4 Commits

Author	SHA1	Message	Date
yy-fighting	0c89fd4fbe	r37 addendum: nomtp sweep raw logs (primary evidence, force-added past *.log ignore)	2026-09-09 09:24:43 +08:00
yy-fighting	4d6dac010c	r37 addendum: B'(nomtp) vs PP+MTP same-profile input/output throughput comparison (i16k/o512 random-ids cc8-64) ... 同栈同脚本 nomtp 对拍：cc8 打平(每token 41.0 vs 41.6ms 拐点实证)、cc16 mtp +6%+TTFT-32%、cc>=32 B' 反超 0.9-5.4%(TPOT 好2x)；图修复把 r36 1.81x 惨败收敛到 ±6% 水平线；生产 r37 mtp 已恢复核验	2026-09-09 09:24:25 +08:00
yy-fighting	911a9a2fb0	r37 addendum: i16k/o512 cc 8-64 sweep (nreq=2cc) on serving container; throughput saturates 130-135 tok/s at cc16+, accept stable; rc=1 root-caused (missing bs_results dir, never existed)	2026-09-09 08:29:51 +08:00
yy-fighting	a3b8f1b2ab	r37: fix PP+MTP verify CUDA graph (pre-planned path missing pp_proxy fill); beats A16 by 40% on killer, now serving on 60.8	2026-09-09 01:24:29 +08:00