From 45de2fd47fcc452ebdb0bf79b0283c938d0460ad Mon Sep 17 00:00:00 2001 From: Frank Date: Tue, 6 Oct 2026 23:39:10 -0700 Subject: [PATCH] perf(run-swift15-bf16): calib at the vendor big-step shape MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit MNBT 65536 / seqs 8: MOE-W4.md:244 measures 640 tok/s at a 32k step vs 930 at 55k — the KLD record's own rate confirmed step size drives disk-tier prefill. 131k resident tokens fits the 3072 pool (~237k tokens/rank at fp8). Cuts the 10M-token calibration from ~4.3 h to ~3 h. --- scripts/run-swift15-bf16.sh | 10 +++++++--- 1 file changed, 7 insertions(+), 3 deletions(-) diff --git a/scripts/run-swift15-bf16.sh b/scripts/run-swift15-bf16.sh index ab30b6d..6aa3aed 100755 --- a/scripts/run-swift15-bf16.sh +++ b/scripts/run-swift15-bf16.sh @@ -63,12 +63,16 @@ case "$MODE" in # unparameterisable threshold. Grams are bf16 activations — engine-invariant. # Same KV ceiling as kld: the disk-tier weights floor (9.57 GiB) plus # ~18 GiB of fixed per-rank claims cap the pool near 3.5 GiB on a 32 GiB - # card. 4 slots x 16k fp8 fits; prefill rate is disk-bound, so fewer - # resident slots only queues pushes, it does not slow the record. + # card — pool minus the 2.17 GiB fixed holds ~237k fp8 tokens a rank, so + # the vendor's big-step calibration shape is affordable: MOE-W4.md:244 + # measures 640 tok/s at a 32k step against 930 at 55k, and the doc's own + # run drives MNBT 65536. 8 slots x 16k = 131k resident tokens fits the + # pool; the step batches 65k. KLD's observed rate confirmed step size, + # not slot count, is the throughput lever. IMAGE="stilldeadcode/radiance@$PIN_120" mkdir -p "$CALIB_DIR" ARGS+=(--port 7770 --vram-kv-mib 3072 - --max-num-seqs 4 --max-num-batched-tokens 32768 + --max-num-seqs 8 --max-num-batched-tokens 65536 --prefix-cache-dir /data/calib-kvcache) ;; probe)