[Artifacts] finalize B300 matrix at reclaim cutoff

2026-09-10 15:56:57 +08:00 · 2026-09-10 15:56:57 +08:00 · 81d17407bc
commit 81d17407bc
parent 9652bfdb9d
23 changed files with 1036 additions and 1055 deletions
--- a/experiments/b300/dsv4_flash_glm53_sglang_text_matrix/README.md
+++ b/experiments/b300/dsv4_flash_glm53_sglang_text_matrix/README.md
@ -1,7 +1,7 @@
 # B300 DeepSeek-V4-Flash / GLM-5.3 SGLang text matrix

-This directory archives the test code and all files written by the active B300
-text-matrix run as of the snapshot time.
+This directory archives the test code and all files written by the B300
+text-matrix run through its manual cutoff and cleanup.

 ## Snapshot

@ -20,9 +20,23 @@ Incremental synchronization:
 - At this point DeepSeek-V4-Flash had produced 53 Low-Latency and 18
  Balanced point JSON files.

-The DeepSeek-V4-Flash matrix was still running when this snapshot was taken.
-Consequently, this is a complete snapshot of files present at that time, not
-the final completed run archive. GLM-5.3 had completed its first pass.
+Final synchronization:
+
+- Stop time: `2026-09-10 07:46:07 UTC` (`2026-09-10 15:46:07 Asia/Shanghai`)
+- Final delta archive SHA-256: `f6ba8dbce5db850d43e7583a40eb8fc7cc95aab91c08fc5352fa69ec1d7d1117`
+- Final status: 185/236 points reached a terminal state (78.39%): 164 pass,
+  3 timeout-pass, 18 fail markers, and 51 pending.
+- Mode progress: DSV4 Low-Latency 54/59, DSV4 Balanced 31/59,
+  GLM-5.3 Low-Latency 54/59, and GLM-5.3 High-Throughput 46/59.
+- Both benchmark tmux sessions were stopped, the test container was removed,
+  and `nvidia-smi` reported no remaining compute process.
+
+The 17 GLM-5.3 fail markers on OSL=1 are validation false negatives: the
+benchmark command returned zero and produced complete JSON, while TPOT is
+mathematically zero because a one-token response has no inter-token interval.
+The DSV4 Low-Latency `512k_1_c1` failure is a real GPU OOM. Raw status markers
+are preserved exactly as emitted; use `final-summary/` together with the point
+JSON and logs when interpreting them.

 ## Layout

@ -32,6 +46,8 @@ the final completed run archive. GLM-5.3 had completed its first pass.
 - `snapshot-20260910-024355/b300-dsv4-glm53-text-matrix/`: raw run
  outputs, point logs, service logs, commands, status markers, and summaries.
 - `snapshot-20260910-024355/*.log`: top-level orchestration logs.
+- `snapshot-20260910-024355/b300-dsv4-glm53-text-matrix/b300-dsv4-glm53-dev4-20260909-084427/final-summary/`:
+  final `summary.md`, `summary.csv`, `comparison.md`, and `progress.json`.

 The repository normally excludes raw benchmark logs. They are intentionally
 included here because this snapshot is an explicit full-run archive.
--- a/experiments/b300/dsv4_flash_glm53_sglang_text_matrix/snapshot-20260910-024355/b300-dsv4-glm53-dev4-20260909-084427-followup.log
+++ b/experiments/b300/dsv4_flash_glm53_sglang_text_matrix/snapshot-20260910-024355/b300-dsv4-glm53-dev4-20260909-084427-followup.log
@ -44,3 +44,188 @@
 [2026-09-10T02:42:10+00:00] warning profile=dsv4-low-latency point=16k_1_c128 zero_ttft_count=4
 [2026-09-10T02:42:10+00:00] pass profile=dsv4-low-latency point=16k_1_c128
 [2026-09-10T02:42:11+00:00] run profile=dsv4-low-latency shape=16k_1 isl=16384 osl=1 c=256 prompts=1280
+[2026-09-10T02:47:47+00:00] warning profile=dsv4-low-latency point=16k_1_c256 zero_ttft_count=7
+[2026-09-10T02:47:47+00:00] pass profile=dsv4-low-latency point=16k_1_c256
+[2026-09-10T02:47:48+00:00] run profile=dsv4-low-latency shape=1k_128 isl=1024 osl=128 c=1 prompts=64
+[2026-09-10T02:48:36+00:00] pass profile=dsv4-low-latency point=1k_128_c1
+[2026-09-10T02:48:37+00:00] run profile=dsv4-low-latency shape=1k_128 isl=1024 osl=128 c=8 prompts=64
+[2026-09-10T02:49:08+00:00] pass profile=dsv4-low-latency point=1k_128_c8
+[2026-09-10T02:49:09+00:00] run profile=dsv4-low-latency shape=1k_128 isl=1024 osl=128 c=32 prompts=160
+[2026-09-10T02:49:45+00:00] pass profile=dsv4-low-latency point=1k_128_c32
+[2026-09-10T02:49:46+00:00] run profile=dsv4-low-latency shape=1k_128 isl=1024 osl=128 c=64 prompts=320
+[2026-09-10T02:50:28+00:00] pass profile=dsv4-low-latency point=1k_128_c64
+[2026-09-10T02:50:29+00:00] run profile=dsv4-low-latency shape=1k_128 isl=1024 osl=128 c=128 prompts=640
+[2026-09-10T02:51:13+00:00] pass profile=dsv4-low-latency point=1k_128_c128
+[2026-09-10T02:51:14+00:00] run profile=dsv4-low-latency shape=1k_128 isl=1024 osl=128 c=256 prompts=1280
+[2026-09-10T02:52:12+00:00] pass profile=dsv4-low-latency point=1k_128_c256
+[2026-09-10T02:52:13+00:00] run profile=dsv4-low-latency shape=1k_4k isl=1024 osl=4096 c=1 prompts=64
+[2026-09-10T02:59:40+00:00] pass profile=dsv4-low-latency point=1k_4k_c1
+[2026-09-10T02:59:41+00:00] run profile=dsv4-low-latency shape=1k_4k isl=1024 osl=4096 c=8 prompts=64
+[2026-09-10T03:01:28+00:00] pass profile=dsv4-low-latency point=1k_4k_c8
+[2026-09-10T03:01:29+00:00] run profile=dsv4-low-latency shape=1k_4k isl=1024 osl=4096 c=32 prompts=160
+[2026-09-10T03:03:21+00:00] pass profile=dsv4-low-latency point=1k_4k_c32
+[2026-09-10T03:03:22+00:00] run profile=dsv4-low-latency shape=1k_4k isl=1024 osl=4096 c=64 prompts=320
+[2026-09-10T03:05:53+00:00] pass profile=dsv4-low-latency point=1k_4k_c64
+[2026-09-10T03:05:54+00:00] run profile=dsv4-low-latency shape=1k_4k isl=1024 osl=4096 c=128 prompts=640
+[2026-09-10T03:09:39+00:00] pass profile=dsv4-low-latency point=1k_4k_c128
+[2026-09-10T03:09:40+00:00] run profile=dsv4-low-latency shape=1k_4k isl=1024 osl=4096 c=256 prompts=1280
+[2026-09-10T03:14:18+00:00] pass profile=dsv4-low-latency point=1k_4k_c256
+[2026-09-10T03:14:19+00:00] run profile=dsv4-low-latency shape=64k_1 isl=65536 osl=1 c=1 prompts=64
+[2026-09-10T03:15:49+00:00] warning profile=dsv4-low-latency point=64k_1_c1 zero_ttft_count=1
+[2026-09-10T03:15:49+00:00] pass profile=dsv4-low-latency point=64k_1_c1
+[2026-09-10T03:15:50+00:00] run profile=dsv4-low-latency shape=64k_1 isl=65536 osl=1 c=4 prompts=64
+[2026-09-10T03:17:15+00:00] warning profile=dsv4-low-latency point=64k_1_c4 zero_ttft_count=1
+[2026-09-10T03:17:15+00:00] pass profile=dsv4-low-latency point=64k_1_c4
+[2026-09-10T03:17:16+00:00] run profile=dsv4-low-latency shape=64k_1 isl=65536 osl=1 c=8 prompts=64
+[2026-09-10T03:18:42+00:00] warning profile=dsv4-low-latency point=64k_1_c8 zero_ttft_count=1
+[2026-09-10T03:18:42+00:00] pass profile=dsv4-low-latency point=64k_1_c8
+[2026-09-10T03:18:43+00:00] run profile=dsv4-low-latency shape=64k_1 isl=65536 osl=1 c=16 prompts=80
+[2026-09-10T03:20:27+00:00] warning profile=dsv4-low-latency point=64k_1_c16 zero_ttft_count=1
+[2026-09-10T03:20:27+00:00] pass profile=dsv4-low-latency point=64k_1_c16
+[2026-09-10T03:20:28+00:00] run profile=dsv4-low-latency shape=64k_1 isl=65536 osl=1 c=32 prompts=160
+[2026-09-10T03:23:43+00:00] warning profile=dsv4-low-latency point=64k_1_c32 zero_ttft_count=1
+[2026-09-10T03:23:43+00:00] pass profile=dsv4-low-latency point=64k_1_c32
+[2026-09-10T03:23:44+00:00] run profile=dsv4-low-latency shape=64k_1 isl=65536 osl=1 c=64 prompts=320
+[2026-09-10T03:30:02+00:00] warning profile=dsv4-low-latency point=64k_1_c64 zero_ttft_count=3
+[2026-09-10T03:30:02+00:00] pass profile=dsv4-low-latency point=64k_1_c64
+[2026-09-10T03:30:03+00:00] run profile=dsv4-low-latency shape=64k_1 isl=65536 osl=1 c=128 prompts=640
+[2026-09-10T03:42:28+00:00] warning profile=dsv4-low-latency point=64k_1_c128 zero_ttft_count=5
+[2026-09-10T03:42:28+00:00] pass profile=dsv4-low-latency point=64k_1_c128
+[2026-09-10T03:42:29+00:00] run profile=dsv4-low-latency shape=64k_512 isl=65536 osl=512 c=1 prompts=64
+[2026-09-10T03:44:52+00:00] pass profile=dsv4-low-latency point=64k_512_c1
+[2026-09-10T03:44:53+00:00] run profile=dsv4-low-latency shape=64k_512 isl=65536 osl=512 c=4 prompts=64
+[2026-09-10T03:46:37+00:00] pass profile=dsv4-low-latency point=64k_512_c4
+[2026-09-10T03:46:38+00:00] run profile=dsv4-low-latency shape=64k_512 isl=65536 osl=512 c=8 prompts=64
+[2026-09-10T03:48:14+00:00] pass profile=dsv4-low-latency point=64k_512_c8
+[2026-09-10T03:48:15+00:00] run profile=dsv4-low-latency shape=64k_512 isl=65536 osl=512 c=16 prompts=80
+[2026-09-10T03:50:06+00:00] pass profile=dsv4-low-latency point=64k_512_c16
+[2026-09-10T03:50:07+00:00] run profile=dsv4-low-latency shape=64k_512 isl=65536 osl=512 c=32 prompts=160
+[2026-09-10T03:53:33+00:00] pass profile=dsv4-low-latency point=64k_512_c32
+[2026-09-10T03:53:34+00:00] run profile=dsv4-low-latency shape=64k_512 isl=65536 osl=512 c=64 prompts=320
+[2026-09-10T04:00:06+00:00] pass profile=dsv4-low-latency point=64k_512_c64
+[2026-09-10T04:00:07+00:00] run profile=dsv4-low-latency shape=128k_1 isl=131072 osl=1 c=1 prompts=64
+[2026-09-10T04:03:11+00:00] pass profile=dsv4-low-latency point=128k_1_c1
+[2026-09-10T04:03:12+00:00] run profile=dsv4-low-latency shape=128k_1 isl=131072 osl=1 c=2 prompts=64
+[2026-09-10T04:06:11+00:00] pass profile=dsv4-low-latency point=128k_1_c2
+[2026-09-10T04:06:12+00:00] run profile=dsv4-low-latency shape=128k_1 isl=131072 osl=1 c=4 prompts=64
+[2026-09-10T04:09:12+00:00] pass profile=dsv4-low-latency point=128k_1_c4
+[2026-09-10T04:09:13+00:00] run profile=dsv4-low-latency shape=128k_1 isl=131072 osl=1 c=8 prompts=64
+[2026-09-10T04:12:12+00:00] pass profile=dsv4-low-latency point=128k_1_c8
+[2026-09-10T04:12:13+00:00] run profile=dsv4-low-latency shape=128k_1 isl=131072 osl=1 c=16 prompts=80
+[2026-09-10T04:15:55+00:00] pass profile=dsv4-low-latency point=128k_1_c16
+[2026-09-10T04:15:56+00:00] run profile=dsv4-low-latency shape=128k_1 isl=131072 osl=1 c=32 prompts=160
+[2026-09-10T04:23:05+00:00] pass profile=dsv4-low-latency point=128k_1_c32
+[2026-09-10T04:23:06+00:00] run profile=dsv4-low-latency shape=128k_1 isl=131072 osl=1 c=64 prompts=320
+[2026-09-10T04:37:16+00:00] warning profile=dsv4-low-latency point=128k_1_c64 zero_ttft_count=2
+[2026-09-10T04:37:16+00:00] pass profile=dsv4-low-latency point=128k_1_c64
+[2026-09-10T04:37:17+00:00] run profile=dsv4-low-latency shape=128k_512 isl=131072 osl=512 c=1 prompts=64
+[2026-09-10T04:42:26+00:00] pass profile=dsv4-low-latency point=128k_512_c1
+[2026-09-10T04:42:27+00:00] run profile=dsv4-low-latency shape=128k_512 isl=131072 osl=512 c=2 prompts=64
+[2026-09-10T04:46:39+00:00] pass profile=dsv4-low-latency point=128k_512_c2
+[2026-09-10T04:46:40+00:00] run profile=dsv4-low-latency shape=128k_512 isl=131072 osl=512 c=4 prompts=64
+[2026-09-10T04:50:19+00:00] pass profile=dsv4-low-latency point=128k_512_c4
+[2026-09-10T04:50:21+00:00] run profile=dsv4-low-latency shape=128k_512 isl=131072 osl=512 c=8 prompts=64
+[2026-09-10T04:53:45+00:00] pass profile=dsv4-low-latency point=128k_512_c8
+[2026-09-10T04:53:46+00:00] run profile=dsv4-low-latency shape=128k_512 isl=131072 osl=512 c=16 prompts=80
+[2026-09-10T04:57:47+00:00] pass profile=dsv4-low-latency point=128k_512_c16
+[2026-09-10T04:57:48+00:00] run profile=dsv4-low-latency shape=128k_512 isl=131072 osl=512 c=32 prompts=160
+[2026-09-10T05:05:25+00:00] pass profile=dsv4-low-latency point=128k_512_c32
+[2026-09-10T05:05:26+00:00] run profile=dsv4-low-latency shape=256k_1 isl=262144 osl=1 c=1 prompts=16
+[2026-09-10T05:07:29+00:00] pass profile=dsv4-low-latency point=256k_1_c1
+[2026-09-10T05:07:31+00:00] run profile=dsv4-low-latency shape=256k_1 isl=262144 osl=1 c=2 prompts=16
+[2026-09-10T05:09:29+00:00] pass profile=dsv4-low-latency point=256k_1_c2
+[2026-09-10T05:09:30+00:00] run profile=dsv4-low-latency shape=256k_1 isl=262144 osl=1 c=4 prompts=16
+[2026-09-10T05:11:28+00:00] pass profile=dsv4-low-latency point=256k_1_c4
+[2026-09-10T05:11:29+00:00] run profile=dsv4-low-latency shape=512k_1 isl=524288 osl=1 c=1 prompts=8
+[2026-09-10T05:12:50+00:00] failed profile=dsv4-low-latency point=512k_1_c1 rc=1
+[2026-09-10T05:12:50+00:00] capacity short-circuit armed profile=dsv4-low-latency shape=512k_1 after c=1
+[2026-09-10T05:12:50+00:00] server failure aborts profile=dsv4-low-latency shape=512k_1
+[2026-09-10T05:12:50+00:00] ERROR: unhealthy before profile=dsv4-low-latency point=1m_boundary_1_c1
+[2026-09-10T05:12:50+00:00] server failure aborts profile=dsv4-low-latency shape=1m_boundary_1
+[2026-09-10T05:13:22+00:00] starting profile=dsv4-balanced model=dsv4-flash-0731
+[2026-09-10T05:13:23+00:00] waiting for server profile=dsv4-balanced elapsed=0s
+[2026-09-10T05:14:24+00:00] waiting for server profile=dsv4-balanced elapsed=60s
+[2026-09-10T05:15:25+00:00] waiting for server profile=dsv4-balanced elapsed=120s
+[2026-09-10T05:16:25+00:00] waiting for server profile=dsv4-balanced elapsed=180s
+[2026-09-10T05:17:26+00:00] waiting for server profile=dsv4-balanced elapsed=240s
+[2026-09-10T05:18:26+00:00] waiting for server profile=dsv4-balanced elapsed=300s
+[2026-09-10T05:19:27+00:00] waiting for server profile=dsv4-balanced elapsed=360s
+[2026-09-10T05:20:27+00:00] waiting for server profile=dsv4-balanced elapsed=420s
+[2026-09-10T05:21:28+00:00] waiting for server profile=dsv4-balanced elapsed=480s
+[2026-09-10T05:22:29+00:00] waiting for server profile=dsv4-balanced elapsed=540s
+[2026-09-10T05:23:00+00:00] server healthy profile=dsv4-balanced elapsed=570s
+[2026-09-10T05:23:18+00:00] run profile=dsv4-balanced shape=16k_512 isl=16384 osl=512 c=1 prompts=64
+[2026-09-10T05:30:32+00:00] pass profile=dsv4-balanced point=16k_512_c1
+[2026-09-10T05:30:33+00:00] run profile=dsv4-balanced shape=16k_512 isl=16384 osl=512 c=8 prompts=64
+[2026-09-10T05:31:50+00:00] pass profile=dsv4-balanced point=16k_512_c8
+[2026-09-10T05:31:51+00:00] run profile=dsv4-balanced shape=16k_512 isl=16384 osl=512 c=32 prompts=160
+[2026-09-10T05:33:19+00:00] pass profile=dsv4-balanced point=16k_512_c32
+[2026-09-10T05:33:20+00:00] run profile=dsv4-balanced shape=16k_512 isl=16384 osl=512 c=64 prompts=320
+[2026-09-10T05:35:32+00:00] pass profile=dsv4-balanced point=16k_512_c64
+[2026-09-10T05:35:33+00:00] run profile=dsv4-balanced shape=16k_512 isl=16384 osl=512 c=128 prompts=640
+[2026-09-10T05:39:14+00:00] pass profile=dsv4-balanced point=16k_512_c128
+[2026-09-10T05:39:15+00:00] run profile=dsv4-balanced shape=16k_512 isl=16384 osl=512 c=256 prompts=1280
+[2026-09-10T05:45:44+00:00] pass profile=dsv4-balanced point=16k_512_c256
+[2026-09-10T05:45:45+00:00] run profile=dsv4-balanced shape=16k_1 isl=16384 osl=1 c=1 prompts=64
+[2026-09-10T05:47:02+00:00] pass profile=dsv4-balanced point=16k_1_c1
+[2026-09-10T05:47:03+00:00] run profile=dsv4-balanced shape=16k_1 isl=16384 osl=1 c=8 prompts=64
+[2026-09-10T05:47:34+00:00] pass profile=dsv4-balanced point=16k_1_c8
+[2026-09-10T05:47:35+00:00] run profile=dsv4-balanced shape=16k_1 isl=16384 osl=1 c=32 prompts=160
+[2026-09-10T05:48:31+00:00] warning profile=dsv4-balanced point=16k_1_c32 zero_ttft_count=2
+[2026-09-10T05:48:31+00:00] pass profile=dsv4-balanced point=16k_1_c32
+[2026-09-10T05:48:32+00:00] run profile=dsv4-balanced shape=16k_1 isl=16384 osl=1 c=64 prompts=320
+[2026-09-10T05:50:09+00:00] warning profile=dsv4-balanced point=16k_1_c64 zero_ttft_count=2
+[2026-09-10T05:50:09+00:00] pass profile=dsv4-balanced point=16k_1_c64
+[2026-09-10T05:50:10+00:00] run profile=dsv4-balanced shape=16k_1 isl=16384 osl=1 c=128 prompts=640
+[2026-09-10T05:53:09+00:00] warning profile=dsv4-balanced point=16k_1_c128 zero_ttft_count=4
+[2026-09-10T05:53:09+00:00] pass profile=dsv4-balanced point=16k_1_c128
+[2026-09-10T05:53:10+00:00] run profile=dsv4-balanced shape=16k_1 isl=16384 osl=1 c=256 prompts=1280
+[2026-09-10T05:58:59+00:00] warning profile=dsv4-balanced point=16k_1_c256 zero_ttft_count=7
+[2026-09-10T05:58:59+00:00] pass profile=dsv4-balanced point=16k_1_c256
+[2026-09-10T05:59:00+00:00] run profile=dsv4-balanced shape=1k_128 isl=1024 osl=128 c=1 prompts=64
+[2026-09-10T06:00:58+00:00] pass profile=dsv4-balanced point=1k_128_c1
+[2026-09-10T06:00:59+00:00] run profile=dsv4-balanced shape=1k_128 isl=1024 osl=128 c=8 prompts=64
+[2026-09-10T06:01:28+00:00] pass profile=dsv4-balanced point=1k_128_c8
+[2026-09-10T06:01:29+00:00] run profile=dsv4-balanced shape=1k_128 isl=1024 osl=128 c=32 prompts=160
+[2026-09-10T06:01:55+00:00] pass profile=dsv4-balanced point=1k_128_c32
+[2026-09-10T06:01:56+00:00] run profile=dsv4-balanced shape=1k_128 isl=1024 osl=128 c=64 prompts=320
+[2026-09-10T06:02:25+00:00] pass profile=dsv4-balanced point=1k_128_c64
+[2026-09-10T06:02:26+00:00] run profile=dsv4-balanced shape=1k_128 isl=1024 osl=128 c=128 prompts=640
+[2026-09-10T06:03:03+00:00] pass profile=dsv4-balanced point=1k_128_c128
+[2026-09-10T06:03:04+00:00] run profile=dsv4-balanced shape=1k_128 isl=1024 osl=128 c=256 prompts=1280
+[2026-09-10T06:03:55+00:00] pass profile=dsv4-balanced point=1k_128_c256
+[2026-09-10T06:03:56+00:00] run profile=dsv4-balanced shape=1k_4k isl=1024 osl=4096 c=1 prompts=64
+run_full_text_matrix.sh: line 247: 1797795 Killed                  timeout --signal=TERM --kill-after=120 "${MAX_POINT_SECONDS}" sudo docker exec "${CONTAINER}" python3 -m sglang.bench_serving --backend sglang --host 127.0.0.1 --port "${PORT}" --model "${MODEL_CONTAINER}" --tokenizer "${MODEL_CONTAINER}" --dataset-name random-ids --tokenize-prompt --random-input-len "${isl}" --random-output-len "${osl}" --random-range-ratio 1.0 --num-prompts "${prompts}" --max-concurrency "${concurrency}" --seed "${SEED}" --disable-tqdm --flush-cache --output-details --output-file "/results/points/${point}.json" > "${console}" 2>&1
+[2026-09-10T06:35:57+00:00] timeout pass profile=dsv4-balanced point=1k_4k_c1
+[2026-09-10T06:35:58+00:00] run profile=dsv4-balanced shape=1k_4k isl=1024 osl=4096 c=8 prompts=64
+[2026-09-10T06:42:24+00:00] pass profile=dsv4-balanced point=1k_4k_c8
+[2026-09-10T06:42:25+00:00] run profile=dsv4-balanced shape=1k_4k isl=1024 osl=4096 c=32 prompts=160
+[2026-09-10T06:47:31+00:00] pass profile=dsv4-balanced point=1k_4k_c32
+[2026-09-10T06:47:32+00:00] run profile=dsv4-balanced shape=1k_4k isl=1024 osl=4096 c=64 prompts=320
+[2026-09-10T06:53:11+00:00] pass profile=dsv4-balanced point=1k_4k_c64
+[2026-09-10T06:53:12+00:00] run profile=dsv4-balanced shape=1k_4k isl=1024 osl=4096 c=128 prompts=640
+[2026-09-10T06:59:43+00:00] pass profile=dsv4-balanced point=1k_4k_c128
+[2026-09-10T06:59:44+00:00] run profile=dsv4-balanced shape=1k_4k isl=1024 osl=4096 c=256 prompts=1280
+[2026-09-10T07:09:13+00:00] pass profile=dsv4-balanced point=1k_4k_c256
+[2026-09-10T07:09:14+00:00] run profile=dsv4-balanced shape=64k_1 isl=65536 osl=1 c=1 prompts=64
+[2026-09-10T07:13:38+00:00] warning profile=dsv4-balanced point=64k_1_c1 zero_ttft_count=1
+[2026-09-10T07:13:38+00:00] pass profile=dsv4-balanced point=64k_1_c1
+[2026-09-10T07:13:39+00:00] run profile=dsv4-balanced shape=64k_1 isl=65536 osl=1 c=4 prompts=64
+[2026-09-10T07:15:06+00:00] warning profile=dsv4-balanced point=64k_1_c4 zero_ttft_count=1
+[2026-09-10T07:15:06+00:00] pass profile=dsv4-balanced point=64k_1_c4
+[2026-09-10T07:15:07+00:00] run profile=dsv4-balanced shape=64k_1 isl=65536 osl=1 c=8 prompts=64
+[2026-09-10T07:16:30+00:00] warning profile=dsv4-balanced point=64k_1_c8 zero_ttft_count=1
+[2026-09-10T07:16:30+00:00] pass profile=dsv4-balanced point=64k_1_c8
+[2026-09-10T07:16:31+00:00] run profile=dsv4-balanced shape=64k_1 isl=65536 osl=1 c=16 prompts=80
+[2026-09-10T07:18:09+00:00] warning profile=dsv4-balanced point=64k_1_c16 zero_ttft_count=1
+[2026-09-10T07:18:09+00:00] pass profile=dsv4-balanced point=64k_1_c16
+[2026-09-10T07:18:10+00:00] run profile=dsv4-balanced shape=64k_1 isl=65536 osl=1 c=32 prompts=160
+[2026-09-10T07:21:11+00:00] warning profile=dsv4-balanced point=64k_1_c32 zero_ttft_count=1
+[2026-09-10T07:21:11+00:00] pass profile=dsv4-balanced point=64k_1_c32
+[2026-09-10T07:21:12+00:00] run profile=dsv4-balanced shape=64k_1 isl=65536 osl=1 c=64 prompts=320
+[2026-09-10T07:26:57+00:00] warning profile=dsv4-balanced point=64k_1_c64 zero_ttft_count=3
+[2026-09-10T07:26:57+00:00] pass profile=dsv4-balanced point=64k_1_c64
+[2026-09-10T07:26:58+00:00] run profile=dsv4-balanced shape=64k_1 isl=65536 osl=1 c=128 prompts=640
+[2026-09-10T07:38:13+00:00] warning profile=dsv4-balanced point=64k_1_c128 zero_ttft_count=5
+[2026-09-10T07:38:13+00:00] pass profile=dsv4-balanced point=64k_1_c128
+[2026-09-10T07:38:14+00:00] run profile=dsv4-balanced shape=64k_512 isl=65536 osl=512 c=1 prompts=64
--- a/experiments/b300/dsv4_flash_glm53_sglang_text_matrix/snapshot-20260910-024355/b300-dsv4-glm53-text-matrix/b300-dsv4-glm53-dev4-20260909-084427/deadline-cleanup.complete
+++ b/experiments/b300/dsv4_flash_glm53_sglang_text_matrix/snapshot-20260910-024355/b300-dsv4-glm53-text-matrix/b300-dsv4-glm53-dev4-20260909-084427/deadline-cleanup.complete
@ -0,0 +1 @@
+2026-09-10T07:45:20+00:00
--- a/experiments/b300/dsv4_flash_glm53_sglang_text_matrix/snapshot-20260910-024355/b300-dsv4-glm53-text-matrix/b300-dsv4-glm53-dev4-20260909-084427/deadline-summary/comparison.md
+++ b/experiments/b300/dsv4_flash_glm53_sglang_text_matrix/snapshot-20260910-024355/b300-dsv4-glm53-text-matrix/b300-dsv4-glm53-dev4-20260909-084427/deadline-summary/comparison.md
@ -0,0 +1,121 @@
+# B300 DeepSeek-V4-Flash / GLM-5.3 Comparison: b300-dsv4-glm53-dev4-20260909-084427
+
+## Run Status
+
+- Finished: 78.39%
+- Pass: 164
+- Timeout pass: 3
+- Invalid: 0
+- Fail: 18
+
+## Service Configuration And Capacity
+
+| Profile | TP | DP | EP | Max running | Max tokens/rank | Chunk | KV dtype | Attention | MoE runner | MoE A2A | Speculative |
+|---|---:|---:|---:|---:|---:|---:|---|---|---|---|---|
+| dsv4-low-latency | - | - | - | - | - | - | - | - | - | - | - |
+| dsv4-balanced | - | - | - | - | - | - | - | - | - | - | - |
+| glm53-low-latency | 8 | 1 | 1 | 48 | 2696000 | 16384 | fp8_e4m3 | dsa | flashinfer_trtllm | none | EAGLE |
+| glm53-high-throughput | 8 | 8 | 8 | 256 | 2542208 | 4096 | fp8_e4m3 | dsa | auto | deepep | EAGLE |
+
+## Throughput Peaks And Their Latency
+
+Only validated PASS points are considered.
+
+| Shape | Mode | Peak input TPS @ C | TTFT P95 (ms) | Peak output TPS @ C | TPOT P95 (ms) |
+|---|---|---:|---:|---:|---:|
+| 16k_512 | dsv4-low-latency | 60485.71 @ C256 | 48385.38 | 1890.18 @ C256 | 226.23 |
+| 16k_512 | dsv4-balanced | 57436.40 @ C256 | 62849.59 | 1794.89 @ C256 | 134.28 |
+| 16k_512 | glm53-low-latency | 24021.20 @ C256 | 144384.42 | 750.66 @ C256 | 102.36 |
+| 16k_512 | glm53-high-throughput | 31889.21 @ C128 | 30890.29 | 996.54 @ C128 | 165.33 |
+| 16k_1 | dsv4-low-latency | 66726.27 @ C32 | 7810.68 | 4.07 @ C32 | 0.00 |
+| 16k_1 | dsv4-balanced | 65093.70 @ C32 | 7943.21 | 3.97 @ C32 | 0.00 |
+| 1k_128 | dsv4-low-latency | 29913.48 @ C256 | 3544.47 | 3739.19 @ C256 | 99.14 |
+| 1k_128 | dsv4-balanced | 36265.26 @ C256 | 4505.51 | 4533.16 @ C256 | 59.65 |
+| 1k_128 | glm53-low-latency | 14308.93 @ C64 | 1848.75 | 1788.62 @ C64 | 40.49 |
+| 1k_128 | glm53-high-throughput | 15193.07 @ C128 | 2569.96 | 1899.13 @ C128 | 87.74 |
+| 1k_4k | dsv4-low-latency | 5283.08 @ C256 | 3658.46 | 21132.32 @ C256 | 17.31 |
+| 1k_4k | dsv4-balanced | 2441.07 @ C256 | 4386.00 | 9764.26 @ C256 | 26.51 |
+| 1k_4k | glm53-low-latency | 1530.74 @ C128 | 56046.13 | 6122.95 @ C128 | 11.08 |
+| 1k_4k | glm53-high-throughput | 2653.31 @ C256 | 4540.61 | 10613.25 @ C256 | 22.36 |
+| 64k_1 | dsv4-low-latency | 58913.97 @ C4 | 4440.74 | 0.90 @ C4 | 0.00 |
+| 64k_1 | dsv4-balanced | 65864.61 @ C16 | 15823.61 | 1.01 @ C16 | 0.00 |
+| 64k_1 | glm53-low-latency | 25336.86 @ C8 | 20633.79 | 0.39 @ C8 | 0.00 |
+| 64k_1 | glm53-high-throughput | 53720.01 @ C8 | 9891.34 | 0.82 @ C8 | 0.00 |
+| 64k_512 | dsv4-low-latency | 56712.78 @ C64 | 55290.66 | 443.07 @ C64 | 238.49 |
+| 64k_512 | glm53-low-latency | 24220.17 @ C64 | 128260.27 | 189.22 @ C64 | 323.81 |
+| 64k_512 | glm53-high-throughput | 33437.72 @ C64 | 57344.18 | 261.23 @ C64 | 329.77 |
+| 128k_1 | dsv4-low-latency | 51351.81 @ C2 | 5095.93 | 0.39 @ C2 | 0.00 |
+| 128k_1 | glm53-low-latency | 22544.03 @ C2 | 11620.01 | 0.17 @ C2 | 0.00 |
+| 128k_512 | dsv4-low-latency | 48253.24 @ C32 | 62228.62 | 188.49 @ C32 | 214.99 |
+| 128k_512 | glm53-low-latency | 21787.55 @ C32 | 142627.99 | 85.11 @ C32 | 339.34 |
+| 128k_512 | glm53-high-throughput | 27503.67 @ C32 | 59905.46 | 107.44 @ C32 | 401.75 |
+| 256k_1 | dsv4-low-latency | 40426.30 @ C2 | 13069.41 | 0.15 @ C2 | 0.00 |
+| 256k_1 | glm53-low-latency | 17620.27 @ C2 | 29920.79 | 0.07 @ C2 | 0.00 |
+| 512k_1 | glm53-low-latency | 12497.10 @ C2 | 84423.37 | 0.02 @ C2 | 0.00 |
+| 512k_1 | glm53-high-throughput | 24677.77 @ C4 | 86777.89 | 0.05 @ C4 | 0.00 |
+| 1m_boundary_1 | glm53-low-latency | 7914.10 @ C4 | 509514.56 | 0.01 @ C4 | 0.00 |
+| 1m_boundary_1 | glm53-high-throughput | 18580.87 @ C4 | 225402.20 | 0.02 @ C4 | 0.00 |
+
+## Capacity And Failure Boundaries
+
+| Mode | Shape | C | Status | Evidence |
+|---|---|---:|---|---|
+| dsv4-low-latency | 512k_1 | 1 | FAIL | `/data/b300-dsv4-glm53-text-matrix/b300-dsv4-glm53-dev4-20260909-084427/dsv4-low-latency/points/512k_1_c1.failed` |
+| dsv4-balanced | 1k_4k | 1 | TIMEOUT_PASS | `/data/b300-dsv4-glm53-text-matrix/b300-dsv4-glm53-dev4-20260909-084427/dsv4-balanced/points/1k_4k_c1.json` |
+| glm53-low-latency | 16k_1 | 1 | FAIL | `/data/b300-dsv4-glm53-text-matrix/b300-dsv4-glm53-dev4-20260909-084427/glm53-low-latency/points/16k_1_c1.json` |
+| glm53-low-latency | 64k_1 | 4 | FAIL | `/data/b300-dsv4-glm53-text-matrix/b300-dsv4-glm53-dev4-20260909-084427/glm53-low-latency/points/64k_1_c4.json` |
+| glm53-low-latency | 64k_1 | 16 | FAIL | `/data/b300-dsv4-glm53-text-matrix/b300-dsv4-glm53-dev4-20260909-084427/glm53-low-latency/points/64k_1_c16.json` |
+| glm53-low-latency | 64k_1 | 32 | FAIL | `/data/b300-dsv4-glm53-text-matrix/b300-dsv4-glm53-dev4-20260909-084427/glm53-low-latency/points/64k_1_c32.json` |
+| glm53-low-latency | 64k_1 | 64 | FAIL | `/data/b300-dsv4-glm53-text-matrix/b300-dsv4-glm53-dev4-20260909-084427/glm53-low-latency/points/64k_1_c64.json` |
+| glm53-low-latency | 64k_1 | 128 | FAIL | `/data/b300-dsv4-glm53-text-matrix/b300-dsv4-glm53-dev4-20260909-084427/glm53-low-latency/points/64k_1_c128.json` |
+| glm53-low-latency | 128k_1 | 8 | FAIL | `/data/b300-dsv4-glm53-text-matrix/b300-dsv4-glm53-dev4-20260909-084427/glm53-low-latency/points/128k_1_c8.json` |
+| glm53-low-latency | 128k_1 | 16 | FAIL | `/data/b300-dsv4-glm53-text-matrix/b300-dsv4-glm53-dev4-20260909-084427/glm53-low-latency/points/128k_1_c16.json` |
+| glm53-low-latency | 128k_1 | 32 | FAIL | `/data/b300-dsv4-glm53-text-matrix/b300-dsv4-glm53-dev4-20260909-084427/glm53-low-latency/points/128k_1_c32.json` |
+| glm53-low-latency | 128k_1 | 64 | TIMEOUT_PASS | `/data/b300-dsv4-glm53-text-matrix/b300-dsv4-glm53-dev4-20260909-084427/glm53-low-latency/points/128k_1_c64.json` |
+| glm53-low-latency | 256k_1 | 4 | FAIL | `/data/b300-dsv4-glm53-text-matrix/b300-dsv4-glm53-dev4-20260909-084427/glm53-low-latency/points/256k_1_c4.json` |
+| glm53-high-throughput | 16k_1 | 1 | FAIL | `/data/b300-dsv4-glm53-text-matrix/b300-dsv4-glm53-dev4-20260909-084427/glm53-high-throughput/points/16k_1_c1.json` |
+| glm53-high-throughput | 1k_4k | 1 | TIMEOUT_PASS | `/data/b300-dsv4-glm53-text-matrix/b300-dsv4-glm53-dev4-20260909-084427/glm53-high-throughput/points/1k_4k_c1.json` |
+| glm53-high-throughput | 64k_1 | 16 | FAIL | `/data/b300-dsv4-glm53-text-matrix/b300-dsv4-glm53-dev4-20260909-084427/glm53-high-throughput/points/64k_1_c16.json` |
+| glm53-high-throughput | 64k_1 | 32 | FAIL | `/data/b300-dsv4-glm53-text-matrix/b300-dsv4-glm53-dev4-20260909-084427/glm53-high-throughput/points/64k_1_c32.json` |
+| glm53-high-throughput | 64k_1 | 64 | FAIL | `/data/b300-dsv4-glm53-text-matrix/b300-dsv4-glm53-dev4-20260909-084427/glm53-high-throughput/points/64k_1_c64.json` |
+| glm53-high-throughput | 64k_1 | 128 | FAIL | `/data/b300-dsv4-glm53-text-matrix/b300-dsv4-glm53-dev4-20260909-084427/glm53-high-throughput/points/64k_1_c128.json` |
+| glm53-high-throughput | 128k_1 | 1 | FAIL | `/data/b300-dsv4-glm53-text-matrix/b300-dsv4-glm53-dev4-20260909-084427/glm53-high-throughput/points/128k_1_c1.json` |
+| glm53-high-throughput | 256k_1 | 1 | FAIL | `/data/b300-dsv4-glm53-text-matrix/b300-dsv4-glm53-dev4-20260909-084427/glm53-high-throughput/points/256k_1_c1.json` |
+
+## Service Error Summary
+
+### dsv4-low-latency
+
+- Matched lines: 319
+- Source: `/data/b300-dsv4-glm53-text-matrix/b300-dsv4-glm53-dev4-20260909-084427/dsv4-low-latency/server.final.log`
+- `[rank0]:[W910 05:05:37.307446434 CUDACachingAllocator.cpp:3933] memory allocation failed with OOM on device 0 while trying to allocate 2415919104 bytes (free: 753532928, total: 287428771840).`
+- `[rank1]:[W910 05:05:37.312576349 CUDACachingAllocator.cpp:3933] memory allocation failed with OOM on device 1 while trying to allocate 2415919104 bytes (free: 724172800, total: 287428771840).`
+- `[rank3]:[W910 05:05:37.312868926 CUDACachingAllocator.cpp:3933] memory allocation failed with OOM on device 3 while trying to allocate 2415919104 bytes (free: 1066008576, total: 287428771840).`
+- `[rank2]:[W910 05:05:37.314989750 CUDACachingAllocator.cpp:3933] memory allocation failed with OOM on device 2 while trying to allocate 2415919104 bytes (free: 724172800, total: 287428771840).`
+- `[rank1]:[W910 05:05:38.764535635 CUDACachingAllocator.cpp:3933] memory allocation failed with OOM on device 1 while trying to allocate 3221225472 bytes (free: 870973440, total: 287428771840).`
+
+### dsv4-balanced
+
+- Matched lines: 0
+- Source: `/data/b300-dsv4-glm53-text-matrix/b300-dsv4-glm53-dev4-20260909-084427/dsv4-balanced/server.final.log`
+
+### glm53-low-latency
+
+- Matched lines: 141
+- Source: `/data/b300-dsv4-glm53-text-matrix/b300-dsv4-glm53-dev4-20260909-084427/glm53-low-latency/server.final.log`
+- `[rank2]:[W909 15:01:40.129936229 CUDACachingAllocator.cpp:3933] memory allocation failed with OOM on device 2 while trying to allocate 8858370048 bytes (free: 2053701632, total: 287428771840).`
+- `[rank3]:[W909 15:01:40.130091786 CUDACachingAllocator.cpp:3933] memory allocation failed with OOM on device 3 while trying to allocate 8858370048 bytes (free: 2116616192, total: 287428771840).`
+- `[rank1]:[W909 15:01:40.130433377 CUDACachingAllocator.cpp:3933] memory allocation failed with OOM on device 1 while trying to allocate 8858370048 bytes (free: 2053701632, total: 287428771840).`
+- `[rank0]:[W909 15:01:40.130656418 CUDACachingAllocator.cpp:3933] memory allocation failed with OOM on device 0 while trying to allocate 8885633024 bytes (free: 2229862400, total: 287428771840).`
+- `[rank7]:[W909 15:01:40.131055935 CUDACachingAllocator.cpp:3933] memory allocation failed with OOM on device 7 while trying to allocate 8994684928 bytes (free: 2714304512, total: 287428771840).`
+
+### glm53-high-throughput
+
+- Matched lines: 586
+- Source: `/data/b300-dsv4-glm53-text-matrix/b300-dsv4-glm53-dev4-20260909-084427/glm53-high-throughput/server.final.log`
+- `/dvs/p4/build/sw/rel/gpgpu/toolkit/r13.0/main_nvshmem/src/host/transport/transport.cpp:nvshmemi_transport_init:282: /dvs/p4/build/sw/rel/gpgpu/toolkit/r13.0/main_nvshmem/src/host/transport/transport.cpp:nvshmemi_transport_init:282: init failed for transport: IBGDAinit failed for transport: IBGDA`
+- `/dvs/p4/build/sw/rel/gpgpu/toolkit/r13.0/main_nvshmem/src/host/transport/transport.cpp:nvshmemi_transport_init:282: init failed for transport: IBGDA`
+- `[rank1]:[W909 19:58:27.770722283 CUDACachingAllocator.cpp:3933] memory allocation failed with OOM on device 1 while trying to allocate 1950351360 bytes (free: 1160314880, total: 287428771840).`
+- `[rank2]:[W909 19:58:42.743872031 CUDACachingAllocator.cpp:3933] memory allocation failed with OOM on device 2 while trying to allocate 1950351360 bytes (free: 724107264, total: 287428771840).`
+- `[rank3]:[W909 19:58:55.386477067 CUDACachingAllocator.cpp:3933] memory allocation failed with OOM on device 3 while trying to allocate 1883242496 bytes (free: 1875443712, total: 287428771840).`
+
--- a/experiments/b300/dsv4_flash_glm53_sglang_text_matrix/snapshot-20260910-024355/b300-dsv4-glm53-text-matrix/b300-dsv4-glm53-dev4-20260909-084427/deadline-summary/progress.json
+++ b/experiments/b300/dsv4_flash_glm53_sglang_text_matrix/snapshot-20260910-024355/b300-dsv4-glm53-text-matrix/b300-dsv4-glm53-dev4-20260909-084427/deadline-summary/progress.json
@ -0,0 +1,47 @@
+{
+  "expected": 236,
+  "fail": 18,
+  "finished_percent": 78.39,
+  "invalid": 0,
+  "modes": {
+    "dsv4-balanced": {
+      "expected": 59,
+      "fail": 0,
+      "finished_percent": 52.54,
+      "invalid": 0,
+      "pass": 30,
+      "pending": 28,
+      "timeout_pass": 1
+    },
+    "dsv4-low-latency": {
+      "expected": 59,
+      "fail": 1,
+      "finished_percent": 91.53,
+      "invalid": 0,
+      "pass": 53,
+      "pending": 5,
+      "timeout_pass": 0
+    },
+    "glm53-high-throughput": {
+      "expected": 59,
+      "fail": 7,
+      "finished_percent": 77.97,
+      "invalid": 0,
+      "pass": 38,
+      "pending": 13,
+      "timeout_pass": 1
+    },
+    "glm53-low-latency": {
+      "expected": 59,
+      "fail": 10,
+      "finished_percent": 91.53,
+      "invalid": 0,
+      "pass": 43,
+      "pending": 5,
+      "timeout_pass": 1
+    }
+  },
+  "pass": 164,
+  "pending": 51,
+  "timeout_pass": 3
+}
--- a/experiments/b300/dsv4_flash_glm53_sglang_text_matrix/snapshot-20260910-024355/b300-dsv4-glm53-text-matrix/b300-dsv4-glm53-dev4-20260909-084427/deadline-summary/summary.md
+++ b/experiments/b300/dsv4_flash_glm53_sglang_text_matrix/snapshot-20260910-024355/b300-dsv4-glm53-text-matrix/b300-dsv4-glm53-dev4-20260909-084427/deadline-summary/summary.md
@ -0,0 +1,200 @@
+# B300 DeepSeek-V4-Flash / GLM-5.3 Text Matrix: b300-dsv4-glm53-dev4-20260909-084427
+
+## Progress
+
+| Mode | Finished | Expected | Pass | Timeout pass | Invalid | Fail | Progress |
+|---|---:|---:|---:|---:|---:|---:|---:|
+| dsv4-low-latency | 54 | 59 | 53 | 0 | 0 | 1 | 91.53% |
+| dsv4-balanced | 31 | 59 | 30 | 1 | 0 | 0 | 52.54% |
+| glm53-low-latency | 54 | 59 | 43 | 1 | 0 | 10 | 91.53% |
+| glm53-high-throughput | 46 | 59 | 38 | 1 | 0 | 7 | 77.97% |
+
+## Results
+
+| Mode | Shape | C | Status | Input TPS | Output TPS | TTFT P95 (ms) | TPOT P95 (ms) |
+|---|---|---:|---|---:|---:|---:|---:|
+| dsv4-low-latency | 16k_512 | 1 | PASS | 13128.54 | 410.27 | 265.87 | 3.07 |
+| dsv4-low-latency | 16k_512 | 8 | PASS | 33936.55 | 1060.52 | 1517.12 | 10.23 |
+| dsv4-low-latency | 16k_512 | 32 | PASS | 49777.89 | 1555.56 | 6230.65 | 31.87 |
+| dsv4-low-latency | 16k_512 | 64 | PASS | 55645.12 | 1738.91 | 12091.63 | 57.79 |
+| dsv4-low-latency | 16k_512 | 128 | PASS | 59108.19 | 1847.13 | 24712.64 | 114.28 |
+| dsv4-low-latency | 16k_512 | 256 | PASS | 60485.71 | 1890.18 | 48385.38 | 226.23 |
+| dsv4-low-latency | 16k_1 | 1 | PASS | 61163.57 | 3.73 | 266.91 | 0.00 |
+| dsv4-low-latency | 16k_1 | 8 | PASS | 66664.03 | 4.07 | 1950.41 | 0.00 |
+| dsv4-low-latency | 16k_1 | 32 | PASS | 66726.27 | 4.07 | 7810.68 | 0.00 |
+| dsv4-low-latency | 16k_1 | 64 | PASS | 66288.99 | 4.05 | 16147.44 | 0.00 |
+| dsv4-low-latency | 16k_1 | 128 | PASS | 66596.05 | 4.06 | 31378.16 | 0.00 |
+| dsv4-low-latency | 16k_1 | 256 | PASS | 66600.10 | 4.06 | 62914.28 | 0.00 |
+| dsv4-low-latency | 1k_128 | 1 | PASS | 2041.91 | 255.24 | 183.12 | 3.53 |
+| dsv4-low-latency | 1k_128 | 8 | PASS | 3578.88 | 447.36 | 3387.56 | 31.29 |
+| dsv4-low-latency | 1k_128 | 32 | PASS | 7184.95 | 898.12 | 1708.71 | 49.89 |
+| dsv4-low-latency | 1k_128 | 64 | PASS | 11724.62 | 1465.58 | 1205.29 | 62.99 |
+| dsv4-low-latency | 1k_128 | 128 | PASS | 20916.51 | 2614.56 | 2101.48 | 68.93 |
+| dsv4-low-latency | 1k_128 | 256 | PASS | 29913.48 | 3739.19 | 3544.47 | 99.14 |
+| dsv4-low-latency | 1k_4k | 1 | PASS | 151.34 | 605.35 | 183.93 | 2.28 |
+| dsv4-low-latency | 1k_4k | 8 | PASS | 709.27 | 2837.08 | 451.52 | 3.52 |
+| dsv4-low-latency | 1k_4k | 32 | PASS | 1679.25 | 6717.00 | 754.10 | 6.34 |
+| dsv4-low-latency | 1k_4k | 64 | PASS | 2459.31 | 9837.23 | 1200.11 | 9.86 |
+| dsv4-low-latency | 1k_4k | 128 | PASS | 3217.13 | 12868.50 | 2188.22 | 14.14 |
+| dsv4-low-latency | 1k_4k | 256 | PASS | 5283.08 | 21132.32 | 3658.46 | 17.31 |
+| dsv4-low-latency | 64k_1 | 1 | PASS | 57079.88 | 0.87 | 1180.60 | 0.00 |
+| dsv4-low-latency | 64k_1 | 4 | PASS | 58913.97 | 0.90 | 4440.74 | 0.00 |
+| dsv4-low-latency | 64k_1 | 8 | PASS | 58818.80 | 0.90 | 8883.26 | 0.00 |
+| dsv4-low-latency | 64k_1 | 16 | PASS | 58871.47 | 0.90 | 17757.38 | 0.00 |
+| dsv4-low-latency | 64k_1 | 32 | PASS | 58821.54 | 0.90 | 35523.52 | 0.00 |
+| dsv4-low-latency | 64k_1 | 64 | PASS | 58852.44 | 0.90 | 71226.16 | 0.00 |
+| dsv4-low-latency | 64k_1 | 128 | PASS | 58701.82 | 0.90 | 142742.12 | 0.00 |
+| dsv4-low-latency | 64k_512 | 1 | PASS | 32785.22 | 256.13 | 1146.10 | 2.91 |
+| dsv4-low-latency | 64k_512 | 4 | PASS | 47367.52 | 370.06 | 3651.92 | 13.00 |
+| dsv4-low-latency | 64k_512 | 8 | PASS | 51470.88 | 402.12 | 8142.53 | 27.01 |
+| dsv4-low-latency | 64k_512 | 16 | PASS | 54390.07 | 424.92 | 13764.60 | 53.61 |
+| dsv4-low-latency | 64k_512 | 32 | PASS | 55570.81 | 434.15 | 27317.01 | 113.22 |
+| dsv4-low-latency | 64k_512 | 64 | PASS | 56712.78 | 443.07 | 55290.66 | 238.49 |
+| dsv4-low-latency | 128k_1 | 1 | PASS | 50299.31 | 0.38 | 2606.88 | 0.00 |
+| dsv4-low-latency | 128k_1 | 2 | PASS | 51351.81 | 0.39 | 5095.93 | 0.00 |
+| dsv4-low-latency | 128k_1 | 4 | PASS | 51297.32 | 0.39 | 10307.95 | 0.00 |
+| dsv4-low-latency | 128k_1 | 8 | PASS | 51293.08 | 0.39 | 20440.51 | 0.00 |
+| dsv4-low-latency | 128k_1 | 16 | PASS | 51305.80 | 0.39 | 40813.75 | 0.00 |
+| dsv4-low-latency | 128k_1 | 32 | PASS | 51293.67 | 0.39 | 81674.88 | 0.00 |
+| dsv4-low-latency | 128k_1 | 64 | PASS | 51257.15 | 0.39 | 163688.22 | 0.00 |
+| dsv4-low-latency | 128k_512 | 1 | PASS | 28968.88 | 113.16 | 2613.09 | 4.47 |
+| dsv4-low-latency | 128k_512 | 2 | PASS | 35668.55 | 139.33 | 2886.10 | 14.56 |
+| dsv4-low-latency | 128k_512 | 4 | PASS | 41221.32 | 161.02 | 5149.64 | 25.35 |
+| dsv4-low-latency | 128k_512 | 8 | PASS | 44544.04 | 174.00 | 13131.55 | 55.16 |
+| dsv4-low-latency | 128k_512 | 16 | PASS | 46962.57 | 183.45 | 31281.83 | 107.27 |
+| dsv4-low-latency | 128k_512 | 32 | PASS | 48253.24 | 188.49 | 62228.62 | 214.99 |
+| dsv4-low-latency | 256k_1 | 1 | PASS | 39875.71 | 0.15 | 6610.03 | 0.00 |
+| dsv4-low-latency | 256k_1 | 2 | PASS | 40426.30 | 0.15 | 13069.41 | 0.00 |
+| dsv4-low-latency | 256k_1 | 4 | PASS | 40371.52 | 0.15 | 26029.44 | 0.00 |
+| dsv4-low-latency | 512k_1 | 1 | FAIL | - | - | - | - |
+| dsv4-balanced | 16k_512 | 1 | PASS | 2517.53 | 78.67 | 1067.18 | 11.26 |
+| dsv4-balanced | 16k_512 | 8 | PASS | 16954.74 | 529.84 | 2578.91 | 12.69 |
+| dsv4-balanced | 16k_512 | 32 | PASS | 36100.44 | 1128.14 | 8087.71 | 25.54 |
+| dsv4-balanced | 16k_512 | 64 | PASS | 45336.60 | 1416.77 | 16042.56 | 41.92 |
+| dsv4-balanced | 16k_512 | 128 | PASS | 52115.96 | 1628.62 | 31955.08 | 73.50 |
+| dsv4-balanced | 16k_512 | 256 | PASS | 57436.40 | 1794.89 | 62849.59 | 134.28 |
+| dsv4-balanced | 16k_1 | 1 | PASS | 16704.37 | 1.02 | 1001.65 | 0.00 |
+| dsv4-balanced | 16k_1 | 8 | PASS | 63912.26 | 3.90 | 2334.27 | 0.00 |
+| dsv4-balanced | 16k_1 | 32 | PASS | 65093.70 | 3.97 | 7943.21 | 0.00 |
+| dsv4-balanced | 16k_1 | 64 | PASS | 64887.20 | 3.96 | 16398.83 | 0.00 |
+| dsv4-balanced | 16k_1 | 128 | PASS | 64875.87 | 3.96 | 32514.97 | 0.00 |
+| dsv4-balanced | 16k_1 | 256 | PASS | 64245.92 | 3.92 | 65033.17 | 0.00 |
+| dsv4-balanced | 1k_128 | 1 | PASS | 643.09 | 80.39 | 342.42 | 10.62 |
+| dsv4-balanced | 1k_128 | 8 | PASS | 4391.78 | 548.97 | 649.01 | 10.73 |
+| dsv4-balanced | 1k_128 | 32 | PASS | 13757.02 | 1719.63 | 943.32 | 12.48 |
+| dsv4-balanced | 1k_128 | 64 | PASS | 21219.84 | 2652.48 | 1451.23 | 17.88 |
+| dsv4-balanced | 1k_128 | 128 | PASS | 29472.67 | 3684.08 | 2841.20 | 27.68 |
+| dsv4-balanced | 1k_128 | 256 | PASS | 36265.26 | 4533.16 | 4505.51 | 59.65 |
+| dsv4-balanced | 1k_4k | 1 | TIMEOUT_PASS | 21.98 | 87.94 | 370.83 | 14.15 |
+| dsv4-balanced | 1k_4k | 8 | PASS | 187.97 | 751.87 | 799.79 | 10.62 |
+| dsv4-balanced | 1k_4k | 32 | PASS | 631.65 | 2526.62 | 916.72 | 12.59 |
+| dsv4-balanced | 1k_4k | 64 | PASS | 1112.31 | 4449.26 | 1374.52 | 14.35 |
+| dsv4-balanced | 1k_4k | 128 | PASS | 1913.20 | 7652.81 | 3162.08 | 16.57 |
+| dsv4-balanced | 1k_4k | 256 | PASS | 2441.07 | 9764.26 | 4386.00 | 26.51 |
+| dsv4-balanced | 64k_1 | 1 | PASS | 17118.25 | 0.26 | 3874.73 | 0.00 |
+| dsv4-balanced | 64k_1 | 4 | PASS | 61303.08 | 0.94 | 4407.47 | 0.00 |
+| dsv4-balanced | 64k_1 | 8 | PASS | 65564.19 | 1.00 | 8356.27 | 0.00 |
+| dsv4-balanced | 64k_1 | 16 | PASS | 65864.61 | 1.01 | 15823.61 | 0.00 |
+| dsv4-balanced | 64k_1 | 32 | PASS | 65504.60 | 1.00 | 32070.29 | 0.00 |
+| dsv4-balanced | 64k_1 | 64 | PASS | 65616.93 | 1.00 | 64081.25 | 0.00 |
+| dsv4-balanced | 64k_1 | 128 | PASS | 65401.37 | 1.00 | 128671.94 | 0.00 |
+| glm53-low-latency | 16k_512 | 1 | PASS | 7882.31 | 246.32 | 612.89 | 4.36 |
+| glm53-low-latency | 16k_512 | 8 | PASS | 16384.46 | 512.01 | 3892.78 | 22.46 |
+| glm53-low-latency | 16k_512 | 32 | PASS | 21303.26 | 665.73 | 15127.36 | 69.03 |
+| glm53-low-latency | 16k_512 | 64 | PASS | 23502.93 | 734.47 | 29490.35 | 105.62 |
+| glm53-low-latency | 16k_512 | 128 | PASS | 23920.46 | 747.51 | 63407.51 | 101.14 |
+| glm53-low-latency | 16k_512 | 256 | PASS | 24021.20 | 750.66 | 144384.42 | 102.36 |
+| glm53-low-latency | 16k_1 | 1 | FAIL | 26686.87 | 1.63 | 606.55 | 0.00 |
+| glm53-low-latency | 1k_128 | 1 | PASS | 2005.88 | 250.73 | 70.19 | 5.55 |
+| glm53-low-latency | 1k_128 | 8 | PASS | 6062.38 | 757.80 | 813.10 | 15.41 |
+| glm53-low-latency | 1k_128 | 32 | PASS | 11946.48 | 1493.31 | 1204.54 | 31.01 |
+| glm53-low-latency | 1k_128 | 64 | PASS | 14308.93 | 1788.62 | 1848.75 | 40.49 |
+| glm53-low-latency | 1k_128 | 128 | PASS | 14124.94 | 1765.62 | 6332.63 | 42.03 |
+| glm53-low-latency | 1k_128 | 256 | PASS | 13944.40 | 1743.05 | 15823.61 | 43.56 |
+| glm53-low-latency | 1k_4k | 1 | PASS | 109.08 | 436.33 | 70.69 | 3.32 |
+| glm53-low-latency | 1k_4k | 8 | PASS | 477.85 | 1911.42 | 435.51 | 5.56 |
+| glm53-low-latency | 1k_4k | 32 | PASS | 1191.77 | 4767.10 | 892.67 | 9.33 |
+| glm53-low-latency | 1k_4k | 64 | PASS | 1458.93 | 5835.71 | 19807.67 | 10.84 |
+| glm53-low-latency | 1k_4k | 128 | PASS | 1530.74 | 6122.95 | 56046.13 | 11.08 |
+| glm53-low-latency | 1k_4k | 256 | PASS | 1526.22 | 6104.87 | 140390.00 | 11.26 |
+| glm53-low-latency | 64k_1 | 1 | PASS | 24918.24 | 0.38 | 2625.80 | 0.00 |
+| glm53-low-latency | 64k_1 | 4 | FAIL | 25330.44 | 0.39 | 10319.69 | 0.00 |
+| glm53-low-latency | 64k_1 | 8 | PASS | 25336.86 | 0.39 | 20633.79 | 0.00 |
+| glm53-low-latency | 64k_1 | 16 | FAIL | 25368.13 | 0.39 | 41257.75 | 0.00 |
+| glm53-low-latency | 64k_1 | 32 | FAIL | 25376.71 | 0.39 | 82506.66 | 0.00 |
+| glm53-low-latency | 64k_1 | 64 | FAIL | 25399.08 | 0.39 | 164948.88 | 0.00 |
+| glm53-low-latency | 64k_1 | 128 | FAIL | 25411.76 | 0.39 | 329925.47 | 0.00 |
+| glm53-low-latency | 64k_512 | 1 | PASS | 15809.05 | 123.51 | 2626.48 | 4.40 |
+| glm53-low-latency | 64k_512 | 4 | PASS | 20932.05 | 163.53 | 5658.47 | 26.76 |
+| glm53-low-latency | 64k_512 | 8 | PASS | 22427.36 | 175.21 | 13072.49 | 61.58 |
+| glm53-low-latency | 64k_512 | 16 | PASS | 23295.48 | 182.00 | 31634.83 | 127.45 |
+| glm53-low-latency | 64k_512 | 32 | PASS | 23917.39 | 186.85 | 62694.72 | 243.87 |
+| glm53-low-latency | 64k_512 | 64 | PASS | 24220.17 | 189.22 | 128260.27 | 323.81 |
+| glm53-low-latency | 128k_1 | 1 | PASS | 22283.97 | 0.17 | 5887.63 | 0.00 |
+| glm53-low-latency | 128k_1 | 2 | PASS | 22544.03 | 0.17 | 11620.01 | 0.00 |
+| glm53-low-latency | 128k_1 | 4 | PASS | 22540.08 | 0.17 | 23234.84 | 0.00 |
+| glm53-low-latency | 128k_1 | 8 | FAIL | 22556.35 | 0.17 | 46447.98 | 0.00 |
+| glm53-low-latency | 128k_1 | 16 | FAIL | 22547.53 | 0.17 | 92876.53 | 0.00 |
+| glm53-low-latency | 128k_1 | 32 | FAIL | 22549.85 | 0.17 | 185810.07 | 0.00 |
+| glm53-low-latency | 128k_1 | 64 | TIMEOUT_PASS | 22567.39 | 0.17 | 371590.24 | 0.00 |
+| glm53-low-latency | 128k_512 | 1 | PASS | 17525.45 | 68.46 | 5894.95 | 4.62 |
+| glm53-low-latency | 128k_512 | 2 | PASS | 19570.06 | 76.45 | 6325.56 | 27.21 |
+| glm53-low-latency | 128k_512 | 4 | PASS | 20526.29 | 80.18 | 11700.22 | 62.53 |
+| glm53-low-latency | 128k_512 | 8 | PASS | 21132.30 | 82.55 | 28726.77 | 145.04 |
+| glm53-low-latency | 128k_512 | 16 | PASS | 21643.46 | 84.54 | 70612.08 | 282.81 |
+| glm53-low-latency | 128k_512 | 32 | PASS | 21787.55 | 85.11 | 142627.99 | 339.34 |
+| glm53-low-latency | 256k_1 | 1 | PASS | 17456.68 | 0.07 | 15141.04 | 0.00 |
+| glm53-low-latency | 256k_1 | 2 | PASS | 17620.27 | 0.07 | 29920.79 | 0.00 |
+| glm53-low-latency | 256k_1 | 4 | FAIL | 17584.82 | 0.07 | 59691.73 | 0.00 |
+| glm53-low-latency | 512k_1 | 1 | PASS | 12420.94 | 0.02 | 42470.03 | 0.00 |
+| glm53-low-latency | 512k_1 | 2 | PASS | 12497.10 | 0.02 | 84423.37 | 0.00 |
+| glm53-low-latency | 512k_1 | 4 | PASS | 12494.82 | 0.02 | 168091.37 | 0.00 |
+| glm53-low-latency | 1m_boundary_1 | 1 | PASS | 7897.30 | 0.01 | 132967.28 | 0.00 |
+| glm53-low-latency | 1m_boundary_1 | 2 | PASS | 7846.52 | 0.01 | 269850.67 | 0.00 |
+| glm53-low-latency | 1m_boundary_1 | 4 | PASS | 7914.10 | 0.01 | 509514.56 | 0.00 |
+| glm53-high-throughput | 16k_512 | 1 | PASS | 1753.70 | 54.80 | 1706.01 | 16.40 |
+| glm53-high-throughput | 16k_512 | 8 | PASS | 7602.48 | 237.58 | 2952.39 | 38.86 |
+| glm53-high-throughput | 16k_512 | 32 | PASS | 15478.02 | 483.69 | 10709.46 | 82.99 |
+| glm53-high-throughput | 16k_512 | 64 | PASS | 22864.39 | 714.51 | 17177.98 | 115.88 |
+| glm53-high-throughput | 16k_512 | 128 | PASS | 31889.21 | 996.54 | 30890.29 | 165.33 |
+| glm53-high-throughput | 16k_512 | 256 | PASS | 25722.47 | 803.83 | 58208.90 | 435.35 |
+| glm53-high-throughput | 16k_1 | 1 | FAIL | 10265.87 | 0.63 | 1635.66 | 0.00 |
+| glm53-high-throughput | 1k_128 | 1 | PASS | 440.61 | 55.08 | 354.14 | 17.25 |
+| glm53-high-throughput | 1k_128 | 8 | PASS | 2216.13 | 277.02 | 658.97 | 33.33 |
+| glm53-high-throughput | 1k_128 | 32 | PASS | 5512.82 | 689.10 | 1351.79 | 55.76 |
+| glm53-high-throughput | 1k_128 | 64 | PASS | 8781.68 | 1097.71 | 1544.35 | 75.33 |
+| glm53-high-throughput | 1k_128 | 128 | PASS | 15193.07 | 1899.13 | 2569.96 | 87.74 |
+| glm53-high-throughput | 1k_128 | 256 | PASS | 14521.72 | 1815.21 | 4585.68 | 192.57 |
+| glm53-high-throughput | 1k_4k | 1 | TIMEOUT_PASS | 15.31 | 61.25 | 722.89 | 21.05 |
+| glm53-high-throughput | 1k_4k | 8 | PASS | 122.82 | 491.26 | 585.67 | 17.31 |
+| glm53-high-throughput | 1k_4k | 32 | PASS | 458.53 | 1834.10 | 964.13 | 18.18 |
+| glm53-high-throughput | 1k_4k | 64 | PASS | 772.99 | 3091.95 | 1480.98 | 20.33 |
+| glm53-high-throughput | 1k_4k | 128 | PASS | 1421.86 | 5687.44 | 2492.89 | 23.05 |
+| glm53-high-throughput | 1k_4k | 256 | PASS | 2653.31 | 10613.25 | 4540.61 | 22.36 |
+| glm53-high-throughput | 64k_1 | 1 | PASS | 10461.13 | 0.16 | 6387.30 | 0.00 |
+| glm53-high-throughput | 64k_1 | 4 | PASS | 31751.47 | 0.48 | 8371.38 | 0.00 |
+| glm53-high-throughput | 64k_1 | 8 | PASS | 53720.01 | 0.82 | 9891.34 | 0.00 |
+| glm53-high-throughput | 64k_1 | 16 | FAIL | 56023.39 | 0.85 | 19477.62 | 0.00 |
+| glm53-high-throughput | 64k_1 | 32 | FAIL | 56599.60 | 0.86 | 36913.14 | 0.00 |
+| glm53-high-throughput | 64k_1 | 64 | FAIL | 56554.68 | 0.86 | 74260.73 | 0.00 |
+| glm53-high-throughput | 64k_1 | 128 | FAIL | 56357.44 | 0.86 | 153394.58 | 0.00 |
+| glm53-high-throughput | 64k_512 | 1 | PASS | 4707.22 | 36.78 | 6471.79 | 16.01 |
+| glm53-high-throughput | 64k_512 | 4 | PASS | 8844.03 | 69.09 | 8240.16 | 54.88 |
+| glm53-high-throughput | 64k_512 | 8 | PASS | 13286.74 | 103.80 | 9928.84 | 96.89 |
+| glm53-high-throughput | 64k_512 | 16 | PASS | 18863.70 | 147.37 | 19171.29 | 135.82 |
+| glm53-high-throughput | 64k_512 | 32 | PASS | 25109.46 | 196.17 | 29462.53 | 206.58 |
+| glm53-high-throughput | 64k_512 | 64 | PASS | 33437.72 | 261.23 | 57344.18 | 329.77 |
+| glm53-high-throughput | 128k_1 | 1 | FAIL | 10100.97 | 0.08 | 13350.24 | 0.00 |
+| glm53-high-throughput | 128k_512 | 1 | PASS | 6350.91 | 24.81 | 13384.88 | 16.13 |
+| glm53-high-throughput | 128k_512 | 2 | PASS | 7862.65 | 30.71 | 13400.23 | 41.95 |
+| glm53-high-throughput | 128k_512 | 4 | PASS | 9986.29 | 39.01 | 16714.05 | 92.60 |
+| glm53-high-throughput | 128k_512 | 8 | PASS | 13503.28 | 52.75 | 20219.46 | 171.97 |
+| glm53-high-throughput | 128k_512 | 16 | PASS | 19923.94 | 77.83 | 39164.93 | 264.45 |
+| glm53-high-throughput | 128k_512 | 32 | PASS | 27503.67 | 107.44 | 59905.46 | 401.75 |
+| glm53-high-throughput | 256k_1 | 1 | FAIL | 9204.74 | 0.04 | 29442.21 | 0.00 |
+| glm53-high-throughput | 512k_1 | 1 | PASS | 7532.88 | 0.01 | 71779.28 | 0.00 |
+| glm53-high-throughput | 512k_1 | 2 | PASS | 14141.46 | 0.03 | 77207.47 | 0.00 |
+| glm53-high-throughput | 512k_1 | 4 | PASS | 24677.77 | 0.05 | 86777.89 | 0.00 |
+| glm53-high-throughput | 1m_boundary_1 | 1 | PASS | 5330.26 | 0.01 | 204815.71 | 0.00 |
+| glm53-high-throughput | 1m_boundary_1 | 2 | PASS | 9956.15 | 0.01 | 214992.66 | 0.00 |
+| glm53-high-throughput | 1m_boundary_1 | 4 | PASS | 18580.87 | 0.02 | 225402.20 | 0.00 |
--- a/experiments/b300/dsv4_flash_glm53_sglang_text_matrix/snapshot-20260910-024355/b300-dsv4-glm53-text-matrix/b300-dsv4-glm53-dev4-20260909-084427/dsv4-balanced/container.inspect.json
+++ b/experiments/b300/dsv4_flash_glm53_sglang_text_matrix/snapshot-20260910-024355/b300-dsv4-glm53-text-matrix/b300-dsv4-glm53-dev4-20260909-084427/dsv4-balanced/container.inspect.json
@ -52,7 +52,9 @@
        "MountLabel": "",
        "ProcessLabel": "",
        "AppArmorProfile": "docker-default",
-        "ExecIDs": null,
+        "ExecIDs": [
+            "f1a2dea8eb25c373aca73053e98b7f35e7a9ca3852f88b68bec48d293884b2b1"
+        ],
        "HostConfig": {
            "Binds": [
                "/data/model/deepseek-ai--DeepSeek-V4-Flash-0731:/models:ro",
@ -445,14 +447,6 @@
            "Name": "overlay2"
        },
        "Mounts": [
-            {
-                "Type": "bind",
-                "Source": "/data/model/deepseek-ai--DeepSeek-V4-Flash-0731",
-                "Destination": "/models",
-                "Mode": "ro",
-                "RW": false,
-                "Propagation": "rprivate"
-            },
            {
                "Type": "bind",
                "Source": "/data/b300-sglang-cache",
@ -476,6 +470,14 @@
                "Mode": "ro",
                "RW": false,
                "Propagation": "rprivate"
+            },
+            {
+                "Type": "bind",
+                "Source": "/data/model/deepseek-ai--DeepSeek-V4-Flash-0731",
+                "Destination": "/models",
+                "Mode": "ro",
+                "RW": false,
+                "Propagation": "rprivate"
            }
        ],
        "Config": {
--- a/experiments/b300/dsv4_flash_glm53_sglang_text_matrix/snapshot-20260910-024355/b300-dsv4-glm53-text-matrix/b300-dsv4-glm53-dev4-20260909-084427/dsv4-balanced/gpu_state.csv
+++ b/experiments/b300/dsv4_flash_glm53_sglang_text_matrix/snapshot-20260910-024355/b300-dsv4-glm53-text-matrix/b300-dsv4-glm53-dev4-20260909-084427/dsv4-balanced/gpu_state.csv
@ -1,9 +1,9 @@
 index, name, memory.total [MiB], memory.used [MiB], memory.free [MiB], utilization.gpu [%], power.draw [W]
-0, NVIDIA Graphics Device, 275040 MiB, 259411 MiB, 14703 MiB, 0 %, 242.62 W
-1, NVIDIA Graphics Device, 275040 MiB, 259603 MiB, 14511 MiB, 0 %, 234.74 W
-2, NVIDIA Graphics Device, 275040 MiB, 259617 MiB, 14497 MiB, 0 %, 235.85 W
-3, NVIDIA Graphics Device, 275040 MiB, 258643 MiB, 15471 MiB, 0 %, 243.30 W
-4, NVIDIA Graphics Device, 275040 MiB, 4 MiB, 274110 MiB, 0 %, 183.98 W
-5, NVIDIA Graphics Device, 275040 MiB, 4 MiB, 274110 MiB, 0 %, 182.88 W
-6, NVIDIA Graphics Device, 275040 MiB, 4 MiB, 274110 MiB, 0 %, 184.01 W
-7, NVIDIA Graphics Device, 275040 MiB, 4 MiB, 274110 MiB, 0 %, 184.09 W
+0, NVIDIA Graphics Device, 275040 MiB, 262083 MiB, 12031 MiB, 0 %, 304.84 W
+1, NVIDIA Graphics Device, 275040 MiB, 261373 MiB, 12741 MiB, 0 %, 295.27 W
+2, NVIDIA Graphics Device, 275040 MiB, 261495 MiB, 12619 MiB, 0 %, 294.86 W
+3, NVIDIA Graphics Device, 275040 MiB, 260091 MiB, 14023 MiB, 0 %, 306.36 W
+4, NVIDIA Graphics Device, 275040 MiB, 4 MiB, 274110 MiB, 0 %, 183.70 W
+5, NVIDIA Graphics Device, 275040 MiB, 4 MiB, 274110 MiB, 0 %, 182.13 W
+6, NVIDIA Graphics Device, 275040 MiB, 4 MiB, 274110 MiB, 0 %, 183.44 W
+7, NVIDIA Graphics Device, 275040 MiB, 4 MiB, 274110 MiB, 0 %, 183.97 W
--- a/experiments/b300/dsv4_flash_glm53_sglang_text_matrix/snapshot-20260910-024355/b300-dsv4-glm53-text-matrix/b300-dsv4-glm53-dev4-20260909-084427/dsv4-balanced/points/64k_1_c128.done
+++ b/experiments/b300/dsv4_flash_glm53_sglang_text_matrix/snapshot-20260910-024355/b300-dsv4-glm53-text-matrix/b300-dsv4-glm53-dev4-20260909-084427/dsv4-balanced/points/64k_1_c128.done
@ -0,0 +1 @@
+2026-09-10T07:38:13+00:00
--- a/experiments/b300/dsv4_flash_glm53_sglang_text_matrix/snapshot-20260910-024355/b300-dsv4-glm53-text-matrix/b300-dsv4-glm53-dev4-20260909-084427/dsv4-balanced/points/64k_1_c128.json
+++ b/experiments/b300/dsv4_flash_glm53_sglang_text_matrix/snapshot-20260910-024355/b300-dsv4-glm53-text-matrix/b300-dsv4-glm53-dev4-20260909-084427/dsv4-balanced/points/64k_1_c128.json
--- a/experiments/b300/dsv4_flash_glm53_sglang_text_matrix/snapshot-20260910-024355/b300-dsv4-glm53-text-matrix/b300-dsv4-glm53-dev4-20260909-084427/dsv4-balanced/points/64k_1_c64.done
+++ b/experiments/b300/dsv4_flash_glm53_sglang_text_matrix/snapshot-20260910-024355/b300-dsv4-glm53-text-matrix/b300-dsv4-glm53-dev4-20260909-084427/dsv4-balanced/points/64k_1_c64.done
@ -0,0 +1 @@
+2026-09-10T07:26:57+00:00
--- a/experiments/b300/dsv4_flash_glm53_sglang_text_matrix/snapshot-20260910-024355/b300-dsv4-glm53-text-matrix/b300-dsv4-glm53-dev4-20260909-084427/dsv4-balanced/points/64k_1_c64.json
+++ b/experiments/b300/dsv4_flash_glm53_sglang_text_matrix/snapshot-20260910-024355/b300-dsv4-glm53-text-matrix/b300-dsv4-glm53-dev4-20260909-084427/dsv4-balanced/points/64k_1_c64.json
--- a/experiments/b300/dsv4_flash_glm53_sglang_text_matrix/snapshot-20260910-024355/b300-dsv4-glm53-text-matrix/b300-dsv4-glm53-dev4-20260909-084427/dsv4-balanced/points/64k_1_c64.log
+++ b/experiments/b300/dsv4_flash_glm53_sglang_text_matrix/snapshot-20260910-024355/b300-dsv4-glm53-text-matrix/b300-dsv4-glm53-dev4-20260909-084427/dsv4-balanced/points/64k_1_c64.log
@ -1,2 +1,59 @@
 /sgl-workspace/sglang/python/sglang/bench_serving.py:13: FutureWarning: `sglang.bench_serving` is deprecated and will be removed in a future release; use `sglang.benchmark.serving` instead (e.g. `python -m sglang.benchmark.serving`).
  warnings.warn(
+benchmark_args=Namespace(backend='sglang', base_url=None, host='127.0.0.1', port=30000, ready_check_timeout_sec=60, dataset_name='random-ids', dataset_path='', dataset_offset=0, agentic_max_turns=None, speed_bench_category=None, speed_bench_output_len=512, model='/models', served_model_name=None, tokenizer='/models', num_prompts=320, sharegpt_output_len=None, sharegpt_context_len=None, random_input_len=65536, random_output_len=1, random_range_ratio=1.0, image_count=1, image_resolution='1080p', random_image_count=False, image_format='jpeg', image_content='random', request_rate=inf, use_trace_timestamps=False, max_concurrency=64, output_file='/results/points/64k_1_c64.json', output_details=True, print_requests=False, disable_tqdm=True, disable_stream=False, return_logprob=False, top_logprobs_num=0, token_ids_logprob=None, logprob_start_len=-1, return_routed_experts=False, cache_report=False, seed=20260904, disable_ignore_eos=False, temperature=0.0, top_p=1.0, extra_request_body=None, apply_chat_template=False, profile=False, plot_throughput=False, profile_activities=['CPU', 'GPU'], profile_start_step=None, profile_steps=None, profile_num_steps=None, profile_by_stage=False, profile_stages=None, profile_output_dir=None, profile_prefix=None, lora_name=None, lora_request_distribution='uniform', lora_zipf_alpha=1.5, prompt_suffix='', pd_separated=False, profile_prefill_url=None, profile_decode_url=None, flush_cache=True, flush_cache_timeout=60.0, warmup_requests=1, tokenize_prompt=True, gsp_num_groups=64, gsp_prompts_per_group=16, gsp_system_prompt_len=2048, gsp_question_len=128, gsp_output_len=256, gsp_range_ratio=1.0, gsp_fast_prepare=False, gsp_send_routing_key=False, gsp_num_turns=1, gsp_ordered=False, gsp_group_distribution='uniform', gsp_zipf_alpha=None, mooncake_slowdown_factor=1.0, mooncake_num_rounds=1, mooncake_workload='conversation', fake_prefill=False, tag=None, header=None)
+Waiting up to 60s for http://127.0.0.1:30000/v1/models to become ready...
+Server ready in 0.0s.
+
+WARNING It is recommended to use the `Chat` or `Instruct` model for benchmarking.
+Because when the tokenizer counts the output tokens, if there is gibberish, it might count incorrectly.
+
+Namespace(backend='sglang', base_url=None, host='127.0.0.1', port=30000, ready_check_timeout_sec=60, dataset_name='random-ids', dataset_path='', dataset_offset=0, agentic_max_turns=None, speed_bench_category=None, speed_bench_output_len=512, model='/models', served_model_name=None, tokenizer='/models', num_prompts=320, sharegpt_output_len=None, sharegpt_context_len=None, random_input_len=65536, random_output_len=1, random_range_ratio=1.0, image_count=1, image_resolution='1080p', random_image_count=False, image_format='jpeg', image_content='random', request_rate=inf, use_trace_timestamps=False, max_concurrency=64, output_file='/results/points/64k_1_c64.json', output_details=True, print_requests=False, disable_tqdm=True, disable_stream=False, return_logprob=False, top_logprobs_num=0, token_ids_logprob=None, logprob_start_len=-1, return_routed_experts=False, cache_report=False, seed=20260904, disable_ignore_eos=False, temperature=0.0, top_p=1.0, extra_request_body=None, apply_chat_template=False, profile=False, plot_throughput=False, profile_activities=['CPU', 'GPU'], profile_start_step=None, profile_steps=None, profile_num_steps=None, profile_by_stage=False, profile_stages=None, profile_output_dir=None, profile_prefix=None, lora_name=None, lora_request_distribution='uniform', lora_zipf_alpha=1.5, prompt_suffix='', pd_separated=False, profile_prefill_url=None, profile_decode_url=None, flush_cache=True, flush_cache_timeout=60.0, warmup_requests=1, tokenize_prompt=True, gsp_num_groups=64, gsp_prompts_per_group=16, gsp_system_prompt_len=2048, gsp_question_len=128, gsp_output_len=256, gsp_range_ratio=1.0, gsp_fast_prepare=False, gsp_send_routing_key=False, gsp_num_turns=1, gsp_ordered=False, gsp_group_distribution='uniform', gsp_zipf_alpha=None, mooncake_slowdown_factor=1.0, mooncake_num_rounds=1, mooncake_workload='conversation', fake_prefill=False, tag=None, header=None)
+
+#Input tokens: 20971520
+#Output tokens: 320
+Starting warmup with 1 sequences...
+Warmup completed with 1 sequences. Starting main benchmark run...
+
+============ Serving Benchmark Result ============
+Backend:                                 sglang    
+Traffic request rate:                    inf       
+Max request concurrency:                 64        
+Successful requests:                     320       
+Benchmark duration (s):                  319.61    
+Total input tokens:                      20971520  
+Total input text tokens:                 20971520  
+Total generated tokens:                  320       
+Total generated tokens (retokenized):    317       
+Request throughput (req/s):              1.00      
+Input token throughput (tok/s):          65616.93  
+Output token throughput (tok/s):         1.00      
+Peak output token throughput (tok/s):    5.00      
+Peak concurrent requests:                68        
+Total token throughput (tok/s):          65617.93  
+Concurrency:                             57.92     
+----------------End-to-End Latency----------------
+Mean E2E Latency (ms):                   57846.31  
+Median E2E Latency (ms):                 63517.85  
+P90 E2E Latency (ms):                    63880.28  
+P95 E2E Latency (ms):                    64084.94  
+P99 E2E Latency (ms):                    64357.08  
+---------------Time to First Token----------------
+Mean TTFT (ms):                          57407.25  
+Median TTFT (ms):                        63517.12  
+P90 TTFT (ms):                           63877.81  
+P95 TTFT (ms):                           64081.25  
+P99 TTFT (ms):                           64357.05  
+-----Time per Output Token (excl. 1st token)------
+Mean TPOT (ms):                          0.00      
+Median TPOT (ms):                        0.00      
+P90 TPOT (ms):                           0.00      
+P95 TPOT (ms):                           0.00      
+P99 TPOT (ms):                           0.00      
+---------------Inter-Token Latency----------------
+Mean ITL (ms):                           0.00      
+Median ITL (ms):                         0.00      
+P90 ITL (ms):                            0.00      
+P95 ITL (ms):                            0.00      
+P99 ITL (ms):                            0.00      
+Max ITL (ms):                            0.00      
+==================================================
--- a/experiments/b300/dsv4_flash_glm53_sglang_text_matrix/snapshot-20260910-024355/b300-dsv4-glm53-text-matrix/b300-dsv4-glm53-dev4-20260909-084427/dsv4-balanced/server.final.log
+++ b/experiments/b300/dsv4_flash_glm53_sglang_text_matrix/snapshot-20260910-024355/b300-dsv4-glm53-text-matrix/b300-dsv4-glm53-dev4-20260909-084427/dsv4-balanced/server.final.log
--- a/experiments/b300/dsv4_flash_glm53_sglang_text_matrix/snapshot-20260910-024355/b300-dsv4-glm53-text-matrix/b300-dsv4-glm53-dev4-20260909-084427/dsv4-balanced/server.log
+++ b/experiments/b300/dsv4_flash_glm53_sglang_text_matrix/snapshot-20260910-024355/b300-dsv4-glm53-text-matrix/b300-dsv4-glm53-dev4-20260909-084427/dsv4-balanced/server.log
--- a/experiments/b300/dsv4_flash_glm53_sglang_text_matrix/snapshot-20260910-024355/b300-dsv4-glm53-text-matrix/b300-dsv4-glm53-dev4-20260909-084427/dsv4-balanced/server_info.json
+++ b/experiments/b300/dsv4_flash_glm53_sglang_text_matrix/snapshot-20260910-024355/b300-dsv4-glm53-text-matrix/b300-dsv4-glm53-dev4-20260909-084427/dsv4-balanced/server_info.json
--- a/experiments/b300/dsv4_flash_glm53_sglang_text_matrix/snapshot-20260910-024355/b300-dsv4-glm53-text-matrix/b300-dsv4-glm53-dev4-20260909-084427/final-summary/comparison.md
+++ b/experiments/b300/dsv4_flash_glm53_sglang_text_matrix/snapshot-20260910-024355/b300-dsv4-glm53-text-matrix/b300-dsv4-glm53-dev4-20260909-084427/final-summary/comparison.md
@ -0,0 +1,121 @@
+# B300 DeepSeek-V4-Flash / GLM-5.3 Comparison: b300-dsv4-glm53-dev4-20260909-084427
+
+## Run Status
+
+- Finished: 78.39%
+- Pass: 164
+- Timeout pass: 3
+- Invalid: 0
+- Fail: 18
+
+## Service Configuration And Capacity
+
+| Profile | TP | DP | EP | Max running | Max tokens/rank | Chunk | KV dtype | Attention | MoE runner | MoE A2A | Speculative |
+|---|---:|---:|---:|---:|---:|---:|---|---|---|---|---|
+| dsv4-low-latency | - | - | - | - | - | - | - | - | - | - | - |
+| dsv4-balanced | - | - | - | - | - | - | - | - | - | - | - |
+| glm53-low-latency | 8 | 1 | 1 | 48 | 2696000 | 16384 | fp8_e4m3 | dsa | flashinfer_trtllm | none | EAGLE |
+| glm53-high-throughput | 8 | 8 | 8 | 256 | 2542208 | 4096 | fp8_e4m3 | dsa | auto | deepep | EAGLE |
+
+## Throughput Peaks And Their Latency
+
+Only validated PASS points are considered.
+
+| Shape | Mode | Peak input TPS @ C | TTFT P95 (ms) | Peak output TPS @ C | TPOT P95 (ms) |
+|---|---|---:|---:|---:|---:|
+| 16k_512 | dsv4-low-latency | 60485.71 @ C256 | 48385.38 | 1890.18 @ C256 | 226.23 |
+| 16k_512 | dsv4-balanced | 57436.40 @ C256 | 62849.59 | 1794.89 @ C256 | 134.28 |
+| 16k_512 | glm53-low-latency | 24021.20 @ C256 | 144384.42 | 750.66 @ C256 | 102.36 |
+| 16k_512 | glm53-high-throughput | 31889.21 @ C128 | 30890.29 | 996.54 @ C128 | 165.33 |
+| 16k_1 | dsv4-low-latency | 66726.27 @ C32 | 7810.68 | 4.07 @ C32 | 0.00 |
+| 16k_1 | dsv4-balanced | 65093.70 @ C32 | 7943.21 | 3.97 @ C32 | 0.00 |
+| 1k_128 | dsv4-low-latency | 29913.48 @ C256 | 3544.47 | 3739.19 @ C256 | 99.14 |
+| 1k_128 | dsv4-balanced | 36265.26 @ C256 | 4505.51 | 4533.16 @ C256 | 59.65 |
+| 1k_128 | glm53-low-latency | 14308.93 @ C64 | 1848.75 | 1788.62 @ C64 | 40.49 |
+| 1k_128 | glm53-high-throughput | 15193.07 @ C128 | 2569.96 | 1899.13 @ C128 | 87.74 |
+| 1k_4k | dsv4-low-latency | 5283.08 @ C256 | 3658.46 | 21132.32 @ C256 | 17.31 |
+| 1k_4k | dsv4-balanced | 2441.07 @ C256 | 4386.00 | 9764.26 @ C256 | 26.51 |
+| 1k_4k | glm53-low-latency | 1530.74 @ C128 | 56046.13 | 6122.95 @ C128 | 11.08 |
+| 1k_4k | glm53-high-throughput | 2653.31 @ C256 | 4540.61 | 10613.25 @ C256 | 22.36 |
+| 64k_1 | dsv4-low-latency | 58913.97 @ C4 | 4440.74 | 0.90 @ C4 | 0.00 |
+| 64k_1 | dsv4-balanced | 65864.61 @ C16 | 15823.61 | 1.01 @ C16 | 0.00 |
+| 64k_1 | glm53-low-latency | 25336.86 @ C8 | 20633.79 | 0.39 @ C8 | 0.00 |
+| 64k_1 | glm53-high-throughput | 53720.01 @ C8 | 9891.34 | 0.82 @ C8 | 0.00 |
+| 64k_512 | dsv4-low-latency | 56712.78 @ C64 | 55290.66 | 443.07 @ C64 | 238.49 |
+| 64k_512 | glm53-low-latency | 24220.17 @ C64 | 128260.27 | 189.22 @ C64 | 323.81 |
+| 64k_512 | glm53-high-throughput | 33437.72 @ C64 | 57344.18 | 261.23 @ C64 | 329.77 |
+| 128k_1 | dsv4-low-latency | 51351.81 @ C2 | 5095.93 | 0.39 @ C2 | 0.00 |
+| 128k_1 | glm53-low-latency | 22544.03 @ C2 | 11620.01 | 0.17 @ C2 | 0.00 |
+| 128k_512 | dsv4-low-latency | 48253.24 @ C32 | 62228.62 | 188.49 @ C32 | 214.99 |
+| 128k_512 | glm53-low-latency | 21787.55 @ C32 | 142627.99 | 85.11 @ C32 | 339.34 |
+| 128k_512 | glm53-high-throughput | 27503.67 @ C32 | 59905.46 | 107.44 @ C32 | 401.75 |
+| 256k_1 | dsv4-low-latency | 40426.30 @ C2 | 13069.41 | 0.15 @ C2 | 0.00 |
+| 256k_1 | glm53-low-latency | 17620.27 @ C2 | 29920.79 | 0.07 @ C2 | 0.00 |
+| 512k_1 | glm53-low-latency | 12497.10 @ C2 | 84423.37 | 0.02 @ C2 | 0.00 |
+| 512k_1 | glm53-high-throughput | 24677.77 @ C4 | 86777.89 | 0.05 @ C4 | 0.00 |
+| 1m_boundary_1 | glm53-low-latency | 7914.10 @ C4 | 509514.56 | 0.01 @ C4 | 0.00 |
+| 1m_boundary_1 | glm53-high-throughput | 18580.87 @ C4 | 225402.20 | 0.02 @ C4 | 0.00 |
+
+## Capacity And Failure Boundaries
+
+| Mode | Shape | C | Status | Evidence |
+|---|---|---:|---|---|
+| dsv4-low-latency | 512k_1 | 1 | FAIL | `/data/b300-dsv4-glm53-text-matrix/b300-dsv4-glm53-dev4-20260909-084427/dsv4-low-latency/points/512k_1_c1.failed` |
+| dsv4-balanced | 1k_4k | 1 | TIMEOUT_PASS | `/data/b300-dsv4-glm53-text-matrix/b300-dsv4-glm53-dev4-20260909-084427/dsv4-balanced/points/1k_4k_c1.json` |
+| glm53-low-latency | 16k_1 | 1 | FAIL | `/data/b300-dsv4-glm53-text-matrix/b300-dsv4-glm53-dev4-20260909-084427/glm53-low-latency/points/16k_1_c1.json` |
+| glm53-low-latency | 64k_1 | 4 | FAIL | `/data/b300-dsv4-glm53-text-matrix/b300-dsv4-glm53-dev4-20260909-084427/glm53-low-latency/points/64k_1_c4.json` |
+| glm53-low-latency | 64k_1 | 16 | FAIL | `/data/b300-dsv4-glm53-text-matrix/b300-dsv4-glm53-dev4-20260909-084427/glm53-low-latency/points/64k_1_c16.json` |
+| glm53-low-latency | 64k_1 | 32 | FAIL | `/data/b300-dsv4-glm53-text-matrix/b300-dsv4-glm53-dev4-20260909-084427/glm53-low-latency/points/64k_1_c32.json` |
+| glm53-low-latency | 64k_1 | 64 | FAIL | `/data/b300-dsv4-glm53-text-matrix/b300-dsv4-glm53-dev4-20260909-084427/glm53-low-latency/points/64k_1_c64.json` |
+| glm53-low-latency | 64k_1 | 128 | FAIL | `/data/b300-dsv4-glm53-text-matrix/b300-dsv4-glm53-dev4-20260909-084427/glm53-low-latency/points/64k_1_c128.json` |
+| glm53-low-latency | 128k_1 | 8 | FAIL | `/data/b300-dsv4-glm53-text-matrix/b300-dsv4-glm53-dev4-20260909-084427/glm53-low-latency/points/128k_1_c8.json` |
+| glm53-low-latency | 128k_1 | 16 | FAIL | `/data/b300-dsv4-glm53-text-matrix/b300-dsv4-glm53-dev4-20260909-084427/glm53-low-latency/points/128k_1_c16.json` |
+| glm53-low-latency | 128k_1 | 32 | FAIL | `/data/b300-dsv4-glm53-text-matrix/b300-dsv4-glm53-dev4-20260909-084427/glm53-low-latency/points/128k_1_c32.json` |
+| glm53-low-latency | 128k_1 | 64 | TIMEOUT_PASS | `/data/b300-dsv4-glm53-text-matrix/b300-dsv4-glm53-dev4-20260909-084427/glm53-low-latency/points/128k_1_c64.json` |
+| glm53-low-latency | 256k_1 | 4 | FAIL | `/data/b300-dsv4-glm53-text-matrix/b300-dsv4-glm53-dev4-20260909-084427/glm53-low-latency/points/256k_1_c4.json` |
+| glm53-high-throughput | 16k_1 | 1 | FAIL | `/data/b300-dsv4-glm53-text-matrix/b300-dsv4-glm53-dev4-20260909-084427/glm53-high-throughput/points/16k_1_c1.json` |
+| glm53-high-throughput | 1k_4k | 1 | TIMEOUT_PASS | `/data/b300-dsv4-glm53-text-matrix/b300-dsv4-glm53-dev4-20260909-084427/glm53-high-throughput/points/1k_4k_c1.json` |
+| glm53-high-throughput | 64k_1 | 16 | FAIL | `/data/b300-dsv4-glm53-text-matrix/b300-dsv4-glm53-dev4-20260909-084427/glm53-high-throughput/points/64k_1_c16.json` |
+| glm53-high-throughput | 64k_1 | 32 | FAIL | `/data/b300-dsv4-glm53-text-matrix/b300-dsv4-glm53-dev4-20260909-084427/glm53-high-throughput/points/64k_1_c32.json` |
+| glm53-high-throughput | 64k_1 | 64 | FAIL | `/data/b300-dsv4-glm53-text-matrix/b300-dsv4-glm53-dev4-20260909-084427/glm53-high-throughput/points/64k_1_c64.json` |
+| glm53-high-throughput | 64k_1 | 128 | FAIL | `/data/b300-dsv4-glm53-text-matrix/b300-dsv4-glm53-dev4-20260909-084427/glm53-high-throughput/points/64k_1_c128.json` |
+| glm53-high-throughput | 128k_1 | 1 | FAIL | `/data/b300-dsv4-glm53-text-matrix/b300-dsv4-glm53-dev4-20260909-084427/glm53-high-throughput/points/128k_1_c1.json` |
+| glm53-high-throughput | 256k_1 | 1 | FAIL | `/data/b300-dsv4-glm53-text-matrix/b300-dsv4-glm53-dev4-20260909-084427/glm53-high-throughput/points/256k_1_c1.json` |
+
+## Service Error Summary
+
+### dsv4-low-latency
+
+- Matched lines: 319
+- Source: `/data/b300-dsv4-glm53-text-matrix/b300-dsv4-glm53-dev4-20260909-084427/dsv4-low-latency/server.final.log`
+- `[rank0]:[W910 05:05:37.307446434 CUDACachingAllocator.cpp:3933] memory allocation failed with OOM on device 0 while trying to allocate 2415919104 bytes (free: 753532928, total: 287428771840).`
+- `[rank1]:[W910 05:05:37.312576349 CUDACachingAllocator.cpp:3933] memory allocation failed with OOM on device 1 while trying to allocate 2415919104 bytes (free: 724172800, total: 287428771840).`
+- `[rank3]:[W910 05:05:37.312868926 CUDACachingAllocator.cpp:3933] memory allocation failed with OOM on device 3 while trying to allocate 2415919104 bytes (free: 1066008576, total: 287428771840).`
+- `[rank2]:[W910 05:05:37.314989750 CUDACachingAllocator.cpp:3933] memory allocation failed with OOM on device 2 while trying to allocate 2415919104 bytes (free: 724172800, total: 287428771840).`
+- `[rank1]:[W910 05:05:38.764535635 CUDACachingAllocator.cpp:3933] memory allocation failed with OOM on device 1 while trying to allocate 3221225472 bytes (free: 870973440, total: 287428771840).`
+
+### dsv4-balanced
+
+- Matched lines: 0
+- Source: `/data/b300-dsv4-glm53-text-matrix/b300-dsv4-glm53-dev4-20260909-084427/dsv4-balanced/server.final.log`
+
+### glm53-low-latency
+
+- Matched lines: 141
+- Source: `/data/b300-dsv4-glm53-text-matrix/b300-dsv4-glm53-dev4-20260909-084427/glm53-low-latency/server.final.log`
+- `[rank2]:[W909 15:01:40.129936229 CUDACachingAllocator.cpp:3933] memory allocation failed with OOM on device 2 while trying to allocate 8858370048 bytes (free: 2053701632, total: 287428771840).`
+- `[rank3]:[W909 15:01:40.130091786 CUDACachingAllocator.cpp:3933] memory allocation failed with OOM on device 3 while trying to allocate 8858370048 bytes (free: 2116616192, total: 287428771840).`
+- `[rank1]:[W909 15:01:40.130433377 CUDACachingAllocator.cpp:3933] memory allocation failed with OOM on device 1 while trying to allocate 8858370048 bytes (free: 2053701632, total: 287428771840).`
+- `[rank0]:[W909 15:01:40.130656418 CUDACachingAllocator.cpp:3933] memory allocation failed with OOM on device 0 while trying to allocate 8885633024 bytes (free: 2229862400, total: 287428771840).`
+- `[rank7]:[W909 15:01:40.131055935 CUDACachingAllocator.cpp:3933] memory allocation failed with OOM on device 7 while trying to allocate 8994684928 bytes (free: 2714304512, total: 287428771840).`
+
+### glm53-high-throughput
+
+- Matched lines: 586
+- Source: `/data/b300-dsv4-glm53-text-matrix/b300-dsv4-glm53-dev4-20260909-084427/glm53-high-throughput/server.final.log`
+- `/dvs/p4/build/sw/rel/gpgpu/toolkit/r13.0/main_nvshmem/src/host/transport/transport.cpp:nvshmemi_transport_init:282: /dvs/p4/build/sw/rel/gpgpu/toolkit/r13.0/main_nvshmem/src/host/transport/transport.cpp:nvshmemi_transport_init:282: init failed for transport: IBGDAinit failed for transport: IBGDA`
+- `/dvs/p4/build/sw/rel/gpgpu/toolkit/r13.0/main_nvshmem/src/host/transport/transport.cpp:nvshmemi_transport_init:282: init failed for transport: IBGDA`
+- `[rank1]:[W909 19:58:27.770722283 CUDACachingAllocator.cpp:3933] memory allocation failed with OOM on device 1 while trying to allocate 1950351360 bytes (free: 1160314880, total: 287428771840).`
+- `[rank2]:[W909 19:58:42.743872031 CUDACachingAllocator.cpp:3933] memory allocation failed with OOM on device 2 while trying to allocate 1950351360 bytes (free: 724107264, total: 287428771840).`
+- `[rank3]:[W909 19:58:55.386477067 CUDACachingAllocator.cpp:3933] memory allocation failed with OOM on device 3 while trying to allocate 1883242496 bytes (free: 1875443712, total: 287428771840).`
+
--- a/experiments/b300/dsv4_flash_glm53_sglang_text_matrix/snapshot-20260910-024355/b300-dsv4-glm53-text-matrix/b300-dsv4-glm53-dev4-20260909-084427/final-summary/progress.json
+++ b/experiments/b300/dsv4_flash_glm53_sglang_text_matrix/snapshot-20260910-024355/b300-dsv4-glm53-text-matrix/b300-dsv4-glm53-dev4-20260909-084427/final-summary/progress.json
@ -0,0 +1,47 @@
+{
+  "expected": 236,
+  "fail": 18,
+  "finished_percent": 78.39,
+  "invalid": 0,
+  "modes": {
+    "dsv4-balanced": {
+      "expected": 59,
+      "fail": 0,
+      "finished_percent": 52.54,
+      "invalid": 0,
+      "pass": 30,
+      "pending": 28,
+      "timeout_pass": 1
+    },
+    "dsv4-low-latency": {
+      "expected": 59,
+      "fail": 1,
+      "finished_percent": 91.53,
+      "invalid": 0,
+      "pass": 53,
+      "pending": 5,
+      "timeout_pass": 0
+    },
+    "glm53-high-throughput": {
+      "expected": 59,
+      "fail": 7,
+      "finished_percent": 77.97,
+      "invalid": 0,
+      "pass": 38,
+      "pending": 13,
+      "timeout_pass": 1
+    },
+    "glm53-low-latency": {
+      "expected": 59,
+      "fail": 10,
+      "finished_percent": 91.53,
+      "invalid": 0,
+      "pass": 43,
+      "pending": 5,
+      "timeout_pass": 1
+    }
+  },
+  "pass": 164,
+  "pending": 51,
+  "timeout_pass": 3
+}
--- a/experiments/b300/dsv4_flash_glm53_sglang_text_matrix/snapshot-20260910-024355/b300-dsv4-glm53-text-matrix/b300-dsv4-glm53-dev4-20260909-084427/final-summary/summary.md
+++ b/experiments/b300/dsv4_flash_glm53_sglang_text_matrix/snapshot-20260910-024355/b300-dsv4-glm53-text-matrix/b300-dsv4-glm53-dev4-20260909-084427/final-summary/summary.md
@ -0,0 +1,200 @@
+# B300 DeepSeek-V4-Flash / GLM-5.3 Text Matrix: b300-dsv4-glm53-dev4-20260909-084427
+
+## Progress
+
+| Mode | Finished | Expected | Pass | Timeout pass | Invalid | Fail | Progress |
+|---|---:|---:|---:|---:|---:|---:|---:|
+| dsv4-low-latency | 54 | 59 | 53 | 0 | 0 | 1 | 91.53% |
+| dsv4-balanced | 31 | 59 | 30 | 1 | 0 | 0 | 52.54% |
+| glm53-low-latency | 54 | 59 | 43 | 1 | 0 | 10 | 91.53% |
+| glm53-high-throughput | 46 | 59 | 38 | 1 | 0 | 7 | 77.97% |
+
+## Results
+
+| Mode | Shape | C | Status | Input TPS | Output TPS | TTFT P95 (ms) | TPOT P95 (ms) |
+|---|---|---:|---|---:|---:|---:|---:|
+| dsv4-low-latency | 16k_512 | 1 | PASS | 13128.54 | 410.27 | 265.87 | 3.07 |
+| dsv4-low-latency | 16k_512 | 8 | PASS | 33936.55 | 1060.52 | 1517.12 | 10.23 |
+| dsv4-low-latency | 16k_512 | 32 | PASS | 49777.89 | 1555.56 | 6230.65 | 31.87 |
+| dsv4-low-latency | 16k_512 | 64 | PASS | 55645.12 | 1738.91 | 12091.63 | 57.79 |
+| dsv4-low-latency | 16k_512 | 128 | PASS | 59108.19 | 1847.13 | 24712.64 | 114.28 |
+| dsv4-low-latency | 16k_512 | 256 | PASS | 60485.71 | 1890.18 | 48385.38 | 226.23 |
+| dsv4-low-latency | 16k_1 | 1 | PASS | 61163.57 | 3.73 | 266.91 | 0.00 |
+| dsv4-low-latency | 16k_1 | 8 | PASS | 66664.03 | 4.07 | 1950.41 | 0.00 |
+| dsv4-low-latency | 16k_1 | 32 | PASS | 66726.27 | 4.07 | 7810.68 | 0.00 |
+| dsv4-low-latency | 16k_1 | 64 | PASS | 66288.99 | 4.05 | 16147.44 | 0.00 |
+| dsv4-low-latency | 16k_1 | 128 | PASS | 66596.05 | 4.06 | 31378.16 | 0.00 |
+| dsv4-low-latency | 16k_1 | 256 | PASS | 66600.10 | 4.06 | 62914.28 | 0.00 |
+| dsv4-low-latency | 1k_128 | 1 | PASS | 2041.91 | 255.24 | 183.12 | 3.53 |
+| dsv4-low-latency | 1k_128 | 8 | PASS | 3578.88 | 447.36 | 3387.56 | 31.29 |
+| dsv4-low-latency | 1k_128 | 32 | PASS | 7184.95 | 898.12 | 1708.71 | 49.89 |
+| dsv4-low-latency | 1k_128 | 64 | PASS | 11724.62 | 1465.58 | 1205.29 | 62.99 |
+| dsv4-low-latency | 1k_128 | 128 | PASS | 20916.51 | 2614.56 | 2101.48 | 68.93 |
+| dsv4-low-latency | 1k_128 | 256 | PASS | 29913.48 | 3739.19 | 3544.47 | 99.14 |
+| dsv4-low-latency | 1k_4k | 1 | PASS | 151.34 | 605.35 | 183.93 | 2.28 |
+| dsv4-low-latency | 1k_4k | 8 | PASS | 709.27 | 2837.08 | 451.52 | 3.52 |
+| dsv4-low-latency | 1k_4k | 32 | PASS | 1679.25 | 6717.00 | 754.10 | 6.34 |
+| dsv4-low-latency | 1k_4k | 64 | PASS | 2459.31 | 9837.23 | 1200.11 | 9.86 |
+| dsv4-low-latency | 1k_4k | 128 | PASS | 3217.13 | 12868.50 | 2188.22 | 14.14 |
+| dsv4-low-latency | 1k_4k | 256 | PASS | 5283.08 | 21132.32 | 3658.46 | 17.31 |
+| dsv4-low-latency | 64k_1 | 1 | PASS | 57079.88 | 0.87 | 1180.60 | 0.00 |
+| dsv4-low-latency | 64k_1 | 4 | PASS | 58913.97 | 0.90 | 4440.74 | 0.00 |
+| dsv4-low-latency | 64k_1 | 8 | PASS | 58818.80 | 0.90 | 8883.26 | 0.00 |
+| dsv4-low-latency | 64k_1 | 16 | PASS | 58871.47 | 0.90 | 17757.38 | 0.00 |
+| dsv4-low-latency | 64k_1 | 32 | PASS | 58821.54 | 0.90 | 35523.52 | 0.00 |
+| dsv4-low-latency | 64k_1 | 64 | PASS | 58852.44 | 0.90 | 71226.16 | 0.00 |
+| dsv4-low-latency | 64k_1 | 128 | PASS | 58701.82 | 0.90 | 142742.12 | 0.00 |
+| dsv4-low-latency | 64k_512 | 1 | PASS | 32785.22 | 256.13 | 1146.10 | 2.91 |
+| dsv4-low-latency | 64k_512 | 4 | PASS | 47367.52 | 370.06 | 3651.92 | 13.00 |
+| dsv4-low-latency | 64k_512 | 8 | PASS | 51470.88 | 402.12 | 8142.53 | 27.01 |
+| dsv4-low-latency | 64k_512 | 16 | PASS | 54390.07 | 424.92 | 13764.60 | 53.61 |
+| dsv4-low-latency | 64k_512 | 32 | PASS | 55570.81 | 434.15 | 27317.01 | 113.22 |
+| dsv4-low-latency | 64k_512 | 64 | PASS | 56712.78 | 443.07 | 55290.66 | 238.49 |
+| dsv4-low-latency | 128k_1 | 1 | PASS | 50299.31 | 0.38 | 2606.88 | 0.00 |
+| dsv4-low-latency | 128k_1 | 2 | PASS | 51351.81 | 0.39 | 5095.93 | 0.00 |
+| dsv4-low-latency | 128k_1 | 4 | PASS | 51297.32 | 0.39 | 10307.95 | 0.00 |
+| dsv4-low-latency | 128k_1 | 8 | PASS | 51293.08 | 0.39 | 20440.51 | 0.00 |
+| dsv4-low-latency | 128k_1 | 16 | PASS | 51305.80 | 0.39 | 40813.75 | 0.00 |
+| dsv4-low-latency | 128k_1 | 32 | PASS | 51293.67 | 0.39 | 81674.88 | 0.00 |
+| dsv4-low-latency | 128k_1 | 64 | PASS | 51257.15 | 0.39 | 163688.22 | 0.00 |
+| dsv4-low-latency | 128k_512 | 1 | PASS | 28968.88 | 113.16 | 2613.09 | 4.47 |
+| dsv4-low-latency | 128k_512 | 2 | PASS | 35668.55 | 139.33 | 2886.10 | 14.56 |
+| dsv4-low-latency | 128k_512 | 4 | PASS | 41221.32 | 161.02 | 5149.64 | 25.35 |
+| dsv4-low-latency | 128k_512 | 8 | PASS | 44544.04 | 174.00 | 13131.55 | 55.16 |
+| dsv4-low-latency | 128k_512 | 16 | PASS | 46962.57 | 183.45 | 31281.83 | 107.27 |
+| dsv4-low-latency | 128k_512 | 32 | PASS | 48253.24 | 188.49 | 62228.62 | 214.99 |
+| dsv4-low-latency | 256k_1 | 1 | PASS | 39875.71 | 0.15 | 6610.03 | 0.00 |
+| dsv4-low-latency | 256k_1 | 2 | PASS | 40426.30 | 0.15 | 13069.41 | 0.00 |
+| dsv4-low-latency | 256k_1 | 4 | PASS | 40371.52 | 0.15 | 26029.44 | 0.00 |
+| dsv4-low-latency | 512k_1 | 1 | FAIL | - | - | - | - |
+| dsv4-balanced | 16k_512 | 1 | PASS | 2517.53 | 78.67 | 1067.18 | 11.26 |
+| dsv4-balanced | 16k_512 | 8 | PASS | 16954.74 | 529.84 | 2578.91 | 12.69 |
+| dsv4-balanced | 16k_512 | 32 | PASS | 36100.44 | 1128.14 | 8087.71 | 25.54 |
+| dsv4-balanced | 16k_512 | 64 | PASS | 45336.60 | 1416.77 | 16042.56 | 41.92 |
+| dsv4-balanced | 16k_512 | 128 | PASS | 52115.96 | 1628.62 | 31955.08 | 73.50 |
+| dsv4-balanced | 16k_512 | 256 | PASS | 57436.40 | 1794.89 | 62849.59 | 134.28 |
+| dsv4-balanced | 16k_1 | 1 | PASS | 16704.37 | 1.02 | 1001.65 | 0.00 |
+| dsv4-balanced | 16k_1 | 8 | PASS | 63912.26 | 3.90 | 2334.27 | 0.00 |
+| dsv4-balanced | 16k_1 | 32 | PASS | 65093.70 | 3.97 | 7943.21 | 0.00 |
+| dsv4-balanced | 16k_1 | 64 | PASS | 64887.20 | 3.96 | 16398.83 | 0.00 |
+| dsv4-balanced | 16k_1 | 128 | PASS | 64875.87 | 3.96 | 32514.97 | 0.00 |
+| dsv4-balanced | 16k_1 | 256 | PASS | 64245.92 | 3.92 | 65033.17 | 0.00 |
+| dsv4-balanced | 1k_128 | 1 | PASS | 643.09 | 80.39 | 342.42 | 10.62 |
+| dsv4-balanced | 1k_128 | 8 | PASS | 4391.78 | 548.97 | 649.01 | 10.73 |
+| dsv4-balanced | 1k_128 | 32 | PASS | 13757.02 | 1719.63 | 943.32 | 12.48 |
+| dsv4-balanced | 1k_128 | 64 | PASS | 21219.84 | 2652.48 | 1451.23 | 17.88 |
+| dsv4-balanced | 1k_128 | 128 | PASS | 29472.67 | 3684.08 | 2841.20 | 27.68 |
+| dsv4-balanced | 1k_128 | 256 | PASS | 36265.26 | 4533.16 | 4505.51 | 59.65 |
+| dsv4-balanced | 1k_4k | 1 | TIMEOUT_PASS | 21.98 | 87.94 | 370.83 | 14.15 |
+| dsv4-balanced | 1k_4k | 8 | PASS | 187.97 | 751.87 | 799.79 | 10.62 |
+| dsv4-balanced | 1k_4k | 32 | PASS | 631.65 | 2526.62 | 916.72 | 12.59 |
+| dsv4-balanced | 1k_4k | 64 | PASS | 1112.31 | 4449.26 | 1374.52 | 14.35 |
+| dsv4-balanced | 1k_4k | 128 | PASS | 1913.20 | 7652.81 | 3162.08 | 16.57 |
+| dsv4-balanced | 1k_4k | 256 | PASS | 2441.07 | 9764.26 | 4386.00 | 26.51 |
+| dsv4-balanced | 64k_1 | 1 | PASS | 17118.25 | 0.26 | 3874.73 | 0.00 |
+| dsv4-balanced | 64k_1 | 4 | PASS | 61303.08 | 0.94 | 4407.47 | 0.00 |
+| dsv4-balanced | 64k_1 | 8 | PASS | 65564.19 | 1.00 | 8356.27 | 0.00 |
+| dsv4-balanced | 64k_1 | 16 | PASS | 65864.61 | 1.01 | 15823.61 | 0.00 |
+| dsv4-balanced | 64k_1 | 32 | PASS | 65504.60 | 1.00 | 32070.29 | 0.00 |
+| dsv4-balanced | 64k_1 | 64 | PASS | 65616.93 | 1.00 | 64081.25 | 0.00 |
+| dsv4-balanced | 64k_1 | 128 | PASS | 65401.37 | 1.00 | 128671.94 | 0.00 |
+| glm53-low-latency | 16k_512 | 1 | PASS | 7882.31 | 246.32 | 612.89 | 4.36 |
+| glm53-low-latency | 16k_512 | 8 | PASS | 16384.46 | 512.01 | 3892.78 | 22.46 |
+| glm53-low-latency | 16k_512 | 32 | PASS | 21303.26 | 665.73 | 15127.36 | 69.03 |
+| glm53-low-latency | 16k_512 | 64 | PASS | 23502.93 | 734.47 | 29490.35 | 105.62 |
+| glm53-low-latency | 16k_512 | 128 | PASS | 23920.46 | 747.51 | 63407.51 | 101.14 |
+| glm53-low-latency | 16k_512 | 256 | PASS | 24021.20 | 750.66 | 144384.42 | 102.36 |
+| glm53-low-latency | 16k_1 | 1 | FAIL | 26686.87 | 1.63 | 606.55 | 0.00 |
+| glm53-low-latency | 1k_128 | 1 | PASS | 2005.88 | 250.73 | 70.19 | 5.55 |
+| glm53-low-latency | 1k_128 | 8 | PASS | 6062.38 | 757.80 | 813.10 | 15.41 |
+| glm53-low-latency | 1k_128 | 32 | PASS | 11946.48 | 1493.31 | 1204.54 | 31.01 |
+| glm53-low-latency | 1k_128 | 64 | PASS | 14308.93 | 1788.62 | 1848.75 | 40.49 |
+| glm53-low-latency | 1k_128 | 128 | PASS | 14124.94 | 1765.62 | 6332.63 | 42.03 |
+| glm53-low-latency | 1k_128 | 256 | PASS | 13944.40 | 1743.05 | 15823.61 | 43.56 |
+| glm53-low-latency | 1k_4k | 1 | PASS | 109.08 | 436.33 | 70.69 | 3.32 |
+| glm53-low-latency | 1k_4k | 8 | PASS | 477.85 | 1911.42 | 435.51 | 5.56 |
+| glm53-low-latency | 1k_4k | 32 | PASS | 1191.77 | 4767.10 | 892.67 | 9.33 |
+| glm53-low-latency | 1k_4k | 64 | PASS | 1458.93 | 5835.71 | 19807.67 | 10.84 |
+| glm53-low-latency | 1k_4k | 128 | PASS | 1530.74 | 6122.95 | 56046.13 | 11.08 |
+| glm53-low-latency | 1k_4k | 256 | PASS | 1526.22 | 6104.87 | 140390.00 | 11.26 |
+| glm53-low-latency | 64k_1 | 1 | PASS | 24918.24 | 0.38 | 2625.80 | 0.00 |
+| glm53-low-latency | 64k_1 | 4 | FAIL | 25330.44 | 0.39 | 10319.69 | 0.00 |
+| glm53-low-latency | 64k_1 | 8 | PASS | 25336.86 | 0.39 | 20633.79 | 0.00 |
+| glm53-low-latency | 64k_1 | 16 | FAIL | 25368.13 | 0.39 | 41257.75 | 0.00 |
+| glm53-low-latency | 64k_1 | 32 | FAIL | 25376.71 | 0.39 | 82506.66 | 0.00 |
+| glm53-low-latency | 64k_1 | 64 | FAIL | 25399.08 | 0.39 | 164948.88 | 0.00 |
+| glm53-low-latency | 64k_1 | 128 | FAIL | 25411.76 | 0.39 | 329925.47 | 0.00 |
+| glm53-low-latency | 64k_512 | 1 | PASS | 15809.05 | 123.51 | 2626.48 | 4.40 |
+| glm53-low-latency | 64k_512 | 4 | PASS | 20932.05 | 163.53 | 5658.47 | 26.76 |
+| glm53-low-latency | 64k_512 | 8 | PASS | 22427.36 | 175.21 | 13072.49 | 61.58 |
+| glm53-low-latency | 64k_512 | 16 | PASS | 23295.48 | 182.00 | 31634.83 | 127.45 |
+| glm53-low-latency | 64k_512 | 32 | PASS | 23917.39 | 186.85 | 62694.72 | 243.87 |
+| glm53-low-latency | 64k_512 | 64 | PASS | 24220.17 | 189.22 | 128260.27 | 323.81 |
+| glm53-low-latency | 128k_1 | 1 | PASS | 22283.97 | 0.17 | 5887.63 | 0.00 |
+| glm53-low-latency | 128k_1 | 2 | PASS | 22544.03 | 0.17 | 11620.01 | 0.00 |
+| glm53-low-latency | 128k_1 | 4 | PASS | 22540.08 | 0.17 | 23234.84 | 0.00 |
+| glm53-low-latency | 128k_1 | 8 | FAIL | 22556.35 | 0.17 | 46447.98 | 0.00 |
+| glm53-low-latency | 128k_1 | 16 | FAIL | 22547.53 | 0.17 | 92876.53 | 0.00 |
+| glm53-low-latency | 128k_1 | 32 | FAIL | 22549.85 | 0.17 | 185810.07 | 0.00 |
+| glm53-low-latency | 128k_1 | 64 | TIMEOUT_PASS | 22567.39 | 0.17 | 371590.24 | 0.00 |
+| glm53-low-latency | 128k_512 | 1 | PASS | 17525.45 | 68.46 | 5894.95 | 4.62 |
+| glm53-low-latency | 128k_512 | 2 | PASS | 19570.06 | 76.45 | 6325.56 | 27.21 |
+| glm53-low-latency | 128k_512 | 4 | PASS | 20526.29 | 80.18 | 11700.22 | 62.53 |
+| glm53-low-latency | 128k_512 | 8 | PASS | 21132.30 | 82.55 | 28726.77 | 145.04 |
+| glm53-low-latency | 128k_512 | 16 | PASS | 21643.46 | 84.54 | 70612.08 | 282.81 |
+| glm53-low-latency | 128k_512 | 32 | PASS | 21787.55 | 85.11 | 142627.99 | 339.34 |
+| glm53-low-latency | 256k_1 | 1 | PASS | 17456.68 | 0.07 | 15141.04 | 0.00 |
+| glm53-low-latency | 256k_1 | 2 | PASS | 17620.27 | 0.07 | 29920.79 | 0.00 |
+| glm53-low-latency | 256k_1 | 4 | FAIL | 17584.82 | 0.07 | 59691.73 | 0.00 |
+| glm53-low-latency | 512k_1 | 1 | PASS | 12420.94 | 0.02 | 42470.03 | 0.00 |
+| glm53-low-latency | 512k_1 | 2 | PASS | 12497.10 | 0.02 | 84423.37 | 0.00 |
+| glm53-low-latency | 512k_1 | 4 | PASS | 12494.82 | 0.02 | 168091.37 | 0.00 |
+| glm53-low-latency | 1m_boundary_1 | 1 | PASS | 7897.30 | 0.01 | 132967.28 | 0.00 |
+| glm53-low-latency | 1m_boundary_1 | 2 | PASS | 7846.52 | 0.01 | 269850.67 | 0.00 |
+| glm53-low-latency | 1m_boundary_1 | 4 | PASS | 7914.10 | 0.01 | 509514.56 | 0.00 |
+| glm53-high-throughput | 16k_512 | 1 | PASS | 1753.70 | 54.80 | 1706.01 | 16.40 |
+| glm53-high-throughput | 16k_512 | 8 | PASS | 7602.48 | 237.58 | 2952.39 | 38.86 |
+| glm53-high-throughput | 16k_512 | 32 | PASS | 15478.02 | 483.69 | 10709.46 | 82.99 |
+| glm53-high-throughput | 16k_512 | 64 | PASS | 22864.39 | 714.51 | 17177.98 | 115.88 |
+| glm53-high-throughput | 16k_512 | 128 | PASS | 31889.21 | 996.54 | 30890.29 | 165.33 |
+| glm53-high-throughput | 16k_512 | 256 | PASS | 25722.47 | 803.83 | 58208.90 | 435.35 |
+| glm53-high-throughput | 16k_1 | 1 | FAIL | 10265.87 | 0.63 | 1635.66 | 0.00 |
+| glm53-high-throughput | 1k_128 | 1 | PASS | 440.61 | 55.08 | 354.14 | 17.25 |
+| glm53-high-throughput | 1k_128 | 8 | PASS | 2216.13 | 277.02 | 658.97 | 33.33 |
+| glm53-high-throughput | 1k_128 | 32 | PASS | 5512.82 | 689.10 | 1351.79 | 55.76 |
+| glm53-high-throughput | 1k_128 | 64 | PASS | 8781.68 | 1097.71 | 1544.35 | 75.33 |
+| glm53-high-throughput | 1k_128 | 128 | PASS | 15193.07 | 1899.13 | 2569.96 | 87.74 |
+| glm53-high-throughput | 1k_128 | 256 | PASS | 14521.72 | 1815.21 | 4585.68 | 192.57 |
+| glm53-high-throughput | 1k_4k | 1 | TIMEOUT_PASS | 15.31 | 61.25 | 722.89 | 21.05 |
+| glm53-high-throughput | 1k_4k | 8 | PASS | 122.82 | 491.26 | 585.67 | 17.31 |
+| glm53-high-throughput | 1k_4k | 32 | PASS | 458.53 | 1834.10 | 964.13 | 18.18 |
+| glm53-high-throughput | 1k_4k | 64 | PASS | 772.99 | 3091.95 | 1480.98 | 20.33 |
+| glm53-high-throughput | 1k_4k | 128 | PASS | 1421.86 | 5687.44 | 2492.89 | 23.05 |
+| glm53-high-throughput | 1k_4k | 256 | PASS | 2653.31 | 10613.25 | 4540.61 | 22.36 |
+| glm53-high-throughput | 64k_1 | 1 | PASS | 10461.13 | 0.16 | 6387.30 | 0.00 |
+| glm53-high-throughput | 64k_1 | 4 | PASS | 31751.47 | 0.48 | 8371.38 | 0.00 |
+| glm53-high-throughput | 64k_1 | 8 | PASS | 53720.01 | 0.82 | 9891.34 | 0.00 |
+| glm53-high-throughput | 64k_1 | 16 | FAIL | 56023.39 | 0.85 | 19477.62 | 0.00 |
+| glm53-high-throughput | 64k_1 | 32 | FAIL | 56599.60 | 0.86 | 36913.14 | 0.00 |
+| glm53-high-throughput | 64k_1 | 64 | FAIL | 56554.68 | 0.86 | 74260.73 | 0.00 |
+| glm53-high-throughput | 64k_1 | 128 | FAIL | 56357.44 | 0.86 | 153394.58 | 0.00 |
+| glm53-high-throughput | 64k_512 | 1 | PASS | 4707.22 | 36.78 | 6471.79 | 16.01 |
+| glm53-high-throughput | 64k_512 | 4 | PASS | 8844.03 | 69.09 | 8240.16 | 54.88 |
+| glm53-high-throughput | 64k_512 | 8 | PASS | 13286.74 | 103.80 | 9928.84 | 96.89 |
+| glm53-high-throughput | 64k_512 | 16 | PASS | 18863.70 | 147.37 | 19171.29 | 135.82 |
+| glm53-high-throughput | 64k_512 | 32 | PASS | 25109.46 | 196.17 | 29462.53 | 206.58 |
+| glm53-high-throughput | 64k_512 | 64 | PASS | 33437.72 | 261.23 | 57344.18 | 329.77 |
+| glm53-high-throughput | 128k_1 | 1 | FAIL | 10100.97 | 0.08 | 13350.24 | 0.00 |
+| glm53-high-throughput | 128k_512 | 1 | PASS | 6350.91 | 24.81 | 13384.88 | 16.13 |
+| glm53-high-throughput | 128k_512 | 2 | PASS | 7862.65 | 30.71 | 13400.23 | 41.95 |
+| glm53-high-throughput | 128k_512 | 4 | PASS | 9986.29 | 39.01 | 16714.05 | 92.60 |
+| glm53-high-throughput | 128k_512 | 8 | PASS | 13503.28 | 52.75 | 20219.46 | 171.97 |
+| glm53-high-throughput | 128k_512 | 16 | PASS | 19923.94 | 77.83 | 39164.93 | 264.45 |
+| glm53-high-throughput | 128k_512 | 32 | PASS | 27503.67 | 107.44 | 59905.46 | 401.75 |
+| glm53-high-throughput | 256k_1 | 1 | FAIL | 9204.74 | 0.04 | 29442.21 | 0.00 |
+| glm53-high-throughput | 512k_1 | 1 | PASS | 7532.88 | 0.01 | 71779.28 | 0.00 |
+| glm53-high-throughput | 512k_1 | 2 | PASS | 14141.46 | 0.03 | 77207.47 | 0.00 |
+| glm53-high-throughput | 512k_1 | 4 | PASS | 24677.77 | 0.05 | 86777.89 | 0.00 |
+| glm53-high-throughput | 1m_boundary_1 | 1 | PASS | 5330.26 | 0.01 | 204815.71 | 0.00 |
+| glm53-high-throughput | 1m_boundary_1 | 2 | PASS | 9956.15 | 0.01 | 214992.66 | 0.00 |
+| glm53-high-throughput | 1m_boundary_1 | 4 | PASS | 18580.87 | 0.02 | 225402.20 | 0.00 |
--- a/experiments/b300/dsv4_flash_glm53_sglang_text_matrix/snapshot-20260910-024355/b300-dsv4-glm53-text-matrix/b300-dsv4-glm53-dev4-20260909-084427/manual-stop.complete
+++ b/experiments/b300/dsv4_flash_glm53_sglang_text_matrix/snapshot-20260910-024355/b300-dsv4-glm53-text-matrix/b300-dsv4-glm53-dev4-20260909-084427/manual-stop.complete
@ -0,0 +1 @@
+2026-09-10T07:46:07Z
--- a/experiments/b300/dsv4_flash_glm53_sglang_text_matrix/snapshot-20260910-024355/b300-dsv4-glm53-text-matrix/b300-dsv4-glm53-dev4-20260909-084427/matrix.log
+++ b/experiments/b300/dsv4_flash_glm53_sglang_text_matrix/snapshot-20260910-024355/b300-dsv4-glm53-text-matrix/b300-dsv4-glm53-dev4-20260909-084427/matrix.log
@ -518,3 +518,9 @@ run_full_text_matrix.sh: line 247: 1797795 Killed                  timeout --sig
 [2026-09-10T07:21:11+00:00] warning profile=dsv4-balanced point=64k_1_c32 zero_ttft_count=1
 [2026-09-10T07:21:11+00:00] pass profile=dsv4-balanced point=64k_1_c32
 [2026-09-10T07:21:12+00:00] run profile=dsv4-balanced shape=64k_1 isl=65536 osl=1 c=64 prompts=320
+[2026-09-10T07:26:57+00:00] warning profile=dsv4-balanced point=64k_1_c64 zero_ttft_count=3
+[2026-09-10T07:26:57+00:00] pass profile=dsv4-balanced point=64k_1_c64
+[2026-09-10T07:26:58+00:00] run profile=dsv4-balanced shape=64k_1 isl=65536 osl=1 c=128 prompts=640
+[2026-09-10T07:38:13+00:00] warning profile=dsv4-balanced point=64k_1_c128 zero_ttft_count=5
+[2026-09-10T07:38:13+00:00] pass profile=dsv4-balanced point=64k_1_c128
+[2026-09-10T07:38:14+00:00] run profile=dsv4-balanced shape=64k_512 isl=65536 osl=512 c=1 prompts=64
--- a/experiments/b300/dsv4_flash_glm53_sglang_text_matrix/snapshot-20260910-024355/b300-dsv4-glm53-text-matrix/b300-dsv4-glm53-dev4-20260909-084427/point_status.csv
+++ b/experiments/b300/dsv4_flash_glm53_sglang_text_matrix/snapshot-20260910-024355/b300-dsv4-glm53-text-matrix/b300-dsv4-glm53-dev4-20260909-084427/point_status.csv
@ -182,3 +182,5 @@ timestamp,profile,point,status,rc
 2026-09-10T07:16:30+00:00,dsv4-balanced,64k_1_c8,PASS,0
 2026-09-10T07:18:09+00:00,dsv4-balanced,64k_1_c16,PASS,0
 2026-09-10T07:21:11+00:00,dsv4-balanced,64k_1_c32,PASS,0
+2026-09-10T07:26:57+00:00,dsv4-balanced,64k_1_c64,PASS,0
+2026-09-10T07:38:13+00:00,dsv4-balanced,64k_1_c128,PASS,0
--- a/experiments/b300/dsv4_flash_glm53_sglang_text_matrix/snapshot-20260910-024355/b300-dsv4-glm53-text-matrix/b300-dsv4-glm53-dev4-20260909-084427/started_points.csv
+++ b/experiments/b300/dsv4_flash_glm53_sglang_text_matrix/snapshot-20260910-024355/b300-dsv4-glm53-text-matrix/b300-dsv4-glm53-dev4-20260909-084427/started_points.csv
@ -183,3 +183,5 @@ timestamp,profile,shape,isl,osl,concurrency,num_prompts
 2026-09-10T07:16:31+00:00,dsv4-balanced,64k_1,65536,1,16,80
 2026-09-10T07:18:10+00:00,dsv4-balanced,64k_1,65536,1,32,160
 2026-09-10T07:21:12+00:00,dsv4-balanced,64k_1,65536,1,64,320
+2026-09-10T07:26:58+00:00,dsv4-balanced,64k_1,65536,1,128,640
+2026-09-10T07:38:14+00:00,dsv4-balanced,64k_512,65536,512,1,64