001 Date 2026-04-16 Category Claude Read 3 min
Three ways to work with Claude. They are not interchangeable. Using the wrong one wastes hours. Here is when to reach for each.
Read article >