5 Commits

Include renames

Author	SHA1	Message	Date
Frank	8551bb006b	frank-large: roll back to gufo Qwen3.8-27B Q4_K_XL, one session x 262144 f16 KV ... frank-small: llama.cpp CUDA (master, sm_86) on MiMo Q6_K GGUF, 2 x 262144, asymmetric KV (f16 K + q8_0 V, flash-attn pinned for the V quant)	2026-10-01 22:40:59 -07:00
Frank	e452a1afbf	harness: tolerate missing container in ASI capture; llama cmdline fallback	2026-10-01 22:11:34 -07:00
Frank	01ba41cab4	harness: llama-server lifecycle in pre-clean and health guard	2026-10-01 21:58:51 -07:00
Frank	cde91c1a00	llama.cpp runtime arm: serve-script runtime switch, harness containerless support, master build	2026-10-01 21:56:30 -07:00
Frank	5df6ef3166	autoresearch: harness setup ... Benchmark entrypoint: bash autoresearch.sh Goal: Maximize frank-large decode throughput on the Strix Halo iGPU (gfx1151) via quantization >=q6/fp6, engine flags, spec-decode shape, or runtime swap — context limits unchanged, KV stays bf16.	2026-10-01 20:41:51 -07:00