AbhishekGhosh1998 commited on
Commit
1d8b8ee
·
verified ·
1 Parent(s): 15dbf92

granularity results tierb-fixed-moonep-full @ 2026-08-30T02:38:06Z from sp-aiinfra-interactive-aghosh-7jtjl-8xb200-moe-cg-0 (part 2)

Browse files
tierb-fixed-moonep-full/steps.tsv CHANGED
@@ -4,5 +4,5 @@ OK moonep tierb 8 2026-08-29T23:44:39+00:00 2026-08-29T23:59:52+00:00 /workspace
4
  OK moonep tierbmem 8 2026-08-29T23:59:59+00:00 2026-08-30T00:09:58+00:00 /workspace/tierb-fixed-moonep-full/moonep/tierbmem
5
  OK echo tierbtrace 8 2026-08-30T00:10:16+00:00 2026-08-30T00:38:30+00:00 /workspace/tierb-fixed-moonep-full/echo/tierbtrace
6
  OK ultraep tierbtrace 8 2026-08-30T00:38:33+00:00 2026-08-30T00:59:28+00:00 /workspace/tierb-fixed-moonep-full/ultraep/tierbtrace
7
- FAIL moonep tierbtrace 8 2026-08-30T00:59:40+00:00 2026-08-30T01:06:12+00:00 /workspace/tierb-fixed-moonep-full/moonep/tierbtrace
8
  OK moonep tierb 8 2026-08-30T01:30:54+00:00 2026-08-30T01:46:01+00:00 /workspace/tierb-fixed-moonep-full/moonep/tierb
 
4
  OK moonep tierbmem 8 2026-08-29T23:59:59+00:00 2026-08-30T00:09:58+00:00 /workspace/tierb-fixed-moonep-full/moonep/tierbmem
5
  OK echo tierbtrace 8 2026-08-30T00:10:16+00:00 2026-08-30T00:38:30+00:00 /workspace/tierb-fixed-moonep-full/echo/tierbtrace
6
  OK ultraep tierbtrace 8 2026-08-30T00:38:33+00:00 2026-08-30T00:59:28+00:00 /workspace/tierb-fixed-moonep-full/ultraep/tierbtrace
7
+ SUPERSEDED moonep tierbtrace 8 2026-08-30T00:59:40+00:00 2026-08-30T01:06:12+00:00 /workspace/tierb-fixed-moonep-full/moonep/tierbtrace
8
  OK moonep tierb 8 2026-08-30T01:30:54+00:00 2026-08-30T01:46:01+00:00 /workspace/tierb-fixed-moonep-full/moonep/tierb
tierb-fixed-moonep-full/steps_attempt1_preserved.tsv ADDED
@@ -0,0 +1,8 @@
 
 
 
 
 
 
 
 
 
1
+ status campaign phase gbs started ended result_dir
2
+ OK moonep tierbprobe 8 2026-08-29T23:33:50+00:00 2026-08-29T23:44:30+00:00 /workspace/tierb-fixed-moonep-full/moonep/tierbprobe
3
+ OK moonep tierb 8 2026-08-29T23:44:39+00:00 2026-08-29T23:59:52+00:00 /workspace/tierb-fixed-moonep-full/moonep/tierb
4
+ OK moonep tierbmem 8 2026-08-29T23:59:59+00:00 2026-08-30T00:09:58+00:00 /workspace/tierb-fixed-moonep-full/moonep/tierbmem
5
+ OK echo tierbtrace 8 2026-08-30T00:10:16+00:00 2026-08-30T00:38:30+00:00 /workspace/tierb-fixed-moonep-full/echo/tierbtrace
6
+ OK ultraep tierbtrace 8 2026-08-30T00:38:33+00:00 2026-08-30T00:59:28+00:00 /workspace/tierb-fixed-moonep-full/ultraep/tierbtrace
7
+ FAIL moonep tierbtrace 8 2026-08-30T00:59:40+00:00 2026-08-30T01:06:12+00:00 /workspace/tierb-fixed-moonep-full/moonep/tierbtrace
8
+ OK moonep tierb 8 2026-08-30T01:30:54+00:00 2026-08-30T01:46:01+00:00 /workspace/tierb-fixed-moonep-full/moonep/tierb
tierb-fixed-moonep-full/study.log CHANGED
@@ -968,3 +968,93 @@ QWEN3_30B moonep 0 none none OK 113.7 40/
968
 
969
  ECHO/UltraEP note: graph-on uses static capacity + device-initiated GEMM while graph-off uses dynamic allocation + host-derived sizes. Their difference is not labelled as pure graph cost.
970
  Already logged in to Hugging Face as: AbhishekGhosh1998
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
968
 
969
  ECHO/UltraEP note: graph-on uses static capacity + device-initiated GEMM while graph-off uses dynamic allocation + host-derived sizes. Their difference is not labelled as pure graph cost.
970
  Already logged in to Hugging Face as: AbhishekGhosh1998
971
+ No files have been modified since last commit. Skipping to prevent empty commit.
972
+ uploaded: /workspace/tierb-fixed-moonep-full -> MoE-CUDA-Graph-Analysis/tierb-b200-gbs8/tierb-fixed-moonep-full
973
+ https://huggingface.co/MoE-CUDA-Graph-Analysis/tierb-b200-gbs8/tree/main
974
+
975
+ === STUDY COMPLETE
976
+ results : /workspace/tierb-fixed-moonep-full
977
+ status : /workspace/tierb-fixed-moonep-full/STATUS.txt
978
+ steps : /workspace/tierb-fixed-moonep-full/steps.tsv
979
+ hub : https://huggingface.co/datasets/MoE-CUDA-Graph-Analysis/tierb-b200-gbs8
980
+
981
+ status campaign phase gbs started ended result_dir
982
+ OK moonep tierbprobe 8 2026-08-29T23:33:50+00:00 2026-08-29T23:44:30+00:00 /workspace/tierb-fixed-moonep-full/moonep/tierbprobe
983
+ OK moonep tierb 8 2026-08-29T23:44:39+00:00 2026-08-29T23:59:52+00:00 /workspace/tierb-fixed-moonep-full/moonep/tierb
984
+ OK moonep tierbmem 8 2026-08-29T23:59:59+00:00 2026-08-30T00:09:58+00:00 /workspace/tierb-fixed-moonep-full/moonep/tierbmem
985
+ OK echo tierbtrace 8 2026-08-30T00:10:16+00:00 2026-08-30T00:38:30+00:00 /workspace/tierb-fixed-moonep-full/echo/tierbtrace
986
+ OK ultraep tierbtrace 8 2026-08-30T00:38:33+00:00 2026-08-30T00:59:28+00:00 /workspace/tierb-fixed-moonep-full/ultraep/tierbtrace
987
+ FAIL moonep tierbtrace 8 2026-08-30T00:59:40+00:00 2026-08-30T01:06:12+00:00 /workspace/tierb-fixed-moonep-full/moonep/tierbtrace
988
+ OK moonep tierb 8 2026-08-30T01:30:54+00:00 2026-08-30T01:46:01+00:00 /workspace/tierb-fixed-moonep-full/moonep/tierb
989
+ study_root: /workspace/tierb-fixed-moonep-full
990
+ gbs: 8
991
+ stages: 9
992
+ nsys: 1
993
+ hf_repo: MoE-CUDA-Graph-Analysis/tierb-b200-gbs8
994
+ hf_remote_root: tierb-fixed-moonep-full
995
+ started: 2026-08-30T02:38:05+00:00
996
+ driver_repo: /workspace/blackwell_run_scripts @ e8bd2bd
997
+
998
+ === STAGE 9 -- Tier B nsys traces (final, trace-only)
999
+
1000
+ === ROLL-UP
1001
+ 36 runs ok=32 oom=0 -> /workspace/tierb-fixed-moonep-full/collected_all.tsv
1002
+
1003
+ (excluded 8 nsys-traced run(s) from the timing comparison)
1004
+
1005
+ phase config seq gbs policy rc cg asy backend iter_ms vs_best
1006
+ -------------------------------------------------------------------------------------
1007
+ tierb DSV2_LITE_L24 8192 8 matched none 1 1 moonep 593.43 1.00x
1008
+
1009
+ tierb QWEN2_57B 4096 8 matched none 1 1 moonep 180.12 1.00x
1010
+
1011
+ tierb QWEN3_235B 8192 8 matched none 1 1 moonep 317.80 1.00x
1012
+
1013
+ tierb QWEN3_30B 4096 8 matched none 1 1 moonep 355.16 1.00x
1014
+
1015
+ tierbmem DSV2_LITE_L24 8192 8 matched none 1 1 moonep 622.80 1.00x
1016
+
1017
+ tierbmem QWEN2_57B 4096 8 matched none 1 1 moonep 196.60 1.00x
1018
+
1019
+ tierbmem QWEN3_235B 8192 8 matched none 1 1 moonep 341.70 1.00x
1020
+
1021
+ tierbmem QWEN3_30B 4096 8 matched none 1 1 moonep 435.30 1.00x
1022
+
1023
+ tierbprobeDSV2_LITE_L24 8192 8 matched none 0 1 moonep 673.94 1.00x
1024
+
1025
+ tierbprobeQWEN2_57B 4096 8 matched none 0 1 moonep 238.91 1.00x
1026
+
1027
+ tierbprobeQWEN3_235B 8192 8 matched none 0 1 moonep 384.76 1.00x
1028
+
1029
+ tierbprobeQWEN3_30B 4096 8 matched none 0 1 moonep 824.23 1.00x
1030
+
1031
+ 8 probe runs -> /workspace/tierb-fixed-moonep-full/probe_summary_all.tsv
1032
+ run arm cap rawExpI preRankI postRankI drop% reqCap
1033
+ --------------------------------------------------------------------------------------------------------------------
1034
+ moonep/tierbprobe/tierbprobe_gbs8/DSV2_LITE_L24__moo natural none 7.5443* 2.1316 1.1058
1035
+ moonep/tierbprobe/tierbprobe_gbs8/DSV2_LITE_L24__moo forced none 1.0286 1.0061 1.0026
1036
+ moonep/tierbprobe/tierbprobe_gbs8/QWEN2_57B__moonep_ natural none 7.4132* 2.3019 1.0979
1037
+ moonep/tierbprobe/tierbprobe_gbs8/QWEN2_57B__moonep_ forced none 1.0348 1.0073 1.0033
1038
+ moonep/tierbprobe/tierbprobe_gbs8/QWEN3_235B__moonep natural none 15.9227* 2.5851 1.0712
1039
+ moonep/tierbprobe/tierbprobe_gbs8/QWEN3_235B__moonep forced none 1.0386 1.0053 1.0025
1040
+ moonep/tierbprobe/tierbprobe_gbs8/QWEN3_30B__moonep_ natural none 15.7078* 2.4672 1.0635
1041
+ moonep/tierbprobe/tierbprobe_gbs8/QWEN3_30B__moonep_ forced none 1.0565 1.0077 1.0036
1042
+
1043
+ * At least one globally aggregated snapshot reached the E/K expert-Imax ceiling.
1044
+ Inspect raw_expert_imax_mean and hottest_expert_token_fraction_mean before generalising.
1045
+
1046
+ ECHO note: echo_logged_pre_rank_metric_valid=0 and echo_logged_load_reduction_valid=0 are intentional.
1047
+ The branch's built-in fields mix source-rank and destination-rank axes; this tool reconstructs both globally.
1048
+ 8 memory-dump runs -> /workspace/tierb-fixed-moonep-full/memory_summary_all.tsv
1049
+ model backend cg allocation capacity status peak GiB dumps
1050
+ ----------------------------------------------------------------------------------
1051
+ DSV2_LITE_L24 moonep 1 none none OK 130.2 48/48
1052
+ DSV2_LITE_L24 moonep 0 none none OK 122.2 40/40
1053
+ QWEN2_57B moonep 1 none none OK 94.5 48/48
1054
+ QWEN2_57B moonep 0 none none OK 93.6 40/40
1055
+ QWEN3_235B moonep 1 none none OK 117.5 48/48
1056
+ QWEN3_235B moonep 0 none none OK 117.3 40/40
1057
+ QWEN3_30B moonep 1 none none OK 114.0 48/48
1058
+ QWEN3_30B moonep 0 none none OK 113.7 40/40
1059
+
1060
+ ECHO/UltraEP note: graph-on uses static capacity + device-initiated GEMM while graph-off uses dynamic allocation + host-derived sizes. Their difference is not labelled as pure graph cost.