noonghunna 579a717201 Experimental Gemma-4 QAT W4A16 vLLM composes + kv-calc int4 fix (#339) ...

Two 🧪 experimental QAT-W4A16 vLLM slugs (vllm/gemma-12b-qat-w4a16-single, vllm/gemma-31b-qat-w4a16-dual) + the standalone kv-calc gemma-12B int4 weight-pricing fix. 12B needs the gemma4-unified-vision-unquant sitecustomize workaround (vLLM #44494, both bugs, self-contained); 31B boots clean on stock vllm-gemma-stable. Full suite 42/0.

2026-06-07 05:26:16 +05:00

..

charts

charts: compose names on x-axis + description legend block below

2026-05-20 02:12:09 +00:00

grammar-eval

feat(llama.cpp): Structured-CoT bounded-thinking compose + grammar-dialect fix (#214)

2026-05-24 11:40:42 +05:00

residency-instrument

docs+scripts: finish <quant>/ path migration across full repo sweep

2026-05-26 16:50:13 +00:00

bump-engine-nightlies.sh

fix(tools): bump engine nightly profiles

2026-05-14 21:39:11 +00:00

kv-calc.py

Experimental Gemma-4 QAT W4A16 vLLM composes + kv-calc int4 fix (#339)

2026-06-07 05:26:16 +05:00