diff --git a/granularity_next_b200/tiera_topk_trace/hf_watcher.log b/granularity_next_b200/tiera_topk_trace/hf_watcher.log index 37c325d66e655075b2fbfbb4259caf466d0d372e..8b6ccce65bbe04a8d807c8880dc2dc1f3e108094 100644 --- a/granularity_next_b200/tiera_topk_trace/hf_watcher.log +++ b/granularity_next_b200/tiera_topk_trace/hf_watcher.log @@ -36,3 +36,6 @@ https://huggingface.co/MoE-CUDA-Graph-Analysis/tierb-b200-gbs8/tree/main Already logged in to Hugging Face as: AbhishekGhosh1998 Watching /workspace/granularity_next_b200/tiera_topk_trace -> HF every 600s (kill to stop) [2026-09-21T05:22:22+00:00] backup pass + uploaded: /workspace/granularity_next_b200/tiera_topk_trace -> MoE-CUDA-Graph-Analysis/tierb-b200-gbs8/granularity_next_b200/tiera_topk_trace +https://huggingface.co/MoE-CUDA-Graph-Analysis/tierb-b200-gbs8/tree/main +[2026-09-21T05:32:24+00:00] backup pass diff --git a/granularity_next_b200/tiera_topk_trace/study.log b/granularity_next_b200/tiera_topk_trace/study.log index bf42dff397f61130af0413242b41a2255bf99461..0e8fc0f468699754208c60c220256711d81ccbf4 100644 --- a/granularity_next_b200/tiera_topk_trace/study.log +++ b/granularity_next_b200/tiera_topk_trace/study.log @@ -143,3 +143,25 @@ driver_repo: /workspace/blackwell_run_scripts @ b8bd100 === STAGE 9 -- Tier B nsys traces (final, trace-only) === STEP ultraep__tierbtrace__gbs16 +============================================================== + CAMPAIGN: ultraep backend=ultraep phase=tierbtrace + Megatron: /workspace/Megatron-UltraEP @ ultraep-bracing-regions-through-kernels + GBS=16 seq=4096 repeats=3 nsys=1 + results -> /workspace/granularity_next_b200/tiera_topk_trace/ultraep/tierbtrace +============================================================== +[05:22:36] building ultraep marker extension (log: /workspace/granularity_next_b200/tiera_topk_trace/ultraep/tierbtrace/tierbtrace_gbs16/trace_extension_build.log) +[05:26:43] marker extension ready: _ultraep_trace_C.cpython-312-x86_64-linux-gnu.so +[05:26:43] TRACE-ONLY G2_K2__ultraep__matched__seq4096__gbs16__cap1p15625 (ranks: 0 1 2 3 4 5 6 7) + [guard] iter 6/10 (70s) + [guard] iter 8/10 (80s) + [guard] iter 10/10 (120s) + [guard] nsys stop (flushing report) +[05:29:15] trace verified -> G2_K2__ultraep__matched__seq4096__gbs16__cap1p15625.nsys-rep (11M, all 8 GPUs + ultraep markers) +[05:29:15] TRACE-ONLY G3_K2__ultraep__matched__seq4096__gbs16__cap1p1875 (ranks: 0 1 2 3 4 5 6 7) + [guard] iter 6/10 (70s) + [guard] iter 8/10 (80s) + [guard] no progress for 60s at iter 8/10 (past profile window); treating as done + [guard] nsys stop (flushing report) + [guard] tearing down process group (outcome=STALLED) +[05:32:23] trace verified -> G3_K2__ultraep__matched__seq4096__gbs16__cap1p1875.nsys-rep (12M, all 8 GPUs + ultraep markers) +[05:32:23] TRACE-ONLY G3_K8__ultraep__matched__seq4096__gbs16__cap1p09375 (ranks: 0 1 2 3 4 5 6 7) diff --git a/granularity_next_b200/tiera_topk_trace/ultraep/tierbtrace/manifest.txt b/granularity_next_b200/tiera_topk_trace/ultraep/tierbtrace/manifest.txt index c84c3f97247758264aa57dab340a9c37e8b48280..f4afd983e21f4fe27babbfbde6f8ae5bebad546a 100644 --- a/granularity_next_b200/tiera_topk_trace/ultraep/tierbtrace/manifest.txt +++ b/granularity_next_b200/tiera_topk_trace/ultraep/tierbtrace/manifest.txt @@ -1,9 +1,9 @@ -timestamp: 2026-09-17T03:22:41+00:00 -host: hp-aiinfra-interactive-aghosh-8-xzthl-8xb200-moe-cg-0 +timestamp: 2026-09-21T05:22:36+00:00 +host: sp-aiinfra-interactive-aghosh-8-mb99k-8xb200-moe-cg-0 gpu: NVIDIA B200 x8 driver: 580.95.05 study_scripts_branch: granularity-ultraep -study_scripts_commit: 78aa53a +study_scripts_commit: 05dd2cb campaign: ultraep backend: ultraep phase: tierbtrace diff --git a/granularity_next_b200/tiera_topk_trace/ultraep/tierbtrace/pip_freeze.txt b/granularity_next_b200/tiera_topk_trace/ultraep/tierbtrace/pip_freeze.txt index 6a22ffad3b53f9b2295632309fa94b175e74ff3a..1c576bdea6c1bb3eb8730d1e2cbd7e12123e4049 100644 --- a/granularity_next_b200/tiera_topk_trace/ultraep/tierbtrace/pip_freeze.txt +++ b/granularity_next_b200/tiera_topk_trace/ultraep/tierbtrace/pip_freeze.txt @@ -75,7 +75,7 @@ h11==0.16.0 hf-xet==1.6.0 httpcore==1.0.9 httpx==0.28.1 -huggingface_hub==1.31.0 +huggingface_hub==1.32.0 hypothesis==6.130.8 idna==3.10 importlib_metadata==8.7.0 diff --git a/granularity_next_b200/tiera_topk_trace/ultraep/tierbtrace/provenance.txt b/granularity_next_b200/tiera_topk_trace/ultraep/tierbtrace/provenance.txt index 8ac54462def86812bda163c8887b5782b50e8783..db84d22669078ebd256dd803480d84d494d91faf 100644 --- a/granularity_next_b200/tiera_topk_trace/ultraep/tierbtrace/provenance.txt +++ b/granularity_next_b200/tiera_topk_trace/ultraep/tierbtrace/provenance.txt @@ -27,3 +27,64 @@ command : run_ladder.sh --phase tierbtrace --gbs 16 --out /workspace/granu 7, 120 MHz, 1965 MHz, 1000.00 W, 28 NVLink topology: + GPU0 GPU1 GPU2 GPU3 GPU4 GPU5 GPU6 GPU7 NIC0 NIC1 NIC2 NIC3 NIC4 NIC5 NIC6 NIC7 NIC8 CPU Affinity NUMA Affinity GPU NUMA ID + GPU0 X NV18 NV18 NV18 NV18 NV18 NV18 NV18 PIX NODE SYS SYS SYS SYS SYS SYS NODE 0-27 0 N/A + GPU1 NV18 X NV18 NV18 NV18 NV18 NV18 NV18 NODE PIX SYS SYS SYS SYS SYS SYS NODE 0-27 0 N/A + GPU2 NV18 NV18 X NV18 NV18 NV18 NV18 NV18 SYS SYS PIX NODE SYS SYS SYS SYS SYS 28-55 1 N/A + GPU3 NV18 NV18 NV18 X NV18 NV18 NV18 NV18 SYS SYS NODE PIX SYS SYS SYS SYS SYS 28-55 1 N/A + GPU4 NV18 NV18 NV18 NV18 X NV18 NV18 NV18 SYS SYS SYS SYS PIX NODE SYS SYS SYS 56-83 2 N/A + GPU5 NV18 NV18 NV18 NV18 NV18 X NV18 NV18 SYS SYS SYS SYS NODE PIX SYS SYS SYS 56-83 2 N/A + GPU6 NV18 NV18 NV18 NV18 NV18 NV18 X NV18 SYS SYS SYS SYS SYS SYS PIX NODE SYS 84-111 3 N/A + GPU7 NV18 NV18 NV18 NV18 NV18 NV18 NV18 X SYS SYS SYS SYS SYS SYS NODE PIX SYS 84-111 3 N/A + NIC0 PIX NODE SYS SYS SYS SYS SYS SYS X NODE SYS SYS SYS SYS SYS SYS NODE + NIC1 NODE PIX SYS SYS SYS SYS SYS SYS NODE X SYS SYS SYS SYS SYS SYS NODE + NIC2 SYS SYS PIX NODE SYS SYS SYS SYS SYS SYS X NODE SYS SYS SYS SYS SYS + NIC3 SYS SYS NODE PIX SYS SYS SYS SYS SYS SYS NODE X SYS SYS SYS SYS SYS + NIC4 SYS SYS SYS SYS PIX NODE SYS SYS SYS SYS SYS SYS X NODE SYS SYS SYS + unavailable + +---- toolchain ----------------------------------------------- + nvcc : release 12.9, V12.9.86 + gcc : gcc (Ubuntu 13.3.0-6ubuntu2~24.04) 13.3.0 + python : 3.12.3 + torch : 2.8.0a0+5228986c39.nv25.06 cuda=12.9 nccl=2.27.3 + torch arch : ['sm_75', 'sm_80', 'sm_86', 'sm_90', 'sm_100', 'sm_120', 'compute_120'] + transformer_engine: 2.10.0.dev0+10f6128d [/usr/local/lib/python3.12/dist-packages/transformer_engine/__init__.py] + deep_ep : 1.2.1+34152ae [/workspace/DeepEP/deep_ep/__init__.py] + ultra_ep : 1.0.0 [/workspace/UltraEP/ultra_ep/__init__.py] + moonep : 0.0.1 [/workspace/MoonEP/moonep/__init__.py] + megatron : 0.14.0rc5 [None] + flash_attn : 2.7.4.post1 [/usr/local/lib/python3.12/dist-packages/flash_attn/__init__.py] + numpy : 1.26.4 [/usr/local/lib/python3.12/dist-packages/numpy/__init__.py] + huggingface_hub: 1.32.0 [/usr/local/lib/python3.12/dist-packages/huggingface_hub/__init__.py] + +---- repositories -------------------------------------------- + Megatron-MoE-Echo bracing-regions-through-kernels 9a23679390d7 + Megatron-UltraEP ultraep-bracing-regions-through-kernels 03d57ef7d476 + UltraEP main 94cab099b44f + DeepEP HEAD 34152ae28f80 (+3 untracked: build output) + MoonEP megatron-ddp-symmetric-params d8bc6667fad7 + blackwell_run_scripts granularity-ultraep 05dd2cbfb1eb + +---- environment --------------------------------------------- + CUDA_ARCH_LIST=7.5 8.0 8.6 9.0 10.0 12.0 + CUDA_DRIVER_VERSION=575.57.08 + CUDA_HOME=/usr/local/cuda + CUDA_MODULE_LOADING=LAZY + CUDA_VERSION=12.9.1.010 + HF_REMOTE_ROOT=granularity_next_b200/tiera_topk_trace + NCCL_VERSION=2.27.3 + NVSHMEM_VERSION=3.2.5 + PYTORCH_BUILD_NUMBER=0 + PYTORCH_BUILD_VERSION=2.8.0a0+5228986 + PYTORCH_HOME=/opt/pytorch/pytorch + PYTORCH_VERSION=2.8.0a0+5228986 + TORCH_ALLOW_TF32_CUBLAS_OVERRIDE=1 + TORCH_CUDA_ARCH_LIST=7.5 8.0 8.6 9.0 10.0 12.0+PTX + TORCH_NCCL_USE_COMM_NONBLOCKING=0 + +---- shared library resolution ------------------------------- + libnvshmem_host.so: not in ldconfig + libnccl.so: /lib/x86_64-linux-gnu/libnccl.so.2 + lrwxrwxrwx 1 root root 20 Sep 21 04:57 /usr/local/lib/python3.12/dist-packages/nvidia/nvshmem/lib/libnvshmem_host.so -> libnvshmem_host.so.3 + -rw-r--r-- 1 root root 179045744 Sep 21 04:54 /usr/local/lib/python3.12/dist-packages/nvidia/nvshmem/lib/libnvshmem_host.so.3 diff --git a/granularity_next_b200/tiera_topk_trace/ultraep/tierbtrace/tierbtrace_gbs16/G2_K2__ultraep__matched__seq4096__gbs16__cap1p15625/nsys/G2_K2__ultraep__matched__seq4096__gbs16__cap1p15625.nsys-rep b/granularity_next_b200/tiera_topk_trace/ultraep/tierbtrace/tierbtrace_gbs16/G2_K2__ultraep__matched__seq4096__gbs16__cap1p15625/nsys/G2_K2__ultraep__matched__seq4096__gbs16__cap1p15625.nsys-rep index e03d5d6bd01bc9e606541ceb9b7ed989f0daf53f..a1982202c9a488fd5c188e9f195ee64e0e05f3e1 100644 --- a/granularity_next_b200/tiera_topk_trace/ultraep/tierbtrace/tierbtrace_gbs16/G2_K2__ultraep__matched__seq4096__gbs16__cap1p15625/nsys/G2_K2__ultraep__matched__seq4096__gbs16__cap1p15625.nsys-rep +++ b/granularity_next_b200/tiera_topk_trace/ultraep/tierbtrace/tierbtrace_gbs16/G2_K2__ultraep__matched__seq4096__gbs16__cap1p15625/nsys/G2_K2__ultraep__matched__seq4096__gbs16__cap1p15625.nsys-rep @@ -1,3 +1,3 @@ version https://git-lfs.github.com/spec/v1 -oid sha256:16d8b184e95b4cc82e363b6b774e38586cdc69ef72bb35e09082e92a2190d7f6 -size 10833175 +oid sha256:95b9bff1008473c2a14f43f44b047326c6ad5bb7512c6ef71354c84edee07304 +size 10756575 diff --git a/granularity_next_b200/tiera_topk_trace/ultraep/tierbtrace/tierbtrace_gbs16/G2_K2__ultraep__matched__seq4096__gbs16__cap1p15625/nsys/train.log b/granularity_next_b200/tiera_topk_trace/ultraep/tierbtrace/tierbtrace_gbs16/G2_K2__ultraep__matched__seq4096__gbs16__cap1p15625/nsys/train.log index 0fa822119d28849c35d1f9d7f338653beec69b57..477033642cba158efd9b28cc0051785f68b21039 100644 --- a/granularity_next_b200/tiera_topk_trace/ultraep/tierbtrace/tierbtrace_gbs16/G2_K2__ultraep__matched__seq4096__gbs16__cap1p15625/nsys/train.log +++ b/granularity_next_b200/tiera_topk_trace/ultraep/tierbtrace/tierbtrace_gbs16/G2_K2__ultraep__matched__seq4096__gbs16__cap1p15625/nsys/train.log @@ -6,12 +6,10 @@ memory_dump=none snapshots=/workspace/granularity_next_b200/tiera_topk_trace/ultraep/tierbtrace/tierbtrace_gbs16/G2_K2__ultraep__matched__seq4096__gbs16__cap1p15625/nsys/memory_snapshots/snapshot.pickle repo=/workspace/Megatron-UltraEP @ 03d57ef7 ============================================================== -W0917 03:26:59.698000 948326 torch/distributed/run.py:766] -W0917 03:26:59.698000 948326 torch/distributed/run.py:766] ***************************************** -W0917 03:26:59.698000 948326 torch/distributed/run.py:766] Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed. -W0917 03:26:59.698000 948326 torch/distributed/run.py:766] ***************************************** -/workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled - warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled") +W0921 05:26:51.825000 24330 torch/distributed/run.py:766] +W0921 05:26:51.825000 24330 torch/distributed/run.py:766] ***************************************** +W0921 05:26:51.825000 24330 torch/distributed/run.py:766] Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed. +W0921 05:26:51.825000 24330 torch/distributed/run.py:766] ***************************************** using world size: 8, data-parallel size: 8, context-parallel size: 1, hierarchical context-parallel sizes: None, tensor-model-parallel size: 1, pipeline-model-parallel size: 1 WARNING: overriding default arguments for tokenizer_type:GPT2BPETokenizer with tokenizer_type:GPT2BPETokenizer Number of virtual stages per pipeline stage: None @@ -572,58 +570,60 @@ WARNING:megatron.core.rerun_state_machine:RerunStateMachine initialized in mode warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled") /workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled") +WARNING: one_logger package is required to enable e2e metrics tracking. please go to https://confluence.nvidia.com/display/MLWFO/Package+Repositories for details to install it /workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled") /workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled") -WARNING: one_logger package is required to enable e2e metrics tracking. please go to https://confluence.nvidia.com/display/MLWFO/Package+Repositories for details to install it /workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled") -[Gloo] Rank 1 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 -[Gloo] Rank 2 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 +/workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled + warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled") +[Gloo] Rank [Gloo] Rank 2 is connected to 70 peer ranks. is connected to Expected number of connected peer ranks is : 77 peer ranks. Expected number of connected peer ranks is : 7 + [Gloo] Rank 3 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 +[Gloo] Rank 1 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 [Gloo] Rank 4 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 -[Gloo] Rank 7[Gloo] Rank is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 -6 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 -[Gloo] Rank 0 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 [Gloo] Rank 5 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 +[Gloo] Rank 6 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 +[Gloo] Rank 7 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 [Gloo] Rank 0 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 [Gloo] Rank 1 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 -[Gloo] Rank [Gloo] Rank 45 is connected to is connected to 77 peer ranks. peer ranks. Expected number of connected peer ranks is : 7Expected number of connected peer ranks is : 7 - [Gloo] Rank 2 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 [Gloo] Rank 3 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 -[Gloo] Rank [Gloo] Rank 76 is connected to is connected to 77 peer ranks. peer ranks. Expected number of connected peer ranks is : Expected number of connected peer ranks is : 77 - -[Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0 -[Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0 -[Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0 -[Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0 -[Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0 +[Gloo] Rank 4 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 +[Gloo] Rank 5 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 +[Gloo] Rank 7 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7[Gloo] Rank +6 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 [Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0 > initialized tensor model parallel with size 1 > initialized pipeline model parallel with size 1 > setting random seeds to 1234 ... [Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0 [Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0 -[rank3]:[W917 03:27:12.759074705 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 3] using GPU 3 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device. -[rank1]:[W917 03:27:12.759291933 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 1] using GPU 1 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device. -[rank5]:[W917 03:27:12.759731800 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 5] using GPU 5 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device. -[rank6]:[W917 03:27:12.760213509 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 6] using GPU 6 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device. -[rank2]:[W917 03:27:12.763662209 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 2] using GPU 2 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device. -[rank4]:[W917 03:27:12.763770125 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 4] using GPU 4 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device. -[rank7]:[W917 03:27:12.763810340 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 7] using GPU 7 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device. +[Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0[Gloo] Rank +0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0 +[Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0 +[Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0 +[Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0 +[rank3]:[W921 05:27:04.261394981 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 3] using GPU 3 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device. +[rank6]:[W921 05:27:04.262518453 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 6] using GPU 6 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device. +[rank2]:[W921 05:27:04.262639934 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 2] using GPU 2 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device. +[rank5]:[W921 05:27:04.262843243 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 5] using GPU 5 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device. +[rank1]:[W921 05:27:04.262845359 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 1] using GPU 1 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device. +[rank7]:[W921 05:27:04.262903971 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 7] using GPU 7 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device. +[rank4]:[W921 05:27:04.262939575 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 4] using GPU 4 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device. > compiling dataset index builder ... make: Entering directory '/workspace/Megatron-UltraEP/megatron/core/datasets' make: Nothing to be done for 'default'. make: Leaving directory '/workspace/Megatron-UltraEP/megatron/core/datasets' ->>> done with dataset index builder. Compilation time: 0.437 seconds +>>> done with dataset index builder. Compilation time: 0.425 seconds WARNING: constraints for invoking optimized fused softmax kernel are not met. We default back to unfused kernel invocations. > compiling and loading fused kernels ... -[rank0]:[W917 03:27:13.386048103 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 0] using GPU 0 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device. ->>> done with compiling and loading fused kernels. Compilation time: 4.013 seconds -time to initialize megatron (seconds): 9.439 -[after megatron is initialized] datetime: 2026-09-17 03:27:19 +[rank0]:[W921 05:27:04.862236191 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 0] using GPU 0 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device. +>>> done with compiling and loading fused kernels. Compilation time: 3.498 seconds +time to initialize megatron (seconds): 10.193 +[after megatron is initialized] datetime: 2026-09-21 05:27:12 building GPT model ... /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect. warnings.warn( @@ -667,11 +667,11 @@ INFO:megatron.core.distributed.param_and_grad_buffer:Number of buckets for gradi Params for bucket 1 (206045184 elements, 206045184 padded size): module.output_layer.weight Params for bucket 2 (42086400 elements, 42086400 padded size): + module.decoder.final_layernorm.weight module.decoder.layers.15.self_attention.linear_qkv.weight module.decoder.layers.15.pre_mlp_layernorm.weight - module.decoder.final_layernorm.weight - module.decoder.layers.15.self_attention.linear_qkv.layer_norm_weight module.decoder.layers.15.mlp.router.weight + module.decoder.layers.15.self_attention.linear_qkv.layer_norm_weight module.decoder.layers.15.self_attention.linear_proj.weight Params for bucket 3 (42082304 elements, 42082304 padded size): module.decoder.layers.14.self_attention.linear_qkv.weight @@ -680,29 +680,29 @@ Params for bucket 3 (42082304 elements, 42082304 padded size): module.decoder.layers.14.self_attention.linear_proj.weight module.decoder.layers.14.self_attention.linear_qkv.layer_norm_weight Params for bucket 4 (42082304 elements, 42082304 padded size): - module.decoder.layers.13.pre_mlp_layernorm.weight module.decoder.layers.13.self_attention.linear_qkv.weight + module.decoder.layers.13.pre_mlp_layernorm.weight module.decoder.layers.13.mlp.router.weight - module.decoder.layers.13.self_attention.linear_proj.weight module.decoder.layers.13.self_attention.linear_qkv.layer_norm_weight + module.decoder.layers.13.self_attention.linear_proj.weight Params for bucket 5 (42082304 elements, 42082304 padded size): - module.decoder.layers.12.self_attention.linear_qkv.weight - module.decoder.layers.12.pre_mlp_layernorm.weight module.decoder.layers.12.mlp.router.weight module.decoder.layers.12.self_attention.linear_proj.weight module.decoder.layers.12.self_attention.linear_qkv.layer_norm_weight + module.decoder.layers.12.self_attention.linear_qkv.weight + module.decoder.layers.12.pre_mlp_layernorm.weight Params for bucket 6 (42082304 elements, 42082304 padded size): - module.decoder.layers.11.self_attention.linear_qkv.weight module.decoder.layers.11.pre_mlp_layernorm.weight + module.decoder.layers.11.self_attention.linear_qkv.weight module.decoder.layers.11.mlp.router.weight - module.decoder.layers.11.self_attention.linear_proj.weight module.decoder.layers.11.self_attention.linear_qkv.layer_norm_weight + module.decoder.layers.11.self_attention.linear_proj.weight Params for bucket 7 (42082304 elements, 42082304 padded size): module.decoder.layers.10.self_attention.linear_qkv.weight module.decoder.layers.10.pre_mlp_layernorm.weight module.decoder.layers.10.mlp.router.weight - module.decoder.layers.10.self_attention.linear_proj.weight module.decoder.layers.10.self_attention.linear_qkv.layer_norm_weight + module.decoder.layers.10.self_attention.linear_proj.weight Params for bucket 8 (42082304 elements, 42082304 padded size): module.decoder.layers.9.self_attention.linear_qkv.weight module.decoder.layers.9.pre_mlp_layernorm.weight @@ -716,60 +716,60 @@ Params for bucket 9 (42082304 elements, 42082304 padded size): module.decoder.layers.8.self_attention.linear_qkv.layer_norm_weight module.decoder.layers.8.self_attention.linear_proj.weight Params for bucket 10 (42082304 elements, 42082304 padded size): + module.decoder.layers.7.pre_mlp_layernorm.weight + module.decoder.layers.7.self_attention.linear_qkv.weight module.decoder.layers.7.mlp.router.weight module.decoder.layers.7.self_attention.linear_proj.weight module.decoder.layers.7.self_attention.linear_qkv.layer_norm_weight - module.decoder.layers.7.self_attention.linear_qkv.weight - module.decoder.layers.7.pre_mlp_layernorm.weight Params for bucket 11 (42082304 elements, 42082304 padded size): module.decoder.layers.6.mlp.router.weight - module.decoder.layers.6.self_attention.linear_proj.weight module.decoder.layers.6.self_attention.linear_qkv.layer_norm_weight + module.decoder.layers.6.self_attention.linear_proj.weight module.decoder.layers.6.self_attention.linear_qkv.weight module.decoder.layers.6.pre_mlp_layernorm.weight Params for bucket 12 (42082304 elements, 42082304 padded size): + module.decoder.layers.5.mlp.router.weight + module.decoder.layers.5.self_attention.linear_qkv.layer_norm_weight module.decoder.layers.5.self_attention.linear_qkv.weight module.decoder.layers.5.self_attention.linear_proj.weight module.decoder.layers.5.pre_mlp_layernorm.weight - module.decoder.layers.5.mlp.router.weight - module.decoder.layers.5.self_attention.linear_qkv.layer_norm_weight Params for bucket 13 (42082304 elements, 42082304 padded size): + module.decoder.layers.4.self_attention.linear_qkv.weight + module.decoder.layers.4.pre_mlp_layernorm.weight module.decoder.layers.4.mlp.router.weight module.decoder.layers.4.self_attention.linear_proj.weight module.decoder.layers.4.self_attention.linear_qkv.layer_norm_weight - module.decoder.layers.4.self_attention.linear_qkv.weight - module.decoder.layers.4.pre_mlp_layernorm.weight Params for bucket 14 (42082304 elements, 42082304 padded size): + module.decoder.layers.3.self_attention.linear_qkv.weight + module.decoder.layers.3.pre_mlp_layernorm.weight module.decoder.layers.3.mlp.router.weight module.decoder.layers.3.self_attention.linear_proj.weight module.decoder.layers.3.self_attention.linear_qkv.layer_norm_weight - module.decoder.layers.3.self_attention.linear_qkv.weight - module.decoder.layers.3.pre_mlp_layernorm.weight Params for bucket 15 (42082304 elements, 42082304 padded size): + module.decoder.layers.2.self_attention.linear_qkv.weight + module.decoder.layers.2.pre_mlp_layernorm.weight module.decoder.layers.2.mlp.router.weight module.decoder.layers.2.self_attention.linear_proj.weight module.decoder.layers.2.self_attention.linear_qkv.layer_norm_weight - module.decoder.layers.2.pre_mlp_layernorm.weight - module.decoder.layers.2.self_attention.linear_qkv.weight Params for bucket 16 (42082304 elements, 42082304 padded size): - module.decoder.layers.1.self_attention.linear_qkv.layer_norm_weight - module.decoder.layers.1.self_attention.linear_proj.weight - module.decoder.layers.1.mlp.router.weight module.decoder.layers.1.self_attention.linear_qkv.weight module.decoder.layers.1.pre_mlp_layernorm.weight + module.decoder.layers.1.mlp.router.weight + module.decoder.layers.1.self_attention.linear_qkv.layer_norm_weight + module.decoder.layers.1.self_attention.linear_proj.weight Params for bucket 17 (42082304 elements, 42082304 padded size): - module.decoder.layers.0.pre_mlp_layernorm.weight + module.decoder.layers.0.self_attention.linear_qkv.weight module.decoder.layers.0.self_attention.linear_proj.weight + module.decoder.layers.0.pre_mlp_layernorm.weight module.decoder.layers.0.mlp.router.weight - module.decoder.layers.0.self_attention.linear_qkv.weight module.decoder.layers.0.self_attention.linear_qkv.layer_norm_weight Params for bucket 18 (206045184 elements, 206045184 padded size): module.embedding.word_embeddings.weight INFO:megatron.core.distributed.param_and_grad_buffer:Number of buckets for gradient all-reduce / reduce-scatter: 56 Params for bucket 1 (44040192 elements, 44040192 padded size): module.decoder.layers.15.mlp.experts.linear_fc2.weight3 - module.decoder.layers.15.mlp.experts.linear_fc2.weight1 module.decoder.layers.15.mlp.experts.linear_fc2.weight2 + module.decoder.layers.15.mlp.experts.linear_fc2.weight1 Params for bucket 2 (44040192 elements, 44040192 padded size): module.decoder.layers.15.mlp.experts.linear_fc2.weight0 module.decoder.layers.15.mlp.experts.linear_fc1.weight3 @@ -781,8 +781,8 @@ Params for bucket 4 (44040192 elements, 44040192 padded size): module.decoder.layers.14.mlp.experts.linear_fc2.weight3 Params for bucket 5 (44040192 elements, 44040192 padded size): module.decoder.layers.14.mlp.experts.linear_fc2.weight2 - module.decoder.layers.14.mlp.experts.linear_fc2.weight0 module.decoder.layers.14.mlp.experts.linear_fc2.weight1 + module.decoder.layers.14.mlp.experts.linear_fc2.weight0 Params for bucket 6 (58720256 elements, 58720256 padded size): module.decoder.layers.14.mlp.experts.linear_fc1.weight3 module.decoder.layers.14.mlp.experts.linear_fc1.weight2 @@ -827,8 +827,8 @@ Params for bucket 18 (44040192 elements, 44040192 padded size): module.decoder.layers.10.mlp.experts.linear_fc2.weight3 Params for bucket 19 (44040192 elements, 44040192 padded size): module.decoder.layers.10.mlp.experts.linear_fc2.weight2 - module.decoder.layers.10.mlp.experts.linear_fc2.weight1 module.decoder.layers.10.mlp.experts.linear_fc2.weight0 + module.decoder.layers.10.mlp.experts.linear_fc2.weight1 Params for bucket 20 (58720256 elements, 58720256 padded size): module.decoder.layers.10.mlp.experts.linear_fc1.weight3 module.decoder.layers.10.mlp.experts.linear_fc1.weight2 @@ -850,8 +850,8 @@ Params for bucket 25 (44040192 elements, 44040192 padded size): module.decoder.layers.8.mlp.experts.linear_fc2.weight3 Params for bucket 26 (44040192 elements, 44040192 padded size): module.decoder.layers.8.mlp.experts.linear_fc2.weight2 - module.decoder.layers.8.mlp.experts.linear_fc2.weight1 module.decoder.layers.8.mlp.experts.linear_fc2.weight0 + module.decoder.layers.8.mlp.experts.linear_fc2.weight1 Params for bucket 27 (58720256 elements, 58720256 padded size): module.decoder.layers.8.mlp.experts.linear_fc1.weight3 module.decoder.layers.8.mlp.experts.linear_fc1.weight2 @@ -883,8 +883,8 @@ Params for bucket 35 (58720256 elements, 58720256 padded size): module.decoder.layers.6.mlp.experts.linear_fc1.weight0 Params for bucket 36 (44040192 elements, 44040192 padded size): module.decoder.layers.5.mlp.experts.linear_fc2.weight3 - module.decoder.layers.5.mlp.experts.linear_fc2.weight1 module.decoder.layers.5.mlp.experts.linear_fc2.weight2 + module.decoder.layers.5.mlp.experts.linear_fc2.weight1 Params for bucket 37 (44040192 elements, 44040192 padded size): module.decoder.layers.5.mlp.experts.linear_fc2.weight0 module.decoder.layers.5.mlp.experts.linear_fc1.weight3 @@ -906,8 +906,8 @@ Params for bucket 42 (58720256 elements, 58720256 padded size): module.decoder.layers.4.mlp.experts.linear_fc1.weight0 Params for bucket 43 (44040192 elements, 44040192 padded size): module.decoder.layers.3.mlp.experts.linear_fc2.weight3 - module.decoder.layers.3.mlp.experts.linear_fc2.weight1 module.decoder.layers.3.mlp.experts.linear_fc2.weight2 + module.decoder.layers.3.mlp.experts.linear_fc2.weight1 Params for bucket 44 (44040192 elements, 44040192 padded size): module.decoder.layers.3.mlp.experts.linear_fc2.weight0 module.decoder.layers.3.mlp.experts.linear_fc1.weight3 @@ -942,17 +942,17 @@ Params for bucket 53 (44040192 elements, 44040192 padded size): module.decoder.layers.0.mlp.experts.linear_fc2.weight3 Params for bucket 54 (44040192 elements, 44040192 padded size): module.decoder.layers.0.mlp.experts.linear_fc2.weight2 - module.decoder.layers.0.mlp.experts.linear_fc2.weight0 module.decoder.layers.0.mlp.experts.linear_fc2.weight1 + module.decoder.layers.0.mlp.experts.linear_fc2.weight0 Params for bucket 55 (58720256 elements, 58720256 padded size): module.decoder.layers.0.mlp.experts.linear_fc1.weight3 module.decoder.layers.0.mlp.experts.linear_fc1.weight2 Params for bucket 56 (58720256 elements, 58720256 padded size): module.decoder.layers.0.mlp.experts.linear_fc1.weight1 module.decoder.layers.0.mlp.experts.linear_fc1.weight0 -INFO:megatron.core.optimizer:Setting up optimizer with config OptimizerConfig(optimizer='adam', lr=0.0001, min_lr=1e-05, decoupled_lr=None, decoupled_min_lr=None, weight_decay=0.1, fp8_recipe='mxfp8', fp16=False, bf16=True, reuse_grad_buf_for_mxfp8_param_ag=False, params_dtype=torch.bfloat16, use_precision_aware_optimizer=False, store_param_remainders=True, main_grads_dtype=torch.float32, main_params_dtype=torch.float32, exp_avg_dtype=torch.float32, exp_avg_sq_dtype=torch.float32, loss_scale=None, initial_loss_scale=4294967296, min_loss_scale=1.0, loss_scale_window=1000, hysteresis=2, adam_beta1=0.9, adam_beta2=0.999, adam_eps=1e-08, sgd_momentum=0.9, use_distributed_optimizer=True, overlap_param_gather_with_optimizer_step=False, optimizer_cpu_offload=False, optimizer_offload_fraction=1.0, use_torch_optimizer_for_cpu_offload=False, overlap_cpu_optimizer_d2h_h2d=False, pin_cpu_grads=True, pin_cpu_params=True, clip_grad=1.0, log_num_zeros_in_grad=False, barrier_with_L1_time=True, timers=, config_logger_dir='') +INFO:megatron.core.optimizer:Setting up optimizer with config OptimizerConfig(optimizer='adam', lr=0.0001, min_lr=1e-05, decoupled_lr=None, decoupled_min_lr=None, weight_decay=0.1, fp8_recipe='mxfp8', fp16=False, bf16=True, reuse_grad_buf_for_mxfp8_param_ag=False, params_dtype=torch.bfloat16, use_precision_aware_optimizer=False, store_param_remainders=True, main_grads_dtype=torch.float32, main_params_dtype=torch.float32, exp_avg_dtype=torch.float32, exp_avg_sq_dtype=torch.float32, loss_scale=None, initial_loss_scale=4294967296, min_loss_scale=1.0, loss_scale_window=1000, hysteresis=2, adam_beta1=0.9, adam_beta2=0.999, adam_eps=1e-08, sgd_momentum=0.9, use_distributed_optimizer=True, overlap_param_gather_with_optimizer_step=False, optimizer_cpu_offload=False, optimizer_offload_fraction=1.0, use_torch_optimizer_for_cpu_offload=False, overlap_cpu_optimizer_d2h_h2d=False, pin_cpu_grads=True, pin_cpu_params=True, clip_grad=1.0, log_num_zeros_in_grad=False, barrier_with_L1_time=True, timers=, config_logger_dir='') INFO:megatron.core.optimizer_param_scheduler:> learning rate decay style: cosine -[after model, optimizer, and learning rate scheduler are built] datetime: 2026-09-17 03:27:22 +[after model, optimizer, and learning rate scheduler are built] datetime: 2026-09-21 05:27:16 > building train, validation, and test datasets ... > datasets target sizes (minimum size): train: 160 @@ -960,27 +960,25 @@ INFO:megatron.core.optimizer_param_scheduler:> learning rate decay style: cosine test: 0 INFO:megatron.core.datasets.blended_megatron_dataset_config:Let split_matrix = [(0, 1.0), None, None] > building train, validation, and test datasets for GPT ... -INFO:megatron.core.datasets.blended_megatron_dataset_builder:Building GPTDataset splits with sizes=(160, 0, 0) and config=GPTDatasetConfig(random_seed=1234, sequence_length=4096, blend=(['/workspace/data/wikitext/wikitext103_gpt2_text_document'], None), blend_per_split=None, split='100,0,0', split_matrix=[(0, 1.0), None, None], num_dataset_builder_threads=1, path_to_cache=None, mmap_bin_files=True, mock=False, tokenizer=, mid_level_dataset_surplus=0.005, reset_position_ids=False, reset_attention_mask=False, eod_mask_loss=False, create_attention_mask=True, drop_last_partial_validation_sequence=True, add_extra_token_to_sequence=True, object_storage_cache_path=None) +INFO:megatron.core.datasets.blended_megatron_dataset_builder:Building GPTDataset splits with sizes=(160, 0, 0) and config=GPTDatasetConfig(random_seed=1234, sequence_length=4096, blend=(['/workspace/data/wikitext/wikitext103_gpt2_text_document'], None), blend_per_split=None, split='100,0,0', split_matrix=[(0, 1.0), None, None], num_dataset_builder_threads=1, path_to_cache=None, mmap_bin_files=True, mock=False, tokenizer=, mid_level_dataset_surplus=0.005, reset_position_ids=False, reset_attention_mask=False, eod_mask_loss=False, create_attention_mask=True, drop_last_partial_validation_sequence=True, add_extra_token_to_sequence=True, object_storage_cache_path=None) INFO:megatron.core.datasets.indexed_dataset:Load the _IndexReader from /workspace/data/wikitext/wikitext103_gpt2_text_document.idx INFO:megatron.core.datasets.indexed_dataset: Extract the sequence lengths INFO:megatron.core.datasets.indexed_dataset: Extract the sequence pointers INFO:megatron.core.datasets.indexed_dataset: Extract the document indices INFO:megatron.core.datasets.indexed_dataset:> total number of sequences: 1165029 INFO:megatron.core.datasets.indexed_dataset:> total number of documents: 1801350 -INFO:megatron.core.datasets.gpt_dataset:Load the GPTDataset train indices -INFO:megatron.core.datasets.gpt_dataset: Load the document index from 68684a515e118253a8572c4517cf2ffe-GPTDataset-train-document_index.npy -INFO:megatron.core.datasets.gpt_dataset: Load the sample index from 68684a515e118253a8572c4517cf2ffe-GPTDataset-train-sample_index.npy -INFO:megatron.core.datasets.gpt_dataset: Load the shuffle index from 68684a515e118253a8572c4517cf2ffe-GPTDataset-train-shuffle_index.npy +INFO:megatron.core.datasets.gpt_dataset:Build and save the GPTDataset train indices INFO:megatron.core.datasets.gpt_dataset:> total number of samples: 29073 +INFO:megatron.core.datasets.gpt_dataset:> total number of epochs: 1 > finished creating GPT datasets ... -[after dataloaders are built] datetime: 2026-09-17 03:27:22 +[after dataloaders are built] datetime: 2026-09-21 05:27:16 done with setup ... (min, max) time across ranks (ms): - model-and-optimizer-setup ......................: (3347.95, 3351.05) - train/valid/test-data-iterators-setup ..........: (113.53, 139.33) + model-and-optimizer-setup ......................: (3579.39, 3581.84) + train/valid/test-data-iterators-setup ..........: (168.94, 187.22) training ... Overwriting rerun_state_machine.current_iteration from -1 to 0... -[before the start of training step] datetime: 2026-09-17 03:27:23 +[before the start of training step] datetime: 2026-09-21 05:27:16 /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. @@ -1035,16 +1033,16 @@ Overwriting rerun_state_machine.current_iteration from -1 to 0... warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") -/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. - warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") -/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. - warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. @@ -1109,39 +1107,50 @@ Overwriting rerun_state_machine.current_iteration from -1 to 0... warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") - [2026-09-17 03:27:52] iteration 1/ 10 | consumed samples: 16 | elapsed time per iteration (ms): 29813.9 | throughput per GPU (TFLOP/s/GPU): 4.2 | learning rate: 2.000000E-05 | global batch size: 16 | lm loss: 1.102902E+01 | loss scale: 1.0 | grad norm: 44.186 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 8832 | min safe capacity: 1.0781 | + [2026-09-21 05:27:49] iteration 1/ 10 | consumed samples: 16 | elapsed time per iteration (ms): 33028.6 | throughput per GPU (TFLOP/s/GPU): 3.8 | learning rate: 2.000000E-05 | global batch size: 16 | lm loss: 1.102902E+01 | loss scale: 1.0 | grad norm: 44.184 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 8832 | min safe capacity: 1.0781 | Number of parameters in transformer block in billions: 23.22 Number of parameters in embedding layers in billions: 0.41 Total number of parameters in billions: 23.63 Number of parameters in most loaded shard in billions: 23.6320 Theoretical memory footprints: weight and optimizer=169029.43 MB -[Rank 0] (after 1 iterations) memory (MB) | allocated: 58124.0458984375 | max allocated: 65718.16796875 | reserved: 79458.0 | max reserved: 79458.0 - [2026-09-17 03:27:53] iteration 2/ 10 | consumed samples: 32 | elapsed time per iteration (ms): 586.2 | throughput per GPU (TFLOP/s/GPU): 214.2 | learning rate: 4.000000E-05 | global batch size: 16 | lm loss: 1.101823E+01 | loss scale: 1.0 | grad norm: 45.729 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 8832 | min safe capacity: 1.0781 | - [2026-09-17 03:27:53] iteration 3/ 10 | consumed samples: 48 | elapsed time per iteration (ms): 434.7 | throughput per GPU (TFLOP/s/GPU): 288.9 | learning rate: 6.000000E-05 | global batch size: 16 | lm loss: 9.972256E+00 | loss scale: 1.0 | grad norm: 157.296 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 8832 | min safe capacity: 1.0781 | +[Rank 0] (after 1 iterations) memory (MB) | allocated: 58124.0458984375 | max allocated: 65718.0 | reserved: 79284.0 | max reserved: 79284.0 + [2026-09-21 05:27:50] iteration 2/ 10 | consumed samples: 32 | elapsed time per iteration (ms): 564.2 | throughput per GPU (TFLOP/s/GPU): 222.6 | learning rate: 4.000000E-05 | global batch size: 16 | lm loss: 1.101823E+01 | loss scale: 1.0 | grad norm: 45.733 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 8832 | min safe capacity: 1.0781 | + [2026-09-21 05:27:50] iteration 3/ 10 | consumed samples: 48 | elapsed time per iteration (ms): 437.4 | throughput per GPU (TFLOP/s/GPU): 287.1 | learning rate: 6.000000E-05 | global batch size: 16 | lm loss: 9.971716E+00 | loss scale: 1.0 | grad norm: 157.169 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 8832 | min safe capacity: 1.0781 | Capture CUDA graph for training!!! CUDA graph capture done!!! - [2026-09-17 03:27:57] iteration 4/ 10 | consumed samples: 64 | elapsed time per iteration (ms): 3115.4 | throughput per GPU (TFLOP/s/GPU): 40.3 | learning rate: 8.000000E-05 | global batch size: 16 | lm loss: 1.311298E+01 | loss scale: 1.0 | grad norm: 41.251 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 8832 | min safe capacity: 1.0781 | - [2026-09-17 03:27:57] iteration 5/ 10 | consumed samples: 80 | elapsed time per iteration (ms): 323.9 | throughput per GPU (TFLOP/s/GPU): 387.7 | learning rate: 1.000000E-04 | global batch size: 16 | lm loss: 1.040733E+01 | loss scale: 1.0 | grad norm: 25.625 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 8832 | min safe capacity: 1.0781 | - [2026-09-17 03:27:57] iteration 6/ 10 | consumed samples: 96 | elapsed time per iteration (ms): 333.2 | throughput per GPU (TFLOP/s/GPU): 376.9 | learning rate: 7.750000E-05 | global batch size: 16 | lm loss: 1.025381E+01 | loss scale: 1.0 | grad norm: 19.172 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 8832 | min safe capacity: 1.0781 | - [2026-09-17 03:28:00] iteration 7/ 10 | consumed samples: 112 | elapsed time per iteration (ms): 3161.7 | throughput per GPU (TFLOP/s/GPU): 39.7 | learning rate: 3.250000E-05 | global batch size: 16 | lm loss: 1.002502E+01 | loss scale: 1.0 | grad norm: 8.664 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 8832 | min safe capacity: 1.0781 | - [2026-09-17 03:28:01] iteration 8/ 10 | consumed samples: 128 | elapsed time per iteration (ms): 342.7 | throughput per GPU (TFLOP/s/GPU): 366.4 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 9.464363E+00 | loss scale: 1.0 | grad norm: 8.108 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 8832 | min safe capacity: 1.0781 | + [2026-09-21 05:27:53] iteration 4/ 10 | consumed samples: 64 | elapsed time per iteration (ms): 3034.2 | throughput per GPU (TFLOP/s/GPU): 41.4 | learning rate: 8.000000E-05 | global batch size: 16 | lm loss: 1.311224E+01 | loss scale: 1.0 | grad norm: 41.298 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 8832 | min safe capacity: 1.0781 | + [2026-09-21 05:27:53] iteration 5/ 10 | consumed samples: 80 | elapsed time per iteration (ms): 325.6 | throughput per GPU (TFLOP/s/GPU): 385.6 | learning rate: 1.000000E-04 | global batch size: 16 | lm loss: 1.040747E+01 | loss scale: 1.0 | grad norm: 25.670 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 8832 | min safe capacity: 1.0781 | + [2026-09-21 05:27:54] iteration 6/ 10 | consumed samples: 96 | elapsed time per iteration (ms): 335.0 | throughput per GPU (TFLOP/s/GPU): 374.9 | learning rate: 7.750000E-05 | global batch size: 16 | lm loss: 1.025583E+01 | loss scale: 1.0 | grad norm: 19.185 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 8832 | min safe capacity: 1.0781 | + [2026-09-21 05:27:57] iteration 7/ 10 | consumed samples: 112 | elapsed time per iteration (ms): 3243.3 | throughput per GPU (TFLOP/s/GPU): 38.7 | learning rate: 3.250000E-05 | global batch size: 16 | lm loss: 1.002962E+01 | loss scale: 1.0 | grad norm: 8.693 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 8832 | min safe capacity: 1.0781 | + [2026-09-21 05:27:57] iteration 8/ 10 | consumed samples: 128 | elapsed time per iteration (ms): 343.8 | throughput per GPU (TFLOP/s/GPU): 365.2 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 9.465693E+00 | loss scale: 1.0 | grad norm: 7.851 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 8832 | min safe capacity: 1.0781 | Capture range started in the application. Capture range ended in the application. -Generating '/tmp/nsys-report-2929.qdstrm' +Generating '/tmp/nsys-report-2397.qdstrm' [1/1] [0% ] G2_K2__ultraep__matched__seq4096__gbs16__cap1p15625.nsys-repProcessing events... - [1/1] [0% ] G2_K2__ultraep__matched__seq4096__gbs16__cap1p15625.nsys-rep [1/1] [0% ] G2_K2__ultraep__matched__seq4096__gbs16__cap1p15625.nsys-rep [1/1] [5% ] G2_K2__ultraep__matched__seq4096__gbs16__cap1p15625.nsys-rep [1/1] [8% ] G2_K2__ultraep__matched__seq4096__gbs16__cap1p15625.nsys-rep [1/1] [11% ] G2_K2__ultraep__matched__seq4096__gbs16__cap1p15625.nsys-rep [1/1] [13% ] G2_K2__ultraep__matched__seq4096__gbs16__cap1p15625.nsys-rep [1/1] [=16% ] G2_K2__ultraep__matched__seq4096__gbs16__cap1p15625.nsys-rep [1/1] [==19% ] G2_K2__ultraep__matched__seq4096__gbs16__cap1p15625.nsys-rep [1/1] [==21% ] G2_K2__ultraep__matched__seq4096__gbs16__cap1p15625.nsys-rep [1/1] [===24% ] G2_K2__ultraep__matched__seq4096__gbs16__cap1p15625.nsys-rep [1/1] [====27% ] G2_K2__ultraep__matched__seq4096__gbs16__cap1p15625.nsys-rep [1/1] [=====29% ] G2_K2__ultraep__matched__seq4096__gbs16__cap1p15625.nsys-rep [1/1] [=====32% ] G2_K2__ultraep__matched__seq4096__gbs16__cap1p15625.nsys-rep [1/1] [======35% ] G2_K2__ultraep__matched__seq4096__gbs16__cap1p15625.nsys-rep [1/1] [=======37% ] G2_K2__ultraep__matched__seq4096__gbs16__cap1p15625.nsys-rep [1/1] [=======38% ] G2_K2__ultraep__matched__seq4096__gbs16__cap1p15625.nsys-rep [1/1] [=======39% ] G2_K2__ultraep__matched__seq4096__gbs16__cap1p15625.nsys-rep [1/1] [========40% ] G2_K2__ultraep__matched__seq4096__gbs16__cap1p15625.nsys-rep [1/1] [========41% ] G2_K2__ultraep__matched__seq4096__gbs16__cap1p15625.nsys-rep [1/1] [========42% ] G2_K2__ultraep__matched__seq4096__gbs16__cap1p15625.nsys-rep [1/1] [=========44% ] G2_K2__ultraep__matched__seq4096__gbs16__cap1p15625.nsys-rep [1/1] [=========45% ] G2_K2__ultraep__matched__seq4096__gbs16__cap1p15625.nsys-rep [1/1] [=========46% ] G2_K2__ultraep__matched__seq4096__gbs16__cap1p15625.nsys-rep [1/1] [==========48% ] G2_K2__ultraep__matched__seq4096__gbs16__cap1p15625.nsys-rep [1/1] [==========49% ] G2_K2__ultraep__matched__seq4096__gbs16__cap1p15625.nsys-rep [1/1] [===========51% ] G2_K2__ultraep__matched__seq4096__gbs16__cap1p15625.nsys-rep [1/1] [===========52% ] G2_K2__ultraep__matched__seq4096__gbs16__cap1p15625.nsys-rep [1/1] [===========53% ] G2_K2__ultraep__matched__seq4096__gbs16__cap1p15625.nsys-rep [1/1] [============55% ] G2_K2__ultraep__matched__seq4096__gbs16__cap1p15625.nsys-rep [1/1] [============56% ] G2_K2__ultraep__matched__seq4096__gbs16__cap1p15625.nsys-rep [1/1] [=============58% ] G2_K2__ultraep__matched__seq4096__gbs16__cap1p15625.nsys-rep [1/1] [=============60% ] G2_K2__ultraep__matched__seq4096__gbs16__cap1p15625.nsys-rep [1/1] [========================100%] G2_K2__ultraep__matched__seq4096__gbs16__cap1p15625.nsys-rep [1/1] [========================100%] G2_K2__ultraep__matched__seq4096__gbs16__cap1p15625.nsys-rep + [1/1] [0% ] G2_K2__ultraep__matched__seq4096__gbs16__cap1p15625.nsys-rep [1/1] [0% ] G2_K2__ultraep__matched__seq4096__gbs16__cap1p15625.nsys-rep [1/1] [5% ] G2_K2__ultraep__matched__seq4096__gbs16__cap1p15625.nsys-rep [1/1] [7% ] G2_K2__ultraep__matched__seq4096__gbs16__cap1p15625.nsys-rep [1/1] [10% ] G2_K2__ultraep__matched__seq4096__gbs16__cap1p15625.nsys-rep [1/1] [12% ] G2_K2__ultraep__matched__seq4096__gbs16__cap1p15625.nsys-rep [1/1] [=15% ] G2_K2__ultraep__matched__seq4096__gbs16__cap1p15625.nsys-rep [1/1] [==18% ] G2_K2__ultraep__matched__seq4096__gbs16__cap1p15625.nsys-rep [1/1] [==20% ] G2_K2__ultraep__matched__seq4096__gbs16__cap1p15625.nsys-rep [1/1] [===23% ] G2_K2__ultraep__matched__seq4096__gbs16__cap1p15625.nsys-rep [1/1] [====25% ] G2_K2__ultraep__matched__seq4096__gbs16__cap1p15625.nsys-rep [1/1] [====28% ] G2_K2__ultraep__matched__seq4096__gbs16__cap1p15625.nsys-rep [1/1] [=====31% ] G2_K2__ultraep__matched__seq4096__gbs16__cap1p15625.nsys-rep [1/1] [======33% ] G2_K2__ultraep__matched__seq4096__gbs16__cap1p15625.nsys-rep [1/1] [======35% ] G2_K2__ultraep__matched__seq4096__gbs16__cap1p15625.nsys-rep [1/1] [=======36% ] G2_K2__ultraep__matched__seq4096__gbs16__cap1p15625.nsys-rep [1/1] [=======37% ] G2_K2__ultraep__matched__seq4096__gbs16__cap1p15625.nsys-rep [1/1] [=======38% ] G2_K2__ultraep__matched__seq4096__gbs16__cap1p15625.nsys-rep [1/1] [=======39% ] G2_K2__ultraep__matched__seq4096__gbs16__cap1p15625.nsys-rep [1/1] [========41% ] G2_K2__ultraep__matched__seq4096__gbs16__cap1p15625.nsys-rep [1/1] [========42% ] G2_K2__ultraep__matched__seq4096__gbs16__cap1p15625.nsys-rep [1/1] [=========43% ] G2_K2__ultraep__matched__seq4096__gbs16__cap1p15625.nsys-rep [1/1] [=========44% ] G2_K2__ultraep__matched__seq4096__gbs16__cap1p15625.nsys-rep [1/1] [=========46% ] G2_K2__ultraep__matched__seq4096__gbs16__cap1p15625.nsys-rep [1/1] [==========47% ] G2_K2__ultraep__matched__seq4096__gbs16__cap1p15625.nsys-rep [1/1] [==========49% ] G2_K2__ultraep__matched__seq4096__gbs16__cap1p15625.nsys-rep [1/1] [===========50% ] G2_K2__ultraep__matched__seq4096__gbs16__cap1p15625.nsys-rep [1/1] [===========52% ] G2_K2__ultraep__matched__seq4096__gbs16__cap1p15625.nsys-rep [1/1] [===========53% ] G2_K2__ultraep__matched__seq4096__gbs16__cap1p15625.nsys-rep [1/1] [============54% ] G2_K2__ultraep__matched__seq4096__gbs16__cap1p15625.nsys-rep [1/1] [============56% ] G2_K2__ultraep__matched__seq4096__gbs16__cap1p15625.nsys-rep [1/1] [=============59% ] G2_K2__ultraep__matched__seq4096__gbs16__cap1p15625.nsys-rep [1/1] [========================100%] G2_K2__ultraep__matched__seq4096__gbs16__cap1p15625.nsys-rep [1/1] [========================100%] G2_K2__ultraep__matched__seq4096__gbs16__cap1p15625.nsys-rep + [2026-09-21 05:28:40] iteration 9/ 10 | consumed samples: 144 | elapsed time per iteration (ms): 42919.9 | throughput per GPU (TFLOP/s/GPU): 2.9 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 9.198565E+00 | loss scale: 1.0 | grad norm: 7.259 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 8832 | min safe capacity: 1.0781 | + [2026-09-21 05:28:41] iteration 10/ 10 | consumed samples: 160 | elapsed time per iteration (ms): 363.9 | throughput per GPU (TFLOP/s/GPU): 345.1 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 9.043821E+00 | loss scale: 1.0 | grad norm: 6.923 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 8832 | min safe capacity: 1.0781 | +[after training is done] datetime: 2026-09-21 05:28:41 +[rank0]:[W921 05:28:42.332826642 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()) +[rank1]:[W921 05:28:43.605743009 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()) +[rank4]:[W921 05:28:43.608715326 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()) +[rank5]:[W921 05:28:43.624970738 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()) +[rank6]:[W921 05:28:44.043986247 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()) +[rank2]:[W921 05:28:44.074015347 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()) +[rank3]:[W921 05:28:44.075061371 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()) +[rank7]:[W921 05:28:44.524534589 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()) Collection stop is not allowed in this state. -W0917 03:29:32.760000 948326 torch/distributed/elastic/agent/server/api.py:719] Received 15 death signal, shutting down workers -W0917 03:29:32.761000 948326 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 948649 closing signal SIGTERM -Generated: - /workspace/granularity_next_b200/tiera_topk_trace/ultraep/tierbtrace/tierbtrace_gbs16/G2_K2__ultraep__matched__seq4096__gbs16__cap1p15625/nsys/G2_K2__ultraep__matched__seq4096__gbs16__cap1p15625.nsys-rep -W0917 03:29:32.765000 948326 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 948650 closing signal SIGTERM -W0917 03:29:32.772000 948326 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 948651 closing signal SIGTERM -W0917 03:29:32.773000 948326 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 948652 closing signal SIGTERM -W0917 03:29:32.777000 948326 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 948653 closing signal SIGTERM -W0917 03:29:32.783000 948326 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 948654 closing signal SIGTERM -W0917 03:29:32.794000 948326 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 948655 closing signal SIGTERM -W0917 03:29:32.804000 948326 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 948656 closing signal SIGTERM +[rank0]:[W921 05:28:46.626373583 observer.cpp:116] Warning: Leaked callback handle: 1 (function operator()) +pure virtual method called +terminate called without an active exception +W0921 05:28:47.337000 24330 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 24442 closing signal SIGTERM +W0921 05:28:47.338000 24330 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 24443 closing signal SIGTERM +W0921 05:28:47.338000 24330 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 24444 closing signal SIGTERM +W0921 05:28:47.338000 24330 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 24445 closing signal SIGTERM +W0921 05:28:47.339000 24330 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 24446 closing signal SIGTERM +W0921 05:28:47.339000 24330 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 24447 closing signal SIGTERM +W0921 05:28:47.340000 24330 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 24448 closing signal SIGTERM +E0921 05:28:49.885000 24330 torch/distributed/elastic/multiprocessing/api.py:874] failed (exitcode: -6) local_rank: 0 (pid: 24441) of binary: /usr/bin/python Traceback (most recent call last): File "/usr/local/bin/torchrun", line 8, in sys.exit(main()) @@ -1156,21 +1165,23 @@ Traceback (most recent call last): File "/usr/local/lib/python3.12/dist-packages/torch/distributed/launcher/api.py", line 139, in __call__ return launch_agent(self._config, self._entrypoint, list(args)) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ - File "/usr/local/lib/python3.12/dist-packages/torch/distributed/launcher/api.py", line 261, in launch_agent - result = agent.run() - ^^^^^^^^^^^ - File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/metrics/api.py", line 138, in wrapper - result = f(*args, **kwargs) - ^^^^^^^^^^^^^^^^^^ - File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/agent/server/api.py", line 711, in run - result = self._invoke_run(role) - ^^^^^^^^^^^^^^^^^^^^^^ - File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/agent/server/api.py", line 870, in _invoke_run - time.sleep(monitor_interval) - File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/multiprocessing/api.py", line 84, in _terminate_process_handler - raise SignalException(f"Process {os.getpid()} got signal: {sigval}", sigval=sigval) -torch.distributed.elastic.multiprocessing.api.SignalException: Process 948326 got signal: 15 -The target application terminated. One or more process it created re-parented. -Waiting for termination of re-parented processes. -Use the `--wait` option to modify this behavior. -There are no active sessions. + File "/usr/local/lib/python3.12/dist-packages/torch/distributed/launcher/api.py", line 270, in launch_agent + raise ChildFailedError( +torch.distributed.elastic.multiprocessing.errors.ChildFailedError: +============================================================ +pretrain_gpt.py FAILED +------------------------------------------------------------ +Failures: + +------------------------------------------------------------ +Root Cause (first observed failure): +[0]: + time : 2026-09-21_05:28:47 + host : sp-aiinfra-interactive-aghosh-8-mb99k-8xb200-moe-cg-0 + rank : 0 (local_rank: 0) + exitcode : -6 (pid: 24441) + error_file: + traceback : Signal 6 (SIGABRT) received by PID 24441 +============================================================ +Generated: + /workspace/granularity_next_b200/tiera_topk_trace/ultraep/tierbtrace/tierbtrace_gbs16/G2_K2__ultraep__matched__seq4096__gbs16__cap1p15625/nsys/G2_K2__ultraep__matched__seq4096__gbs16__cap1p15625.nsys-rep diff --git a/granularity_next_b200/tiera_topk_trace/ultraep/tierbtrace/tierbtrace_gbs16/G3_K2__ultraep__matched__seq4096__gbs16__cap1p1875/nsys/G3_K2__ultraep__matched__seq4096__gbs16__cap1p1875.nsys-rep b/granularity_next_b200/tiera_topk_trace/ultraep/tierbtrace/tierbtrace_gbs16/G3_K2__ultraep__matched__seq4096__gbs16__cap1p1875/nsys/G3_K2__ultraep__matched__seq4096__gbs16__cap1p1875.nsys-rep index b2b8a2df3cc0197c0f58276dddb1d0c2b0866357..c7449b43be12ff4773f146cd4547383106ad28b6 100644 --- a/granularity_next_b200/tiera_topk_trace/ultraep/tierbtrace/tierbtrace_gbs16/G3_K2__ultraep__matched__seq4096__gbs16__cap1p1875/nsys/G3_K2__ultraep__matched__seq4096__gbs16__cap1p1875.nsys-rep +++ b/granularity_next_b200/tiera_topk_trace/ultraep/tierbtrace/tierbtrace_gbs16/G3_K2__ultraep__matched__seq4096__gbs16__cap1p1875/nsys/G3_K2__ultraep__matched__seq4096__gbs16__cap1p1875.nsys-rep @@ -1,3 +1,3 @@ version https://git-lfs.github.com/spec/v1 -oid sha256:b72936061abb85f8abaf8716a77beb00af3bf28399bb4c5824e9a1ef123b6169 -size 12226799 +oid sha256:aa5ad8c4f434530d5f7e35a6d1c4637caaf159082ca5fa7b61f6f8487d392cb2 +size 11939180 diff --git a/granularity_next_b200/tiera_topk_trace/ultraep/tierbtrace/tierbtrace_gbs16/G3_K2__ultraep__matched__seq4096__gbs16__cap1p1875/nsys/train.log b/granularity_next_b200/tiera_topk_trace/ultraep/tierbtrace/tierbtrace_gbs16/G3_K2__ultraep__matched__seq4096__gbs16__cap1p1875/nsys/train.log index 163437bbed5fa33d1d425be78219fdfd917d2d74..68b9ca4ac7cd8750007fd619988ac75c76f5e51f 100644 --- a/granularity_next_b200/tiera_topk_trace/ultraep/tierbtrace/tierbtrace_gbs16/G3_K2__ultraep__matched__seq4096__gbs16__cap1p1875/nsys/train.log +++ b/granularity_next_b200/tiera_topk_trace/ultraep/tierbtrace/tierbtrace_gbs16/G3_K2__ultraep__matched__seq4096__gbs16__cap1p1875/nsys/train.log @@ -6,12 +6,10 @@ memory_dump=none snapshots=/workspace/granularity_next_b200/tiera_topk_trace/ultraep/tierbtrace/tierbtrace_gbs16/G3_K2__ultraep__matched__seq4096__gbs16__cap1p1875/nsys/memory_snapshots/snapshot.pickle repo=/workspace/Megatron-UltraEP @ 03d57ef7 ============================================================== -W0917 03:29:57.172000 960711 torch/distributed/run.py:766] -W0917 03:29:57.172000 960711 torch/distributed/run.py:766] ***************************************** -W0917 03:29:57.172000 960711 torch/distributed/run.py:766] Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed. -W0917 03:29:57.172000 960711 torch/distributed/run.py:766] ***************************************** -/workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled - warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled") +W0921 05:29:24.331000 40554 torch/distributed/run.py:766] +W0921 05:29:24.331000 40554 torch/distributed/run.py:766] ***************************************** +W0921 05:29:24.331000 40554 torch/distributed/run.py:766] Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed. +W0921 05:29:24.331000 40554 torch/distributed/run.py:766] ***************************************** /workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled") /workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled @@ -23,8 +21,6 @@ WARNING: overriding default arguments for tokenizer_type:GPT2BPETokenizer Number of virtual stages per pipeline stage: None accumulate and all-reduce gradients in fp32 for bfloat16 data type. using torch.bfloat16 for parameters ... -/workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled - warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled") /workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled") ------------------------ arguments ------------------------ @@ -573,28 +569,35 @@ INFO:megatron.core.num_microbatches_calculator:setting number of microbatches to > building GPT2BPETokenizer tokenizer ... /workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled") +/workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled + warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled") +WARNING: one_logger package is required to enable e2e metrics tracking. please go to https://confluence.nvidia.com/display/MLWFO/Package+Repositories for details to install it /workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled") > padded vocab (size: 50257) with 47 dummy tokens (new size: 50304) WARNING:megatron.core.rerun_state_machine:RerunStateMachine initialized in mode RerunMode.VALIDATE_RESULTS > initializing torch distributed ... -WARNING: one_logger package is required to enable e2e metrics tracking. please go to https://confluence.nvidia.com/display/MLWFO/Package+Repositories for details to install it +/workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled + warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled") +[Gloo] Rank 0 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 [Gloo] Rank 1 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 -[Gloo] Rank [Gloo] Rank 0 is connected to 37 is connected to peer ranks. Expected number of connected peer ranks is : 77 peer ranks. Expected number of connected peer ranks is : 7 - [Gloo] Rank 2 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 +[Gloo] Rank [Gloo] Rank 6 is connected to 7 peer ranks. Expected number of connected peer ranks is : 73 is connected to +7 peer ranks. Expected number of connected peer ranks is : 7 [Gloo] Rank 4 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 [Gloo] Rank 5 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 -[Gloo] Rank [Gloo] Rank 7 is connected to 76 peer ranks. Expected number of connected peer ranks is : is connected to 77 peer ranks. -Expected number of connected peer ranks is : 7 +[Gloo] Rank 7 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 [Gloo] Rank 0 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 [Gloo] Rank 1 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 [Gloo] Rank 2 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 -[Gloo] Rank [Gloo] Rank 5 is connected to 37 is connected to peer ranks. 7Expected number of connected peer ranks is : peer ranks. 7Expected number of connected peer ranks is : 7 - +[Gloo] Rank 3 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 [Gloo] Rank 4 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 -[Gloo] Rank 6 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 +[Gloo] Rank 5 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 [Gloo] Rank 7 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 +[Gloo] Rank 6 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 +[Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0 +[Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0 +[Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0 [Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0 [Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0 [Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0 @@ -603,27 +606,24 @@ Expected number of connected peer ranks is : 7 > setting random seeds to 1234 ... [Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0 [Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0 -[Gloo] Rank 0 is connected to 0[Gloo] Rank peer ranks. Expected number of connected peer ranks is : 0 -0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0 -[Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0 -[rank3]:[W917 03:30:09.927160051 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 3] using GPU 3 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device. -[rank1]:[W917 03:30:09.927160874 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 1] using GPU 1 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device. -[rank5]:[W917 03:30:09.927340509 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 5] using GPU 5 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device. -[rank6]:[W917 03:30:09.927344996 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 6] using GPU 6 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device. -[rank4]:[W917 03:30:09.927925286 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 4] using GPU 4 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device. -[rank7]:[W917 03:30:09.927971518 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 7] using GPU 7 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device. -[rank2]:[W917 03:30:09.928149080 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 2] using GPU 2 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device. +[rank4]:[W921 05:29:37.103162947 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 4] using GPU 4 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device. +[rank2]:[W921 05:29:37.103195924 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 2] using GPU 2 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device. +[rank1]:[W921 05:29:37.108298812 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 1] using GPU 1 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device. +[rank7]:[W921 05:29:37.108324544 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 7] using GPU 7 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device. +[rank6]:[W921 05:29:37.108372298 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 6] using GPU 6 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device. +[rank5]:[W921 05:29:37.108439648 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 5] using GPU 5 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device. +[rank3]:[W921 05:29:37.108449582 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 3] using GPU 3 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device. > compiling dataset index builder ... make: Entering directory '/workspace/Megatron-UltraEP/megatron/core/datasets' make: Nothing to be done for 'default'. make: Leaving directory '/workspace/Megatron-UltraEP/megatron/core/datasets' ->>> done with dataset index builder. Compilation time: 0.437 seconds +>>> done with dataset index builder. Compilation time: 0.425 seconds WARNING: constraints for invoking optimized fused softmax kernel are not met. We default back to unfused kernel invocations. > compiling and loading fused kernels ... -[rank0]:[W917 03:30:10.553623237 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 0] using GPU 0 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device. ->>> done with compiling and loading fused kernels. Compilation time: 3.679 seconds -time to initialize megatron (seconds): 9.143 -[after megatron is initialized] datetime: 2026-09-17 03:30:16 +[rank0]:[W921 05:29:37.720116691 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 0] using GPU 0 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device. +>>> done with compiling and loading fused kernels. Compilation time: 4.130 seconds +time to initialize megatron (seconds): 9.722 +[after megatron is initialized] datetime: 2026-09-21 05:29:44 building GPT model ... /usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect. warnings.warn( @@ -667,12 +667,12 @@ INFO:megatron.core.distributed.param_and_grad_buffer:Number of buckets for gradi Params for bucket 1 (206045184 elements, 206045184 padded size): module.output_layer.weight Params for bucket 2 (42217472 elements, 42217472 padded size): - module.decoder.layers.15.self_attention.linear_qkv.layer_norm_weight + module.decoder.final_layernorm.weight module.decoder.layers.15.mlp.router.weight + module.decoder.layers.15.self_attention.linear_qkv.layer_norm_weight module.decoder.layers.15.self_attention.linear_proj.weight - module.decoder.final_layernorm.weight - module.decoder.layers.15.self_attention.linear_qkv.weight module.decoder.layers.15.pre_mlp_layernorm.weight + module.decoder.layers.15.self_attention.linear_qkv.weight Params for bucket 3 (42213376 elements, 42213376 padded size): module.decoder.layers.14.mlp.router.weight module.decoder.layers.14.self_attention.linear_qkv.layer_norm_weight @@ -681,16 +681,16 @@ Params for bucket 3 (42213376 elements, 42213376 padded size): module.decoder.layers.14.pre_mlp_layernorm.weight Params for bucket 4 (42213376 elements, 42213376 padded size): module.decoder.layers.13.mlp.router.weight - module.decoder.layers.13.self_attention.linear_qkv.layer_norm_weight module.decoder.layers.13.self_attention.linear_proj.weight + module.decoder.layers.13.self_attention.linear_qkv.layer_norm_weight module.decoder.layers.13.self_attention.linear_qkv.weight module.decoder.layers.13.pre_mlp_layernorm.weight Params for bucket 5 (42213376 elements, 42213376 padded size): module.decoder.layers.12.mlp.router.weight module.decoder.layers.12.self_attention.linear_proj.weight module.decoder.layers.12.self_attention.linear_qkv.layer_norm_weight - module.decoder.layers.12.pre_mlp_layernorm.weight module.decoder.layers.12.self_attention.linear_qkv.weight + module.decoder.layers.12.pre_mlp_layernorm.weight Params for bucket 6 (42213376 elements, 42213376 padded size): module.decoder.layers.11.mlp.router.weight module.decoder.layers.11.self_attention.linear_proj.weight @@ -698,17 +698,17 @@ Params for bucket 6 (42213376 elements, 42213376 padded size): module.decoder.layers.11.self_attention.linear_qkv.weight module.decoder.layers.11.pre_mlp_layernorm.weight Params for bucket 7 (42213376 elements, 42213376 padded size): + module.decoder.layers.10.self_attention.linear_qkv.weight + module.decoder.layers.10.pre_mlp_layernorm.weight module.decoder.layers.10.mlp.router.weight module.decoder.layers.10.self_attention.linear_proj.weight module.decoder.layers.10.self_attention.linear_qkv.layer_norm_weight - module.decoder.layers.10.pre_mlp_layernorm.weight - module.decoder.layers.10.self_attention.linear_qkv.weight Params for bucket 8 (42213376 elements, 42213376 padded size): - module.decoder.layers.9.mlp.router.weight - module.decoder.layers.9.self_attention.linear_qkv.layer_norm_weight - module.decoder.layers.9.self_attention.linear_proj.weight module.decoder.layers.9.self_attention.linear_qkv.weight module.decoder.layers.9.pre_mlp_layernorm.weight + module.decoder.layers.9.mlp.router.weight + module.decoder.layers.9.self_attention.linear_proj.weight + module.decoder.layers.9.self_attention.linear_qkv.layer_norm_weight Params for bucket 9 (42213376 elements, 42213376 padded size): module.decoder.layers.8.self_attention.linear_qkv.weight module.decoder.layers.8.pre_mlp_layernorm.weight @@ -716,41 +716,41 @@ Params for bucket 9 (42213376 elements, 42213376 padded size): module.decoder.layers.8.self_attention.linear_qkv.layer_norm_weight module.decoder.layers.8.self_attention.linear_proj.weight Params for bucket 10 (42213376 elements, 42213376 padded size): - module.decoder.layers.7.pre_mlp_layernorm.weight module.decoder.layers.7.self_attention.linear_qkv.weight + module.decoder.layers.7.pre_mlp_layernorm.weight module.decoder.layers.7.mlp.router.weight - module.decoder.layers.7.self_attention.linear_proj.weight module.decoder.layers.7.self_attention.linear_qkv.layer_norm_weight + module.decoder.layers.7.self_attention.linear_proj.weight Params for bucket 11 (42213376 elements, 42213376 padded size): - module.decoder.layers.6.self_attention.linear_qkv.weight - module.decoder.layers.6.pre_mlp_layernorm.weight module.decoder.layers.6.mlp.router.weight - module.decoder.layers.6.self_attention.linear_qkv.layer_norm_weight module.decoder.layers.6.self_attention.linear_proj.weight + module.decoder.layers.6.self_attention.linear_qkv.layer_norm_weight + module.decoder.layers.6.self_attention.linear_qkv.weight + module.decoder.layers.6.pre_mlp_layernorm.weight Params for bucket 12 (42213376 elements, 42213376 padded size): - module.decoder.layers.5.self_attention.linear_proj.weight - module.decoder.layers.5.pre_mlp_layernorm.weight - module.decoder.layers.5.self_attention.linear_qkv.weight module.decoder.layers.5.mlp.router.weight module.decoder.layers.5.self_attention.linear_qkv.layer_norm_weight + module.decoder.layers.5.self_attention.linear_qkv.weight + module.decoder.layers.5.self_attention.linear_proj.weight + module.decoder.layers.5.pre_mlp_layernorm.weight Params for bucket 13 (42213376 elements, 42213376 padded size): - module.decoder.layers.4.mlp.router.weight - module.decoder.layers.4.self_attention.linear_qkv.layer_norm_weight - module.decoder.layers.4.self_attention.linear_proj.weight module.decoder.layers.4.self_attention.linear_qkv.weight module.decoder.layers.4.pre_mlp_layernorm.weight + module.decoder.layers.4.mlp.router.weight + module.decoder.layers.4.self_attention.linear_proj.weight + module.decoder.layers.4.self_attention.linear_qkv.layer_norm_weight Params for bucket 14 (42213376 elements, 42213376 padded size): module.decoder.layers.3.mlp.router.weight - module.decoder.layers.3.self_attention.linear_proj.weight module.decoder.layers.3.self_attention.linear_qkv.layer_norm_weight - module.decoder.layers.3.pre_mlp_layernorm.weight + module.decoder.layers.3.self_attention.linear_proj.weight module.decoder.layers.3.self_attention.linear_qkv.weight + module.decoder.layers.3.pre_mlp_layernorm.weight Params for bucket 15 (42213376 elements, 42213376 padded size): module.decoder.layers.2.mlp.router.weight - module.decoder.layers.2.self_attention.linear_qkv.layer_norm_weight module.decoder.layers.2.self_attention.linear_proj.weight - module.decoder.layers.2.self_attention.linear_qkv.weight + module.decoder.layers.2.self_attention.linear_qkv.layer_norm_weight module.decoder.layers.2.pre_mlp_layernorm.weight + module.decoder.layers.2.self_attention.linear_qkv.weight Params for bucket 16 (42213376 elements, 42213376 padded size): module.decoder.layers.1.mlp.router.weight module.decoder.layers.1.self_attention.linear_proj.weight @@ -758,21 +758,21 @@ Params for bucket 16 (42213376 elements, 42213376 padded size): module.decoder.layers.1.self_attention.linear_qkv.weight module.decoder.layers.1.pre_mlp_layernorm.weight Params for bucket 17 (42213376 elements, 42213376 padded size): - module.decoder.layers.0.pre_mlp_layernorm.weight - module.decoder.layers.0.self_attention.linear_qkv.layer_norm_weight - module.decoder.layers.0.self_attention.linear_qkv.weight module.decoder.layers.0.mlp.router.weight + module.decoder.layers.0.self_attention.linear_qkv.layer_norm_weight module.decoder.layers.0.self_attention.linear_proj.weight + module.decoder.layers.0.self_attention.linear_qkv.weight + module.decoder.layers.0.pre_mlp_layernorm.weight Params for bucket 18 (206045184 elements, 206045184 padded size): module.embedding.word_embeddings.weight INFO:megatron.core.distributed.param_and_grad_buffer:Number of buckets for gradient all-reduce / reduce-scatter: 64 Params for bucket 1 (44040192 elements, 44040192 padded size): - module.decoder.layers.15.mlp.experts.linear_fc2.weight4 - module.decoder.layers.15.mlp.experts.linear_fc2.weight6 - module.decoder.layers.15.mlp.experts.linear_fc2.weight2 module.decoder.layers.15.mlp.experts.linear_fc2.weight7 module.decoder.layers.15.mlp.experts.linear_fc2.weight3 module.decoder.layers.15.mlp.experts.linear_fc2.weight5 + module.decoder.layers.15.mlp.experts.linear_fc2.weight6 + module.decoder.layers.15.mlp.experts.linear_fc2.weight4 + module.decoder.layers.15.mlp.experts.linear_fc2.weight2 Params for bucket 2 (44040192 elements, 44040192 padded size): module.decoder.layers.15.mlp.experts.linear_fc2.weight1 module.decoder.layers.15.mlp.experts.linear_fc2.weight0 @@ -796,8 +796,8 @@ Params for bucket 5 (44040192 elements, 44040192 padded size): Params for bucket 6 (44040192 elements, 44040192 padded size): module.decoder.layers.14.mlp.experts.linear_fc2.weight1 module.decoder.layers.14.mlp.experts.linear_fc2.weight0 - module.decoder.layers.14.mlp.experts.linear_fc1.weight7 module.decoder.layers.14.mlp.experts.linear_fc1.weight6 + module.decoder.layers.14.mlp.experts.linear_fc1.weight7 Params for bucket 7 (44040192 elements, 44040192 padded size): module.decoder.layers.14.mlp.experts.linear_fc1.weight5 module.decoder.layers.14.mlp.experts.linear_fc1.weight4 @@ -828,20 +828,20 @@ Params for bucket 12 (44040192 elements, 44040192 padded size): module.decoder.layers.13.mlp.experts.linear_fc1.weight1 Params for bucket 13 (44040192 elements, 44040192 padded size): module.decoder.layers.12.mlp.experts.linear_fc2.weight7 - module.decoder.layers.12.mlp.experts.linear_fc2.weight5 module.decoder.layers.12.mlp.experts.linear_fc2.weight3 - module.decoder.layers.12.mlp.experts.linear_fc2.weight6 + module.decoder.layers.12.mlp.experts.linear_fc2.weight5 module.decoder.layers.12.mlp.experts.linear_fc2.weight4 + module.decoder.layers.12.mlp.experts.linear_fc2.weight6 module.decoder.layers.12.mlp.experts.linear_fc2.weight2 Params for bucket 14 (44040192 elements, 44040192 padded size): module.decoder.layers.12.mlp.experts.linear_fc2.weight1 - module.decoder.layers.12.mlp.experts.linear_fc1.weight6 - module.decoder.layers.12.mlp.experts.linear_fc2.weight0 module.decoder.layers.12.mlp.experts.linear_fc1.weight7 + module.decoder.layers.12.mlp.experts.linear_fc2.weight0 + module.decoder.layers.12.mlp.experts.linear_fc1.weight6 Params for bucket 15 (44040192 elements, 44040192 padded size): module.decoder.layers.12.mlp.experts.linear_fc1.weight5 - module.decoder.layers.12.mlp.experts.linear_fc1.weight3 module.decoder.layers.12.mlp.experts.linear_fc1.weight4 + module.decoder.layers.12.mlp.experts.linear_fc1.weight3 Params for bucket 16 (44040192 elements, 44040192 padded size): module.decoder.layers.12.mlp.experts.linear_fc1.weight2 module.decoder.layers.12.mlp.experts.linear_fc1.weight0 @@ -850,13 +850,13 @@ Params for bucket 17 (44040192 elements, 44040192 padded size): module.decoder.layers.11.mlp.experts.linear_fc2.weight7 module.decoder.layers.11.mlp.experts.linear_fc2.weight5 module.decoder.layers.11.mlp.experts.linear_fc2.weight3 - module.decoder.layers.11.mlp.experts.linear_fc2.weight4 module.decoder.layers.11.mlp.experts.linear_fc2.weight6 + module.decoder.layers.11.mlp.experts.linear_fc2.weight4 module.decoder.layers.11.mlp.experts.linear_fc2.weight2 Params for bucket 18 (44040192 elements, 44040192 padded size): module.decoder.layers.11.mlp.experts.linear_fc2.weight1 - module.decoder.layers.11.mlp.experts.linear_fc1.weight7 module.decoder.layers.11.mlp.experts.linear_fc1.weight6 + module.decoder.layers.11.mlp.experts.linear_fc1.weight7 module.decoder.layers.11.mlp.experts.linear_fc2.weight0 Params for bucket 19 (44040192 elements, 44040192 padded size): module.decoder.layers.11.mlp.experts.linear_fc1.weight5 @@ -867,37 +867,37 @@ Params for bucket 20 (44040192 elements, 44040192 padded size): module.decoder.layers.11.mlp.experts.linear_fc1.weight1 module.decoder.layers.11.mlp.experts.linear_fc1.weight0 Params for bucket 21 (44040192 elements, 44040192 padded size): - module.decoder.layers.10.mlp.experts.linear_fc2.weight7 - module.decoder.layers.10.mlp.experts.linear_fc2.weight3 - module.decoder.layers.10.mlp.experts.linear_fc2.weight5 module.decoder.layers.10.mlp.experts.linear_fc2.weight6 module.decoder.layers.10.mlp.experts.linear_fc2.weight4 module.decoder.layers.10.mlp.experts.linear_fc2.weight2 + module.decoder.layers.10.mlp.experts.linear_fc2.weight7 + module.decoder.layers.10.mlp.experts.linear_fc2.weight5 + module.decoder.layers.10.mlp.experts.linear_fc2.weight3 Params for bucket 22 (44040192 elements, 44040192 padded size): module.decoder.layers.10.mlp.experts.linear_fc2.weight1 + module.decoder.layers.10.mlp.experts.linear_fc1.weight6 module.decoder.layers.10.mlp.experts.linear_fc2.weight0 module.decoder.layers.10.mlp.experts.linear_fc1.weight7 - module.decoder.layers.10.mlp.experts.linear_fc1.weight6 Params for bucket 23 (44040192 elements, 44040192 padded size): module.decoder.layers.10.mlp.experts.linear_fc1.weight5 module.decoder.layers.10.mlp.experts.linear_fc1.weight3 module.decoder.layers.10.mlp.experts.linear_fc1.weight4 Params for bucket 24 (44040192 elements, 44040192 padded size): module.decoder.layers.10.mlp.experts.linear_fc1.weight2 - module.decoder.layers.10.mlp.experts.linear_fc1.weight0 module.decoder.layers.10.mlp.experts.linear_fc1.weight1 + module.decoder.layers.10.mlp.experts.linear_fc1.weight0 Params for bucket 25 (44040192 elements, 44040192 padded size): + module.decoder.layers.9.mlp.experts.linear_fc2.weight6 + module.decoder.layers.9.mlp.experts.linear_fc2.weight4 + module.decoder.layers.9.mlp.experts.linear_fc2.weight2 module.decoder.layers.9.mlp.experts.linear_fc2.weight7 module.decoder.layers.9.mlp.experts.linear_fc2.weight3 module.decoder.layers.9.mlp.experts.linear_fc2.weight5 - module.decoder.layers.9.mlp.experts.linear_fc2.weight4 - module.decoder.layers.9.mlp.experts.linear_fc2.weight6 - module.decoder.layers.9.mlp.experts.linear_fc2.weight2 Params for bucket 26 (44040192 elements, 44040192 padded size): module.decoder.layers.9.mlp.experts.linear_fc2.weight1 + module.decoder.layers.9.mlp.experts.linear_fc1.weight6 module.decoder.layers.9.mlp.experts.linear_fc2.weight0 module.decoder.layers.9.mlp.experts.linear_fc1.weight7 - module.decoder.layers.9.mlp.experts.linear_fc1.weight6 Params for bucket 27 (44040192 elements, 44040192 padded size): module.decoder.layers.9.mlp.experts.linear_fc1.weight5 module.decoder.layers.9.mlp.experts.linear_fc1.weight3 @@ -907,92 +907,92 @@ Params for bucket 28 (44040192 elements, 44040192 padded size): module.decoder.layers.9.mlp.experts.linear_fc1.weight0 module.decoder.layers.9.mlp.experts.linear_fc1.weight1 Params for bucket 29 (44040192 elements, 44040192 padded size): - module.decoder.layers.8.mlp.experts.linear_fc2.weight7 - module.decoder.layers.8.mlp.experts.linear_fc2.weight5 - module.decoder.layers.8.mlp.experts.linear_fc2.weight3 - module.decoder.layers.8.mlp.experts.linear_fc2.weight6 module.decoder.layers.8.mlp.experts.linear_fc2.weight4 + module.decoder.layers.8.mlp.experts.linear_fc2.weight6 module.decoder.layers.8.mlp.experts.linear_fc2.weight2 + module.decoder.layers.8.mlp.experts.linear_fc2.weight7 + module.decoder.layers.8.mlp.experts.linear_fc2.weight3 + module.decoder.layers.8.mlp.experts.linear_fc2.weight5 Params for bucket 30 (44040192 elements, 44040192 padded size): module.decoder.layers.8.mlp.experts.linear_fc2.weight1 module.decoder.layers.8.mlp.experts.linear_fc1.weight7 - module.decoder.layers.8.mlp.experts.linear_fc2.weight0 module.decoder.layers.8.mlp.experts.linear_fc1.weight6 + module.decoder.layers.8.mlp.experts.linear_fc2.weight0 Params for bucket 31 (44040192 elements, 44040192 padded size): module.decoder.layers.8.mlp.experts.linear_fc1.weight5 module.decoder.layers.8.mlp.experts.linear_fc1.weight4 module.decoder.layers.8.mlp.experts.linear_fc1.weight3 Params for bucket 32 (44040192 elements, 44040192 padded size): module.decoder.layers.8.mlp.experts.linear_fc1.weight2 - module.decoder.layers.8.mlp.experts.linear_fc1.weight0 module.decoder.layers.8.mlp.experts.linear_fc1.weight1 + module.decoder.layers.8.mlp.experts.linear_fc1.weight0 Params for bucket 33 (44040192 elements, 44040192 padded size): - module.decoder.layers.7.mlp.experts.linear_fc2.weight7 - module.decoder.layers.7.mlp.experts.linear_fc2.weight3 - module.decoder.layers.7.mlp.experts.linear_fc2.weight5 - module.decoder.layers.7.mlp.experts.linear_fc2.weight4 module.decoder.layers.7.mlp.experts.linear_fc2.weight6 + module.decoder.layers.7.mlp.experts.linear_fc2.weight4 module.decoder.layers.7.mlp.experts.linear_fc2.weight2 + module.decoder.layers.7.mlp.experts.linear_fc2.weight7 + module.decoder.layers.7.mlp.experts.linear_fc2.weight5 + module.decoder.layers.7.mlp.experts.linear_fc2.weight3 Params for bucket 34 (44040192 elements, 44040192 padded size): module.decoder.layers.7.mlp.experts.linear_fc2.weight1 - module.decoder.layers.7.mlp.experts.linear_fc1.weight6 - module.decoder.layers.7.mlp.experts.linear_fc2.weight0 module.decoder.layers.7.mlp.experts.linear_fc1.weight7 + module.decoder.layers.7.mlp.experts.linear_fc2.weight0 + module.decoder.layers.7.mlp.experts.linear_fc1.weight6 Params for bucket 35 (44040192 elements, 44040192 padded size): module.decoder.layers.7.mlp.experts.linear_fc1.weight5 - module.decoder.layers.7.mlp.experts.linear_fc1.weight3 module.decoder.layers.7.mlp.experts.linear_fc1.weight4 + module.decoder.layers.7.mlp.experts.linear_fc1.weight3 Params for bucket 36 (44040192 elements, 44040192 padded size): module.decoder.layers.7.mlp.experts.linear_fc1.weight2 - module.decoder.layers.7.mlp.experts.linear_fc1.weight0 module.decoder.layers.7.mlp.experts.linear_fc1.weight1 + module.decoder.layers.7.mlp.experts.linear_fc1.weight0 Params for bucket 37 (44040192 elements, 44040192 padded size): - module.decoder.layers.6.mlp.experts.linear_fc2.weight4 - module.decoder.layers.6.mlp.experts.linear_fc2.weight6 - module.decoder.layers.6.mlp.experts.linear_fc2.weight2 module.decoder.layers.6.mlp.experts.linear_fc2.weight7 module.decoder.layers.6.mlp.experts.linear_fc2.weight5 module.decoder.layers.6.mlp.experts.linear_fc2.weight3 + module.decoder.layers.6.mlp.experts.linear_fc2.weight4 + module.decoder.layers.6.mlp.experts.linear_fc2.weight6 + module.decoder.layers.6.mlp.experts.linear_fc2.weight2 Params for bucket 38 (44040192 elements, 44040192 padded size): module.decoder.layers.6.mlp.experts.linear_fc2.weight1 + module.decoder.layers.6.mlp.experts.linear_fc2.weight0 module.decoder.layers.6.mlp.experts.linear_fc1.weight7 module.decoder.layers.6.mlp.experts.linear_fc1.weight6 - module.decoder.layers.6.mlp.experts.linear_fc2.weight0 Params for bucket 39 (44040192 elements, 44040192 padded size): module.decoder.layers.6.mlp.experts.linear_fc1.weight5 module.decoder.layers.6.mlp.experts.linear_fc1.weight4 module.decoder.layers.6.mlp.experts.linear_fc1.weight3 Params for bucket 40 (44040192 elements, 44040192 padded size): module.decoder.layers.6.mlp.experts.linear_fc1.weight2 - module.decoder.layers.6.mlp.experts.linear_fc1.weight1 module.decoder.layers.6.mlp.experts.linear_fc1.weight0 + module.decoder.layers.6.mlp.experts.linear_fc1.weight1 Params for bucket 41 (44040192 elements, 44040192 padded size): - module.decoder.layers.5.mlp.experts.linear_fc2.weight6 - module.decoder.layers.5.mlp.experts.linear_fc2.weight4 - module.decoder.layers.5.mlp.experts.linear_fc2.weight2 module.decoder.layers.5.mlp.experts.linear_fc2.weight7 module.decoder.layers.5.mlp.experts.linear_fc2.weight5 module.decoder.layers.5.mlp.experts.linear_fc2.weight3 + module.decoder.layers.5.mlp.experts.linear_fc2.weight4 + module.decoder.layers.5.mlp.experts.linear_fc2.weight6 + module.decoder.layers.5.mlp.experts.linear_fc2.weight2 Params for bucket 42 (44040192 elements, 44040192 padded size): module.decoder.layers.5.mlp.experts.linear_fc2.weight1 - module.decoder.layers.5.mlp.experts.linear_fc1.weight6 - module.decoder.layers.5.mlp.experts.linear_fc2.weight0 module.decoder.layers.5.mlp.experts.linear_fc1.weight7 + module.decoder.layers.5.mlp.experts.linear_fc2.weight0 + module.decoder.layers.5.mlp.experts.linear_fc1.weight6 Params for bucket 43 (44040192 elements, 44040192 padded size): module.decoder.layers.5.mlp.experts.linear_fc1.weight5 - module.decoder.layers.5.mlp.experts.linear_fc1.weight3 module.decoder.layers.5.mlp.experts.linear_fc1.weight4 + module.decoder.layers.5.mlp.experts.linear_fc1.weight3 Params for bucket 44 (44040192 elements, 44040192 padded size): module.decoder.layers.5.mlp.experts.linear_fc1.weight2 - module.decoder.layers.5.mlp.experts.linear_fc1.weight0 module.decoder.layers.5.mlp.experts.linear_fc1.weight1 + module.decoder.layers.5.mlp.experts.linear_fc1.weight0 Params for bucket 45 (44040192 elements, 44040192 padded size): - module.decoder.layers.4.mlp.experts.linear_fc2.weight7 - module.decoder.layers.4.mlp.experts.linear_fc2.weight5 - module.decoder.layers.4.mlp.experts.linear_fc2.weight3 module.decoder.layers.4.mlp.experts.linear_fc2.weight4 module.decoder.layers.4.mlp.experts.linear_fc2.weight6 module.decoder.layers.4.mlp.experts.linear_fc2.weight2 + module.decoder.layers.4.mlp.experts.linear_fc2.weight7 + module.decoder.layers.4.mlp.experts.linear_fc2.weight3 + module.decoder.layers.4.mlp.experts.linear_fc2.weight5 Params for bucket 46 (44040192 elements, 44040192 padded size): module.decoder.layers.4.mlp.experts.linear_fc2.weight1 module.decoder.layers.4.mlp.experts.linear_fc1.weight7 @@ -1007,57 +1007,57 @@ Params for bucket 48 (44040192 elements, 44040192 padded size): module.decoder.layers.4.mlp.experts.linear_fc1.weight1 module.decoder.layers.4.mlp.experts.linear_fc1.weight0 Params for bucket 49 (44040192 elements, 44040192 padded size): - module.decoder.layers.3.mlp.experts.linear_fc2.weight7 - module.decoder.layers.3.mlp.experts.linear_fc2.weight3 - module.decoder.layers.3.mlp.experts.linear_fc2.weight5 module.decoder.layers.3.mlp.experts.linear_fc2.weight6 module.decoder.layers.3.mlp.experts.linear_fc2.weight4 module.decoder.layers.3.mlp.experts.linear_fc2.weight2 + module.decoder.layers.3.mlp.experts.linear_fc2.weight7 + module.decoder.layers.3.mlp.experts.linear_fc2.weight3 + module.decoder.layers.3.mlp.experts.linear_fc2.weight5 Params for bucket 50 (44040192 elements, 44040192 padded size): module.decoder.layers.3.mlp.experts.linear_fc2.weight1 - module.decoder.layers.3.mlp.experts.linear_fc2.weight0 module.decoder.layers.3.mlp.experts.linear_fc1.weight7 + module.decoder.layers.3.mlp.experts.linear_fc2.weight0 module.decoder.layers.3.mlp.experts.linear_fc1.weight6 Params for bucket 51 (44040192 elements, 44040192 padded size): module.decoder.layers.3.mlp.experts.linear_fc1.weight5 - module.decoder.layers.3.mlp.experts.linear_fc1.weight3 module.decoder.layers.3.mlp.experts.linear_fc1.weight4 + module.decoder.layers.3.mlp.experts.linear_fc1.weight3 Params for bucket 52 (44040192 elements, 44040192 padded size): module.decoder.layers.3.mlp.experts.linear_fc1.weight2 - module.decoder.layers.3.mlp.experts.linear_fc1.weight0 module.decoder.layers.3.mlp.experts.linear_fc1.weight1 + module.decoder.layers.3.mlp.experts.linear_fc1.weight0 Params for bucket 53 (44040192 elements, 44040192 padded size): module.decoder.layers.2.mlp.experts.linear_fc2.weight7 module.decoder.layers.2.mlp.experts.linear_fc2.weight5 module.decoder.layers.2.mlp.experts.linear_fc2.weight3 - module.decoder.layers.2.mlp.experts.linear_fc2.weight4 module.decoder.layers.2.mlp.experts.linear_fc2.weight6 + module.decoder.layers.2.mlp.experts.linear_fc2.weight4 module.decoder.layers.2.mlp.experts.linear_fc2.weight2 Params for bucket 54 (44040192 elements, 44040192 padded size): module.decoder.layers.2.mlp.experts.linear_fc2.weight1 + module.decoder.layers.2.mlp.experts.linear_fc2.weight0 module.decoder.layers.2.mlp.experts.linear_fc1.weight7 module.decoder.layers.2.mlp.experts.linear_fc1.weight6 - module.decoder.layers.2.mlp.experts.linear_fc2.weight0 Params for bucket 55 (44040192 elements, 44040192 padded size): module.decoder.layers.2.mlp.experts.linear_fc1.weight5 - module.decoder.layers.2.mlp.experts.linear_fc1.weight4 module.decoder.layers.2.mlp.experts.linear_fc1.weight3 + module.decoder.layers.2.mlp.experts.linear_fc1.weight4 Params for bucket 56 (44040192 elements, 44040192 padded size): module.decoder.layers.2.mlp.experts.linear_fc1.weight2 module.decoder.layers.2.mlp.experts.linear_fc1.weight0 module.decoder.layers.2.mlp.experts.linear_fc1.weight1 Params for bucket 57 (44040192 elements, 44040192 padded size): module.decoder.layers.1.mlp.experts.linear_fc2.weight7 - module.decoder.layers.1.mlp.experts.linear_fc2.weight5 module.decoder.layers.1.mlp.experts.linear_fc2.weight3 - module.decoder.layers.1.mlp.experts.linear_fc2.weight6 + module.decoder.layers.1.mlp.experts.linear_fc2.weight5 module.decoder.layers.1.mlp.experts.linear_fc2.weight4 + module.decoder.layers.1.mlp.experts.linear_fc2.weight6 module.decoder.layers.1.mlp.experts.linear_fc2.weight2 Params for bucket 58 (44040192 elements, 44040192 padded size): module.decoder.layers.1.mlp.experts.linear_fc2.weight1 + module.decoder.layers.1.mlp.experts.linear_fc1.weight7 module.decoder.layers.1.mlp.experts.linear_fc1.weight6 module.decoder.layers.1.mlp.experts.linear_fc2.weight0 - module.decoder.layers.1.mlp.experts.linear_fc1.weight7 Params for bucket 59 (44040192 elements, 44040192 padded size): module.decoder.layers.1.mlp.experts.linear_fc1.weight5 module.decoder.layers.1.mlp.experts.linear_fc1.weight3 @@ -1067,16 +1067,16 @@ Params for bucket 60 (44040192 elements, 44040192 padded size): module.decoder.layers.1.mlp.experts.linear_fc1.weight0 module.decoder.layers.1.mlp.experts.linear_fc1.weight1 Params for bucket 61 (44040192 elements, 44040192 padded size): - module.decoder.layers.0.mlp.experts.linear_fc2.weight6 module.decoder.layers.0.mlp.experts.linear_fc2.weight4 + module.decoder.layers.0.mlp.experts.linear_fc2.weight6 module.decoder.layers.0.mlp.experts.linear_fc2.weight2 module.decoder.layers.0.mlp.experts.linear_fc2.weight7 module.decoder.layers.0.mlp.experts.linear_fc2.weight5 module.decoder.layers.0.mlp.experts.linear_fc2.weight3 Params for bucket 62 (44040192 elements, 44040192 padded size): module.decoder.layers.0.mlp.experts.linear_fc2.weight1 - module.decoder.layers.0.mlp.experts.linear_fc1.weight6 module.decoder.layers.0.mlp.experts.linear_fc2.weight0 + module.decoder.layers.0.mlp.experts.linear_fc1.weight6 module.decoder.layers.0.mlp.experts.linear_fc1.weight7 Params for bucket 63 (44040192 elements, 44040192 padded size): module.decoder.layers.0.mlp.experts.linear_fc1.weight5 @@ -1086,9 +1086,9 @@ Params for bucket 64 (44040192 elements, 44040192 padded size): module.decoder.layers.0.mlp.experts.linear_fc1.weight2 module.decoder.layers.0.mlp.experts.linear_fc1.weight1 module.decoder.layers.0.mlp.experts.linear_fc1.weight0 -INFO:megatron.core.optimizer:Setting up optimizer with config OptimizerConfig(optimizer='adam', lr=0.0001, min_lr=1e-05, decoupled_lr=None, decoupled_min_lr=None, weight_decay=0.1, fp8_recipe='mxfp8', fp16=False, bf16=True, reuse_grad_buf_for_mxfp8_param_ag=False, params_dtype=torch.bfloat16, use_precision_aware_optimizer=False, store_param_remainders=True, main_grads_dtype=torch.float32, main_params_dtype=torch.float32, exp_avg_dtype=torch.float32, exp_avg_sq_dtype=torch.float32, loss_scale=None, initial_loss_scale=4294967296, min_loss_scale=1.0, loss_scale_window=1000, hysteresis=2, adam_beta1=0.9, adam_beta2=0.999, adam_eps=1e-08, sgd_momentum=0.9, use_distributed_optimizer=True, overlap_param_gather_with_optimizer_step=False, optimizer_cpu_offload=False, optimizer_offload_fraction=1.0, use_torch_optimizer_for_cpu_offload=False, overlap_cpu_optimizer_d2h_h2d=False, pin_cpu_grads=True, pin_cpu_params=True, clip_grad=1.0, log_num_zeros_in_grad=False, barrier_with_L1_time=True, timers=, config_logger_dir='') +INFO:megatron.core.optimizer:Setting up optimizer with config OptimizerConfig(optimizer='adam', lr=0.0001, min_lr=1e-05, decoupled_lr=None, decoupled_min_lr=None, weight_decay=0.1, fp8_recipe='mxfp8', fp16=False, bf16=True, reuse_grad_buf_for_mxfp8_param_ag=False, params_dtype=torch.bfloat16, use_precision_aware_optimizer=False, store_param_remainders=True, main_grads_dtype=torch.float32, main_params_dtype=torch.float32, exp_avg_dtype=torch.float32, exp_avg_sq_dtype=torch.float32, loss_scale=None, initial_loss_scale=4294967296, min_loss_scale=1.0, loss_scale_window=1000, hysteresis=2, adam_beta1=0.9, adam_beta2=0.999, adam_eps=1e-08, sgd_momentum=0.9, use_distributed_optimizer=True, overlap_param_gather_with_optimizer_step=False, optimizer_cpu_offload=False, optimizer_offload_fraction=1.0, use_torch_optimizer_for_cpu_offload=False, overlap_cpu_optimizer_d2h_h2d=False, pin_cpu_grads=True, pin_cpu_params=True, clip_grad=1.0, log_num_zeros_in_grad=False, barrier_with_L1_time=True, timers=, config_logger_dir='') INFO:megatron.core.optimizer_param_scheduler:> learning rate decay style: cosine -[after model, optimizer, and learning rate scheduler are built] datetime: 2026-09-17 03:30:19 +[after model, optimizer, and learning rate scheduler are built] datetime: 2026-09-21 05:29:47 > building train, validation, and test datasets ... > datasets target sizes (minimum size): train: 160 @@ -1096,7 +1096,7 @@ INFO:megatron.core.optimizer_param_scheduler:> learning rate decay style: cosine test: 0 INFO:megatron.core.datasets.blended_megatron_dataset_config:Let split_matrix = [(0, 1.0), None, None] > building train, validation, and test datasets for GPT ... -INFO:megatron.core.datasets.blended_megatron_dataset_builder:Building GPTDataset splits with sizes=(160, 0, 0) and config=GPTDatasetConfig(random_seed=1234, sequence_length=4096, blend=(['/workspace/data/wikitext/wikitext103_gpt2_text_document'], None), blend_per_split=None, split='100,0,0', split_matrix=[(0, 1.0), None, None], num_dataset_builder_threads=1, path_to_cache=None, mmap_bin_files=True, mock=False, tokenizer=, mid_level_dataset_surplus=0.005, reset_position_ids=False, reset_attention_mask=False, eod_mask_loss=False, create_attention_mask=True, drop_last_partial_validation_sequence=True, add_extra_token_to_sequence=True, object_storage_cache_path=None) +INFO:megatron.core.datasets.blended_megatron_dataset_builder:Building GPTDataset splits with sizes=(160, 0, 0) and config=GPTDatasetConfig(random_seed=1234, sequence_length=4096, blend=(['/workspace/data/wikitext/wikitext103_gpt2_text_document'], None), blend_per_split=None, split='100,0,0', split_matrix=[(0, 1.0), None, None], num_dataset_builder_threads=1, path_to_cache=None, mmap_bin_files=True, mock=False, tokenizer=, mid_level_dataset_surplus=0.005, reset_position_ids=False, reset_attention_mask=False, eod_mask_loss=False, create_attention_mask=True, drop_last_partial_validation_sequence=True, add_extra_token_to_sequence=True, object_storage_cache_path=None) INFO:megatron.core.datasets.indexed_dataset:Load the _IndexReader from /workspace/data/wikitext/wikitext103_gpt2_text_document.idx INFO:megatron.core.datasets.indexed_dataset: Extract the sequence lengths INFO:megatron.core.datasets.indexed_dataset: Extract the sequence pointers @@ -1109,14 +1109,14 @@ INFO:megatron.core.datasets.gpt_dataset: Load the sample index from 68684a515e11 INFO:megatron.core.datasets.gpt_dataset: Load the shuffle index from 68684a515e118253a8572c4517cf2ffe-GPTDataset-train-shuffle_index.npy INFO:megatron.core.datasets.gpt_dataset:> total number of samples: 29073 > finished creating GPT datasets ... -[after dataloaders are built] datetime: 2026-09-17 03:30:19 +[after dataloaders are built] datetime: 2026-09-21 05:29:47 done with setup ... (min, max) time across ranks (ms): - model-and-optimizer-setup ......................: (3225.24, 3230.28) - train/valid/test-data-iterators-setup ..........: (108.05, 137.65) + model-and-optimizer-setup ......................: (3612.13, 3618.21) + train/valid/test-data-iterators-setup ..........: (108.00, 131.25) training ... Overwriting rerun_state_machine.current_iteration from -1 to 0... -[before the start of training step] datetime: 2026-09-17 03:30:19 +[before the start of training step] datetime: 2026-09-21 05:29:47 /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. @@ -1139,12 +1139,12 @@ Overwriting rerun_state_machine.current_iteration from -1 to 0... warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") -/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. - warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. @@ -1167,18 +1167,14 @@ Overwriting rerun_state_machine.current_iteration from -1 to 0... warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") -/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. - warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") -/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. - warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") -/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. - warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. @@ -1187,6 +1183,10 @@ Overwriting rerun_state_machine.current_iteration from -1 to 0... warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. + warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") +/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. + warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. @@ -1245,39 +1245,39 @@ Overwriting rerun_state_machine.current_iteration from -1 to 0... warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") /workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") - [2026-09-17 03:30:50] iteration 1/ 10 | consumed samples: 16 | elapsed time per iteration (ms): 31014.6 | throughput per GPU (TFLOP/s/GPU): 2.9 | learning rate: 2.000000E-05 | global batch size: 16 | lm loss: 1.102587E+01 | loss scale: 1.0 | grad norm: 46.862 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 9216 | min safe capacity: 1.1250 | + [2026-09-21 05:30:20] iteration 1/ 10 | consumed samples: 16 | elapsed time per iteration (ms): 32858.0 | throughput per GPU (TFLOP/s/GPU): 2.8 | learning rate: 2.000000E-05 | global batch size: 16 | lm loss: 1.102587E+01 | loss scale: 1.0 | grad norm: 46.865 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 9216 | min safe capacity: 1.1250 | Number of parameters in transformer block in billions: 23.22 Number of parameters in embedding layers in billions: 0.41 Total number of parameters in billions: 23.63 Number of parameters in most loaded shard in billions: 23.6320 Theoretical memory footprints: weight and optimizer=169029.43 MB -[Rank 0] (after 1 iterations) memory (MB) | allocated: 58124.50390625 | max allocated: 64453.58203125 | reserved: 80694.0 | max reserved: 80694.0 - [2026-09-17 03:30:51] iteration 2/ 10 | consumed samples: 32 | elapsed time per iteration (ms): 948.0 | throughput per GPU (TFLOP/s/GPU): 95.9 | learning rate: 4.000000E-05 | global batch size: 16 | lm loss: 1.101135E+01 | loss scale: 1.0 | grad norm: 47.659 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 9216 | min safe capacity: 1.1250 | - [2026-09-17 03:30:52] iteration 3/ 10 | consumed samples: 48 | elapsed time per iteration (ms): 389.4 | throughput per GPU (TFLOP/s/GPU): 233.5 | learning rate: 6.000000E-05 | global batch size: 16 | lm loss: 1.024894E+01 | loss scale: 1.0 | grad norm: 187.969 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 9216 | min safe capacity: 1.1250 | +[Rank 0] (after 1 iterations) memory (MB) | allocated: 58124.75390625 | max allocated: 64480.37890625 | reserved: 80670.0 | max reserved: 80670.0 + [2026-09-21 05:30:21] iteration 2/ 10 | consumed samples: 32 | elapsed time per iteration (ms): 565.1 | throughput per GPU (TFLOP/s/GPU): 160.9 | learning rate: 4.000000E-05 | global batch size: 16 | lm loss: 1.101135E+01 | loss scale: 1.0 | grad norm: 47.659 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 9216 | min safe capacity: 1.1250 | + [2026-09-21 05:30:21] iteration 3/ 10 | consumed samples: 48 | elapsed time per iteration (ms): 390.3 | throughput per GPU (TFLOP/s/GPU): 233.0 | learning rate: 6.000000E-05 | global batch size: 16 | lm loss: 1.024875E+01 | loss scale: 1.0 | grad norm: 187.935 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 9216 | min safe capacity: 1.1250 | Capture CUDA graph for training!!! CUDA graph capture done!!! - [2026-09-17 03:30:56] iteration 4/ 10 | consumed samples: 64 | elapsed time per iteration (ms): 3909.6 | throughput per GPU (TFLOP/s/GPU): 23.3 | learning rate: 8.000000E-05 | global batch size: 16 | lm loss: 1.308593E+01 | loss scale: 1.0 | grad norm: 44.378 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 9216 | min safe capacity: 1.1250 | - [2026-09-17 03:30:56] iteration 5/ 10 | consumed samples: 80 | elapsed time per iteration (ms): 250.6 | throughput per GPU (TFLOP/s/GPU): 362.9 | learning rate: 1.000000E-04 | global batch size: 16 | lm loss: 1.040534E+01 | loss scale: 1.0 | grad norm: 20.456 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 9216 | min safe capacity: 1.1250 | - [2026-09-17 03:30:56] iteration 6/ 10 | consumed samples: 96 | elapsed time per iteration (ms): 251.5 | throughput per GPU (TFLOP/s/GPU): 361.6 | learning rate: 7.750000E-05 | global batch size: 16 | lm loss: 1.022187E+01 | loss scale: 1.0 | grad norm: 31.356 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 9216 | min safe capacity: 1.1250 | - [2026-09-17 03:30:59] iteration 7/ 10 | consumed samples: 112 | elapsed time per iteration (ms): 3070.8 | throughput per GPU (TFLOP/s/GPU): 29.6 | learning rate: 3.250000E-05 | global batch size: 16 | lm loss: 9.913305E+00 | loss scale: 1.0 | grad norm: 9.749 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 9216 | min safe capacity: 1.1250 | - [2026-09-17 03:30:59] iteration 8/ 10 | consumed samples: 128 | elapsed time per iteration (ms): 270.2 | throughput per GPU (TFLOP/s/GPU): 336.6 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 9.461720E+00 | loss scale: 1.0 | grad norm: 8.259 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 9216 | min safe capacity: 1.1250 | + [2026-09-21 05:30:25] iteration 4/ 10 | consumed samples: 64 | elapsed time per iteration (ms): 3753.4 | throughput per GPU (TFLOP/s/GPU): 24.2 | learning rate: 8.000000E-05 | global batch size: 16 | lm loss: 1.308636E+01 | loss scale: 1.0 | grad norm: 44.344 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 9216 | min safe capacity: 1.1250 | + [2026-09-21 05:30:25] iteration 5/ 10 | consumed samples: 80 | elapsed time per iteration (ms): 250.6 | throughput per GPU (TFLOP/s/GPU): 362.9 | learning rate: 1.000000E-04 | global batch size: 16 | lm loss: 1.040598E+01 | loss scale: 1.0 | grad norm: 20.458 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 9216 | min safe capacity: 1.1250 | + [2026-09-21 05:30:26] iteration 6/ 10 | consumed samples: 96 | elapsed time per iteration (ms): 251.7 | throughput per GPU (TFLOP/s/GPU): 361.4 | learning rate: 7.750000E-05 | global batch size: 16 | lm loss: 1.022778E+01 | loss scale: 1.0 | grad norm: 31.694 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 9216 | min safe capacity: 1.1250 | + [2026-09-21 05:30:29] iteration 7/ 10 | consumed samples: 112 | elapsed time per iteration (ms): 3026.7 | throughput per GPU (TFLOP/s/GPU): 30.0 | learning rate: 3.250000E-05 | global batch size: 16 | lm loss: 9.919262E+00 | loss scale: 1.0 | grad norm: 9.716 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 9216 | min safe capacity: 1.1250 | + [2026-09-21 05:30:29] iteration 8/ 10 | consumed samples: 128 | elapsed time per iteration (ms): 263.7 | throughput per GPU (TFLOP/s/GPU): 344.9 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 9.460276E+00 | loss scale: 1.0 | grad norm: 8.308 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 9216 | min safe capacity: 1.1250 | Capture range started in the application. Capture range ended in the application. -Generating '/tmp/nsys-report-5702.qdstrm' +Generating '/tmp/nsys-report-c7be.qdstrm' [1/1] [0% ] G3_K2__ultraep__matched__seq4096__gbs16__cap1p1875.nsys-repProcessing events... - [1/1] [0% ] G3_K2__ultraep__matched__seq4096__gbs16__cap1p1875.nsys-rep [1/1] [0% ] G3_K2__ultraep__matched__seq4096__gbs16__cap1p1875.nsys-rep [1/1] [5% ] G3_K2__ultraep__matched__seq4096__gbs16__cap1p1875.nsys-rep [1/1] [7% ] G3_K2__ultraep__matched__seq4096__gbs16__cap1p1875.nsys-rep [1/1] [10% ] G3_K2__ultraep__matched__seq4096__gbs16__cap1p1875.nsys-rep [1/1] [12% ] G3_K2__ultraep__matched__seq4096__gbs16__cap1p1875.nsys-rep [1/1] [14% ] G3_K2__ultraep__matched__seq4096__gbs16__cap1p1875.nsys-rep [1/1] [=17% ] G3_K2__ultraep__matched__seq4096__gbs16__cap1p1875.nsys-rep [1/1] [==19% ] G3_K2__ultraep__matched__seq4096__gbs16__cap1p1875.nsys-rep [1/1] [===22% ] G3_K2__ultraep__matched__seq4096__gbs16__cap1p1875.nsys-rep [1/1] [===24% ] G3_K2__ultraep__matched__seq4096__gbs16__cap1p1875.nsys-rep [1/1] [====27% ] G3_K2__ultraep__matched__seq4096__gbs16__cap1p1875.nsys-rep [1/1] [=====29% ] G3_K2__ultraep__matched__seq4096__gbs16__cap1p1875.nsys-rep [1/1] [=====31% ] G3_K2__ultraep__matched__seq4096__gbs16__cap1p1875.nsys-rep [1/1] [======34% ] G3_K2__ultraep__matched__seq4096__gbs16__cap1p1875.nsys-rep [1/1] [=======36% ] G3_K2__ultraep__matched__seq4096__gbs16__cap1p1875.nsys-rep [1/1] [=======37% ] G3_K2__ultraep__matched__seq4096__gbs16__cap1p1875.nsys-rep [1/1] [=======38% ] G3_K2__ultraep__matched__seq4096__gbs16__cap1p1875.nsys-rep [1/1] [=======39% ] G3_K2__ultraep__matched__seq4096__gbs16__cap1p1875.nsys-rep [1/1] [========40% ] G3_K2__ultraep__matched__seq4096__gbs16__cap1p1875.nsys-rep [1/1] [========41% ] G3_K2__ultraep__matched__seq4096__gbs16__cap1p1875.nsys-rep [1/1] [========42% ] G3_K2__ultraep__matched__seq4096__gbs16__cap1p1875.nsys-rep [1/1] [=========43% ] G3_K2__ultraep__matched__seq4096__gbs16__cap1p1875.nsys-rep [1/1] [=========44% ] G3_K2__ultraep__matched__seq4096__gbs16__cap1p1875.nsys-rep [1/1] [=========45% ] G3_K2__ultraep__matched__seq4096__gbs16__cap1p1875.nsys-rep [1/1] [=========46% ] G3_K2__ultraep__matched__seq4096__gbs16__cap1p1875.nsys-rep [1/1] [==========47% ] G3_K2__ultraep__matched__seq4096__gbs16__cap1p1875.nsys-rep [1/1] [==========48% ] G3_K2__ultraep__matched__seq4096__gbs16__cap1p1875.nsys-rep [1/1] [===========50% ] G3_K2__ultraep__matched__seq4096__gbs16__cap1p1875.nsys-rep [1/1] [===========51% ] G3_K2__ultraep__matched__seq4096__gbs16__cap1p1875.nsys-rep [1/1] [===========53% ] G3_K2__ultraep__matched__seq4096__gbs16__cap1p1875.nsys-rep [1/1] [============54% ] G3_K2__ultraep__matched__seq4096__gbs16__cap1p1875.nsys-rep [1/1] [============55% ] G3_K2__ultraep__matched__seq4096__gbs16__cap1p1875.nsys-rep [1/1] [============56% ] G3_K2__ultraep__matched__seq4096__gbs16__cap1p1875.nsys-rep [1/1] [=============58% ] G3_K2__ultraep__matched__seq4096__gbs16__cap1p1875.nsys-rep [1/1] [=============59% ] G3_K2__ultraep__matched__seq4096__gbs16__cap1p1875.nsys-rep [1/1] [==============63% ] G3_K2__ultraep__matched__seq4096__gbs16__cap1p1875.nsys-rep [1/1] [========================100%] G3_K2__ultraep__matched__seq4096__gbs16__cap1p1875.nsys-rep [1/1] [========================100%] G3_K2__ultraep__matched__seq4096__gbs16__cap1p1875.nsys-rep + [1/1] [0% ] G3_K2__ultraep__matched__seq4096__gbs16__cap1p1875.nsys-rep [1/1] [0% ] G3_K2__ultraep__matched__seq4096__gbs16__cap1p1875.nsys-rep [1/1] [7% ] G3_K2__ultraep__matched__seq4096__gbs16__cap1p1875.nsys-rep [1/1] [9% ] G3_K2__ultraep__matched__seq4096__gbs16__cap1p1875.nsys-rep [1/1] [11% ] G3_K2__ultraep__matched__seq4096__gbs16__cap1p1875.nsys-rep [1/1] [14% ] G3_K2__ultraep__matched__seq4096__gbs16__cap1p1875.nsys-rep [1/1] [=16% ] G3_K2__ultraep__matched__seq4096__gbs16__cap1p1875.nsys-rep [1/1] [==19% ] G3_K2__ultraep__matched__seq4096__gbs16__cap1p1875.nsys-rep [1/1] [==21% ] G3_K2__ultraep__matched__seq4096__gbs16__cap1p1875.nsys-rep [1/1] [===23% ] G3_K2__ultraep__matched__seq4096__gbs16__cap1p1875.nsys-rep [1/1] [====26% ] G3_K2__ultraep__matched__seq4096__gbs16__cap1p1875.nsys-rep [1/1] [====28% ] G3_K2__ultraep__matched__seq4096__gbs16__cap1p1875.nsys-rep [1/1] [=====31% ] G3_K2__ultraep__matched__seq4096__gbs16__cap1p1875.nsys-rep [1/1] [======33% ] G3_K2__ultraep__matched__seq4096__gbs16__cap1p1875.nsys-rep [1/1] [======35% ] G3_K2__ultraep__matched__seq4096__gbs16__cap1p1875.nsys-rep [1/1] [=======36% ] G3_K2__ultraep__matched__seq4096__gbs16__cap1p1875.nsys-rep [1/1] [=======37% ] G3_K2__ultraep__matched__seq4096__gbs16__cap1p1875.nsys-rep [1/1] [=======38% ] G3_K2__ultraep__matched__seq4096__gbs16__cap1p1875.nsys-rep [1/1] [=======39% ] G3_K2__ultraep__matched__seq4096__gbs16__cap1p1875.nsys-rep [1/1] [========40% ] G3_K2__ultraep__matched__seq4096__gbs16__cap1p1875.nsys-rep [1/1] [========41% ] G3_K2__ultraep__matched__seq4096__gbs16__cap1p1875.nsys-rep [1/1] [========42% ] G3_K2__ultraep__matched__seq4096__gbs16__cap1p1875.nsys-rep [1/1] [=========43% ] G3_K2__ultraep__matched__seq4096__gbs16__cap1p1875.nsys-rep [1/1] [=========44% ] G3_K2__ultraep__matched__seq4096__gbs16__cap1p1875.nsys-rep [1/1] [=========45% ] G3_K2__ultraep__matched__seq4096__gbs16__cap1p1875.nsys-rep [1/1] [==========47% ] G3_K2__ultraep__matched__seq4096__gbs16__cap1p1875.nsys-rep [1/1] [==========48% ] G3_K2__ultraep__matched__seq4096__gbs16__cap1p1875.nsys-rep [1/1] [==========49% ] G3_K2__ultraep__matched__seq4096__gbs16__cap1p1875.nsys-rep [1/1] [===========51% ] G3_K2__ultraep__matched__seq4096__gbs16__cap1p1875.nsys-rep [1/1] [===========52% ] G3_K2__ultraep__matched__seq4096__gbs16__cap1p1875.nsys-rep [1/1] [===========53% ] G3_K2__ultraep__matched__seq4096__gbs16__cap1p1875.nsys-rep [1/1] [============54% ] G3_K2__ultraep__matched__seq4096__gbs16__cap1p1875.nsys-rep [1/1] [============56% ] G3_K2__ultraep__matched__seq4096__gbs16__cap1p1875.nsys-rep [1/1] [============57% ] G3_K2__ultraep__matched__seq4096__gbs16__cap1p1875.nsys-rep [1/1] [=============58% ] G3_K2__ultraep__matched__seq4096__gbs16__cap1p1875.nsys-rep [1/1] [==============62% ] G3_K2__ultraep__matched__seq4096__gbs16__cap1p1875.nsys-rep [1/1] [========================100%] G3_K2__ultraep__matched__seq4096__gbs16__cap1p1875.nsys-rep [1/1] [========================100%] G3_K2__ultraep__matched__seq4096__gbs16__cap1p1875.nsys-rep Collection stop is not allowed in this state. -W0917 03:32:40.178000 960711 torch/distributed/elastic/agent/server/api.py:719] Received 15 death signal, shutting down workers -W0917 03:32:40.179000 960711 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 960827 closing signal SIGTERM +W0921 05:32:07.016000 40554 torch/distributed/elastic/agent/server/api.py:719] Received 15 death signal, shutting down workers +W0921 05:32:07.016000 40554 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 40665 closing signal SIGTERM Generated: /workspace/granularity_next_b200/tiera_topk_trace/ultraep/tierbtrace/tierbtrace_gbs16/G3_K2__ultraep__matched__seq4096__gbs16__cap1p1875/nsys/G3_K2__ultraep__matched__seq4096__gbs16__cap1p1875.nsys-rep -W0917 03:32:40.180000 960711 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 960828 closing signal SIGTERM -W0917 03:32:40.187000 960711 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 960829 closing signal SIGTERM -W0917 03:32:40.189000 960711 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 960830 closing signal SIGTERM -W0917 03:32:40.196000 960711 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 960831 closing signal SIGTERM -W0917 03:32:40.201000 960711 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 960832 closing signal SIGTERM -W0917 03:32:40.214000 960711 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 960833 closing signal SIGTERM -W0917 03:32:40.216000 960711 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 960834 closing signal SIGTERM +W0921 05:32:07.018000 40554 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 40666 closing signal SIGTERM +W0921 05:32:07.021000 40554 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 40667 closing signal SIGTERM +W0921 05:32:07.022000 40554 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 40668 closing signal SIGTERM +W0921 05:32:07.027000 40554 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 40669 closing signal SIGTERM +W0921 05:32:07.028000 40554 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 40670 closing signal SIGTERM +W0921 05:32:07.029000 40554 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 40671 closing signal SIGTERM +W0921 05:32:07.035000 40554 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 40672 closing signal SIGTERM Traceback (most recent call last): File "/usr/local/bin/torchrun", line 8, in sys.exit(main()) @@ -1305,7 +1305,7 @@ Traceback (most recent call last): time.sleep(monitor_interval) File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/multiprocessing/api.py", line 84, in _terminate_process_handler raise SignalException(f"Process {os.getpid()} got signal: {sigval}", sigval=sigval) -torch.distributed.elastic.multiprocessing.api.SignalException: Process 960711 got signal: 15 +torch.distributed.elastic.multiprocessing.api.SignalException: Process 40554 got signal: 15 The target application terminated. One or more process it created re-parented. Waiting for termination of re-parented processes. Use the `--wait` option to modify this behavior. diff --git a/granularity_next_b200/tiera_topk_trace/ultraep/tierbtrace/tierbtrace_gbs16/G3_K8__ultraep__matched__seq4096__gbs16__cap1p09375/nsys/train.log b/granularity_next_b200/tiera_topk_trace/ultraep/tierbtrace/tierbtrace_gbs16/G3_K8__ultraep__matched__seq4096__gbs16__cap1p09375/nsys/train.log index 5a166681c85f7383bab0e1b495f2314f032e6193..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391 100644 --- a/granularity_next_b200/tiera_topk_trace/ultraep/tierbtrace/tierbtrace_gbs16/G3_K8__ultraep__matched__seq4096__gbs16__cap1p09375/nsys/train.log +++ b/granularity_next_b200/tiera_topk_trace/ultraep/tierbtrace/tierbtrace_gbs16/G3_K8__ultraep__matched__seq4096__gbs16__cap1p09375/nsys/train.log @@ -1,1312 +0,0 @@ -============================================================== - G3_K8 / ultraep (slots/rank=8 policy=matched) - E=64 K=8 H'=1792 m=8 - L=16 seq=4096 gbs=16 (2 microbatches) - cap_mode=static cap=1.09375 cg=1 recompute=none - memory_dump=none snapshots=/workspace/granularity_next_b200/tiera_topk_trace/ultraep/tierbtrace/tierbtrace_gbs16/G3_K8__ultraep__matched__seq4096__gbs16__cap1p09375/nsys/memory_snapshots/snapshot.pickle - repo=/workspace/Megatron-UltraEP @ 03d57ef7 -============================================================== -W0917 03:33:05.052000 969559 torch/distributed/run.py:766] -W0917 03:33:05.052000 969559 torch/distributed/run.py:766] ***************************************** -W0917 03:33:05.052000 969559 torch/distributed/run.py:766] Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed. -W0917 03:33:05.052000 969559 torch/distributed/run.py:766] ***************************************** -/workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled - warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled") -using world size: 8, data-parallel size: 8, context-parallel size: 1, hierarchical context-parallel sizes: None, tensor-model-parallel size: 1, pipeline-model-parallel size: 1 -WARNING: overriding default arguments for tokenizer_type:GPT2BPETokenizer with tokenizer_type:GPT2BPETokenizer -Number of virtual stages per pipeline stage: None -accumulate and all-reduce gradients in fp32 for bfloat16 data type. -using torch.bfloat16 for parameters ... -/workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled - warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled") ------------------------- arguments ------------------------ - account_for_embedding_in_pipeline_split ......... False - account_for_loss_in_pipeline_split .............. False - accumulate_allreduce_grads_in_fp32 .............. True - adam_beta1 ...................................... 0.9 - adam_beta2 ...................................... 0.999 - adam_eps ........................................ 1e-08 - add_bias_linear ................................. False - add_position_embedding .......................... False - add_qkv_bias .................................... False - adlr_autoresume ................................. False - adlr_autoresume_interval ........................ 1000 - align_grad_reduce ............................... True - align_param_gather .............................. False - app_tag_run_name ................................ None - app_tag_run_version ............................. 0.0.0 - apply_layernorm_1p .............................. False - apply_query_key_layer_scaling ................... False - apply_residual_connection_post_layernorm ........ False - apply_rope_fusion ............................... True - async_save ...................................... None - async_tensor_model_parallel_allreduce ........... True - attention_backend ............................... AttnBackend.auto - attention_dropout ............................... 0.0 - attention_softmax_in_fp32 ....................... False - auto_detect_ckpt_format ......................... False - barrier_with_L1_time ............................ True - bert_binary_head ................................ True - bert_embedder_type .............................. megatron - bert_load ....................................... None - bf16 ............................................ True - bias_dropout_fusion ............................. True - bias_gelu_fusion ................................ False - bias_swiglu_fusion .............................. True - biencoder_projection_dim ........................ 0 - biencoder_shared_query_context_model ............ False - block_data_path ................................. None - calc_ft_timeouts ................................ False - calculate_per_token_loss ........................ False - check_for_large_grads ........................... False - check_for_nan_in_loss_and_grad .................. False - check_for_spiky_loss ............................ False - check_weight_hash_across_dp_replicas_interval ... None - ckpt_assume_constant_structure .................. False - ckpt_convert_format ............................. None - ckpt_convert_save ............................... None - ckpt_convert_update_legacy_dist_opt_format ...... False - ckpt_format ..................................... torch_dist - ckpt_fully_parallel_load ........................ False - ckpt_fully_parallel_save ........................ True - ckpt_fully_parallel_save_deprecated ............. False - ckpt_step ....................................... None - classes_fraction ................................ 1.0 - clip_grad ....................................... 1.0 - clone_scatter_output_in_embedding ............... True - config_logger_dir ............................... - consumed_train_samples .......................... 0 - consumed_valid_samples .......................... 0 - context_parallel_size ........................... 1 - cp_comm_type .................................... ['p2p'] - create_attention_mask_in_dataloader ............. True - cross_entropy_fusion_impl ....................... native - cross_entropy_loss_fusion ....................... False - cuda_graph_scope ................................ full_iteration - cuda_graph_warmup_steps ......................... 3 - data_args_path .................................. None - data_cache_path ................................. None - data_parallel_random_init ....................... False - data_parallel_sharding_strategy ................. no_shard - data_parallel_size .............................. 8 - data_path ....................................... ['/workspace/data/wikitext/wikitext103_gpt2_text_document'] - data_per_class_fraction ......................... 1.0 - data_sharding ................................... True - dataloader_type ................................. single - ddp_average_in_collective ....................... False - ddp_bucket_size ................................. None - ddp_num_buckets ................................. None - ddp_pad_buckets_for_high_nccl_busbw ............. False - decoder_first_pipeline_num_layers ............... None - decoder_last_pipeline_num_layers ................ None - decoder_num_layers .............................. None - decoder_seq_length .............................. None - decoupled_lr .................................... None - decoupled_min_lr ................................ None - decrease_batch_size_if_needed ................... False - defer_embedding_wgrad_compute ................... False - delay_wgrad_compute ............................. False - deprecated_use_mcore_models ..................... True - deterministic_mode .............................. False - dino_bottleneck_size ............................ 256 - dino_freeze_last_layer .......................... 1 - dino_head_hidden_size ........................... 2048 - dino_local_crops_number ......................... 10 - dino_local_img_size ............................. 96 - dino_norm_last_layer ............................ False - dino_teacher_temp ............................... 0.07 - dino_warmup_teacher_temp ........................ 0.04 - dino_warmup_teacher_temp_epochs ................. 30 - disable_bf16_reduced_precision_matmul ........... False - disable_mamba_mem_eff_path ...................... False - disable_straggler_on_startup .................... False - dist_ckpt_format_deprecated ..................... None - dist_ckpt_strictness ............................ assume_ok_unexpected - distribute_saved_activations .................... False - distributed_backend ............................. nccl - distributed_timeout_minutes ..................... 10 - embedding_init_method_std ....................... None - embedding_path .................................. None - empty_unused_memory_level ....................... 0 - enable_cuda_graph ............................... True - enable_experimental ............................. False - enable_ft_package ............................... False - enable_gloo_process_groups ...................... True - enable_msc ...................................... True - enable_one_logger ............................... True - encoder_num_layers .............................. 16 - encoder_seq_length .............................. 4096 - end_weight_decay ................................ 0.1 - eod_mask_loss ................................... False - error_injection_rate ............................ 0 - error_injection_type ............................ transient_error - eval_interval ................................... 1000000 - eval_iters ...................................... 0 - evidence_data_path .............................. None - exit_duration_in_mins ........................... None - exit_interval ................................... None - exit_on_missing_checkpoint ...................... False - exit_signal_handler ............................. False - exp_avg_dtype ................................... torch.float32 - exp_avg_sq_dtype ................................ torch.float32 - expert_model_parallel_size ...................... 8 - expert_tensor_parallel_size ..................... 1 - export_force_local_attention .................... False - export_kd_cfg ................................... None - export_kd_teacher_ckpt_format ................... None - export_kd_teacher_load .......................... None - export_kv_cache_quant ........................... False - export_legacy_megatron .......................... False - export_model_type ............................... GPTModel - export_moe_apply_probs_on_input ................. False - export_qk_l2_norm ............................... False - export_quant_cfg ................................ None - export_real_quant_cfg ........................... None - export_te_mcore_model ........................... False - external_cuda_graph ............................. False - ffn_hidden_size ................................. 14336 - finetune ........................................ False - finetune_data_split ............................. train - finetune_hf_dataset ............................. None - first_last_layers_bf16 .......................... False - flash_decode .................................... False - fp16 ............................................ False - fp16_lm_cross_entropy ........................... False - fp32_residual_connection ........................ False - fp8 ............................................. e4m3 - fp8_amax_compute_algo ........................... most_recent - fp8_amax_history_len ............................ 1 - fp8_interval .................................... 1 - fp8_margin ...................................... 0 - fp8_param_gather ................................ False - fp8_recipe ...................................... mxfp8 - fp8_wgrad ....................................... True - fsdp_double_buffer .............................. False - global_batch_size ............................... 16 - grad_reduce_in_bf16 ............................. False - gradient_accumulation_fusion .................... True - gradient_reduce_div_fusion ...................... True - group_query_attention ........................... True - head_lr_mult .................................... 1.0 - heterogeneous_layers_config_encoded_json ........ None - heterogeneous_layers_config_path ................ None - hidden_dropout .................................. 0.0 - hidden_size ..................................... 4096 - hierarchical_context_parallel_sizes ............. None - high_priority_stream_groups ..................... [] - hybrid_attention_ratio .......................... 0.0 - hybrid_mlp_ratio ................................ 0.0 - hybrid_override_pattern ......................... None - hysteresis ...................................... 2 - ict_head_size ................................... None - ict_load ........................................ None - img_h ........................................... 224 - img_w ........................................... 224 - indexer_batch_size .............................. 128 - indexer_log_interval ............................ 1000 - inference_batch_times_seqlen_threshold .......... -1 - inference_dynamic_batching ...................... False - inference_dynamic_batching_buffer_guaranteed_fraction 0.2 - inference_dynamic_batching_buffer_overflow_factor None - inference_dynamic_batching_buffer_size_gb ....... 40.0 - inference_dynamic_batching_chunk_size ........... 256 - inference_dynamic_batching_max_requests_override None - inference_dynamic_batching_max_tokens_override .. None - inference_dynamic_batching_num_cuda_graphs ...... 16 - inference_max_batch_size ........................ 8 - inference_max_seq_length ........................ 2560 - inference_rng_tracker ........................... False - init_method_std ................................. 0.01 - init_method_xavier_uniform ...................... False - init_model_with_meta_device ..................... False - initial_loss_scale .............................. 4294967296 - inprocess_active_world_size ..................... 8 - inprocess_barrier_timeout ....................... 120 - inprocess_completion_timeout .................... 120 - inprocess_empty_cuda_cache ...................... False - inprocess_granularity ........................... node - inprocess_hard_timeout .......................... 90 - inprocess_heartbeat_interval .................... 30 - inprocess_heartbeat_timeout ..................... 60 - inprocess_last_call_wait ........................ 1 - inprocess_max_iterations ........................ None - inprocess_monitor_process_interval .............. 1.0 - inprocess_monitor_thread_interval ............... 1.0 - inprocess_progress_watchdog_interval ............ 1.0 - inprocess_restart ............................... False - inprocess_soft_timeout .......................... 60 - inprocess_termination_grace_time ................ 1 - is_hybrid_model ................................. False - iter_per_epoch .................................. 1250 - iterations_to_skip .............................. [] - keep_fp8_transpose_cache_when_using_custom_fsdp . False - kitchen_config_file ............................. None - kitchen_recipe_number ........................... None - kv_channels ..................................... 128 - kv_lora_rank .................................... 32 - lazy_mpu_init ................................... None - load ............................................ None - load_main_params_from_ckpt ...................... None - load_model_opt_format ........................... False - local_rank ...................................... 0 - log_energy ...................................... False - log_interval .................................... 1 - log_loss_scale_to_tensorboard ................... True - log_memory_to_tensorboard ....................... False - log_num_zeros_in_grad ........................... False - log_params_norm ................................. False - log_progress .................................... False - log_straggler ................................... False - log_throughput .................................. True - log_timers_to_tensorboard ....................... False - log_validation_ppl_to_tensorboard ............... False - log_world_size_to_tensorboard ................... False - logging_level ................................... None - loss_scale ...................................... None - loss_scale_window ............................... 1000 - lr .............................................. 0.0001 - lr_decay_iters .................................. 8 - lr_decay_samples ................................ None - lr_decay_style .................................. cosine - lr_warmup_fraction .............................. None - lr_warmup_init .................................. 0.0 - lr_warmup_iters ................................. 5 - lr_warmup_samples ............................... 0 - lr_wsd_decay_iters .............................. None - lr_wsd_decay_samples ............................ None - lr_wsd_decay_style .............................. exponential - main_grads_dtype ................................ torch.float32 - main_params_dtype ............................... torch.float32 - make_vocab_size_divisible_by .................... 128 - mamba_head_dim .................................. 64 - mamba_num_groups ................................ 8 - mamba_num_heads ................................. None - mamba_state_dim ................................. 128 - manual_gc ....................................... False - manual_gc_eval .................................. True - manual_gc_interval .............................. 0 - mask_factor ..................................... 1.0 - mask_prob ....................................... 0.15 - mask_type ....................................... random - masked_softmax_fusion ........................... False - max_position_embeddings ......................... 32768 - max_tokens_to_oom ............................... 12000 - memory_snapshot_path ............................ snapshot.pickle - merge_file ...................................... /workspace/data/wikitext/gpt2-merges.txt - micro_batch_size ................................ 1 - microbatch_group_size_per_vp_stage .............. None - mid_level_dataset_surplus ....................... 0.005 - min_loss_scale .................................. 1.0 - min_lr .......................................... 1e-05 - mlp_chunks_for_prefill .......................... 1 - mmap_bin_files .................................. True - mock_data ....................................... False - moe_apply_probs_on_input ........................ False - moe_aux_loss_coeff .............................. 0.01 - moe_capacity_monitor ............................ True - moe_deepep_num_sms .............................. 20 - moe_echo_enable_random_offloading ............... False - moe_echo_expert_dispatch_overlap ................ False - moe_echo_expert_dispatcher_type ................. hybridep - moe_echo_recompute_expert_dispatch .............. False - moe_enable_deepep ............................... False - moe_enable_echo ................................. False - moe_enable_hybridep ............................. True - moe_enable_ultraep .............................. True - moe_expert_capacity_factor ...................... None - moe_extended_tp ................................. False - moe_ffn_hidden_size ............................. 1792 - moe_grouped_gemm ................................ True - moe_hybridep_num_sms ............................ 16 - moe_input_jitter_eps ............................ None - moe_layer_freq .................................. 1 - moe_layer_recompute ............................. False - moe_num_echo_experts ............................ None - moe_num_redundant_experts_per_rank .............. 8 - moe_pad_expert_input_to_capacity ................ False - moe_per_layer_logging ........................... False - moe_permute_fusion .............................. False - moe_received_token_capacity ..................... 1.09375 - moe_router_bias_update_rate ..................... 0.001 - moe_router_dtype ................................ fp32 - moe_router_enable_expert_bias ................... False - moe_router_force_load_balancing ................. False - moe_router_group_topk ........................... None - moe_router_load_balancing_type .................. aux_loss - moe_router_num_groups ........................... None - moe_router_padding_for_fp8 ...................... False - moe_router_pre_softmax .......................... False - moe_router_score_function ....................... softmax - moe_router_topk ................................. 8 - moe_router_topk_scaling_factor .................. None - moe_shared_expert_intermediate_size ............. None - moe_shared_expert_overlap ....................... False - moe_token_dispatcher_type ....................... flex - moe_token_drop_policy ........................... probs - moe_ultraep_max_microbatches .................... None - moe_ultraep_overlap_weight_sync ................. True - moe_ultraep_weight_sync_join .................... before_dispatch - moe_upcycling_granularity ....................... 1 - moe_use_device_initiated_grouped_gemm ........... True - moe_use_legacy_grouped_gemm ..................... False - moe_use_upcycling ............................... False - moe_z_loss_coeff ................................ None - mrope_section ................................... None - mscale .......................................... 1.0 - mscale_all_dim .................................. 0.0 - mtp_loss_scaling_factor ......................... 0.1 - mtp_num_layers .................................. None - multi_latent_attention .......................... False - nccl_all_reduce_for_prefill ..................... False - nccl_communicator_config_path ................... None - nccl_ub ......................................... False - no_load_optim ................................... None - no_load_rng ..................................... None - no_persist_layer_norm ........................... False - no_rope_freq .................................... None - no_save_optim ................................... None - no_save_rng ..................................... None - non_persistent_ckpt_type ........................ None - non_persistent_global_ckpt_dir .................. None - non_persistent_local_ckpt_algo .................. fully_parallel - non_persistent_local_ckpt_dir ................... None - non_persistent_save_interval .................... None - norm_epsilon .................................... 1e-05 - normalization ................................... RMSNorm - num_attention_heads ............................. 32 - num_channels .................................... 3 - num_classes ..................................... 1000 - num_dataset_builder_threads ..................... 1 - num_distributed_optimizer_instances ............. 1 - num_experts ..................................... 64 - num_layers ...................................... 16 - num_layers_at_end_in_bf16 ....................... 1 - num_layers_at_start_in_bf16 ..................... 1 - num_layers_per_virtual_pipeline_stage ........... None - num_query_groups ................................ 8 - num_virtual_stages_per_pipeline_rank ............ None - num_workers ..................................... 2 - object_storage_cache_path ....................... None - one_logger_async ................................ False - one_logger_project .............................. megatron-lm - one_logger_run_name ............................. None - onnx_safe ....................................... None - openai_gelu ..................................... False - optimizer ....................................... adam - optimizer_cpu_offload ........................... False - optimizer_offload_fraction ...................... 1.0 - output_bert_embeddings .......................... False - overlap_cpu_optimizer_d2h_h2d ................... False - overlap_grad_reduce ............................. True - overlap_moe_expert_parallel_comm ................ False - overlap_p2p_comm ................................ False - overlap_p2p_comm_warmup_flush ................... False - overlap_param_gather ............................ True - overlap_param_gather_with_optimizer_step ........ False - override_opt_param_scheduler .................... False - params_dtype .................................... torch.bfloat16 - patch_dim ....................................... 16 - per_split_data_args_path ........................ None - perform_initialization .......................... True - pin_cpu_grads ................................... True - pin_cpu_params .................................. True - pipeline_model_parallel_comm_backend ............ None - pipeline_model_parallel_layout .................. None - pipeline_model_parallel_size .................... 1 - position_embedding_type ......................... rope - pretrained_checkpoint ........................... None - profile ......................................... True - profile_ranks ................................... [0, 1, 2, 3, 4, 5, 6, 7] - profile_step_end ................................ 8 - profile_step_start .............................. 6 - q_lora_rank ..................................... None - qk_head_dim ..................................... 128 - qk_l2_norm ...................................... False - qk_layernorm .................................... False - qk_pos_emb_head_dim ............................. 64 - query_in_block_prob ............................. 0.1 - rampup_batch_size ............................... None - rank ............................................ 0 - recompute_granularity ........................... None - recompute_method ................................ None - recompute_modules ............................... None - recompute_num_layers ............................ None - record_memory_history ........................... False - relative_attention_max_distance ................. 128 - relative_attention_num_buckets .................. 32 - replication ..................................... False - replication_factor .............................. 2 - replication_jump ................................ None - rerun_mode ...................................... validate_results - reset_attention_mask ............................ False - reset_position_ids .............................. False - result_rejected_tracker_filename ................ None - retriever_report_topk_accuracies ................ [] - retriever_score_scaling ......................... False - retriever_seq_length ............................ 256 - retro_add_retriever ............................. False - retro_attention_gate ............................ 1 - retro_cyclic_train_iters ........................ None - retro_encoder_attention_dropout ................. 0.1 - retro_encoder_hidden_dropout .................... 0.1 - retro_encoder_layers ............................ 2 - retro_num_neighbors ............................. 2 - retro_num_retrieved_chunks ...................... 2 - retro_project_dir ............................... None - retro_verify_neighbor_count ..................... True - reuse_grad_buf_for_mxfp8_param_ag ............... False - rope_scaling_factor ............................. 8.0 - rope_type ....................................... None - rotary_base ..................................... 1000000 - rotary_interleaved .............................. False - rotary_percent .................................. 1.0 - rotary_scaling_factor ........................... 1.0 - rotary_seq_len_interpolation_factor ............. None - run_workload_inspector_server ................... False - sample_rate ..................................... 1.0 - save ............................................ None - save_interval ................................... None - save_retain_interval ............................ None - scatter_gather_tensors_in_pipeline .............. True - seed ............................................ 1234 - seq_length ...................................... 4096 - sequence_parallel ............................... False - sft ............................................. False - sft_tokenizer_prompt_format ..................... nemotron-h-aligned - sgd_momentum .................................... 0.9 - short_seq_prob .................................. 0.1 - skip_train ...................................... False - skipped_train_samples ........................... 0 - spec ............................................ None - split ........................................... 100,0,0 - squared_relu .................................... False - start_weight_decay .............................. 0.1 - straggler_ctrlr_port ............................ 65535 - straggler_minmax_count .......................... 1 - suggested_communication_unit_size ............... None - swiglu .......................................... True - swin_backbone_type .............................. tiny - symmetric_ar_type ............................... None - te_rng_tracker .................................. True - tensor_model_parallel_size ...................... 1 - tensorboard_dir ................................. None - tensorboard_log_interval ........................ 1 - tensorboard_queue_size .......................... 1000 - test_data_path .................................. None - test_mode ....................................... False - tiktoken_num_special_tokens ..................... 1000 - tiktoken_pattern ................................ None - tiktoken_special_tokens ......................... None - timing_log_level ................................ 0 - timing_log_option ............................... minmax - titles_data_path ................................ None - tokenizer_model ................................. None - tokenizer_type .................................. GPT2BPETokenizer - torch_fsdp2_reshard_after_forward ............... True - tp_comm_bootstrap_backend ....................... nccl - tp_comm_bulk_dgrad .............................. True - tp_comm_bulk_wgrad .............................. True - tp_comm_overlap ................................. False - tp_comm_overlap_ag .............................. True - tp_comm_overlap_cfg ............................. None - tp_comm_overlap_rs .............................. True - tp_comm_overlap_rs_dgrad ........................ False - tp_comm_split_ag ................................ True - tp_comm_split_rs ................................ True - train_data_path ................................. None - train_iters ..................................... 10 - train_samples ................................... None - train_sync_interval ............................. None - transformer_impl ................................ transformer_engine - transformer_pipeline_model_parallel_size ........ 1 - untie_embeddings_and_output_weights ............. True - use_checkpoint_args ............................. False - use_checkpoint_opt_param_scheduler .............. False - use_cpu_initialization .......................... None - use_custom_fsdp ................................. False - use_dist_ckpt ................................... True - use_dist_ckpt_deprecated ........................ False - use_distributed_optimizer ....................... True - use_flash_attn .................................. False - use_legacy_models ............................... False - use_mp_args_from_checkpoint_args ................ False - use_one_sent_docs ............................... False - use_persistent_ckpt_worker ...................... False - use_precision_aware_optimizer ................... False - use_pytorch_profiler ............................ False - use_ring_exchange_p2p ........................... False - use_rope_scaling ................................ False - use_rotary_position_embeddings .................. False - use_sharp ....................................... False - use_tokenizer_model_from_checkpoint_args ........ True - use_torch_fsdp2 ................................. False - use_torch_optimizer_for_cpu_offload ............. False - use_tp_pp_dp_mapping ............................ False - v_head_dim ...................................... 128 - valid_data_path ................................. None - variable_seq_lengths ............................ False - virtual_pipeline_model_parallel_size ............ None - vision_backbone_type ............................ vit - vision_pretraining .............................. False - vision_pretraining_type ......................... classify - vocab_extra_ids ................................. 0 - vocab_file ...................................... /workspace/data/wikitext/gpt2-vocab.json - vocab_size ...................................... None - wandb_exp_name .................................. - wandb_project ................................... - wandb_save_dir .................................. - weight_decay .................................... 0.1 - weight_decay_incr_style ......................... constant - wgrad_deferral_limit ............................ 0 - world_size ...................................... 8 - yaml_cfg ........................................ None --------------------- end of arguments --------------------- -INFO:megatron.core.num_microbatches_calculator:setting number of microbatches to constant 2 -> building GPT2BPETokenizer tokenizer ... -/workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled - warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled") -/workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled - warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled") -/workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled - warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled") - > padded vocab (size: 50257) with 47 dummy tokens (new size: 50304) -WARNING:megatron.core.rerun_state_machine:RerunStateMachine initialized in mode RerunMode.VALIDATE_RESULTS -> initializing torch distributed ... -WARNING: one_logger package is required to enable e2e metrics tracking. please go to https://confluence.nvidia.com/display/MLWFO/Package+Repositories for details to install it -/workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled - warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled") -/workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled - warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled") -/workspace/Megatron-UltraEP/megatron/training/arguments.py:831: UserWarning: Disabling sequence parallelism because tensor model parallelism is disabled - warnings.warn("Disabling sequence parallelism because tensor model parallelism is disabled") -[Gloo] Rank 0 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 -[Gloo] Rank 1 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 -[Gloo] Rank 2 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 -[Gloo] Rank 3 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 -[Gloo] Rank 6 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 -[Gloo] Rank 4 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 -[Gloo] Rank 5 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 -[Gloo] Rank 7 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 -[Gloo] Rank 0 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 -[Gloo] Rank 1 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 -[Gloo] Rank 2 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 -[Gloo] Rank 3 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 -[Gloo] Rank 7 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7[Gloo] Rank -4 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 -[Gloo] Rank [Gloo] Rank 5 is connected to 7 peer ranks. Expected number of connected peer ranks is : 67 is connected to -7 peer ranks. Expected number of connected peer ranks is : 7 -[Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0 -[Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0 -[Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0 -[Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0 -[Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0 -[Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0 -[Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0 -[Gloo] Rank 0 is connected to 0 peer ranks. Expected number of connected peer ranks is : 0 -> initialized tensor model parallel with size 1 -> initialized pipeline model parallel with size 1 -> setting random seeds to 1234 ... -[rank5]:[W917 03:33:17.749694598 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 5] using GPU 5 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device. -[rank1]:[W917 03:33:17.749736865 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 1] using GPU 1 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device. -[rank3]:[W917 03:33:17.749771189 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 3] using GPU 3 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device. -[rank2]:[W917 03:33:17.749792831 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 2] using GPU 2 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device. -[rank7]:[W917 03:33:17.753345276 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 7] using GPU 7 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device. -[rank4]:[W917 03:33:17.754063808 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 4] using GPU 4 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device. -[rank6]:[W917 03:33:17.754496723 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 6] using GPU 6 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device. -> compiling dataset index builder ... -make: Entering directory '/workspace/Megatron-UltraEP/megatron/core/datasets' -make: Nothing to be done for 'default'. -make: Leaving directory '/workspace/Megatron-UltraEP/megatron/core/datasets' ->>> done with dataset index builder. Compilation time: 0.429 seconds -WARNING: constraints for invoking optimized fused softmax kernel are not met. We default back to unfused kernel invocations. -> compiling and loading fused kernels ... -[rank0]:[W917 03:33:18.359796766 ProcessGroupNCCL.cpp:4789] [PG ID 0 PG GUID 0 Rank 0] using GPU 0 as device used by this process is currently unknown. This can potentially cause a hang if this rank to GPU mapping is incorrect. You can specify device_id in init_process_group() to force use of a particular device. ->>> done with compiling and loading fused kernels. Compilation time: 3.438 seconds -time to initialize megatron (seconds): 8.825 -[after megatron is initialized] datetime: 2026-09-17 03:33:23 -building GPT model ... -/usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect. - warnings.warn( -/usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect. - warnings.warn( -/usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect. - warnings.warn( -/usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect. - warnings.warn( -/usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect. - warnings.warn( -/usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect. - warnings.warn( -/usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect. - warnings.warn( -/usr/local/lib/python3.12/dist-packages/transformer_engine/pytorch/cpu_offload.py:708: DeprecationWarning: Offloading weights is deprecated. Using offload_weights=True does not have any effect. - warnings.warn( -[INFO] Use MNNVL fabric: False -[WARN] MAX_NUM_NVL_PEERS is not set. Using detected value 8. -NVSHMEM v3.7.2 -/dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed. - -/dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed. - -/dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed. - -/dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed. - -/dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed. - -/dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed. - -/dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed. - -/dvs/p4/build/sw/rel/gpgpu/toolkit/r12.9/main_nvshmem/src/modules/transport/ibrc/ibrc.cpp:1754: error status: 7 (NVSHMEMX_ERROR_INTERNAL) Device enumeration failed. - -UltraEP: NVL domain size <= 8; forcing weight_sync plan mode to direct (was adaptive). - > number of parameters on (tensor, pipeline) model parallel rank (0, 0): 6724653056 -INFO:megatron.core.distributed.distributed_data_parallel:Setting up DistributedDataParallel with config DistributedDataParallelConfig(grad_reduce_in_fp32=True, overlap_grad_reduce=True, overlap_param_gather=True, align_param_gather=False, use_distributed_optimizer=True, num_distributed_optimizer_instances=1, check_for_nan_in_grad=False, check_for_large_grads=False, bucket_size=40000000, pad_buckets_for_high_nccl_busbw=False, average_in_collective=False, fp8_param_gather=False, reuse_grad_buf_for_mxfp8_param_ag=False, use_custom_fsdp=False, data_parallel_sharding_strategy='no_shard', gradient_reduce_div_fusion=True, suggested_communication_unit_size=None, preserve_fp32_weights=True, keep_fp8_transpose_cache_when_using_custom_fsdp=False, nccl_ub=False, fsdp_double_buffer=False) -INFO:megatron.core.distributed.param_and_grad_buffer:Number of buckets for gradient all-reduce / reduce-scatter: 18 -Params for bucket 1 (206045184 elements, 206045184 padded size): - module.output_layer.weight -Params for bucket 2 (42217472 elements, 42217472 padded size): - module.decoder.final_layernorm.weight - module.decoder.layers.15.self_attention.linear_qkv.weight - module.decoder.layers.15.pre_mlp_layernorm.weight - module.decoder.layers.15.mlp.router.weight - module.decoder.layers.15.self_attention.linear_qkv.layer_norm_weight - module.decoder.layers.15.self_attention.linear_proj.weight -Params for bucket 3 (42213376 elements, 42213376 padded size): - module.decoder.layers.14.self_attention.linear_qkv.weight - module.decoder.layers.14.pre_mlp_layernorm.weight - module.decoder.layers.14.mlp.router.weight - module.decoder.layers.14.self_attention.linear_qkv.layer_norm_weight - module.decoder.layers.14.self_attention.linear_proj.weight -Params for bucket 4 (42213376 elements, 42213376 padded size): - module.decoder.layers.13.pre_mlp_layernorm.weight - module.decoder.layers.13.self_attention.linear_qkv.weight - module.decoder.layers.13.mlp.router.weight - module.decoder.layers.13.self_attention.linear_qkv.layer_norm_weight - module.decoder.layers.13.self_attention.linear_proj.weight -Params for bucket 5 (42213376 elements, 42213376 padded size): - module.decoder.layers.12.mlp.router.weight - module.decoder.layers.12.self_attention.linear_proj.weight - module.decoder.layers.12.self_attention.linear_qkv.layer_norm_weight - module.decoder.layers.12.pre_mlp_layernorm.weight - module.decoder.layers.12.self_attention.linear_qkv.weight -Params for bucket 6 (42213376 elements, 42213376 padded size): - module.decoder.layers.11.mlp.router.weight - module.decoder.layers.11.self_attention.linear_proj.weight - module.decoder.layers.11.self_attention.linear_qkv.layer_norm_weight - module.decoder.layers.11.self_attention.linear_qkv.weight - module.decoder.layers.11.pre_mlp_layernorm.weight -Params for bucket 7 (42213376 elements, 42213376 padded size): - module.decoder.layers.10.mlp.router.weight - module.decoder.layers.10.self_attention.linear_proj.weight - module.decoder.layers.10.self_attention.linear_qkv.layer_norm_weight - module.decoder.layers.10.pre_mlp_layernorm.weight - module.decoder.layers.10.self_attention.linear_qkv.weight -Params for bucket 8 (42213376 elements, 42213376 padded size): - module.decoder.layers.9.pre_mlp_layernorm.weight - module.decoder.layers.9.self_attention.linear_qkv.weight - module.decoder.layers.9.mlp.router.weight - module.decoder.layers.9.self_attention.linear_proj.weight - module.decoder.layers.9.self_attention.linear_qkv.layer_norm_weight -Params for bucket 9 (42213376 elements, 42213376 padded size): - module.decoder.layers.8.self_attention.linear_qkv.weight - module.decoder.layers.8.pre_mlp_layernorm.weight - module.decoder.layers.8.mlp.router.weight - module.decoder.layers.8.self_attention.linear_qkv.layer_norm_weight - module.decoder.layers.8.self_attention.linear_proj.weight -Params for bucket 10 (42213376 elements, 42213376 padded size): - module.decoder.layers.7.mlp.router.weight - module.decoder.layers.7.self_attention.linear_qkv.layer_norm_weight - module.decoder.layers.7.self_attention.linear_proj.weight - module.decoder.layers.7.self_attention.linear_qkv.weight - module.decoder.layers.7.pre_mlp_layernorm.weight -Params for bucket 11 (42213376 elements, 42213376 padded size): - module.decoder.layers.6.mlp.router.weight - module.decoder.layers.6.self_attention.linear_proj.weight - module.decoder.layers.6.self_attention.linear_qkv.layer_norm_weight - module.decoder.layers.6.self_attention.linear_qkv.weight - module.decoder.layers.6.pre_mlp_layernorm.weight -Params for bucket 12 (42213376 elements, 42213376 padded size): - module.decoder.layers.5.self_attention.linear_qkv.weight - module.decoder.layers.5.self_attention.linear_proj.weight - module.decoder.layers.5.pre_mlp_layernorm.weight - module.decoder.layers.5.mlp.router.weight - module.decoder.layers.5.self_attention.linear_qkv.layer_norm_weight -Params for bucket 13 (42213376 elements, 42213376 padded size): - module.decoder.layers.4.self_attention.linear_qkv.weight - module.decoder.layers.4.pre_mlp_layernorm.weight - module.decoder.layers.4.mlp.router.weight - module.decoder.layers.4.self_attention.linear_proj.weight - module.decoder.layers.4.self_attention.linear_qkv.layer_norm_weight -Params for bucket 14 (42213376 elements, 42213376 padded size): - module.decoder.layers.3.mlp.router.weight - module.decoder.layers.3.self_attention.linear_proj.weight - module.decoder.layers.3.self_attention.linear_qkv.layer_norm_weight - module.decoder.layers.3.pre_mlp_layernorm.weight - module.decoder.layers.3.self_attention.linear_qkv.weight -Params for bucket 15 (42213376 elements, 42213376 padded size): - module.decoder.layers.2.mlp.router.weight - module.decoder.layers.2.self_attention.linear_qkv.layer_norm_weight - module.decoder.layers.2.self_attention.linear_proj.weight - module.decoder.layers.2.self_attention.linear_qkv.weight - module.decoder.layers.2.pre_mlp_layernorm.weight -Params for bucket 16 (42213376 elements, 42213376 padded size): - module.decoder.layers.1.mlp.router.weight - module.decoder.layers.1.self_attention.linear_proj.weight - module.decoder.layers.1.self_attention.linear_qkv.layer_norm_weight - module.decoder.layers.1.self_attention.linear_qkv.weight - module.decoder.layers.1.pre_mlp_layernorm.weight -Params for bucket 17 (42213376 elements, 42213376 padded size): - module.decoder.layers.0.pre_mlp_layernorm.weight - module.decoder.layers.0.self_attention.linear_proj.weight - module.decoder.layers.0.self_attention.linear_qkv.weight - module.decoder.layers.0.mlp.router.weight - module.decoder.layers.0.self_attention.linear_qkv.layer_norm_weight -Params for bucket 18 (206045184 elements, 206045184 padded size): - module.embedding.word_embeddings.weight -INFO:megatron.core.distributed.param_and_grad_buffer:Number of buckets for gradient all-reduce / reduce-scatter: 64 -Params for bucket 1 (44040192 elements, 44040192 padded size): - module.decoder.layers.15.mlp.experts.linear_fc2.weight6 - module.decoder.layers.15.mlp.experts.linear_fc2.weight4 - module.decoder.layers.15.mlp.experts.linear_fc2.weight2 - module.decoder.layers.15.mlp.experts.linear_fc2.weight7 - module.decoder.layers.15.mlp.experts.linear_fc2.weight5 - module.decoder.layers.15.mlp.experts.linear_fc2.weight3 -Params for bucket 2 (44040192 elements, 44040192 padded size): - module.decoder.layers.15.mlp.experts.linear_fc2.weight1 - module.decoder.layers.15.mlp.experts.linear_fc1.weight6 - module.decoder.layers.15.mlp.experts.linear_fc2.weight0 - module.decoder.layers.15.mlp.experts.linear_fc1.weight7 -Params for bucket 3 (44040192 elements, 44040192 padded size): - module.decoder.layers.15.mlp.experts.linear_fc1.weight5 - module.decoder.layers.15.mlp.experts.linear_fc1.weight3 - module.decoder.layers.15.mlp.experts.linear_fc1.weight4 -Params for bucket 4 (44040192 elements, 44040192 padded size): - module.decoder.layers.15.mlp.experts.linear_fc1.weight2 - module.decoder.layers.15.mlp.experts.linear_fc1.weight0 - module.decoder.layers.15.mlp.experts.linear_fc1.weight1 -Params for bucket 5 (44040192 elements, 44040192 padded size): - module.decoder.layers.14.mlp.experts.linear_fc2.weight6 - module.decoder.layers.14.mlp.experts.linear_fc2.weight4 - module.decoder.layers.14.mlp.experts.linear_fc2.weight2 - module.decoder.layers.14.mlp.experts.linear_fc2.weight7 - module.decoder.layers.14.mlp.experts.linear_fc2.weight3 - module.decoder.layers.14.mlp.experts.linear_fc2.weight5 -Params for bucket 6 (44040192 elements, 44040192 padded size): - module.decoder.layers.14.mlp.experts.linear_fc2.weight1 - module.decoder.layers.14.mlp.experts.linear_fc2.weight0 - module.decoder.layers.14.mlp.experts.linear_fc1.weight6 - module.decoder.layers.14.mlp.experts.linear_fc1.weight7 -Params for bucket 7 (44040192 elements, 44040192 padded size): - module.decoder.layers.14.mlp.experts.linear_fc1.weight5 - module.decoder.layers.14.mlp.experts.linear_fc1.weight4 - module.decoder.layers.14.mlp.experts.linear_fc1.weight3 -Params for bucket 8 (44040192 elements, 44040192 padded size): - module.decoder.layers.14.mlp.experts.linear_fc1.weight2 - module.decoder.layers.14.mlp.experts.linear_fc1.weight1 - module.decoder.layers.14.mlp.experts.linear_fc1.weight0 -Params for bucket 9 (44040192 elements, 44040192 padded size): - module.decoder.layers.13.mlp.experts.linear_fc2.weight7 - module.decoder.layers.13.mlp.experts.linear_fc2.weight5 - module.decoder.layers.13.mlp.experts.linear_fc2.weight3 - module.decoder.layers.13.mlp.experts.linear_fc2.weight4 - module.decoder.layers.13.mlp.experts.linear_fc2.weight6 - module.decoder.layers.13.mlp.experts.linear_fc2.weight2 -Params for bucket 10 (44040192 elements, 44040192 padded size): - module.decoder.layers.13.mlp.experts.linear_fc2.weight1 - module.decoder.layers.13.mlp.experts.linear_fc2.weight0 - module.decoder.layers.13.mlp.experts.linear_fc1.weight6 - module.decoder.layers.13.mlp.experts.linear_fc1.weight7 -Params for bucket 11 (44040192 elements, 44040192 padded size): - module.decoder.layers.13.mlp.experts.linear_fc1.weight5 - module.decoder.layers.13.mlp.experts.linear_fc1.weight3 - module.decoder.layers.13.mlp.experts.linear_fc1.weight4 -Params for bucket 12 (44040192 elements, 44040192 padded size): - module.decoder.layers.13.mlp.experts.linear_fc1.weight2 - module.decoder.layers.13.mlp.experts.linear_fc1.weight0 - module.decoder.layers.13.mlp.experts.linear_fc1.weight1 -Params for bucket 13 (44040192 elements, 44040192 padded size): - module.decoder.layers.12.mlp.experts.linear_fc2.weight6 - module.decoder.layers.12.mlp.experts.linear_fc2.weight4 - module.decoder.layers.12.mlp.experts.linear_fc2.weight2 - module.decoder.layers.12.mlp.experts.linear_fc2.weight7 - module.decoder.layers.12.mlp.experts.linear_fc2.weight5 - module.decoder.layers.12.mlp.experts.linear_fc2.weight3 -Params for bucket 14 (44040192 elements, 44040192 padded size): - module.decoder.layers.12.mlp.experts.linear_fc2.weight1 - module.decoder.layers.12.mlp.experts.linear_fc1.weight7 - module.decoder.layers.12.mlp.experts.linear_fc1.weight6 - module.decoder.layers.12.mlp.experts.linear_fc2.weight0 -Params for bucket 15 (44040192 elements, 44040192 padded size): - module.decoder.layers.12.mlp.experts.linear_fc1.weight5 - module.decoder.layers.12.mlp.experts.linear_fc1.weight4 - module.decoder.layers.12.mlp.experts.linear_fc1.weight3 -Params for bucket 16 (44040192 elements, 44040192 padded size): - module.decoder.layers.12.mlp.experts.linear_fc1.weight2 - module.decoder.layers.12.mlp.experts.linear_fc1.weight1 - module.decoder.layers.12.mlp.experts.linear_fc1.weight0 -Params for bucket 17 (44040192 elements, 44040192 padded size): - module.decoder.layers.11.mlp.experts.linear_fc2.weight7 - module.decoder.layers.11.mlp.experts.linear_fc2.weight5 - module.decoder.layers.11.mlp.experts.linear_fc2.weight3 - module.decoder.layers.11.mlp.experts.linear_fc2.weight4 - module.decoder.layers.11.mlp.experts.linear_fc2.weight6 - module.decoder.layers.11.mlp.experts.linear_fc2.weight2 -Params for bucket 18 (44040192 elements, 44040192 padded size): - module.decoder.layers.11.mlp.experts.linear_fc2.weight1 - module.decoder.layers.11.mlp.experts.linear_fc1.weight7 - module.decoder.layers.11.mlp.experts.linear_fc1.weight6 - module.decoder.layers.11.mlp.experts.linear_fc2.weight0 -Params for bucket 19 (44040192 elements, 44040192 padded size): - module.decoder.layers.11.mlp.experts.linear_fc1.weight5 - module.decoder.layers.11.mlp.experts.linear_fc1.weight4 - module.decoder.layers.11.mlp.experts.linear_fc1.weight3 -Params for bucket 20 (44040192 elements, 44040192 padded size): - module.decoder.layers.11.mlp.experts.linear_fc1.weight2 - module.decoder.layers.11.mlp.experts.linear_fc1.weight1 - module.decoder.layers.11.mlp.experts.linear_fc1.weight0 -Params for bucket 21 (44040192 elements, 44040192 padded size): - module.decoder.layers.10.mlp.experts.linear_fc2.weight7 - module.decoder.layers.10.mlp.experts.linear_fc2.weight3 - module.decoder.layers.10.mlp.experts.linear_fc2.weight5 - module.decoder.layers.10.mlp.experts.linear_fc2.weight6 - module.decoder.layers.10.mlp.experts.linear_fc2.weight4 - module.decoder.layers.10.mlp.experts.linear_fc2.weight2 -Params for bucket 22 (44040192 elements, 44040192 padded size): - module.decoder.layers.10.mlp.experts.linear_fc2.weight1 - module.decoder.layers.10.mlp.experts.linear_fc2.weight0 - module.decoder.layers.10.mlp.experts.linear_fc1.weight7 - module.decoder.layers.10.mlp.experts.linear_fc1.weight6 -Params for bucket 23 (44040192 elements, 44040192 padded size): - module.decoder.layers.10.mlp.experts.linear_fc1.weight5 - module.decoder.layers.10.mlp.experts.linear_fc1.weight3 - module.decoder.layers.10.mlp.experts.linear_fc1.weight4 -Params for bucket 24 (44040192 elements, 44040192 padded size): - module.decoder.layers.10.mlp.experts.linear_fc1.weight2 - module.decoder.layers.10.mlp.experts.linear_fc1.weight0 - module.decoder.layers.10.mlp.experts.linear_fc1.weight1 -Params for bucket 25 (44040192 elements, 44040192 padded size): - module.decoder.layers.9.mlp.experts.linear_fc2.weight7 - module.decoder.layers.9.mlp.experts.linear_fc2.weight3 - module.decoder.layers.9.mlp.experts.linear_fc2.weight5 - module.decoder.layers.9.mlp.experts.linear_fc2.weight4 - module.decoder.layers.9.mlp.experts.linear_fc2.weight6 - module.decoder.layers.9.mlp.experts.linear_fc2.weight2 -Params for bucket 26 (44040192 elements, 44040192 padded size): - module.decoder.layers.9.mlp.experts.linear_fc2.weight1 - module.decoder.layers.9.mlp.experts.linear_fc2.weight0 - module.decoder.layers.9.mlp.experts.linear_fc1.weight7 - module.decoder.layers.9.mlp.experts.linear_fc1.weight6 -Params for bucket 27 (44040192 elements, 44040192 padded size): - module.decoder.layers.9.mlp.experts.linear_fc1.weight5 - module.decoder.layers.9.mlp.experts.linear_fc1.weight3 - module.decoder.layers.9.mlp.experts.linear_fc1.weight4 -Params for bucket 28 (44040192 elements, 44040192 padded size): - module.decoder.layers.9.mlp.experts.linear_fc1.weight2 - module.decoder.layers.9.mlp.experts.linear_fc1.weight0 - module.decoder.layers.9.mlp.experts.linear_fc1.weight1 -Params for bucket 29 (44040192 elements, 44040192 padded size): - module.decoder.layers.8.mlp.experts.linear_fc2.weight4 - module.decoder.layers.8.mlp.experts.linear_fc2.weight6 - module.decoder.layers.8.mlp.experts.linear_fc2.weight2 - module.decoder.layers.8.mlp.experts.linear_fc2.weight7 - module.decoder.layers.8.mlp.experts.linear_fc2.weight5 - module.decoder.layers.8.mlp.experts.linear_fc2.weight3 -Params for bucket 30 (44040192 elements, 44040192 padded size): - module.decoder.layers.8.mlp.experts.linear_fc2.weight1 - module.decoder.layers.8.mlp.experts.linear_fc1.weight7 - module.decoder.layers.8.mlp.experts.linear_fc2.weight0 - module.decoder.layers.8.mlp.experts.linear_fc1.weight6 -Params for bucket 31 (44040192 elements, 44040192 padded size): - module.decoder.layers.8.mlp.experts.linear_fc1.weight5 - module.decoder.layers.8.mlp.experts.linear_fc1.weight4 - module.decoder.layers.8.mlp.experts.linear_fc1.weight3 -Params for bucket 32 (44040192 elements, 44040192 padded size): - module.decoder.layers.8.mlp.experts.linear_fc1.weight2 - module.decoder.layers.8.mlp.experts.linear_fc1.weight0 - module.decoder.layers.8.mlp.experts.linear_fc1.weight1 -Params for bucket 33 (44040192 elements, 44040192 padded size): - module.decoder.layers.7.mlp.experts.linear_fc2.weight4 - module.decoder.layers.7.mlp.experts.linear_fc2.weight6 - module.decoder.layers.7.mlp.experts.linear_fc2.weight2 - module.decoder.layers.7.mlp.experts.linear_fc2.weight7 - module.decoder.layers.7.mlp.experts.linear_fc2.weight5 - module.decoder.layers.7.mlp.experts.linear_fc2.weight3 -Params for bucket 34 (44040192 elements, 44040192 padded size): - module.decoder.layers.7.mlp.experts.linear_fc2.weight1 - module.decoder.layers.7.mlp.experts.linear_fc1.weight7 - module.decoder.layers.7.mlp.experts.linear_fc1.weight6 - module.decoder.layers.7.mlp.experts.linear_fc2.weight0 -Params for bucket 35 (44040192 elements, 44040192 padded size): - module.decoder.layers.7.mlp.experts.linear_fc1.weight5 - module.decoder.layers.7.mlp.experts.linear_fc1.weight4 - module.decoder.layers.7.mlp.experts.linear_fc1.weight3 -Params for bucket 36 (44040192 elements, 44040192 padded size): - module.decoder.layers.7.mlp.experts.linear_fc1.weight2 - module.decoder.layers.7.mlp.experts.linear_fc1.weight0 - module.decoder.layers.7.mlp.experts.linear_fc1.weight1 -Params for bucket 37 (44040192 elements, 44040192 padded size): - module.decoder.layers.6.mlp.experts.linear_fc2.weight7 - module.decoder.layers.6.mlp.experts.linear_fc2.weight5 - module.decoder.layers.6.mlp.experts.linear_fc2.weight3 - module.decoder.layers.6.mlp.experts.linear_fc2.weight6 - module.decoder.layers.6.mlp.experts.linear_fc2.weight4 - module.decoder.layers.6.mlp.experts.linear_fc2.weight2 -Params for bucket 38 (44040192 elements, 44040192 padded size): - module.decoder.layers.6.mlp.experts.linear_fc2.weight1 - module.decoder.layers.6.mlp.experts.linear_fc1.weight6 - module.decoder.layers.6.mlp.experts.linear_fc2.weight0 - module.decoder.layers.6.mlp.experts.linear_fc1.weight7 -Params for bucket 39 (44040192 elements, 44040192 padded size): - module.decoder.layers.6.mlp.experts.linear_fc1.weight5 - module.decoder.layers.6.mlp.experts.linear_fc1.weight3 - module.decoder.layers.6.mlp.experts.linear_fc1.weight4 -Params for bucket 40 (44040192 elements, 44040192 padded size): - module.decoder.layers.6.mlp.experts.linear_fc1.weight2 - module.decoder.layers.6.mlp.experts.linear_fc1.weight0 - module.decoder.layers.6.mlp.experts.linear_fc1.weight1 -Params for bucket 41 (44040192 elements, 44040192 padded size): - module.decoder.layers.5.mlp.experts.linear_fc2.weight7 - module.decoder.layers.5.mlp.experts.linear_fc2.weight3 - module.decoder.layers.5.mlp.experts.linear_fc2.weight5 - module.decoder.layers.5.mlp.experts.linear_fc2.weight6 - module.decoder.layers.5.mlp.experts.linear_fc2.weight4 - module.decoder.layers.5.mlp.experts.linear_fc2.weight2 -Params for bucket 42 (44040192 elements, 44040192 padded size): - module.decoder.layers.5.mlp.experts.linear_fc2.weight1 - module.decoder.layers.5.mlp.experts.linear_fc1.weight7 - module.decoder.layers.5.mlp.experts.linear_fc2.weight0 - module.decoder.layers.5.mlp.experts.linear_fc1.weight6 -Params for bucket 43 (44040192 elements, 44040192 padded size): - module.decoder.layers.5.mlp.experts.linear_fc1.weight5 - module.decoder.layers.5.mlp.experts.linear_fc1.weight4 - module.decoder.layers.5.mlp.experts.linear_fc1.weight3 -Params for bucket 44 (44040192 elements, 44040192 padded size): - module.decoder.layers.5.mlp.experts.linear_fc1.weight2 - module.decoder.layers.5.mlp.experts.linear_fc1.weight0 - module.decoder.layers.5.mlp.experts.linear_fc1.weight1 -Params for bucket 45 (44040192 elements, 44040192 padded size): - module.decoder.layers.4.mlp.experts.linear_fc2.weight7 - module.decoder.layers.4.mlp.experts.linear_fc2.weight3 - module.decoder.layers.4.mlp.experts.linear_fc2.weight5 - module.decoder.layers.4.mlp.experts.linear_fc2.weight4 - module.decoder.layers.4.mlp.experts.linear_fc2.weight6 - module.decoder.layers.4.mlp.experts.linear_fc2.weight2 -Params for bucket 46 (44040192 elements, 44040192 padded size): - module.decoder.layers.4.mlp.experts.linear_fc2.weight1 - module.decoder.layers.4.mlp.experts.linear_fc2.weight0 - module.decoder.layers.4.mlp.experts.linear_fc1.weight7 - module.decoder.layers.4.mlp.experts.linear_fc1.weight6 -Params for bucket 47 (44040192 elements, 44040192 padded size): - module.decoder.layers.4.mlp.experts.linear_fc1.weight5 - module.decoder.layers.4.mlp.experts.linear_fc1.weight3 - module.decoder.layers.4.mlp.experts.linear_fc1.weight4 -Params for bucket 48 (44040192 elements, 44040192 padded size): - module.decoder.layers.4.mlp.experts.linear_fc1.weight2 - module.decoder.layers.4.mlp.experts.linear_fc1.weight0 - module.decoder.layers.4.mlp.experts.linear_fc1.weight1 -Params for bucket 49 (44040192 elements, 44040192 padded size): - module.decoder.layers.3.mlp.experts.linear_fc2.weight7 - module.decoder.layers.3.mlp.experts.linear_fc2.weight3 - module.decoder.layers.3.mlp.experts.linear_fc2.weight5 - module.decoder.layers.3.mlp.experts.linear_fc2.weight6 - module.decoder.layers.3.mlp.experts.linear_fc2.weight4 - module.decoder.layers.3.mlp.experts.linear_fc2.weight2 -Params for bucket 50 (44040192 elements, 44040192 padded size): - module.decoder.layers.3.mlp.experts.linear_fc2.weight1 - module.decoder.layers.3.mlp.experts.linear_fc2.weight0 - module.decoder.layers.3.mlp.experts.linear_fc1.weight7 - module.decoder.layers.3.mlp.experts.linear_fc1.weight6 -Params for bucket 51 (44040192 elements, 44040192 padded size): - module.decoder.layers.3.mlp.experts.linear_fc1.weight5 - module.decoder.layers.3.mlp.experts.linear_fc1.weight3 - module.decoder.layers.3.mlp.experts.linear_fc1.weight4 -Params for bucket 52 (44040192 elements, 44040192 padded size): - module.decoder.layers.3.mlp.experts.linear_fc1.weight2 - module.decoder.layers.3.mlp.experts.linear_fc1.weight0 - module.decoder.layers.3.mlp.experts.linear_fc1.weight1 -Params for bucket 53 (44040192 elements, 44040192 padded size): - module.decoder.layers.2.mlp.experts.linear_fc2.weight7 - module.decoder.layers.2.mlp.experts.linear_fc2.weight5 - module.decoder.layers.2.mlp.experts.linear_fc2.weight3 - module.decoder.layers.2.mlp.experts.linear_fc2.weight4 - module.decoder.layers.2.mlp.experts.linear_fc2.weight6 - module.decoder.layers.2.mlp.experts.linear_fc2.weight2 -Params for bucket 54 (44040192 elements, 44040192 padded size): - module.decoder.layers.2.mlp.experts.linear_fc2.weight1 - module.decoder.layers.2.mlp.experts.linear_fc1.weight7 - module.decoder.layers.2.mlp.experts.linear_fc1.weight6 - module.decoder.layers.2.mlp.experts.linear_fc2.weight0 -Params for bucket 55 (44040192 elements, 44040192 padded size): - module.decoder.layers.2.mlp.experts.linear_fc1.weight5 - module.decoder.layers.2.mlp.experts.linear_fc1.weight4 - module.decoder.layers.2.mlp.experts.linear_fc1.weight3 -Params for bucket 56 (44040192 elements, 44040192 padded size): - module.decoder.layers.2.mlp.experts.linear_fc1.weight2 - module.decoder.layers.2.mlp.experts.linear_fc1.weight0 - module.decoder.layers.2.mlp.experts.linear_fc1.weight1 -Params for bucket 57 (44040192 elements, 44040192 padded size): - module.decoder.layers.1.mlp.experts.linear_fc2.weight7 - module.decoder.layers.1.mlp.experts.linear_fc2.weight5 - module.decoder.layers.1.mlp.experts.linear_fc2.weight3 - module.decoder.layers.1.mlp.experts.linear_fc2.weight6 - module.decoder.layers.1.mlp.experts.linear_fc2.weight4 - module.decoder.layers.1.mlp.experts.linear_fc2.weight2 -Params for bucket 58 (44040192 elements, 44040192 padded size): - module.decoder.layers.1.mlp.experts.linear_fc2.weight1 - module.decoder.layers.1.mlp.experts.linear_fc2.weight0 - module.decoder.layers.1.mlp.experts.linear_fc1.weight7 - module.decoder.layers.1.mlp.experts.linear_fc1.weight6 -Params for bucket 59 (44040192 elements, 44040192 padded size): - module.decoder.layers.1.mlp.experts.linear_fc1.weight5 - module.decoder.layers.1.mlp.experts.linear_fc1.weight3 - module.decoder.layers.1.mlp.experts.linear_fc1.weight4 -Params for bucket 60 (44040192 elements, 44040192 padded size): - module.decoder.layers.1.mlp.experts.linear_fc1.weight2 - module.decoder.layers.1.mlp.experts.linear_fc1.weight0 - module.decoder.layers.1.mlp.experts.linear_fc1.weight1 -Params for bucket 61 (44040192 elements, 44040192 padded size): - module.decoder.layers.0.mlp.experts.linear_fc2.weight6 - module.decoder.layers.0.mlp.experts.linear_fc2.weight4 - module.decoder.layers.0.mlp.experts.linear_fc2.weight2 - module.decoder.layers.0.mlp.experts.linear_fc2.weight7 - module.decoder.layers.0.mlp.experts.linear_fc2.weight5 - module.decoder.layers.0.mlp.experts.linear_fc2.weight3 -Params for bucket 62 (44040192 elements, 44040192 padded size): - module.decoder.layers.0.mlp.experts.linear_fc2.weight1 - module.decoder.layers.0.mlp.experts.linear_fc1.weight6 - module.decoder.layers.0.mlp.experts.linear_fc2.weight0 - module.decoder.layers.0.mlp.experts.linear_fc1.weight7 -Params for bucket 63 (44040192 elements, 44040192 padded size): - module.decoder.layers.0.mlp.experts.linear_fc1.weight5 - module.decoder.layers.0.mlp.experts.linear_fc1.weight4 - module.decoder.layers.0.mlp.experts.linear_fc1.weight3 -Params for bucket 64 (44040192 elements, 44040192 padded size): - module.decoder.layers.0.mlp.experts.linear_fc1.weight2 - module.decoder.layers.0.mlp.experts.linear_fc1.weight0 - module.decoder.layers.0.mlp.experts.linear_fc1.weight1 -INFO:megatron.core.optimizer:Setting up optimizer with config OptimizerConfig(optimizer='adam', lr=0.0001, min_lr=1e-05, decoupled_lr=None, decoupled_min_lr=None, weight_decay=0.1, fp8_recipe='mxfp8', fp16=False, bf16=True, reuse_grad_buf_for_mxfp8_param_ag=False, params_dtype=torch.bfloat16, use_precision_aware_optimizer=False, store_param_remainders=True, main_grads_dtype=torch.float32, main_params_dtype=torch.float32, exp_avg_dtype=torch.float32, exp_avg_sq_dtype=torch.float32, loss_scale=None, initial_loss_scale=4294967296, min_loss_scale=1.0, loss_scale_window=1000, hysteresis=2, adam_beta1=0.9, adam_beta2=0.999, adam_eps=1e-08, sgd_momentum=0.9, use_distributed_optimizer=True, overlap_param_gather_with_optimizer_step=False, optimizer_cpu_offload=False, optimizer_offload_fraction=1.0, use_torch_optimizer_for_cpu_offload=False, overlap_cpu_optimizer_d2h_h2d=False, pin_cpu_grads=True, pin_cpu_params=True, clip_grad=1.0, log_num_zeros_in_grad=False, barrier_with_L1_time=True, timers=, config_logger_dir='') -INFO:megatron.core.optimizer_param_scheduler:> learning rate decay style: cosine -[after model, optimizer, and learning rate scheduler are built] datetime: 2026-09-17 03:33:27 -> building train, validation, and test datasets ... - > datasets target sizes (minimum size): - train: 160 - validation: 0 - test: 0 -INFO:megatron.core.datasets.blended_megatron_dataset_config:Let split_matrix = [(0, 1.0), None, None] -> building train, validation, and test datasets for GPT ... -INFO:megatron.core.datasets.blended_megatron_dataset_builder:Building GPTDataset splits with sizes=(160, 0, 0) and config=GPTDatasetConfig(random_seed=1234, sequence_length=4096, blend=(['/workspace/data/wikitext/wikitext103_gpt2_text_document'], None), blend_per_split=None, split='100,0,0', split_matrix=[(0, 1.0), None, None], num_dataset_builder_threads=1, path_to_cache=None, mmap_bin_files=True, mock=False, tokenizer=, mid_level_dataset_surplus=0.005, reset_position_ids=False, reset_attention_mask=False, eod_mask_loss=False, create_attention_mask=True, drop_last_partial_validation_sequence=True, add_extra_token_to_sequence=True, object_storage_cache_path=None) -INFO:megatron.core.datasets.indexed_dataset:Load the _IndexReader from /workspace/data/wikitext/wikitext103_gpt2_text_document.idx -INFO:megatron.core.datasets.indexed_dataset: Extract the sequence lengths -INFO:megatron.core.datasets.indexed_dataset: Extract the sequence pointers -INFO:megatron.core.datasets.indexed_dataset: Extract the document indices -INFO:megatron.core.datasets.indexed_dataset:> total number of sequences: 1165029 -INFO:megatron.core.datasets.indexed_dataset:> total number of documents: 1801350 -INFO:megatron.core.datasets.gpt_dataset:Load the GPTDataset train indices -INFO:megatron.core.datasets.gpt_dataset: Load the document index from 68684a515e118253a8572c4517cf2ffe-GPTDataset-train-document_index.npy -INFO:megatron.core.datasets.gpt_dataset: Load the sample index from 68684a515e118253a8572c4517cf2ffe-GPTDataset-train-sample_index.npy -INFO:megatron.core.datasets.gpt_dataset: Load the shuffle index from 68684a515e118253a8572c4517cf2ffe-GPTDataset-train-shuffle_index.npy -INFO:megatron.core.datasets.gpt_dataset:> total number of samples: 29073 -> finished creating GPT datasets ... -[after dataloaders are built] datetime: 2026-09-17 03:33:27 -done with setup ... -(min, max) time across ranks (ms): - model-and-optimizer-setup ......................: (3188.55, 3194.23) - train/valid/test-data-iterators-setup ..........: (107.22, 135.77) -training ... -Overwriting rerun_state_machine.current_iteration from -1 to 0... -[before the start of training step] datetime: 2026-09-17 03:33:27 -/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. - warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") -/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. - warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") -/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. - warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") -/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. - warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") -/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. - warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") -/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. - warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") -/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. - warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") -/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. - warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") -/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. - warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") -/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. - warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") -/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. - warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") -/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. - warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") -/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. - warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") -/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. - warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") -/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. - warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") -/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. - warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") -/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. - warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") -/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. - warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") -/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. - warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") -/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. - warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") -/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. - warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") -/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. - warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") -/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. - warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") -/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. - warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") -/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. - warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") -/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. - warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") -/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. - warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") -/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. - warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") -/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. - warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") -/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. - warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") -/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. - warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") -/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. - warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") -/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. - warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") -/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. - warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") -/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. - warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") -/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. - warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") -/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. - warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") -/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. - warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") -/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. - warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") -/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. - warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") -/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. - warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") -/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. - warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") -/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. - warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") -/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. - warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") -/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. - warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") -/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. - warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") -/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. - warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") -/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:482: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. - warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") -/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. - warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") -/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. - warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") -/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. - warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") -/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. - warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") -/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. - warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") -/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. - warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") -/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. - warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") -/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:334: UserWarning: The num_dispatched_tokens is deprecated, it will be removed in the future. - warnings.warn("The num_dispatched_tokens is deprecated, it will be removed in the future.") -/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. - warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") -/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. - warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") -/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. - warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") -/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. - warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") -/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. - warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") -/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. - warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") -/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. - warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") -/workspace/DeepEP/deep_ep/hybrid_ep_buffer.py:336: UserWarning: The use_host_meta is deprecated, it will be removed in the future. - warnings.warn("The use_host_meta is deprecated, it will be removed in the future.") - [2026-09-17 03:33:57] iteration 1/ 10 | consumed samples: 16 | elapsed time per iteration (ms): 30515.0 | throughput per GPU (TFLOP/s/GPU): 6.4 | learning rate: 2.000000E-05 | global batch size: 16 | lm loss: 1.102374E+01 | loss scale: 1.0 | grad norm: 44.521 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 33920 | min safe capacity: 1.0352 | -Number of parameters in transformer block in billions: 23.22 -Number of parameters in embedding layers in billions: 0.41 -Total number of parameters in billions: 23.63 -Number of parameters in most loaded shard in billions: 23.6320 -Theoretical memory footprints: weight and optimizer=169029.43 MB -[Rank 0] (after 1 iterations) memory (MB) | allocated: 58125.50390625 | max allocated: 69789.0546875 | reserved: 76118.0 | max reserved: 76118.0 - [2026-09-17 03:33:58] iteration 2/ 10 | consumed samples: 32 | elapsed time per iteration (ms): 1083.3 | throughput per GPU (TFLOP/s/GPU): 179.9 | learning rate: 4.000000E-05 | global batch size: 16 | lm loss: 1.101671E+01 | loss scale: 1.0 | grad norm: 46.188 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 33920 | min safe capacity: 1.0352 | - [2026-09-17 03:33:59] iteration 3/ 10 | consumed samples: 48 | elapsed time per iteration (ms): 437.1 | throughput per GPU (TFLOP/s/GPU): 445.7 | learning rate: 6.000000E-05 | global batch size: 16 | lm loss: 1.046371E+01 | loss scale: 1.0 | grad norm: 211.813 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 33920 | min safe capacity: 1.0352 | -Capture CUDA graph for training!!! -CUDA graph capture done!!! - [2026-09-17 03:34:02] iteration 4/ 10 | consumed samples: 64 | elapsed time per iteration (ms): 3593.9 | throughput per GPU (TFLOP/s/GPU): 54.2 | learning rate: 8.000000E-05 | global batch size: 16 | lm loss: 1.314799E+01 | loss scale: 1.0 | grad norm: 39.078 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 33920 | min safe capacity: 1.0352 | - [2026-09-17 03:34:03] iteration 5/ 10 | consumed samples: 80 | elapsed time per iteration (ms): 360.8 | throughput per GPU (TFLOP/s/GPU): 540.0 | learning rate: 1.000000E-04 | global batch size: 16 | lm loss: 1.038756E+01 | loss scale: 1.0 | grad norm: 17.511 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 33920 | min safe capacity: 1.0352 | - [2026-09-17 03:34:03] iteration 6/ 10 | consumed samples: 96 | elapsed time per iteration (ms): 383.9 | throughput per GPU (TFLOP/s/GPU): 507.6 | learning rate: 7.750000E-05 | global batch size: 16 | lm loss: 1.033255E+01 | loss scale: 1.0 | grad norm: 29.428 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 34176 | min safe capacity: 1.0430 | - [2026-09-17 03:34:06] iteration 7/ 10 | consumed samples: 112 | elapsed time per iteration (ms): 3303.0 | throughput per GPU (TFLOP/s/GPU): 59.0 | learning rate: 3.250000E-05 | global batch size: 16 | lm loss: 1.003726E+01 | loss scale: 1.0 | grad norm: 7.175 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 34176 | min safe capacity: 1.0430 | - [2026-09-17 03:34:07] iteration 8/ 10 | consumed samples: 128 | elapsed time per iteration (ms): 371.2 | throughput per GPU (TFLOP/s/GPU): 524.8 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 9.514786E+00 | loss scale: 1.0 | grad norm: 5.913 | number of skipped iterations: 0 | number of nan iterations: 0 | peak recv tokens: 34176 | min safe capacity: 1.0430 | -Capture range started in the application. -Capture range ended in the application. -Generating '/tmp/nsys-report-2835.qdstrm' - [1/1] [0% ] G3_K8__ultraep__matched__seq4096__gbs16__cap1p09375.nsys-repProcessing events... - [1/1] [0% ] G3_K8__ultraep__matched__seq4096__gbs16__cap1p09375.nsys-rep [1/1] [0% ] G3_K8__ultraep__matched__seq4096__gbs16__cap1p09375.nsys-rep [1/1] [5% ] G3_K8__ultraep__matched__seq4096__gbs16__cap1p09375.nsys-rep [1/1] [7% ] G3_K8__ultraep__matched__seq4096__gbs16__cap1p09375.nsys-rep [1/1] [9% ] G3_K8__ultraep__matched__seq4096__gbs16__cap1p09375.nsys-rep [1/1] [12% ] G3_K8__ultraep__matched__seq4096__gbs16__cap1p09375.nsys-rep [1/1] [14% ] G3_K8__ultraep__matched__seq4096__gbs16__cap1p09375.nsys-rep [1/1] [=16% ] G3_K8__ultraep__matched__seq4096__gbs16__cap1p09375.nsys-rep [1/1] [==19% ] G3_K8__ultraep__matched__seq4096__gbs16__cap1p09375.nsys-rep [1/1] [==21% ] G3_K8__ultraep__matched__seq4096__gbs16__cap1p09375.nsys-rep [1/1] [===23% ] G3_K8__ultraep__matched__seq4096__gbs16__cap1p09375.nsys-rep [1/1] [====26% ] G3_K8__ultraep__matched__seq4096__gbs16__cap1p09375.nsys-rep [1/1] [====28% ] G3_K8__ultraep__matched__seq4096__gbs16__cap1p09375.nsys-rep [1/1] [=====30% ] G3_K8__ultraep__matched__seq4096__gbs16__cap1p09375.nsys-rep [1/1] [======33% ] G3_K8__ultraep__matched__seq4096__gbs16__cap1p09375.nsys-rep [1/1] [======34% ] G3_K8__ultraep__matched__seq4096__gbs16__cap1p09375.nsys-rep [1/1] [======35% ] G3_K8__ultraep__matched__seq4096__gbs16__cap1p09375.nsys-rep [1/1] [=======36% ] G3_K8__ultraep__matched__seq4096__gbs16__cap1p09375.nsys-rep [1/1] [=======37% ] G3_K8__ultraep__matched__seq4096__gbs16__cap1p09375.nsys-rep [1/1] [=======38% ] G3_K8__ultraep__matched__seq4096__gbs16__cap1p09375.nsys-rep [1/1] [=======39% ] G3_K8__ultraep__matched__seq4096__gbs16__cap1p09375.nsys-rep [1/1] [========40% ] G3_K8__ultraep__matched__seq4096__gbs16__cap1p09375.nsys-rep [1/1] [========41% ] G3_K8__ultraep__matched__seq4096__gbs16__cap1p09375.nsys-rep [1/1] [========42% ] G3_K8__ultraep__matched__seq4096__gbs16__cap1p09375.nsys-rep [1/1] [=========43% ] G3_K8__ultraep__matched__seq4096__gbs16__cap1p09375.nsys-rep [1/1] [=========45% ] G3_K8__ultraep__matched__seq4096__gbs16__cap1p09375.nsys-rep [1/1] [=========46% ] G3_K8__ultraep__matched__seq4096__gbs16__cap1p09375.nsys-rep [1/1] [==========47% ] G3_K8__ultraep__matched__seq4096__gbs16__cap1p09375.nsys-rep [1/1] [==========48% ] G3_K8__ultraep__matched__seq4096__gbs16__cap1p09375.nsys-rep [1/1] [===========50% ] G3_K8__ultraep__matched__seq4096__gbs16__cap1p09375.nsys-rep [1/1] [===========51% ] G3_K8__ultraep__matched__seq4096__gbs16__cap1p09375.nsys-rep [1/1] [===========52% ] G3_K8__ultraep__matched__seq4096__gbs16__cap1p09375.nsys-rep [1/1] [===========53% ] G3_K8__ultraep__matched__seq4096__gbs16__cap1p09375.nsys-rep [1/1] [============54% ] G3_K8__ultraep__matched__seq4096__gbs16__cap1p09375.nsys-rep [1/1] [============56% ] G3_K8__ultraep__matched__seq4096__gbs16__cap1p09375.nsys-rep [1/1] [============57% ] G3_K8__ultraep__matched__seq4096__gbs16__cap1p09375.nsys-rep [1/1] [==============61% ] G3_K8__ultraep__matched__seq4096__gbs16__cap1p09375.nsys-rep [1/1] [========================100%] G3_K8__ultraep__matched__seq4096__gbs16__cap1p09375.nsys-rep [1/1] [========================100%] G3_K8__ultraep__matched__seq4096__gbs16__cap1p09375.nsys-rep -Collection stop is not allowed in this state. -W0917 03:35:37.900000 969559 torch/distributed/elastic/agent/server/api.py:719] Received 15 death signal, shutting down workers -W0917 03:35:37.901000 969559 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 969792 closing signal SIGTERM -Generated: - /workspace/granularity_next_b200/tiera_topk_trace/ultraep/tierbtrace/tierbtrace_gbs16/G3_K8__ultraep__matched__seq4096__gbs16__cap1p09375/nsys/G3_K8__ultraep__matched__seq4096__gbs16__cap1p09375.nsys-rep -W0917 03:35:37.906000 969559 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 969793 closing signal SIGTERM -W0917 03:35:37.908000 969559 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 969795 closing signal SIGTERM -W0917 03:35:37.919000 969559 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 969796 closing signal SIGTERM -W0917 03:35:37.921000 969559 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 969798 closing signal SIGTERM -W0917 03:35:37.933000 969559 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 969799 closing signal SIGTERM -W0917 03:35:37.935000 969559 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 969800 closing signal SIGTERM -W0917 03:35:37.939000 969559 torch/distributed/elastic/multiprocessing/api.py:900] Sending process 969801 closing signal SIGTERM -Traceback (most recent call last): - File "/usr/local/bin/torchrun", line 8, in - sys.exit(main()) - ^^^^^^ - File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/multiprocessing/errors/__init__.py", line 355, in wrapper - return f(*args, **kwargs) - ^^^^^^^^^^^^^^^^^^ - File "/usr/local/lib/python3.12/dist-packages/torch/distributed/run.py", line 892, in main - run(args) - File "/usr/local/lib/python3.12/dist-packages/torch/distributed/run.py", line 883, in run - elastic_launch( - File "/usr/local/lib/python3.12/dist-packages/torch/distributed/launcher/api.py", line 139, in __call__ - return launch_agent(self._config, self._entrypoint, list(args)) - ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ - File "/usr/local/lib/python3.12/dist-packages/torch/distributed/launcher/api.py", line 261, in launch_agent - result = agent.run() - ^^^^^^^^^^^ - File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/metrics/api.py", line 138, in wrapper - result = f(*args, **kwargs) - ^^^^^^^^^^^^^^^^^^ - File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/agent/server/api.py", line 711, in run - result = self._invoke_run(role) - ^^^^^^^^^^^^^^^^^^^^^^ - File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/agent/server/api.py", line 870, in _invoke_run - time.sleep(monitor_interval) - File "/usr/local/lib/python3.12/dist-packages/torch/distributed/elastic/multiprocessing/api.py", line 84, in _terminate_process_handler - raise SignalException(f"Process {os.getpid()} got signal: {sigval}", sigval=sigval) -torch.distributed.elastic.multiprocessing.api.SignalException: Process 969559 got signal: 15 -The target application terminated. One or more process it created re-parented. -Waiting for termination of re-parented processes. -Use the `--wait` option to modify this behavior. -There are no active sessions. diff --git a/granularity_next_b200/tiera_topk_trace/ultraep/tierbtrace/tierbtrace_gbs16/trace_results.tsv b/granularity_next_b200/tiera_topk_trace/ultraep/tierbtrace/tierbtrace_gbs16/trace_results.tsv index 2ca0de1646c17bf427b13ced641d613518aa11c5..fb0db3b164187208f74a23f3b63487dd92fad1cf 100644 --- a/granularity_next_b200/tiera_topk_trace/ultraep/tierbtrace/tierbtrace_gbs16/trace_results.tsv +++ b/granularity_next_b200/tiera_topk_trace/ultraep/tierbtrace/tierbtrace_gbs16/trace_results.tsv @@ -1,5 +1,3 @@ model backend capacity status report async_overlap_json G2_K2 ultraep 1.15625 OK /workspace/granularity_next_b200/tiera_topk_trace/ultraep/tierbtrace/tierbtrace_gbs16/G2_K2__ultraep__matched__seq4096__gbs16__cap1p15625/nsys/G2_K2__ultraep__matched__seq4096__gbs16__cap1p15625.nsys-rep G3_K2 ultraep 1.1875 OK /workspace/granularity_next_b200/tiera_topk_trace/ultraep/tierbtrace/tierbtrace_gbs16/G3_K2__ultraep__matched__seq4096__gbs16__cap1p1875/nsys/G3_K2__ultraep__matched__seq4096__gbs16__cap1p1875.nsys-rep -G3_K8 ultraep 1.09375 OK /workspace/granularity_next_b200/tiera_topk_trace/ultraep/tierbtrace/tierbtrace_gbs16/G3_K8__ultraep__matched__seq4096__gbs16__cap1p09375/nsys/G3_K8__ultraep__matched__seq4096__gbs16__cap1p09375.nsys-rep -G4_K2 ultraep 1.265625 OK /workspace/granularity_next_b200/tiera_topk_trace/ultraep/tierbtrace/tierbtrace_gbs16/G4_K2__ultraep__matched__seq4096__gbs16__cap1p265625/nsys/G4_K2__ultraep__matched__seq4096__gbs16__cap1p265625.nsys-rep