Block a user
97c208cfdb
b300eq: TP2PP4 MRR64 retest (pass-1 + ordered v2 + MRR48 control) — c64 gains +27%/+19%, TTFT collapse, -14% 1K c32 config cost attributed
b9e8c0c4a2
CURRENT.md 60.1: two campaign windows restored; D-upgrade recipe is D-config-only; A-baseline verdicts (image/autotune negative, IPC@TP8 real but patch-coupled)
2370d33c73
glm53 dsv4-migration A-baseline (TP8+EAGLE) campaign: image negative, autotune split-negative, IPC@TP8 real win +4.6~9.1% but net-neutral vs production (patch/image coupling); eager-trap profile + day-drift finding (60.1, 09-10)
c5d91ceafe
b300-equivalent matrix: E7b high-concurrency retest (MRR64 + decode-graph buckets 1-64) - graph-drop cliff fixed, report numbers overwritten in place
d063bd63c9
b300-equivalent matrix: add provenance.md (gpu inventory, vram peaks, qg verdicts, in-service preserve/restore record)
21fcca5d16
b300-equivalent matrix: dual-plan (TP2PP4-D vs E7b) full B300 scenario replication on 60.8 - 41 valid points, divide at C=8, E7b usable window <=C8 (MRR16+graph-drop), boundary 256K/512K/896K TP2PP4-only, 4-5x absolute gap vs B300 narrowing to ~2x at boundary prefill; hicache host-layer cold-cache pitfall documented; in-service container preserved-renamed-restored and verified (09-10)
e3476aff86
glm53 dsv4-migration bench: autotune+latest-image winner (+1.2~3.2% all 5 pts, A/B/A confirmed), PCIe-IPC pack negative (-0.4~-3.5%), page-mark kernel N/A for GLM (60.1, 09-09)
f15003d2f5
hit90 bench: add aggregated summaries (41 scored points) + md5 manifest (70 raw logs on 60.8)
ad1853f49b
hit90 scenario bench: TP4PP2-nomtp@0.90 winner, DP-attention/DCP verdicts (60.8 serving, 60.5 v3 delivered, 09-09)
92517e87f0
128k low-cc capacity topology: TP2PP4-nomtp winner (60.8 serving, 60.5 v2 delivered, 09-09)
4d6dac010c
r37 addendum: B'(nomtp) vs PP+MTP same-profile input/output throughput comparison (i16k/o512 random-ids cc8-64)
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)
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
ffda226f5a
PP+MTP r36 de-GLOO: fix + verdict (PP+MTP loses i8k 1.81x to B', graph mode correctness-broken, A16 restored)
123023b6ae
feat(pro6000/GLM-5.3): i8k/o1k/c16 三方案对比压测入库（A/A16/B 轮换实测 + 60.8 转 A16 在役 + CURRENT.md 台账修正 60.2/60.3/60.6/60.8）
b3165a1d3c
feat(pro6000/GLM-5.3): 部署方案入库（TP8+EAGLE 生产标准 / 场景二高并发变体 / TP4PP2+IndexCache / E7b CAR 实验补丁 + deploy profiles）
9e56401384
PP+MTP deepdive r3: race bisect (mask tooling, r34 candidate), decode-round quantification (gloo rendezvous stalls + AR spin dominate; 3-source hypothesis refuted), bench-profile crash forensics
5ed30006a5
[Experiment] GLM-5.3-NVFP4 TP4PP2 round-2 optimization: all config-level quick wins refuted (2026-09-07)