granularity results tierb-fixed-moonep-full/moonep/tierbtrace @ 2026-08-30T02:50:00Z from sp-aiinfra-interactive-aghosh-7jtjl-8xb200-moe-cg-0
Browse files- .gitattributes +1 -0
- tierb-fixed-moonep-full/moonep/tierbtrace/tierbtrace_gbs8/QWEN2_57B__moonep__matched__seq4096__gbs8__capnone__async/nsys/QWEN2_57B__moonep__matched__seq4096__gbs8__capnone__async.nsys-rep +3 -0
- tierb-fixed-moonep-full/moonep/tierbtrace/tierbtrace_gbs8/QWEN2_57B__moonep__matched__seq4096__gbs8__capnone__async/nsys/moonep_async_overlap.json +9 -0
- tierb-fixed-moonep-full/moonep/tierbtrace/tierbtrace_gbs8/QWEN2_57B__moonep__matched__seq4096__gbs8__capnone__async/nsys/moonep_async_overlap.txt +416 -0
- tierb-fixed-moonep-full/moonep/tierbtrace/tierbtrace_gbs8/QWEN2_57B__moonep__matched__seq4096__gbs8__capnone__async/nsys/moonep_async_overlap_summary.txt +1 -0
- tierb-fixed-moonep-full/moonep/tierbtrace/tierbtrace_gbs8/QWEN2_57B__moonep__matched__seq4096__gbs8__capnone__async/nsys/train.log +81 -0
- tierb-fixed-moonep-full/moonep/tierbtrace/tierbtrace_gbs8/QWEN2_57B__moonep__matched__seq4096__gbs8__capnone__async/nsys/verify_trace_ranks.log +6 -0
- tierb-fixed-moonep-full/moonep/tierbtrace/tierbtrace_gbs8/trace_results.tsv +1 -0
.gitattributes
CHANGED
|
@@ -66,3 +66,4 @@ tierb-fixed-moonep-full/ultraep/tierbtrace/tierbtrace_gbs8/QWEN2_57B__ultraep__m
|
|
| 66 |
tierb-fixed-moonep-full/moonep/tierbtrace/tierbtrace_gbs8/QWEN3_235B__moonep__matched__seq8192__gbs8__capnone__async/nsys/QWEN3_235B__moonep__matched__seq8192__gbs8__capnone__async.nsys-rep filter=lfs diff=lfs merge=lfs -text
|
| 67 |
tierb-fixed-moonep-full/moonep/tierbtrace/tierbtrace_gbs8/DSV2_LITE_L24__moonep__matched__seq8192__gbs8__capnone__async/nsys/DSV2_LITE_L24__moonep__matched__seq8192__gbs8__capnone__async.nsys-rep filter=lfs diff=lfs merge=lfs -text
|
| 68 |
tierb-fixed-moonep-full/moonep/tierbtrace/tierbtrace_gbs8/QWEN3_30B__moonep__matched__seq4096__gbs8__capnone__async/nsys/QWEN3_30B__moonep__matched__seq4096__gbs8__capnone__async.nsys-rep filter=lfs diff=lfs merge=lfs -text
|
|
|
|
|
|
| 66 |
tierb-fixed-moonep-full/moonep/tierbtrace/tierbtrace_gbs8/QWEN3_235B__moonep__matched__seq8192__gbs8__capnone__async/nsys/QWEN3_235B__moonep__matched__seq8192__gbs8__capnone__async.nsys-rep filter=lfs diff=lfs merge=lfs -text
|
| 67 |
tierb-fixed-moonep-full/moonep/tierbtrace/tierbtrace_gbs8/DSV2_LITE_L24__moonep__matched__seq8192__gbs8__capnone__async/nsys/DSV2_LITE_L24__moonep__matched__seq8192__gbs8__capnone__async.nsys-rep filter=lfs diff=lfs merge=lfs -text
|
| 68 |
tierb-fixed-moonep-full/moonep/tierbtrace/tierbtrace_gbs8/QWEN3_30B__moonep__matched__seq4096__gbs8__capnone__async/nsys/QWEN3_30B__moonep__matched__seq4096__gbs8__capnone__async.nsys-rep filter=lfs diff=lfs merge=lfs -text
|
| 69 |
+
tierb-fixed-moonep-full/moonep/tierbtrace/tierbtrace_gbs8/QWEN2_57B__moonep__matched__seq4096__gbs8__capnone__async/nsys/QWEN2_57B__moonep__matched__seq4096__gbs8__capnone__async.nsys-rep filter=lfs diff=lfs merge=lfs -text
|
tierb-fixed-moonep-full/moonep/tierbtrace/tierbtrace_gbs8/QWEN2_57B__moonep__matched__seq4096__gbs8__capnone__async/nsys/QWEN2_57B__moonep__matched__seq4096__gbs8__capnone__async.nsys-rep
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:cf9964e0453e497d76c08d8f285d046dc10a49a585c88e71e7a4c1e46d0d9d7d
|
| 3 |
+
size 5824677
|
tierb-fixed-moonep-full/moonep/tierbtrace/tierbtrace_gbs8/QWEN2_57B__moonep__matched__seq4096__gbs8__capnone__async/nsys/moonep_async_overlap.json
ADDED
|
@@ -0,0 +1,9 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"devices": 8,
|
| 3 |
+
"reduce_busy_ms_per_iteration": 5.654125,
|
| 4 |
+
"overlapped_ms_per_iteration": 5.428875,
|
| 5 |
+
"exposed_ms_per_iteration": 0.225125,
|
| 6 |
+
"overlap_pct": 96.01618287533438,
|
| 7 |
+
"minimum_overlap_pct": 0.1,
|
| 8 |
+
"status": "PASS"
|
| 9 |
+
}
|
tierb-fixed-moonep-full/moonep/tierbtrace/tierbtrace_gbs8/QWEN2_57B__moonep__matched__seq4096__gbs8__capnone__async/nsys/moonep_async_overlap.txt
ADDED
|
@@ -0,0 +1,416 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
|
| 2 |
+
=== kernels on device 0, by total time ===
|
| 3 |
+
43.488 ms device_kernel
|
| 4 |
+
41.759 ms multi_tensor_apply_kernel
|
| 5 |
+
40.659 ms ncclDevKernel_ReduceScatter_Sum_f32_RING_LL
|
| 6 |
+
29.947 ms elementwise_kernel
|
| 7 |
+
28.850 ms vectorized_elementwise_kernel
|
| 8 |
+
24.950 ms cast_mxfp8_2D_kernel
|
| 9 |
+
24.932 ms ncclDevKernel_AllGather_RING_LL
|
| 10 |
+
15.149 ms kernel_cutlass_kernel_moonepdispatchDispatchKernel_object_at__tensorptrbf16gmemalign16o40963584i64i64_tensorptrbf16gmemalign16o2785283584i64i64_tensorptri32gmemalign16o327681_tensorptri32_0
|
| 11 |
+
14.862 ms nvjet_qqtst_256x128_128x5_2x4_2cta_h_bz_Avec32UE8M0_Bvec32UE8M0_NNT
|
| 12 |
+
13.157 ms kernel_cutlass_kernel_moonepcombineCombineKernel_object_at__tensorptrbf16gmemalign16o2785283584i64i64_tensorptrbf16gmemalign16o40963584i64i64_tensorptri32gmemalign16o327681_tensorptri32gm_0
|
| 13 |
+
8.971 ms nvjet_qqsss_128x256_128x6_2x1_2cta_v_bz_Avec32UE8M0_Bvec32UE8M0_NTT
|
| 14 |
+
8.888 ms kernel_cutlass_kernel_moonepinter_rank_syncInterRankSyncKernel_object_at__tensorptri32gmemalign16o41943041_tensorptri32gmemalign16o10___0
|
| 15 |
+
8.660 ms nvjet_qqtst_128x256_128x6_2x1_2cta_v_bz_Avec32UE8M0_Bvec32UE8M0_TNT
|
| 16 |
+
7.547 ms nvjet_qqtst_256x128_128x5_2x4_2cta_h_bz_Avec32UE8M0_Bvec32UE8M0_TNT
|
| 17 |
+
REDUCE 7.402 ms kernel_cutlass_kernel_moonepgrad_reduceGradReduceKernel_object_at__tensorptrf32gmemalign16o3276803584i64i64_tensorptrf32gmemalign16o3276803584i64i64_CopyAtom_ThrID10_TVLayoutSrc11638401_T_0
|
| 18 |
+
6.964 ms nvjet_qqsss_128x256_128x6_2x2_2cta_h_bz_Avec32UE8M0_Bvec32UE8M0_NTT
|
| 19 |
+
6.883 ms cudnn_generated_fort_native_sdpa_sm100_flash_bprop_f16_knob_31_128x128x128_1x4x1_kernel0_0
|
| 20 |
+
6.099 ms unrolled_elementwise_kernel
|
| 21 |
+
5.699 ms triton_poi_fused__to_copy_cat_0
|
| 22 |
+
4.962 ms triton_poi_fused_cat_0
|
| 23 |
+
3.958 ms nvjet_qqtst_128x256_128x6_2x1_2cta_v_bz_Avec32UE8M0_Bvec32UE8M0_NNT
|
| 24 |
+
REDUCE 3.769 ms kernel_cutlass_kernel_moonepgrad_reduceGradReduceKernel_object_at__tensorptrf32gmemalign16o2293762560i64i64_tensorptrf32gmemalign16o2293762560i64i64_CopyAtom_ThrID10_TVLayoutSrc11638401_T_0
|
| 25 |
+
3.501 ms setGroupedGemmWgradArguments
|
| 26 |
+
3.205 ms ncclDevKernel_AllReduce_Sum_f32_RING_LL
|
| 27 |
+
3.194 ms kernel_cutlass_kernel_moonepprefetchPrefetchKernel_object_at__CopyAtom_ThrID10_TVLayoutSrc11638401_TVLayoutDst11638401_Valuetypebf16_tensor00o327680358411i640_CopyAtom_ThrID10_TVLayoutSrc_0
|
| 28 |
+
2.911 ms rmsnorm_bwd_general_kernel
|
| 29 |
+
2.218 ms cudnn_generated_fort_native_sdpa_sm100_flash_fprop_f16_knob_7_128x128x128_4x1x1_kernel0_0
|
| 30 |
+
2.110 ms triton_red_fused__to_copy_mul_silu_sum_1
|
| 31 |
+
1.810 ms nvjet_tst_320x192_64x4_2x2_2cta_h_bz_NNT
|
| 32 |
+
1.761 ms kernel_cutlass_kernel_moonepprefetchPrefetchKernel_object_at__CopyAtom_ThrID10_TVLayoutSrc11638401_TVLayoutDst11638401_Valuetypebf16_tensor00o229376256011i640_CopyAtom_ThrID10_TVLayoutSrc_0
|
| 33 |
+
1.701 ms nvjet_tss_128x256_64x6_2x1_2cta_v_badd_NTT
|
| 34 |
+
1.564 ms triton_poi_fused__to_copy_mul_silu_0
|
| 35 |
+
1.528 ms triton_poi_fused_mul_silu_0
|
| 36 |
+
1.525 ms nvjet_tst_128x256_64x6_2x1_2cta_v_bz_TNT
|
| 37 |
+
1.475 ms swizzle_col_scaling_kernel
|
| 38 |
+
1.407 ms reduce_kernel
|
| 39 |
+
1.312 ms rmsnorm_fwd_general_kernel
|
| 40 |
+
1.178 ms kernel_cutlass_kernel_moonepplanningPlanningKernel_object_at__tensorptri32_gmem_align16_o_641_tensorptri32_gmem_align16_o_327681_tensorptri32_gmem_align16_o_41943041_tensorptri32_gmem_ali_0
|
| 41 |
+
1.088 ms multi_tensor_swizzle_row_scaling_kernel
|
| 42 |
+
1.076 ms _scatter_gather_elementwise_kernel
|
| 43 |
+
|
| 44 |
+
=== QWEN2_57B__moonep__matched__seq4096__gbs8__capnone__async.sqlite (device 0, 2 iteration(s)) ===
|
| 45 |
+
reduce_grad kernels : 40 (20/iter)
|
| 46 |
+
reduce_grad sum : 5.586 ms/iter
|
| 47 |
+
reduce_grad busy (union) : 5.586 ms/iter
|
| 48 |
+
... overlapped with compute: 5.370 ms/iter ( 96.1 %)
|
| 49 |
+
... exposed on critical path: 0.216 ms/iter ( 3.9 %)
|
| 50 |
+
other GPU work (union) : 169.985 ms/iter
|
| 51 |
+
traced wall : 188.270 ms/iter
|
| 52 |
+
--> time hidden by overlap : 5.370 ms/iter
|
| 53 |
+
|
| 54 |
+
=== kernels on device 1, by total time ===
|
| 55 |
+
43.299 ms device_kernel
|
| 56 |
+
41.917 ms multi_tensor_apply_kernel
|
| 57 |
+
40.230 ms ncclDevKernel_ReduceScatter_Sum_f32_RING_LL
|
| 58 |
+
30.588 ms elementwise_kernel
|
| 59 |
+
28.878 ms vectorized_elementwise_kernel
|
| 60 |
+
24.874 ms cast_mxfp8_2D_kernel
|
| 61 |
+
23.761 ms ncclDevKernel_AllGather_RING_LL
|
| 62 |
+
15.118 ms kernel_cutlass_kernel_moonepdispatchDispatchKernel_object_at__tensorptrbf16gmemalign16o40963584i64i64_tensorptrbf16gmemalign16o2785283584i64i64_tensorptri32gmemalign16o327681_tensorptri32_0
|
| 63 |
+
15.038 ms nvjet_qqtst_256x128_128x5_2x4_2cta_h_bz_Avec32UE8M0_Bvec32UE8M0_NNT
|
| 64 |
+
13.366 ms kernel_cutlass_kernel_moonepcombineCombineKernel_object_at__tensorptrbf16gmemalign16o2785283584i64i64_tensorptrbf16gmemalign16o40963584i64i64_tensorptri32gmemalign16o327681_tensorptri32gm_0
|
| 65 |
+
8.768 ms nvjet_qqsss_128x256_128x6_2x1_2cta_v_bz_Avec32UE8M0_Bvec32UE8M0_NTT
|
| 66 |
+
8.489 ms nvjet_qqtst_128x256_128x6_2x1_2cta_v_bz_Avec32UE8M0_Bvec32UE8M0_TNT
|
| 67 |
+
7.563 ms nvjet_qqtst_256x128_128x5_2x4_2cta_h_bz_Avec32UE8M0_Bvec32UE8M0_TNT
|
| 68 |
+
REDUCE 7.395 ms kernel_cutlass_kernel_moonepgrad_reduceGradReduceKernel_object_at__tensorptrf32gmemalign16o3276803584i64i64_tensorptrf32gmemalign16o3276803584i64i64_CopyAtom_ThrID10_TVLayoutSrc11638401_T_0
|
| 69 |
+
7.139 ms kernel_cutlass_kernel_moonepinter_rank_syncInterRankSyncKernel_object_at__tensorptri32gmemalign16o41943041_tensorptri32gmemalign16o10___0
|
| 70 |
+
6.971 ms nvjet_qqsss_128x256_128x6_2x2_2cta_h_bz_Avec32UE8M0_Bvec32UE8M0_NTT
|
| 71 |
+
6.811 ms cudnn_generated_fort_native_sdpa_sm100_flash_bprop_f16_knob_31_128x128x128_1x4x1_kernel0_0
|
| 72 |
+
6.149 ms unrolled_elementwise_kernel
|
| 73 |
+
5.673 ms triton_poi_fused__to_copy_cat_0
|
| 74 |
+
4.930 ms triton_poi_fused_cat_0
|
| 75 |
+
4.109 ms ncclDevKernel_AllReduce_Sum_f32_RING_LL
|
| 76 |
+
4.037 ms kernel_cutlass_kernel_moonepprefetchPrefetchKernel_object_at__CopyAtom_ThrID10_TVLayoutSrc11638401_TVLayoutDst11638401_Valuetypebf16_tensor00o327680358411i640_CopyAtom_ThrID10_TVLayoutSrc_0
|
| 77 |
+
REDUCE 3.982 ms kernel_cutlass_kernel_moonepgrad_reduceGradReduceKernel_object_at__tensorptrf32gmemalign16o2293762560i64i64_tensorptrf32gmemalign16o2293762560i64i64_CopyAtom_ThrID10_TVLayoutSrc11638401_T_0
|
| 78 |
+
3.930 ms nvjet_qqtst_128x256_128x6_2x1_2cta_v_bz_Avec32UE8M0_Bvec32UE8M0_NNT
|
| 79 |
+
3.480 ms setGroupedGemmWgradArguments
|
| 80 |
+
3.029 ms rmsnorm_bwd_general_kernel
|
| 81 |
+
2.223 ms cudnn_generated_fort_native_sdpa_sm100_flash_fprop_f16_knob_7_128x128x128_4x1x1_kernel0_0
|
| 82 |
+
2.120 ms triton_red_fused__to_copy_mul_silu_sum_1
|
| 83 |
+
2.017 ms kernel_cutlass_kernel_moonepprefetchPrefetchKernel_object_at__CopyAtom_ThrID10_TVLayoutSrc11638401_TVLayoutDst11638401_Valuetypebf16_tensor00o229376256011i640_CopyAtom_ThrID10_TVLayoutSrc_0
|
| 84 |
+
1.799 ms nvjet_tst_320x192_64x4_2x2_2cta_h_bz_NNT
|
| 85 |
+
1.715 ms nvjet_tss_128x256_64x6_2x1_2cta_v_badd_NTT
|
| 86 |
+
1.576 ms triton_poi_fused__to_copy_mul_silu_0
|
| 87 |
+
1.538 ms triton_poi_fused_mul_silu_0
|
| 88 |
+
1.512 ms nvjet_tst_128x256_64x6_2x1_2cta_v_bz_TNT
|
| 89 |
+
1.451 ms swizzle_col_scaling_kernel
|
| 90 |
+
1.394 ms reduce_kernel
|
| 91 |
+
1.310 ms rmsnorm_fwd_general_kernel
|
| 92 |
+
1.220 ms kernel_cutlass_kernel_moonepplanningPlanningKernel_object_at__tensorptri32_gmem_align16_o_641_tensorptri32_gmem_align16_o_327681_tensorptri32_gmem_align16_o_41943041_tensorptri32_gmem_ali_0
|
| 93 |
+
1.108 ms multi_tensor_swizzle_row_scaling_kernel
|
| 94 |
+
1.065 ms _scatter_gather_elementwise_kernel
|
| 95 |
+
|
| 96 |
+
=== QWEN2_57B__moonep__matched__seq4096__gbs8__capnone__async.sqlite (device 1, 2 iteration(s)) ===
|
| 97 |
+
reduce_grad kernels : 40 (20/iter)
|
| 98 |
+
reduce_grad sum : 5.688 ms/iter
|
| 99 |
+
reduce_grad busy (union) : 5.688 ms/iter
|
| 100 |
+
... overlapped with compute: 5.451 ms/iter ( 95.8 %)
|
| 101 |
+
... exposed on critical path: 0.238 ms/iter ( 4.2 %)
|
| 102 |
+
other GPU work (union) : 170.206 ms/iter
|
| 103 |
+
traced wall : 188.135 ms/iter
|
| 104 |
+
--> time hidden by overlap : 5.451 ms/iter
|
| 105 |
+
|
| 106 |
+
=== kernels on device 2, by total time ===
|
| 107 |
+
43.195 ms device_kernel
|
| 108 |
+
41.886 ms multi_tensor_apply_kernel
|
| 109 |
+
41.672 ms ncclDevKernel_ReduceScatter_Sum_f32_RING_LL
|
| 110 |
+
30.465 ms elementwise_kernel
|
| 111 |
+
28.768 ms vectorized_elementwise_kernel
|
| 112 |
+
24.836 ms cast_mxfp8_2D_kernel
|
| 113 |
+
24.210 ms ncclDevKernel_AllGather_RING_LL
|
| 114 |
+
15.126 ms kernel_cutlass_kernel_moonepdispatchDispatchKernel_object_at__tensorptrbf16gmemalign16o40963584i64i64_tensorptrbf16gmemalign16o2785283584i64i64_tensorptri32gmemalign16o327681_tensorptri32_0
|
| 115 |
+
14.745 ms nvjet_qqtst_256x128_128x5_2x4_2cta_h_bz_Avec32UE8M0_Bvec32UE8M0_NNT
|
| 116 |
+
13.450 ms kernel_cutlass_kernel_moonepcombineCombineKernel_object_at__tensorptrbf16gmemalign16o2785283584i64i64_tensorptrbf16gmemalign16o40963584i64i64_tensorptri32gmemalign16o327681_tensorptri32gm_0
|
| 117 |
+
8.838 ms nvjet_qqsss_128x256_128x6_2x1_2cta_v_bz_Avec32UE8M0_Bvec32UE8M0_NTT
|
| 118 |
+
8.504 ms nvjet_qqtst_128x256_128x6_2x1_2cta_v_bz_Avec32UE8M0_Bvec32UE8M0_TNT
|
| 119 |
+
7.798 ms kernel_cutlass_kernel_moonepinter_rank_syncInterRankSyncKernel_object_at__tensorptri32gmemalign16o41943041_tensorptri32gmemalign16o10___0
|
| 120 |
+
7.619 ms nvjet_qqtst_256x128_128x5_2x4_2cta_h_bz_Avec32UE8M0_Bvec32UE8M0_TNT
|
| 121 |
+
REDUCE 7.207 ms kernel_cutlass_kernel_moonepgrad_reduceGradReduceKernel_object_at__tensorptrf32gmemalign16o3276803584i64i64_tensorptrf32gmemalign16o3276803584i64i64_CopyAtom_ThrID10_TVLayoutSrc11638401_T_0
|
| 122 |
+
6.936 ms nvjet_qqsss_128x256_128x6_2x2_2cta_h_bz_Avec32UE8M0_Bvec32UE8M0_NTT
|
| 123 |
+
6.841 ms cudnn_generated_fort_native_sdpa_sm100_flash_bprop_f16_knob_31_128x128x128_1x4x1_kernel0_0
|
| 124 |
+
6.202 ms unrolled_elementwise_kernel
|
| 125 |
+
5.663 ms triton_poi_fused__to_copy_cat_0
|
| 126 |
+
5.615 ms ncclDevKernel_AllReduce_Sum_f32_RING_LL
|
| 127 |
+
4.893 ms triton_poi_fused_cat_0
|
| 128 |
+
REDUCE 4.086 ms kernel_cutlass_kernel_moonepgrad_reduceGradReduceKernel_object_at__tensorptrf32gmemalign16o2293762560i64i64_tensorptrf32gmemalign16o2293762560i64i64_CopyAtom_ThrID10_TVLayoutSrc11638401_T_0
|
| 129 |
+
4.057 ms kernel_cutlass_kernel_moonepprefetchPrefetchKernel_object_at__CopyAtom_ThrID10_TVLayoutSrc11638401_TVLayoutDst11638401_Valuetypebf16_tensor00o327680358411i640_CopyAtom_ThrID10_TVLayoutSrc_0
|
| 130 |
+
3.880 ms nvjet_qqtst_128x256_128x6_2x1_2cta_v_bz_Avec32UE8M0_Bvec32UE8M0_NNT
|
| 131 |
+
3.474 ms setGroupedGemmWgradArguments
|
| 132 |
+
3.058 ms rmsnorm_bwd_general_kernel
|
| 133 |
+
2.238 ms cudnn_generated_fort_native_sdpa_sm100_flash_fprop_f16_knob_7_128x128x128_4x1x1_kernel0_0
|
| 134 |
+
2.121 ms triton_red_fused__to_copy_mul_silu_sum_1
|
| 135 |
+
1.913 ms kernel_cutlass_kernel_moonepprefetchPrefetchKernel_object_at__CopyAtom_ThrID10_TVLayoutSrc11638401_TVLayoutDst11638401_Valuetypebf16_tensor00o229376256011i640_CopyAtom_ThrID10_TVLayoutSrc_0
|
| 136 |
+
1.821 ms nvjet_tst_320x192_64x4_2x2_2cta_h_bz_NNT
|
| 137 |
+
1.661 ms nvjet_tss_128x256_64x6_2x1_2cta_v_badd_NTT
|
| 138 |
+
1.572 ms triton_poi_fused__to_copy_mul_silu_0
|
| 139 |
+
1.569 ms triton_poi_fused_mul_silu_0
|
| 140 |
+
1.500 ms nvjet_tst_128x256_64x6_2x1_2cta_v_bz_TNT
|
| 141 |
+
1.454 ms swizzle_col_scaling_kernel
|
| 142 |
+
1.413 ms reduce_kernel
|
| 143 |
+
1.295 ms rmsnorm_fwd_general_kernel
|
| 144 |
+
1.210 ms kernel_cutlass_kernel_moonepplanningPlanningKernel_object_at__tensorptri32_gmem_align16_o_641_tensorptri32_gmem_align16_o_327681_tensorptri32_gmem_align16_o_41943041_tensorptri32_gmem_ali_0
|
| 145 |
+
1.099 ms multi_tensor_swizzle_row_scaling_kernel
|
| 146 |
+
1.076 ms _scatter_gather_elementwise_kernel
|
| 147 |
+
|
| 148 |
+
=== QWEN2_57B__moonep__matched__seq4096__gbs8__capnone__async.sqlite (device 2, 2 iteration(s)) ===
|
| 149 |
+
reduce_grad kernels : 40 (20/iter)
|
| 150 |
+
reduce_grad sum : 5.647 ms/iter
|
| 151 |
+
reduce_grad busy (union) : 5.647 ms/iter
|
| 152 |
+
... overlapped with compute: 5.429 ms/iter ( 96.2 %)
|
| 153 |
+
... exposed on critical path: 0.217 ms/iter ( 3.8 %)
|
| 154 |
+
other GPU work (union) : 171.338 ms/iter
|
| 155 |
+
traced wall : 188.244 ms/iter
|
| 156 |
+
--> time hidden by overlap : 5.429 ms/iter
|
| 157 |
+
|
| 158 |
+
=== kernels on device 3, by total time ===
|
| 159 |
+
44.105 ms device_kernel
|
| 160 |
+
41.994 ms multi_tensor_apply_kernel
|
| 161 |
+
39.302 ms ncclDevKernel_ReduceScatter_Sum_f32_RING_LL
|
| 162 |
+
30.685 ms elementwise_kernel
|
| 163 |
+
29.089 ms vectorized_elementwise_kernel
|
| 164 |
+
24.919 ms cast_mxfp8_2D_kernel
|
| 165 |
+
23.562 ms ncclDevKernel_AllGather_RING_LL
|
| 166 |
+
15.200 ms nvjet_qqtst_256x128_128x5_2x4_2cta_h_bz_Avec32UE8M0_Bvec32UE8M0_NNT
|
| 167 |
+
15.124 ms kernel_cutlass_kernel_moonepdispatchDispatchKernel_object_at__tensorptrbf16gmemalign16o40963584i64i64_tensorptrbf16gmemalign16o2785283584i64i64_tensorptri32gmemalign16o327681_tensorptri32_0
|
| 168 |
+
13.189 ms kernel_cutlass_kernel_moonepcombineCombineKernel_object_at__tensorptrbf16gmemalign16o2785283584i64i64_tensorptrbf16gmemalign16o40963584i64i64_tensorptri32gmemalign16o327681_tensorptri32gm_0
|
| 169 |
+
8.847 ms nvjet_qqsss_128x256_128x6_2x1_2cta_v_bz_Avec32UE8M0_Bvec32UE8M0_NTT
|
| 170 |
+
8.585 ms nvjet_qqtst_128x256_128x6_2x1_2cta_v_bz_Avec32UE8M0_Bvec32UE8M0_TNT
|
| 171 |
+
7.630 ms nvjet_qqtst_256x128_128x5_2x4_2cta_h_bz_Avec32UE8M0_Bvec32UE8M0_TNT
|
| 172 |
+
REDUCE 7.401 ms kernel_cutlass_kernel_moonepgrad_reduceGradReduceKernel_object_at__tensorptrf32gmemalign16o3276803584i64i64_tensorptrf32gmemalign16o3276803584i64i64_CopyAtom_ThrID10_TVLayoutSrc11638401_T_0
|
| 173 |
+
7.023 ms nvjet_qqsss_128x256_128x6_2x2_2cta_h_bz_Avec32UE8M0_Bvec32UE8M0_NTT
|
| 174 |
+
6.847 ms cudnn_generated_fort_native_sdpa_sm100_flash_bprop_f16_knob_31_128x128x128_1x4x1_kernel0_0
|
| 175 |
+
6.231 ms unrolled_elementwise_kernel
|
| 176 |
+
5.685 ms triton_poi_fused__to_copy_cat_0
|
| 177 |
+
5.420 ms ncclDevKernel_AllReduce_Sum_f32_RING_LL
|
| 178 |
+
4.965 ms triton_poi_fused_cat_0
|
| 179 |
+
4.917 ms kernel_cutlass_kernel_moonepinter_rank_syncInterRankSyncKernel_object_at__tensorptri32gmemalign16o41943041_tensorptri32gmemalign16o10___0
|
| 180 |
+
4.377 ms kernel_cutlass_kernel_moonepprefetchPrefetchKernel_object_at__CopyAtom_ThrID10_TVLayoutSrc11638401_TVLayoutDst11638401_Valuetypebf16_tensor00o327680358411i640_CopyAtom_ThrID10_TVLayoutSrc_0
|
| 181 |
+
3.995 ms nvjet_qqtst_128x256_128x6_2x1_2cta_v_bz_Avec32UE8M0_Bvec32UE8M0_NNT
|
| 182 |
+
REDUCE 3.895 ms kernel_cutlass_kernel_moonepgrad_reduceGradReduceKernel_object_at__tensorptrf32gmemalign16o2293762560i64i64_tensorptrf32gmemalign16o2293762560i64i64_CopyAtom_ThrID10_TVLayoutSrc11638401_T_0
|
| 183 |
+
3.379 ms setGroupedGemmWgradArguments
|
| 184 |
+
3.051 ms rmsnorm_bwd_general_kernel
|
| 185 |
+
2.243 ms cudnn_generated_fort_native_sdpa_sm100_flash_fprop_f16_knob_7_128x128x128_4x1x1_kernel0_0
|
| 186 |
+
2.113 ms triton_red_fused__to_copy_mul_silu_sum_1
|
| 187 |
+
2.011 ms kernel_cutlass_kernel_moonepprefetchPrefetchKernel_object_at__CopyAtom_ThrID10_TVLayoutSrc11638401_TVLayoutDst11638401_Valuetypebf16_tensor00o229376256011i640_CopyAtom_ThrID10_TVLayoutSrc_0
|
| 188 |
+
1.826 ms nvjet_tst_320x192_64x4_2x2_2cta_h_bz_NNT
|
| 189 |
+
1.743 ms nvjet_tss_128x256_64x6_2x1_2cta_v_badd_NTT
|
| 190 |
+
1.579 ms triton_poi_fused__to_copy_mul_silu_0
|
| 191 |
+
1.536 ms nvjet_tst_128x256_64x6_2x1_2cta_v_bz_TNT
|
| 192 |
+
1.524 ms triton_poi_fused_mul_silu_0
|
| 193 |
+
1.475 ms swizzle_col_scaling_kernel
|
| 194 |
+
1.414 ms reduce_kernel
|
| 195 |
+
1.314 ms rmsnorm_fwd_general_kernel
|
| 196 |
+
1.207 ms kernel_cutlass_kernel_moonepplanningPlanningKernel_object_at__tensorptri32_gmem_align16_o_641_tensorptri32_gmem_align16_o_327681_tensorptri32_gmem_align16_o_41943041_tensorptri32_gmem_ali_0
|
| 197 |
+
1.139 ms multi_tensor_swizzle_row_scaling_kernel
|
| 198 |
+
1.078 ms _scatter_gather_elementwise_kernel
|
| 199 |
+
|
| 200 |
+
=== QWEN2_57B__moonep__matched__seq4096__gbs8__capnone__async.sqlite (device 3, 2 iteration(s)) ===
|
| 201 |
+
reduce_grad kernels : 40 (20/iter)
|
| 202 |
+
reduce_grad sum : 5.648 ms/iter
|
| 203 |
+
reduce_grad busy (union) : 5.648 ms/iter
|
| 204 |
+
... overlapped with compute: 5.424 ms/iter ( 96.0 %)
|
| 205 |
+
... exposed on critical path: 0.224 ms/iter ( 4.0 %)
|
| 206 |
+
other GPU work (union) : 171.016 ms/iter
|
| 207 |
+
traced wall : 188.252 ms/iter
|
| 208 |
+
--> time hidden by overlap : 5.424 ms/iter
|
| 209 |
+
|
| 210 |
+
=== kernels on device 4, by total time ===
|
| 211 |
+
44.301 ms device_kernel
|
| 212 |
+
41.787 ms multi_tensor_apply_kernel
|
| 213 |
+
39.259 ms ncclDevKernel_ReduceScatter_Sum_f32_RING_LL
|
| 214 |
+
30.678 ms elementwise_kernel
|
| 215 |
+
28.893 ms vectorized_elementwise_kernel
|
| 216 |
+
24.818 ms cast_mxfp8_2D_kernel
|
| 217 |
+
23.267 ms ncclDevKernel_AllGather_RING_LL
|
| 218 |
+
15.239 ms nvjet_qqtst_256x128_128x5_2x4_2cta_h_bz_Avec32UE8M0_Bvec32UE8M0_NNT
|
| 219 |
+
15.101 ms kernel_cutlass_kernel_moonepdispatchDispatchKernel_object_at__tensorptrbf16gmemalign16o40963584i64i64_tensorptrbf16gmemalign16o2785283584i64i64_tensorptri32gmemalign16o327681_tensorptri32_0
|
| 220 |
+
13.362 ms kernel_cutlass_kernel_moonepcombineCombineKernel_object_at__tensorptrbf16gmemalign16o2785283584i64i64_tensorptrbf16gmemalign16o40963584i64i64_tensorptri32gmemalign16o327681_tensorptri32gm_0
|
| 221 |
+
8.906 ms nvjet_qqsss_128x256_128x6_2x1_2cta_v_bz_Avec32UE8M0_Bvec32UE8M0_NTT
|
| 222 |
+
8.530 ms nvjet_qqtst_128x256_128x6_2x1_2cta_v_bz_Avec32UE8M0_Bvec32UE8M0_TNT
|
| 223 |
+
7.635 ms nvjet_qqtst_256x128_128x5_2x4_2cta_h_bz_Avec32UE8M0_Bvec32UE8M0_TNT
|
| 224 |
+
REDUCE 7.536 ms kernel_cutlass_kernel_moonepgrad_reduceGradReduceKernel_object_at__tensorptrf32gmemalign16o3276803584i64i64_tensorptrf32gmemalign16o3276803584i64i64_CopyAtom_ThrID10_TVLayoutSrc11638401_T_0
|
| 225 |
+
7.064 ms nvjet_qqsss_128x256_128x6_2x2_2cta_h_bz_Avec32UE8M0_Bvec32UE8M0_NTT
|
| 226 |
+
6.862 ms cudnn_generated_fort_native_sdpa_sm100_flash_bprop_f16_knob_31_128x128x128_1x4x1_kernel0_0
|
| 227 |
+
6.193 ms unrolled_elementwise_kernel
|
| 228 |
+
5.690 ms triton_poi_fused__to_copy_cat_0
|
| 229 |
+
5.678 ms ncclDevKernel_AllReduce_Sum_f32_RING_LL
|
| 230 |
+
4.958 ms triton_poi_fused_cat_0
|
| 231 |
+
4.700 ms kernel_cutlass_kernel_moonepinter_rank_syncInterRankSyncKernel_object_at__tensorptri32gmemalign16o41943041_tensorptri32gmemalign16o10___0
|
| 232 |
+
4.565 ms kernel_cutlass_kernel_moonepprefetchPrefetchKernel_object_at__CopyAtom_ThrID10_TVLayoutSrc11638401_TVLayoutDst11638401_Valuetypebf16_tensor00o327680358411i640_CopyAtom_ThrID10_TVLayoutSrc_0
|
| 233 |
+
3.995 ms nvjet_qqtst_128x256_128x6_2x1_2cta_v_bz_Avec32UE8M0_Bvec32UE8M0_NNT
|
| 234 |
+
REDUCE 3.871 ms kernel_cutlass_kernel_moonepgrad_reduceGradReduceKernel_object_at__tensorptrf32gmemalign16o2293762560i64i64_tensorptrf32gmemalign16o2293762560i64i64_CopyAtom_ThrID10_TVLayoutSrc11638401_T_0
|
| 235 |
+
3.360 ms setGroupedGemmWgradArguments
|
| 236 |
+
3.038 ms rmsnorm_bwd_general_kernel
|
| 237 |
+
2.234 ms cudnn_generated_fort_native_sdpa_sm100_flash_fprop_f16_knob_7_128x128x128_4x1x1_kernel0_0
|
| 238 |
+
2.119 ms triton_red_fused__to_copy_mul_silu_sum_1
|
| 239 |
+
2.068 ms kernel_cutlass_kernel_moonepprefetchPrefetchKernel_object_at__CopyAtom_ThrID10_TVLayoutSrc11638401_TVLayoutDst11638401_Valuetypebf16_tensor00o229376256011i640_CopyAtom_ThrID10_TVLayoutSrc_0
|
| 240 |
+
1.817 ms nvjet_tst_320x192_64x4_2x2_2cta_h_bz_NNT
|
| 241 |
+
1.736 ms nvjet_tss_128x256_64x6_2x1_2cta_v_badd_NTT
|
| 242 |
+
1.582 ms triton_poi_fused__to_copy_mul_silu_0
|
| 243 |
+
1.543 ms nvjet_tst_128x256_64x6_2x1_2cta_v_bz_TNT
|
| 244 |
+
1.497 ms triton_poi_fused_mul_silu_0
|
| 245 |
+
1.475 ms swizzle_col_scaling_kernel
|
| 246 |
+
1.418 ms reduce_kernel
|
| 247 |
+
1.311 ms rmsnorm_fwd_general_kernel
|
| 248 |
+
1.209 ms kernel_cutlass_kernel_moonepplanningPlanningKernel_object_at__tensorptri32_gmem_align16_o_641_tensorptri32_gmem_align16_o_327681_tensorptri32_gmem_align16_o_41943041_tensorptri32_gmem_ali_0
|
| 249 |
+
1.122 ms multi_tensor_swizzle_row_scaling_kernel
|
| 250 |
+
1.079 ms _scatter_gather_elementwise_kernel
|
| 251 |
+
|
| 252 |
+
=== QWEN2_57B__moonep__matched__seq4096__gbs8__capnone__async.sqlite (device 4, 2 iteration(s)) ===
|
| 253 |
+
reduce_grad kernels : 40 (20/iter)
|
| 254 |
+
reduce_grad sum : 5.704 ms/iter
|
| 255 |
+
reduce_grad busy (union) : 5.704 ms/iter
|
| 256 |
+
... overlapped with compute: 5.484 ms/iter ( 96.2 %)
|
| 257 |
+
... exposed on critical path: 0.219 ms/iter ( 3.8 %)
|
| 258 |
+
other GPU work (union) : 171.090 ms/iter
|
| 259 |
+
traced wall : 188.344 ms/iter
|
| 260 |
+
--> time hidden by overlap : 5.484 ms/iter
|
| 261 |
+
|
| 262 |
+
=== kernels on device 5, by total time ===
|
| 263 |
+
43.196 ms device_kernel
|
| 264 |
+
41.914 ms multi_tensor_apply_kernel
|
| 265 |
+
41.601 ms ncclDevKernel_ReduceScatter_Sum_f32_RING_LL
|
| 266 |
+
30.508 ms elementwise_kernel
|
| 267 |
+
28.676 ms vectorized_elementwise_kernel
|
| 268 |
+
26.302 ms ncclDevKernel_AllGather_RING_LL
|
| 269 |
+
25.015 ms cast_mxfp8_2D_kernel
|
| 270 |
+
15.130 ms kernel_cutlass_kernel_moonepdispatchDispatchKernel_object_at__tensorptrbf16gmemalign16o40963584i64i64_tensorptrbf16gmemalign16o2785283584i64i64_tensorptri32gmemalign16o327681_tensorptri32_0
|
| 271 |
+
14.913 ms nvjet_qqtst_256x128_128x5_2x4_2cta_h_bz_Avec32UE8M0_Bvec32UE8M0_NNT
|
| 272 |
+
13.165 ms kernel_cutlass_kernel_moonepcombineCombineKernel_object_at__tensorptrbf16gmemalign16o2785283584i64i64_tensorptrbf16gmemalign16o40963584i64i64_tensorptri32gmemalign16o327681_tensorptri32gm_0
|
| 273 |
+
9.320 ms kernel_cutlass_kernel_moonepinter_rank_syncInterRankSyncKernel_object_at__tensorptri32gmemalign16o41943041_tensorptri32gmemalign16o10___0
|
| 274 |
+
8.864 ms nvjet_qqsss_128x256_128x6_2x1_2cta_v_bz_Avec32UE8M0_Bvec32UE8M0_NTT
|
| 275 |
+
8.713 ms nvjet_qqtst_128x256_128x6_2x1_2cta_v_bz_Avec32UE8M0_Bvec32UE8M0_TNT
|
| 276 |
+
7.551 ms nvjet_qqtst_256x128_128x5_2x4_2cta_h_bz_Avec32UE8M0_Bvec32UE8M0_TNT
|
| 277 |
+
REDUCE 7.255 ms kernel_cutlass_kernel_moonepgrad_reduceGradReduceKernel_object_at__tensorptrf32gmemalign16o3276803584i64i64_tensorptrf32gmemalign16o3276803584i64i64_CopyAtom_ThrID10_TVLayoutSrc11638401_T_0
|
| 278 |
+
6.934 ms nvjet_qqsss_128x256_128x6_2x2_2cta_h_bz_Avec32UE8M0_Bvec32UE8M0_NTT
|
| 279 |
+
6.813 ms cudnn_generated_fort_native_sdpa_sm100_flash_bprop_f16_knob_31_128x128x128_1x4x1_kernel0_0
|
| 280 |
+
6.141 ms unrolled_elementwise_kernel
|
| 281 |
+
5.664 ms triton_poi_fused__to_copy_cat_0
|
| 282 |
+
5.632 ms ncclDevKernel_AllReduce_Sum_f32_RING_LL
|
| 283 |
+
4.904 ms triton_poi_fused_cat_0
|
| 284 |
+
3.904 ms nvjet_qqtst_128x256_128x6_2x1_2cta_v_bz_Avec32UE8M0_Bvec32UE8M0_NNT
|
| 285 |
+
REDUCE 3.861 ms kernel_cutlass_kernel_moonepgrad_reduceGradReduceKernel_object_at__tensorptrf32gmemalign16o2293762560i64i64_tensorptrf32gmemalign16o2293762560i64i64_CopyAtom_ThrID10_TVLayoutSrc11638401_T_0
|
| 286 |
+
3.444 ms setGroupedGemmWgradArguments
|
| 287 |
+
3.023 ms rmsnorm_bwd_general_kernel
|
| 288 |
+
2.682 ms kernel_cutlass_kernel_moonepprefetchPrefetchKernel_object_at__CopyAtom_ThrID10_TVLayoutSrc11638401_TVLayoutDst11638401_Valuetypebf16_tensor00o327680358411i640_CopyAtom_ThrID10_TVLayoutSrc_0
|
| 289 |
+
2.237 ms cudnn_generated_fort_native_sdpa_sm100_flash_fprop_f16_knob_7_128x128x128_4x1x1_kernel0_0
|
| 290 |
+
2.118 ms triton_red_fused__to_copy_mul_silu_sum_1
|
| 291 |
+
1.813 ms nvjet_tst_320x192_64x4_2x2_2cta_h_bz_NNT
|
| 292 |
+
1.677 ms nvjet_tss_128x256_64x6_2x1_2cta_v_badd_NTT
|
| 293 |
+
1.576 ms triton_poi_fused__to_copy_mul_silu_0
|
| 294 |
+
1.555 ms triton_poi_fused_mul_silu_0
|
| 295 |
+
1.509 ms nvjet_tst_128x256_64x6_2x1_2cta_v_bz_TNT
|
| 296 |
+
1.445 ms swizzle_col_scaling_kernel
|
| 297 |
+
1.396 ms reduce_kernel
|
| 298 |
+
1.307 ms rmsnorm_fwd_general_kernel
|
| 299 |
+
1.286 ms kernel_cutlass_kernel_moonepprefetchPrefetchKernel_object_at__CopyAtom_ThrID10_TVLayoutSrc11638401_TVLayoutDst11638401_Valuetypebf16_tensor00o229376256011i640_CopyAtom_ThrID10_TVLayoutSrc_0
|
| 300 |
+
1.219 ms kernel_cutlass_kernel_moonepplanningPlanningKernel_object_at__tensorptri32_gmem_align16_o_641_tensorptri32_gmem_align16_o_327681_tensorptri32_gmem_align16_o_41943041_tensorptri32_gmem_ali_0
|
| 301 |
+
1.082 ms multi_tensor_swizzle_row_scaling_kernel
|
| 302 |
+
1.065 ms _scatter_gather_elementwise_kernel
|
| 303 |
+
|
| 304 |
+
=== QWEN2_57B__moonep__matched__seq4096__gbs8__capnone__async.sqlite (device 5, 2 iteration(s)) ===
|
| 305 |
+
reduce_grad kernels : 40 (20/iter)
|
| 306 |
+
reduce_grad sum : 5.558 ms/iter
|
| 307 |
+
reduce_grad busy (union) : 5.558 ms/iter
|
| 308 |
+
... overlapped with compute: 5.326 ms/iter ( 95.8 %)
|
| 309 |
+
... exposed on critical path: 0.232 ms/iter ( 4.2 %)
|
| 310 |
+
other GPU work (union) : 171.203 ms/iter
|
| 311 |
+
traced wall : 188.270 ms/iter
|
| 312 |
+
--> time hidden by overlap : 5.326 ms/iter
|
| 313 |
+
|
| 314 |
+
=== kernels on device 6, by total time ===
|
| 315 |
+
43.311 ms device_kernel
|
| 316 |
+
41.861 ms multi_tensor_apply_kernel
|
| 317 |
+
40.487 ms ncclDevKernel_ReduceScatter_Sum_f32_RING_LL
|
| 318 |
+
30.548 ms elementwise_kernel
|
| 319 |
+
28.861 ms vectorized_elementwise_kernel
|
| 320 |
+
24.975 ms cast_mxfp8_2D_kernel
|
| 321 |
+
24.815 ms ncclDevKernel_AllGather_RING_LL
|
| 322 |
+
15.104 ms kernel_cutlass_kernel_moonepdispatchDispatchKernel_object_at__tensorptrbf16gmemalign16o40963584i64i64_tensorptrbf16gmemalign16o2785283584i64i64_tensorptri32gmemalign16o327681_tensorptri32_0
|
| 323 |
+
14.875 ms nvjet_qqtst_256x128_128x5_2x4_2cta_h_bz_Avec32UE8M0_Bvec32UE8M0_NNT
|
| 324 |
+
13.106 ms kernel_cutlass_kernel_moonepcombineCombineKernel_object_at__tensorptrbf16gmemalign16o2785283584i64i64_tensorptrbf16gmemalign16o40963584i64i64_tensorptri32gmemalign16o327681_tensorptri32gm_0
|
| 325 |
+
8.831 ms nvjet_qqsss_128x256_128x6_2x1_2cta_v_bz_Avec32UE8M0_Bvec32UE8M0_NTT
|
| 326 |
+
8.633 ms nvjet_qqtst_128x256_128x6_2x1_2cta_v_bz_Avec32UE8M0_Bvec32UE8M0_TNT
|
| 327 |
+
8.282 ms kernel_cutlass_kernel_moonepinter_rank_syncInterRankSyncKernel_object_at__tensorptri32gmemalign16o41943041_tensorptri32gmemalign16o10___0
|
| 328 |
+
7.567 ms nvjet_qqtst_256x128_128x5_2x4_2cta_h_bz_Avec32UE8M0_Bvec32UE8M0_TNT
|
| 329 |
+
REDUCE 7.415 ms kernel_cutlass_kernel_moonepgrad_reduceGradReduceKernel_object_at__tensorptrf32gmemalign16o3276803584i64i64_tensorptrf32gmemalign16o3276803584i64i64_CopyAtom_ThrID10_TVLayoutSrc11638401_T_0
|
| 330 |
+
7.025 ms nvjet_qqsss_128x256_128x6_2x2_2cta_h_bz_Avec32UE8M0_Bvec32UE8M0_NTT
|
| 331 |
+
6.843 ms cudnn_generated_fort_native_sdpa_sm100_flash_bprop_f16_knob_31_128x128x128_1x4x1_kernel0_0
|
| 332 |
+
6.189 ms unrolled_elementwise_kernel
|
| 333 |
+
5.681 ms triton_poi_fused__to_copy_cat_0
|
| 334 |
+
4.923 ms triton_poi_fused_cat_0
|
| 335 |
+
REDUCE 4.024 ms kernel_cutlass_kernel_moonepgrad_reduceGradReduceKernel_object_at__tensorptrf32gmemalign16o2293762560i64i64_tensorptrf32gmemalign16o2293762560i64i64_CopyAtom_ThrID10_TVLayoutSrc11638401_T_0
|
| 336 |
+
3.955 ms nvjet_qqtst_128x256_128x6_2x1_2cta_v_bz_Avec32UE8M0_Bvec32UE8M0_NNT
|
| 337 |
+
3.448 ms setGroupedGemmWgradArguments
|
| 338 |
+
3.072 ms kernel_cutlass_kernel_moonepprefetchPrefetchKernel_object_at__CopyAtom_ThrID10_TVLayoutSrc11638401_TVLayoutDst11638401_Valuetypebf16_tensor00o327680358411i640_CopyAtom_ThrID10_TVLayoutSrc_0
|
| 339 |
+
3.067 ms rmsnorm_bwd_general_kernel
|
| 340 |
+
2.215 ms cudnn_generated_fort_native_sdpa_sm100_flash_fprop_f16_knob_7_128x128x128_4x1x1_kernel0_0
|
| 341 |
+
2.117 ms triton_red_fused__to_copy_mul_silu_sum_1
|
| 342 |
+
1.802 ms nvjet_tst_320x192_64x4_2x2_2cta_h_bz_NNT
|
| 343 |
+
1.739 ms ncclDevKernel_AllReduce_Sum_f32_RING_LL
|
| 344 |
+
1.705 ms nvjet_tss_128x256_64x6_2x1_2cta_v_badd_NTT
|
| 345 |
+
1.580 ms triton_poi_fused__to_copy_mul_silu_0
|
| 346 |
+
1.563 ms triton_poi_fused_mul_silu_0
|
| 347 |
+
1.518 ms nvjet_tst_128x256_64x6_2x1_2cta_v_bz_TNT
|
| 348 |
+
1.470 ms kernel_cutlass_kernel_moonepprefetchPrefetchKernel_object_at__CopyAtom_ThrID10_TVLayoutSrc11638401_TVLayoutDst11638401_Valuetypebf16_tensor00o229376256011i640_CopyAtom_ThrID10_TVLayoutSrc_0
|
| 349 |
+
1.457 ms swizzle_col_scaling_kernel
|
| 350 |
+
1.397 ms reduce_kernel
|
| 351 |
+
1.312 ms rmsnorm_fwd_general_kernel
|
| 352 |
+
1.220 ms kernel_cutlass_kernel_moonepplanningPlanningKernel_object_at__tensorptri32_gmem_align16_o_641_tensorptri32_gmem_align16_o_327681_tensorptri32_gmem_align16_o_41943041_tensorptri32_gmem_ali_0
|
| 353 |
+
1.078 ms _scatter_gather_elementwise_kernel
|
| 354 |
+
1.077 ms multi_tensor_swizzle_row_scaling_kernel
|
| 355 |
+
|
| 356 |
+
=== QWEN2_57B__moonep__matched__seq4096__gbs8__capnone__async.sqlite (device 6, 2 iteration(s)) ===
|
| 357 |
+
reduce_grad kernels : 40 (20/iter)
|
| 358 |
+
reduce_grad sum : 5.719 ms/iter
|
| 359 |
+
reduce_grad busy (union) : 5.719 ms/iter
|
| 360 |
+
... overlapped with compute: 5.491 ms/iter ( 96.0 %)
|
| 361 |
+
... exposed on critical path: 0.228 ms/iter ( 4.0 %)
|
| 362 |
+
other GPU work (union) : 168.985 ms/iter
|
| 363 |
+
traced wall : 188.376 ms/iter
|
| 364 |
+
--> time hidden by overlap : 5.491 ms/iter
|
| 365 |
+
|
| 366 |
+
=== kernels on device 7, by total time ===
|
| 367 |
+
43.569 ms device_kernel
|
| 368 |
+
41.940 ms multi_tensor_apply_kernel
|
| 369 |
+
40.673 ms ncclDevKernel_ReduceScatter_Sum_f32_RING_LL
|
| 370 |
+
30.640 ms elementwise_kernel
|
| 371 |
+
28.767 ms vectorized_elementwise_kernel
|
| 372 |
+
24.843 ms cast_mxfp8_2D_kernel
|
| 373 |
+
23.460 ms ncclDevKernel_AllGather_RING_LL
|
| 374 |
+
15.104 ms kernel_cutlass_kernel_moonepdispatchDispatchKernel_object_at__tensorptrbf16gmemalign16o40963584i64i64_tensorptrbf16gmemalign16o2785283584i64i64_tensorptri32gmemalign16o327681_tensorptri32_0
|
| 375 |
+
15.014 ms nvjet_qqtst_256x128_128x5_2x4_2cta_h_bz_Avec32UE8M0_Bvec32UE8M0_NNT
|
| 376 |
+
13.108 ms kernel_cutlass_kernel_moonepcombineCombineKernel_object_at__tensorptrbf16gmemalign16o2785283584i64i64_tensorptrbf16gmemalign16o40963584i64i64_tensorptri32gmemalign16o327681_tensorptri32gm_0
|
| 377 |
+
8.765 ms nvjet_qqsss_128x256_128x6_2x1_2cta_v_bz_Avec32UE8M0_Bvec32UE8M0_NTT
|
| 378 |
+
8.455 ms nvjet_qqtst_128x256_128x6_2x1_2cta_v_bz_Avec32UE8M0_Bvec32UE8M0_TNT
|
| 379 |
+
7.578 ms nvjet_qqtst_256x128_128x5_2x4_2cta_h_bz_Avec32UE8M0_Bvec32UE8M0_TNT
|
| 380 |
+
REDUCE 7.547 ms kernel_cutlass_kernel_moonepgrad_reduceGradReduceKernel_object_at__tensorptrf32gmemalign16o3276803584i64i64_tensorptrf32gmemalign16o3276803584i64i64_CopyAtom_ThrID10_TVLayoutSrc11638401_T_0
|
| 381 |
+
6.921 ms nvjet_qqsss_128x256_128x6_2x2_2cta_h_bz_Avec32UE8M0_Bvec32UE8M0_NTT
|
| 382 |
+
6.828 ms cudnn_generated_fort_native_sdpa_sm100_flash_bprop_f16_knob_31_128x128x128_1x4x1_kernel0_0
|
| 383 |
+
6.727 ms kernel_cutlass_kernel_moonepinter_rank_syncInterRankSyncKernel_object_at__tensorptri32gmemalign16o41943041_tensorptri32gmemalign16o10___0
|
| 384 |
+
6.176 ms unrolled_elementwise_kernel
|
| 385 |
+
5.687 ms triton_poi_fused__to_copy_cat_0
|
| 386 |
+
5.324 ms ncclDevKernel_AllReduce_Sum_f32_RING_LL
|
| 387 |
+
4.933 ms triton_poi_fused_cat_0
|
| 388 |
+
4.305 ms kernel_cutlass_kernel_moonepprefetchPrefetchKernel_object_at__CopyAtom_ThrID10_TVLayoutSrc11638401_TVLayoutDst11638401_Valuetypebf16_tensor00o327680358411i640_CopyAtom_ThrID10_TVLayoutSrc_0
|
| 389 |
+
3.869 ms nvjet_qqtst_128x256_128x6_2x1_2cta_v_bz_Avec32UE8M0_Bvec32UE8M0_NNT
|
| 390 |
+
REDUCE 3.818 ms kernel_cutlass_kernel_moonepgrad_reduceGradReduceKernel_object_at__tensorptrf32gmemalign16o2293762560i64i64_tensorptrf32gmemalign16o2293762560i64i64_CopyAtom_ThrID10_TVLayoutSrc11638401_T_0
|
| 391 |
+
3.345 ms setGroupedGemmWgradArguments
|
| 392 |
+
3.058 ms rmsnorm_bwd_general_kernel
|
| 393 |
+
2.214 ms cudnn_generated_fort_native_sdpa_sm100_flash_fprop_f16_knob_7_128x128x128_4x1x1_kernel0_0
|
| 394 |
+
2.120 ms kernel_cutlass_kernel_moonepprefetchPrefetchKernel_object_at__CopyAtom_ThrID10_TVLayoutSrc11638401_TVLayoutDst11638401_Valuetypebf16_tensor00o229376256011i640_CopyAtom_ThrID10_TVLayoutSrc_0
|
| 395 |
+
2.117 ms triton_red_fused__to_copy_mul_silu_sum_1
|
| 396 |
+
1.820 ms nvjet_tst_320x192_64x4_2x2_2cta_h_bz_NNT
|
| 397 |
+
1.696 ms nvjet_tss_128x256_64x6_2x1_2cta_v_badd_NTT
|
| 398 |
+
1.573 ms triton_poi_fused__to_copy_mul_silu_0
|
| 399 |
+
1.562 ms triton_poi_fused_mul_silu_0
|
| 400 |
+
1.508 ms nvjet_tst_128x256_64x6_2x1_2cta_v_bz_TNT
|
| 401 |
+
1.467 ms swizzle_col_scaling_kernel
|
| 402 |
+
1.407 ms reduce_kernel
|
| 403 |
+
1.311 ms rmsnorm_fwd_general_kernel
|
| 404 |
+
1.209 ms kernel_cutlass_kernel_moonepplanningPlanningKernel_object_at__tensorptri32_gmem_align16_o_641_tensorptri32_gmem_align16_o_327681_tensorptri32_gmem_align16_o_41943041_tensorptri32_gmem_ali_0
|
| 405 |
+
1.104 ms multi_tensor_swizzle_row_scaling_kernel
|
| 406 |
+
1.084 ms _scatter_gather_elementwise_kernel
|
| 407 |
+
|
| 408 |
+
=== QWEN2_57B__moonep__matched__seq4096__gbs8__capnone__async.sqlite (device 7, 2 iteration(s)) ===
|
| 409 |
+
reduce_grad kernels : 40 (20/iter)
|
| 410 |
+
reduce_grad sum : 5.683 ms/iter
|
| 411 |
+
reduce_grad busy (union) : 5.683 ms/iter
|
| 412 |
+
... overlapped with compute: 5.456 ms/iter ( 96.0 %)
|
| 413 |
+
... exposed on critical path: 0.227 ms/iter ( 4.0 %)
|
| 414 |
+
other GPU work (union) : 171.022 ms/iter
|
| 415 |
+
traced wall : 188.320 ms/iter
|
| 416 |
+
--> time hidden by overlap : 5.456 ms/iter
|
tierb-fixed-moonep-full/moonep/tierbtrace/tierbtrace_gbs8/QWEN2_57B__moonep__matched__seq4096__gbs8__capnone__async/nsys/moonep_async_overlap_summary.txt
ADDED
|
@@ -0,0 +1 @@
|
|
|
|
|
|
|
| 1 |
+
devices=8 reduce=5.654 ms/iter overlap=5.429 ms/iter (96.0%) exposed=0.225 ms/iter status=PASS
|
tierb-fixed-moonep-full/moonep/tierbtrace/tierbtrace_gbs8/QWEN2_57B__moonep__matched__seq4096__gbs8__capnone__async/nsys/train.log
CHANGED
|
@@ -993,3 +993,84 @@ INFO:moonep.api:MoonEP comm buffer sizes:
|
|
| 993 |
[MoonEP selfcheck 2/4] dropped groups empty OK | covered rows [0, 1, 2, 3, 4, 5, 6, 7, 64, 65, 66, 67, 68, 69, 70, 71] sizes [9856, 3456, 3072, 4736, 512, 5632, 1536, 896, 3712, 0, 0, 0, 0, 0, 0, 0] | duplicated experts [[9, -1, -1, -1, -1, -1, -1, -1], [52, -1, -1, -1, -1, -1, -1, -1], [49, -1, -1, -1, -1, -1, -1, -1], [12, -1, -1, -1, -1, -1, -1, -1], [50, -1, -1, -1, -1, -1, -1, -1], [13, -1, -1, -1, -1, -1, -1, -1], [-1, -1, -1, -1, -1, -1, -1, -1], [28, -1, -1, -1, -1, -1, -1, -1]]
|
| 994 |
[MoonEP selfcheck 3/4] dropped groups empty OK | covered rows [0, 1, 2, 3, 4, 5, 6, 7, 64, 65, 66, 67, 68, 69, 70, 71] sizes [1792, 5120, 4608, 6144, 2560, 2816, 6400, 3712, 0, 0, 0, 0, 0, 0, 0, 0] | duplicated experts [[-1, -1, -1, -1, -1, -1, -1, -1], [52, -1, -1, -1, -1, -1, -1, -1], [36, -1, -1, -1, -1, -1, -1, -1], [36, -1, -1, -1, -1, -1, -1, -1], [54, -1, -1, -1, -1, -1, -1, -1], [50, -1, -1, -1, -1, -1, -1, -1], [6, -1, -1, -1, -1, -1, -1, -1], [34, -1, -1, -1, -1, -1, -1, -1]]
|
| 995 |
[MoonEP selfcheck 4/4] dropped groups empty OK | covered rows [0, 1, 2, 3, 4, 5, 6, 7, 64, 65, 66, 67, 68, 69, 70, 71] sizes [2048, 2176, 768, 8448, 1664, 2048, 9344, 3072, 3712, 0, 0, 0, 0, 0, 0, 0] | duplicated experts [[30, -1, -1, -1, -1, -1, -1, -1], [50, -1, -1, -1, -1, -1, -1, -1], [1, -1, -1, -1, -1, -1, -1, -1], [50, -1, -1, -1, -1, -1, -1, -1], [-1, -1, -1, -1, -1, -1, -1, -1], [56, -1, -1, -1, -1, -1, -1, -1], [37, -1, -1, -1, -1, -1, -1, -1], [9, -1, -1, -1, -1, -1, -1, -1]]
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 993 |
[MoonEP selfcheck 2/4] dropped groups empty OK | covered rows [0, 1, 2, 3, 4, 5, 6, 7, 64, 65, 66, 67, 68, 69, 70, 71] sizes [9856, 3456, 3072, 4736, 512, 5632, 1536, 896, 3712, 0, 0, 0, 0, 0, 0, 0] | duplicated experts [[9, -1, -1, -1, -1, -1, -1, -1], [52, -1, -1, -1, -1, -1, -1, -1], [49, -1, -1, -1, -1, -1, -1, -1], [12, -1, -1, -1, -1, -1, -1, -1], [50, -1, -1, -1, -1, -1, -1, -1], [13, -1, -1, -1, -1, -1, -1, -1], [-1, -1, -1, -1, -1, -1, -1, -1], [28, -1, -1, -1, -1, -1, -1, -1]]
|
| 994 |
[MoonEP selfcheck 3/4] dropped groups empty OK | covered rows [0, 1, 2, 3, 4, 5, 6, 7, 64, 65, 66, 67, 68, 69, 70, 71] sizes [1792, 5120, 4608, 6144, 2560, 2816, 6400, 3712, 0, 0, 0, 0, 0, 0, 0, 0] | duplicated experts [[-1, -1, -1, -1, -1, -1, -1, -1], [52, -1, -1, -1, -1, -1, -1, -1], [36, -1, -1, -1, -1, -1, -1, -1], [36, -1, -1, -1, -1, -1, -1, -1], [54, -1, -1, -1, -1, -1, -1, -1], [50, -1, -1, -1, -1, -1, -1, -1], [6, -1, -1, -1, -1, -1, -1, -1], [34, -1, -1, -1, -1, -1, -1, -1]]
|
| 995 |
[MoonEP selfcheck 4/4] dropped groups empty OK | covered rows [0, 1, 2, 3, 4, 5, 6, 7, 64, 65, 66, 67, 68, 69, 70, 71] sizes [2048, 2176, 768, 8448, 1664, 2048, 9344, 3072, 3712, 0, 0, 0, 0, 0, 0, 0] | duplicated experts [[30, -1, -1, -1, -1, -1, -1, -1], [50, -1, -1, -1, -1, -1, -1, -1], [1, -1, -1, -1, -1, -1, -1, -1], [50, -1, -1, -1, -1, -1, -1, -1], [-1, -1, -1, -1, -1, -1, -1, -1], [56, -1, -1, -1, -1, -1, -1, -1], [37, -1, -1, -1, -1, -1, -1, -1], [9, -1, -1, -1, -1, -1, -1, -1]]
|
| 996 |
+
[2026-08-30 02:48:33] iteration 1/ 10 | consumed samples: 8 | elapsed time per iteration (ms): 20490.0 | throughput per GPU (TFLOP/s/GPU): 6.0 | learning rate: 2.000000E-05 | global batch size: 8 | lm loss: 1.097928E+01 | loss scale: 1.0 | grad norm: 29.871 | number of skipped iterations: 0 | number of nan iterations: 0 |
|
| 997 |
+
Number of parameters in transformer block in billions: 20.11
|
| 998 |
+
Number of parameters in embedding layers in billions: 0.36
|
| 999 |
+
Total number of parameters in billions: 20.47
|
| 1000 |
+
Number of parameters in most loaded shard in billions: 20.4724
|
| 1001 |
+
Theoretical memory footprints: weight and optimizer=146430.14 MB
|
| 1002 |
+
[Rank 0] (after 1 iterations) memory (MB) | allocated: 60534.46875 | max allocated: 60534.46923828125 | reserved: 62900.0 | max reserved: 62900.0
|
| 1003 |
+
[2026-08-30 02:48:33] iteration 2/ 10 | consumed samples: 16 | elapsed time per iteration (ms): 247.2 | throughput per GPU (TFLOP/s/GPU): 499.6 | learning rate: 4.000000E-05 | global batch size: 8 | lm loss: 1.097878E+01 | loss scale: 1.0 | grad norm: 30.630 | number of skipped iterations: 0 | number of nan iterations: 0 |
|
| 1004 |
+
[2026-08-30 02:48:33] iteration 3/ 10 | consumed samples: 24 | elapsed time per iteration (ms): 209.9 | throughput per GPU (TFLOP/s/GPU): 588.3 | learning rate: 6.000000E-05 | global batch size: 8 | lm loss: 9.537870E+00 | loss scale: 1.0 | grad norm: 32.328 | number of skipped iterations: 0 | number of nan iterations: 0 |
|
| 1005 |
+
Capture CUDA graph for training!!!
|
| 1006 |
+
CUDA graph capture done!!!
|
| 1007 |
+
[2026-08-30 02:48:34] iteration 4/ 10 | consumed samples: 32 | elapsed time per iteration (ms): 783.5 | throughput per GPU (TFLOP/s/GPU): 157.6 | learning rate: 8.000000E-05 | global batch size: 8 | lm loss: 1.108238E+01 | loss scale: 1.0 | grad norm: 106.897 | number of skipped iterations: 0 | number of nan iterations: 0 |
|
| 1008 |
+
[2026-08-30 02:48:34] iteration 5/ 10 | consumed samples: 40 | elapsed time per iteration (ms): 180.8 | throughput per GPU (TFLOP/s/GPU): 682.8 | learning rate: 1.000000E-04 | global batch size: 8 | lm loss: 9.997957E+00 | loss scale: 1.0 | grad norm: 26.196 | number of skipped iterations: 0 | number of nan iterations: 0 |
|
| 1009 |
+
[2026-08-30 02:48:34] iteration 6/ 10 | consumed samples: 48 | elapsed time per iteration (ms): 180.7 | throughput per GPU (TFLOP/s/GPU): 683.3 | learning rate: 7.750000E-05 | global batch size: 8 | lm loss: 1.049965E+01 | loss scale: 1.0 | grad norm: 8.726 | number of skipped iterations: 0 | number of nan iterations: 0 |
|
| 1010 |
+
[2026-08-30 02:48:37] iteration 7/ 10 | consumed samples: 56 | elapsed time per iteration (ms): 2917.7 | throughput per GPU (TFLOP/s/GPU): 42.3 | learning rate: 3.250000E-05 | global batch size: 8 | lm loss: 9.442690E+00 | loss scale: 1.0 | grad norm: 5.413 | number of skipped iterations: 0 | number of nan iterations: 0 |
|
| 1011 |
+
[2026-08-30 02:48:37] iteration 8/ 10 | consumed samples: 64 | elapsed time per iteration (ms): 186.2 | throughput per GPU (TFLOP/s/GPU): 663.0 | learning rate: 1.000000E-05 | global batch size: 8 | lm loss: 8.964504E+00 | loss scale: 1.0 | grad norm: 4.556 | number of skipped iterations: 0 | number of nan iterations: 0 |
|
| 1012 |
+
Capture range started in the application.
|
| 1013 |
+
Capture range ended in the application.
|
| 1014 |
+
Generating '/tmp/nsys-report-5085.qdstrm'
|
| 1015 |
+
|
| 1016 |
+
|
| 1017 |
+
[2026-08-30 02:49:18] iteration 9/ 10 | consumed samples: 72 | elapsed time per iteration (ms): 40624.4 | throughput per GPU (TFLOP/s/GPU): 3.0 | learning rate: 1.000000E-05 | global batch size: 8 | lm loss: 8.748680E+00 | loss scale: 1.0 | grad norm: 4.300 | number of skipped iterations: 0 | number of nan iterations: 0 |
|
| 1018 |
+
[2026-08-30 02:49:18] iteration 10/ 10 | consumed samples: 80 | elapsed time per iteration (ms): 185.0 | throughput per GPU (TFLOP/s/GPU): 667.5 | learning rate: 1.000000E-05 | global batch size: 8 | lm loss: 8.688493E+00 | loss scale: 1.0 | grad norm: 4.428 | number of skipped iterations: 0 | number of nan iterations: 0 |
|
| 1019 |
+
[after training is done] datetime: 2026-08-30 02:49:18
|
| 1020 |
+
[rank0]:[W830 02:49:20.407650089 ProcessGroupNCCL.cpp:1505] Warning: WARNING: destroy_process_group() was not called before program exit, which can leak resources. For more info, please see https://pytorch.org/docs/stable/distributed.html#shutdown (function operator())
|
| 1021 |
+
[rank1]:[W830 02:49:21.483277895 ProcessGroupNCCL.cpp:1505] Warning: WARNING: destroy_process_group() was not called before program exit, which can leak resources. For more info, please see https://pytorch.org/docs/stable/distributed.html#shutdown (function operator())
|
| 1022 |
+
[rank5]:[W830 02:49:21.533068668 ProcessGroupNCCL.cpp:1505] Warning: WARNING: destroy_process_group() was not called before program exit, which can leak resources. For more info, please see https://pytorch.org/docs/stable/distributed.html#shutdown (function operator())
|
| 1023 |
+
[rank3]:[W830 02:49:21.533582983 ProcessGroupNCCL.cpp:1505] Warning: WARNING: destroy_process_group() was not called before program exit, which can leak resources. For more info, please see https://pytorch.org/docs/stable/distributed.html#shutdown (function operator())
|
| 1024 |
+
[rank2]:[W830 02:49:21.533642158 ProcessGroupNCCL.cpp:1505] Warning: WARNING: destroy_process_group() was not called before program exit, which can leak resources. For more info, please see https://pytorch.org/docs/stable/distributed.html#shutdown (function operator())
|
| 1025 |
+
[rank6]:[W830 02:49:21.536994791 ProcessGroupNCCL.cpp:1505] Warning: WARNING: destroy_process_group() was not called before program exit, which can leak resources. For more info, please see https://pytorch.org/docs/stable/distributed.html#shutdown (function operator())
|
| 1026 |
+
[rank4]:[W830 02:49:21.829376424 ProcessGroupNCCL.cpp:1505] Warning: WARNING: destroy_process_group() was not called before program exit, which can leak resources. For more info, please see https://pytorch.org/docs/stable/distributed.html#shutdown (function operator())
|
| 1027 |
+
[rank7]:[W830 02:49:22.342994989 ProcessGroupNCCL.cpp:1505] Warning: WARNING: destroy_process_group() was not called before program exit, which can leak resources. For more info, please see https://pytorch.org/docs/stable/distributed.html#shutdown (function operator())
|
| 1028 |
+
[rank1]:[W830 02:49:24.454973039 observer.cpp:116] Warning: Leaked callback handle: 1 (function operator())
|
| 1029 |
+
pure virtual method called
|
| 1030 |
+
terminate called without an active exception
|
| 1031 |
+
[rank5]:[W830 02:49:24.487316028 observer.cpp:116] Warning: Leaked callback handle: 1 (function operator())
|
| 1032 |
+
pure virtual method called
|
| 1033 |
+
terminate called without an active exception
|
| 1034 |
+
W0830 02:49:24.782000 429583 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 429792 closing signal SIGTERM
|
| 1035 |
+
W0830 02:49:24.783000 429583 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 429794 closing signal SIGTERM
|
| 1036 |
+
W0830 02:49:24.784000 429583 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 429795 closing signal SIGTERM
|
| 1037 |
+
W0830 02:49:24.785000 429583 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 429796 closing signal SIGTERM
|
| 1038 |
+
W0830 02:49:24.785000 429583 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 429797 closing signal SIGTERM
|
| 1039 |
+
W0830 02:49:24.786000 429583 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 429798 closing signal SIGTERM
|
| 1040 |
+
W0830 02:49:24.786000 429583 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 429799 closing signal SIGTERM
|
| 1041 |
+
E0830 02:49:26.755000 429583 torch/distributed/elastic/multiprocessing/api.py:874] failed (exitcode: -6) local_rank: 1 (pid: 429793) of binary: /usr/bin/python
|
| 1042 |
+
Traceback (most recent call last):
|
| 1043 |
+
File "/usr/local/bin/torchrun", line 8, in <module>
|
| 1044 |
+
sys.exit(main())
|
| 1045 |
+
^^^^^^
|
| 1046 |
+
File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/multiprocessing/errors/__init__.py", line 355, in wrapper
|
| 1047 |
+
return f(*args, **kwargs)
|
| 1048 |
+
^^^^^^^^^^^^^^^^^^
|
| 1049 |
+
File "/usr/local/lib/python3.12/dist-packages/torch/distributed/run.py", line 892, in main
|
| 1050 |
+
run(args)
|
| 1051 |
+
File "/usr/local/lib/python3.12/dist-packages/torch/distributed/run.py", line 883, in run
|
| 1052 |
+
elastic_launch(
|
| 1053 |
+
File "/usr/local/lib/python3.12/dist-packages/torch/distributed/launcher/api.py", line 139, in __call__
|
| 1054 |
+
return launch_agent(self._config, self._entrypoint, list(args))
|
| 1055 |
+
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
|
| 1056 |
+
File "/usr/local/lib/python3.12/dist-packages/torch/distributed/launcher/api.py", line 270, in launch_agent
|
| 1057 |
+
raise ChildFailedError(
|
| 1058 |
+
torch.distributed.elastic.multiprocessing.errors.ChildFailedError:
|
| 1059 |
+
============================================================
|
| 1060 |
+
pretrain_gpt.py FAILED
|
| 1061 |
+
------------------------------------------------------------
|
| 1062 |
+
Failures:
|
| 1063 |
+
<NO_OTHER_FAILURES>
|
| 1064 |
+
------------------------------------------------------------
|
| 1065 |
+
Root Cause (first observed failure):
|
| 1066 |
+
[0]:
|
| 1067 |
+
time : 2026-08-30_02:49:24
|
| 1068 |
+
host : sp-aiinfra-interactive-aghosh-7jtjl-8xb200-moe-cg-0
|
| 1069 |
+
rank : 1 (local_rank: 1)
|
| 1070 |
+
exitcode : -6 (pid: 429793)
|
| 1071 |
+
error_file: <N/A>
|
| 1072 |
+
traceback : Signal 6 (SIGABRT) received by PID 429793
|
| 1073 |
+
============================================================
|
| 1074 |
+
Generated:
|
| 1075 |
+
/workspace/tierb-fixed-moonep-full/moonep/tierbtrace/tierbtrace_gbs8/QWEN2_57B__moonep__matched__seq4096__gbs8__capnone__async/nsys/QWEN2_57B__moonep__matched__seq4096__gbs8__capnone__async.nsys-rep
|
| 1076 |
+
There are no active sessions.
|
tierb-fixed-moonep-full/moonep/tierbtrace/tierbtrace_gbs8/QWEN2_57B__moonep__matched__seq4096__gbs8__capnone__async/nsys/verify_trace_ranks.log
ADDED
|
@@ -0,0 +1,6 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
report : /workspace/tierb-fixed-moonep-full/moonep/tierbtrace/tierbtrace_gbs8/QWEN2_57B__moonep__matched__seq4096__gbs8__capnone__async/nsys/QWEN2_57B__moonep__matched__seq4096__gbs8__capnone__async.nsys-rep
|
| 2 |
+
kernels : 58910
|
| 3 |
+
devices : 8 of 8 -> [0, 1, 2, 3, 4, 5, 6, 7]
|
| 4 |
+
processes: 1
|
| 5 |
+
markers : 5440 moonep_trace__ kernels on devices [0, 1, 2, 3, 4, 5, 6, 7]
|
| 6 |
+
PASS: all ranks present and backend markers emitted
|
tierb-fixed-moonep-full/moonep/tierbtrace/tierbtrace_gbs8/trace_results.tsv
CHANGED
|
@@ -2,3 +2,4 @@ model backend capacity status report async_overlap_json
|
|
| 2 |
QWEN3_235B moonep none OK /workspace/tierb-fixed-moonep-full/moonep/tierbtrace/tierbtrace_gbs8/QWEN3_235B__moonep__matched__seq8192__gbs8__capnone__async/nsys/QWEN3_235B__moonep__matched__seq8192__gbs8__capnone__async.nsys-rep /workspace/tierb-fixed-moonep-full/moonep/tierbtrace/tierbtrace_gbs8/QWEN3_235B__moonep__matched__seq8192__gbs8__capnone__async/nsys/moonep_async_overlap.json
|
| 3 |
QWEN3_30B moonep none OK /workspace/tierb-fixed-moonep-full/moonep/tierbtrace/tierbtrace_gbs8/QWEN3_30B__moonep__matched__seq4096__gbs8__capnone__async/nsys/QWEN3_30B__moonep__matched__seq4096__gbs8__capnone__async.nsys-rep /workspace/tierb-fixed-moonep-full/moonep/tierbtrace/tierbtrace_gbs8/QWEN3_30B__moonep__matched__seq4096__gbs8__capnone__async/nsys/moonep_async_overlap.json
|
| 4 |
DSV2_LITE_L24 moonep none OK /workspace/tierb-fixed-moonep-full/moonep/tierbtrace/tierbtrace_gbs8/DSV2_LITE_L24__moonep__matched__seq8192__gbs8__capnone__async/nsys/DSV2_LITE_L24__moonep__matched__seq8192__gbs8__capnone__async.nsys-rep /workspace/tierb-fixed-moonep-full/moonep/tierbtrace/tierbtrace_gbs8/DSV2_LITE_L24__moonep__matched__seq8192__gbs8__capnone__async/nsys/moonep_async_overlap.json
|
|
|
|
|
|
| 2 |
QWEN3_235B moonep none OK /workspace/tierb-fixed-moonep-full/moonep/tierbtrace/tierbtrace_gbs8/QWEN3_235B__moonep__matched__seq8192__gbs8__capnone__async/nsys/QWEN3_235B__moonep__matched__seq8192__gbs8__capnone__async.nsys-rep /workspace/tierb-fixed-moonep-full/moonep/tierbtrace/tierbtrace_gbs8/QWEN3_235B__moonep__matched__seq8192__gbs8__capnone__async/nsys/moonep_async_overlap.json
|
| 3 |
QWEN3_30B moonep none OK /workspace/tierb-fixed-moonep-full/moonep/tierbtrace/tierbtrace_gbs8/QWEN3_30B__moonep__matched__seq4096__gbs8__capnone__async/nsys/QWEN3_30B__moonep__matched__seq4096__gbs8__capnone__async.nsys-rep /workspace/tierb-fixed-moonep-full/moonep/tierbtrace/tierbtrace_gbs8/QWEN3_30B__moonep__matched__seq4096__gbs8__capnone__async/nsys/moonep_async_overlap.json
|
| 4 |
DSV2_LITE_L24 moonep none OK /workspace/tierb-fixed-moonep-full/moonep/tierbtrace/tierbtrace_gbs8/DSV2_LITE_L24__moonep__matched__seq8192__gbs8__capnone__async/nsys/DSV2_LITE_L24__moonep__matched__seq8192__gbs8__capnone__async.nsys-rep /workspace/tierb-fixed-moonep-full/moonep/tierbtrace/tierbtrace_gbs8/DSV2_LITE_L24__moonep__matched__seq8192__gbs8__capnone__async/nsys/moonep_async_overlap.json
|
| 5 |
+
QWEN2_57B moonep none OK /workspace/tierb-fixed-moonep-full/moonep/tierbtrace/tierbtrace_gbs8/QWEN2_57B__moonep__matched__seq4096__gbs8__capnone__async/nsys/QWEN2_57B__moonep__matched__seq4096__gbs8__capnone__async.nsys-rep /workspace/tierb-fixed-moonep-full/moonep/tierbtrace/tierbtrace_gbs8/QWEN2_57B__moonep__matched__seq4096__gbs8__capnone__async/nsys/moonep_async_overlap.json
|