diff --git a/tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/G3__echo__one__seq4096__gbs16__cap1p072265625/rep2/mem_samples.csv b/tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/G3__echo__one__seq4096__gbs16__cap1p072265625/rep2/mem_samples.csv index 38c736966af786d820af322dc0d2173ac791eaf6..0121d74808e64c12ac732ce4b9f8b075217d849b 100644 --- a/tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/G3__echo__one__seq4096__gbs16__cap1p072265625/rep2/mem_samples.csv +++ b/tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/G3__echo__one__seq4096__gbs16__cap1p072265625/rep2/mem_samples.csv @@ -142,3 +142,107 @@ 5, 68605 6, 68605 7, 68605 +0, 68587 +1, 68605 +2, 68605 +3, 68605 +4, 68659 +5, 68605 +6, 68605 +7, 68605 +0, 68587 +1, 68605 +2, 68605 +3, 68605 +4, 68659 +5, 68605 +6, 68605 +7, 68605 +0, 68587 +1, 68605 +2, 68605 +3, 68605 +4, 68659 +5, 68605 +6, 68605 +7, 68605 +0, 78681 +1, 78677 +2, 78705 +3, 78695 +4, 78739 +5, 78701 +6, 78349 +7, 78681 +0, 79493 +1, 79505 +2, 79573 +3, 79531 +4, 79615 +5, 79567 +6, 79237 +7, 79539 +0, 80441 +1, 80445 +2, 80445 +3, 80445 +4, 80275 +5, 80445 +6, 80333 +7, 80445 +0, 81053 +1, 81075 +2, 81077 +3, 81089 +4, 80925 +5, 81101 +6, 81001 +7, 81101 +0, 72605 +1, 73265 +2, 70671 +3, 72123 +4, 72567 +5, 71461 +6, 69185 +7, 72349 +0, 116195 +1, 116087 +2, 116087 +3, 116087 +4, 116303 +5, 116087 +6, 116423 +7, 116087 +0, 116195 +1, 116087 +2, 116087 +3, 116087 +4, 116303 +5, 116087 +6, 116423 +7, 116087 +0, 115659 +1, 115813 +2, 115487 +3, 115167 +4, 115447 +5, 115813 +6, 115887 +7, 115813 +0, 1833 +1, 1834 +2, 2090 +3, 115167 +4, 3114 +5, 1834 +6, 1834 +7, 1834 +0, 0 +1, 0 +2, 0 +3, 0 +4, 0 +5, 0 +6, 0 +7, 0 diff --git a/tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/G3__echo__one__seq4096__gbs16__cap1p072265625/rep2/train.log b/tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/G3__echo__one__seq4096__gbs16__cap1p072265625/rep2/train.log index b3fd763057cbba780780061fe84038076c771da7..32c84a2e20e249c64002c458ce4c21198d6c2b9f 100644 --- a/tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/G3__echo__one__seq4096__gbs16__cap1p072265625/rep2/train.log +++ b/tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/G3__echo__one__seq4096__gbs16__cap1p072265625/rep2/train.log @@ -1171,3 +1171,73 @@ 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: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.") + [2026-08-30 06:30:51] iteration 1/ 12 | consumed samples: 16 | elapsed time per iteration (ms): 38327.2 | throughput per GPU (TFLOP/s/GPU): 8.7 | learning rate: 2.000000E-05 | global batch size: 16 | lm loss: 1.103551E+01 | loss scale: 1.0 | grad norm: 44.037 | number of skipped iterations: 0 | number of nan iterations: 0 |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: 58851.314453125 | max allocated: 68669.056640625 | reserved: 73028.0 | max reserved: 73028.0 + [2026-08-30 06:30:52] iteration 2/ 12 | consumed samples: 32 | elapsed time per iteration (ms): 593.5 | throughput per GPU (TFLOP/s/GPU): 561.7 | learning rate: 4.000000E-05 | global batch size: 16 | lm loss: 1.102915E+01 | loss scale: 1.0 | grad norm: 45.934 | number of skipped iterations: 0 | number of nan iterations: 0 | + [2026-08-30 06:30:52] iteration 3/ 12 | consumed samples: 48 | elapsed time per iteration (ms): 500.3 | throughput per GPU (TFLOP/s/GPU): 666.4 | learning rate: 6.000000E-05 | global batch size: 16 | lm loss: 1.045494E+01 | loss scale: 1.0 | grad norm: 206.498 | number of skipped iterations: 0 | number of nan iterations: 0 | +Capture CUDA graph for training!!! +CUDA graph capture done!!! + [2026-08-30 06:30:55] iteration 4/ 12 | consumed samples: 64 | elapsed time per iteration (ms): 2359.7 | throughput per GPU (TFLOP/s/GPU): 141.3 | learning rate: 8.000000E-05 | global batch size: 16 | lm loss: 1.318256E+01 | loss scale: 1.0 | grad norm: 36.176 | number of skipped iterations: 0 | number of nan iterations: 0 | + [2026-08-30 06:30:55] iteration 5/ 12 | consumed samples: 80 | elapsed time per iteration (ms): 413.3 | throughput per GPU (TFLOP/s/GPU): 806.7 | learning rate: 1.000000E-04 | global batch size: 16 | lm loss: 1.039611E+01 | loss scale: 1.0 | grad norm: 16.077 | number of skipped iterations: 0 | number of nan iterations: 0 | + [2026-08-30 06:30:55] iteration 6/ 12 | consumed samples: 96 | elapsed time per iteration (ms): 412.1 | throughput per GPU (TFLOP/s/GPU): 809.1 | learning rate: 7.750000E-05 | global batch size: 16 | lm loss: 1.042192E+01 | loss scale: 1.0 | grad norm: 36.479 | number of skipped iterations: 0 | number of nan iterations: 0 | + [2026-08-30 06:30:56] iteration 7/ 12 | consumed samples: 112 | elapsed time per iteration (ms): 415.6 | throughput per GPU (TFLOP/s/GPU): 802.2 | learning rate: 3.250000E-05 | global batch size: 16 | lm loss: 1.002654E+01 | loss scale: 1.0 | grad norm: 7.845 | number of skipped iterations: 0 | number of nan iterations: 0 | + [2026-08-30 06:30:56] iteration 8/ 12 | consumed samples: 128 | elapsed time per iteration (ms): 414.6 | throughput per GPU (TFLOP/s/GPU): 804.2 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 9.529053E+00 | loss scale: 1.0 | grad norm: 5.793 | number of skipped iterations: 0 | number of nan iterations: 0 | + [2026-08-30 06:30:57] iteration 9/ 12 | consumed samples: 144 | elapsed time per iteration (ms): 412.4 | throughput per GPU (TFLOP/s/GPU): 808.5 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 9.294894E+00 | loss scale: 1.0 | grad norm: 5.640 | number of skipped iterations: 0 | number of nan iterations: 0 | + [2026-08-30 06:30:57] iteration 10/ 12 | consumed samples: 160 | elapsed time per iteration (ms): 415.1 | throughput per GPU (TFLOP/s/GPU): 803.1 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 9.137465E+00 | loss scale: 1.0 | grad norm: 5.465 | number of skipped iterations: 0 | number of nan iterations: 0 | + [2026-08-30 06:30:57] iteration 11/ 12 | consumed samples: 176 | elapsed time per iteration (ms): 414.0 | throughput per GPU (TFLOP/s/GPU): 805.3 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 9.016632E+00 | loss scale: 1.0 | grad norm: 5.265 | number of skipped iterations: 0 | number of nan iterations: 0 | + [2026-08-30 06:30:58] iteration 12/ 12 | consumed samples: 192 | elapsed time per iteration (ms): 412.5 | throughput per GPU (TFLOP/s/GPU): 808.1 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 8.893719E+00 | loss scale: 1.0 | grad norm: 5.342 | number of skipped iterations: 0 | number of nan iterations: 0 | +[after training is done] datetime: 2026-08-30 06:30:58 +[rank0]:[W830 06:30:59.676447285 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()) diff --git a/tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/G3__echo__one__seq4096__gbs16__cap1p072265625/rep3/mem_samples.csv b/tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/G3__echo__one__seq4096__gbs16__cap1p072265625/rep3/mem_samples.csv new file mode 100644 index 0000000000000000000000000000000000000000..91950a2e9d6e3864432d0986c344d1283209b668 --- /dev/null +++ b/tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/G3__echo__one__seq4096__gbs16__cap1p072265625/rep3/mem_samples.csv @@ -0,0 +1,248 @@ +0, 0 +1, 0 +2, 0 +3, 0 +4, 0 +5, 0 +6, 0 +7, 0 +0, 0 +1, 0 +2, 0 +3, 0 +4, 0 +5, 0 +6, 0 +7, 0 +0, 1 +1, 1 +2, 1 +3, 1 +4, 1 +5, 1 +6, 1 +7, 1 +0, 4 +1, 4 +2, 4 +3, 4 +4, 1 +5, 1 +6, 1 +7, 1 +0, 4 +1, 4 +2, 4 +3, 4 +4, 4 +5, 4 +6, 4 +7, 4 +0, 821 +1, 819 +2, 819 +3, 821 +4, 819 +5, 819 +6, 819 +7, 819 +0, 1537 +1, 1537 +2, 1537 +3, 1537 +4, 1537 +5, 1537 +6, 1537 +7, 1537 +0, 2231 +1, 2231 +2, 2231 +3, 2231 +4, 2231 +5, 2231 +6, 2231 +7, 2231 +0, 38591 +1, 38609 +2, 38595 +3, 38597 +4, 38661 +5, 38623 +6, 38621 +7, 38617 +0, 38893 +1, 38885 +2, 38915 +3, 38891 +4, 38885 +5, 38935 +6, 38921 +7, 38931 +0, 41457 +1, 41441 +2, 41441 +3, 41441 +4, 41441 +5, 41441 +6, 41441 +7, 41441 +0, 41487 +1, 41471 +2, 41471 +3, 41471 +4, 41471 +5, 41471 +6, 41471 +7, 41471 +0, 41583 +1, 41557 +2, 41569 +3, 41571 +4, 41581 +5, 41593 +6, 41579 +7, 41581 +0, 42407 +1, 42391 +2, 42391 +3, 42391 +4, 42391 +5, 42391 +6, 42391 +7, 42391 +0, 42957 +1, 42941 +2, 42941 +3, 42941 +4, 42941 +5, 42941 +6, 42941 +7, 42941 +0, 43997 +1, 43981 +2, 43981 +3, 43981 +4, 43981 +5, 43981 +6, 43981 +7, 43981 +0, 44795 +1, 44779 +2, 44779 +3, 44779 +4, 44779 +5, 44779 +6, 44779 +7, 44779 +0, 68587 +1, 68605 +2, 68605 +3, 68605 +4, 68659 +5, 68605 +6, 68605 +7, 68605 +0, 68587 +1, 68605 +2, 68605 +3, 68605 +4, 68659 +5, 68605 +6, 68605 +7, 68605 +0, 68587 +1, 68605 +2, 68605 +3, 68605 +4, 68659 +5, 68605 +6, 68605 +7, 68605 +0, 68587 +1, 68605 +2, 68605 +3, 68605 +4, 68659 +5, 68605 +6, 68605 +7, 68605 +0, 78301 +1, 78665 +2, 78679 +3, 78649 +4, 78725 +5, 78699 +6, 78689 +7, 78725 +0, 79117 +1, 79479 +2, 79497 +3, 79529 +4, 79577 +5, 79543 +6, 79553 +7, 79567 +0, 80105 +1, 80445 +2, 80445 +3, 80445 +4, 80499 +5, 80445 +6, 80445 +7, 80445 +0, 80737 +1, 81073 +2, 81077 +3, 81063 +4, 81161 +5, 81111 +6, 81133 +7, 81093 +0, 82297 +1, 73885 +2, 73879 +3, 78741 +4, 74135 +5, 73879 +6, 74099 +7, 74171 +0, 116531 +1, 116087 +2, 116087 +3, 116311 +4, 116303 +5, 116087 +6, 116311 +7, 116311 +0, 116531 +1, 116087 +2, 116087 +3, 116311 +4, 116303 +5, 116087 +6, 116311 +7, 116311 +0, 115611 +1, 115167 +2, 115171 +3, 115455 +4, 116029 +5, 115487 +6, 115391 +7, 115391 +0, 1831 +1, 2090 +2, 115167 +3, 3114 +4, 1834 +5, 1834 +6, 1834 +7, 1834 +0, 0 +1, 0 +2, 0 +3, 0 +4, 0 +5, 0 +6, 0 +7, 0 diff --git a/tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/G3__echo__one__seq4096__gbs16__cap1p072265625/rep3/run.env b/tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/G3__echo__one__seq4096__gbs16__cap1p072265625/rep3/run.env new file mode 100644 index 0000000000000000000000000000000000000000..7ee8893f372854eff94562c2a689fa84b52ea66e --- /dev/null +++ b/tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/G3__echo__one__seq4096__gbs16__cap1p072265625/rep3/run.env @@ -0,0 +1,70 @@ +backend=echo +config=G3 +family=tierA +repo=/workspace/Megatron-MoE-Echo +branch=bracing-regions-through-kernels +commit=9a236793 +num_layers=16 +hidden=4096 +ffn_hidden=14336 +moe_ffn_hidden=1792 +num_experts=64 +topk=16 +experts_per_rank=8 +expert_params=22020096 +heads=32 +kv_groups=8 +kv_channels= +qk_layernorm=0 +shared_ffn=0 +shared_expert_overlap=0 +moe_token_dispatcher=flex +mla=0 +dense_first_layers=0 +moe_layers=16 +rotary_base=1000000 +rotary_scaling= +qkv_bias=0 +router_balancing=aux_loss +aux_loss_coeff=1e-2 +router_pre_softmax=0 +router_fusion=0 +router_dtype=fp32 +device_initiated_grouped_gemm=1 +norm_epsilon=1e-5 +seq_length=4096 +micro_batch_size=1 +global_batch_size=16 +num_microbatches=2 +assignments_per_rank=65536 +mean_tokens_per_expert=8192 +slot_policy=one +slots_per_rank=1 +capacity_mode=static +capacity=1.072265625 +cuda_graph=1 +probe_mode=0 +async_grad_reduce=0 +branch_role=timing +memory_dump_mode=none +memory_snapshot_path= +cuda_graph_warmup_steps=3 +recompute=none +force_balance=0 +train_iters=12 +ep_size=8 +echo_dispatch_overlap=1 +echo_dispatch_recompute=0 +ultraep_quota_min_tokens= +ultraep_weight_sync_join= +profile_enable=0 +kernel_trace=0 +fp8_mode=mxfp8 +fp8_recipe=mxfp8 +training_seed=1234 +learning_rate=1e-4 +min_learning_rate=1.0e-5 +lr_warmup_iters=5 +state_dump_enable=0 +state_dump_dir= +moonep_assert_no_pending_reduce=0 diff --git a/tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/G3__echo__one__seq4096__gbs16__cap1p072265625/rep3/train.log b/tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/G3__echo__one__seq4096__gbs16__cap1p072265625/rep3/train.log new file mode 100644 index 0000000000000000000000000000000000000000..964a27853790c9e7cfe7922a926eb0414dbec84c --- /dev/null +++ b/tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/G3__echo__one__seq4096__gbs16__cap1p072265625/rep3/train.log @@ -0,0 +1,1243 @@ +============================================================== + G3 / echo (slots/rank=1 policy=one) + E=64 K=16 H'=1792 m=8 + L=16 seq=4096 gbs=16 (2 microbatches) + cap_mode=static cap=1.072265625 cg=1 recompute=none + memory_dump=none snapshots=/workspace/tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/G3__echo__one__seq4096__gbs16__cap1p072265625/rep3/memory_snapshots/snapshot.pickle + repo=/workspace/Megatron-MoE-Echo @ 9a236793 +============================================================== +W0830 06:31:08.968000 615915 torch/distributed/run.py:766] +W0830 06:31:08.968000 615915 torch/distributed/run.py:766] ***************************************** +W0830 06:31:08.968000 615915 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. +W0830 06:31:08.968000 615915 torch/distributed/run.py:766] ***************************************** +/workspace/Megatron-MoE-Echo/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-MoE-Echo/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_deepep_num_sms .............................. 20 + moe_echo_enable_random_offloading ............... False + moe_echo_expert_dispatch_overlap ................ True + moe_echo_expert_dispatcher_type ................. hybridep + moe_echo_recompute_expert_dispatch .............. False + moe_enable_deepep ............................... False + moe_enable_echo ................................. True + moe_enable_hybridep ............................. 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 ............................ 8 + moe_pad_expert_input_to_capacity ................ False + moe_per_layer_logging ........................... False + moe_permute_fusion .............................. False + moe_received_token_capacity ..................... 1.072265625 + 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 ................................. 16 + 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_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 ......................................... False + profile_ranks ................................... [0] + profile_step_end ................................ 12 + profile_step_start .............................. 10 + 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 ..................................... 12 + 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 ... + > 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 ... +/workspace/Megatron-MoE-Echo/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-MoE-Echo/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-MoE-Echo/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-MoE-Echo/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-MoE-Echo/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-MoE-Echo/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 2 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 1 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 4 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. is connected to 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 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 4 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 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 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[Gloo] Rank peer ranks. 0Expected number of connected peer ranks is : is connected to 0[Gloo] Rank 0 +0 peer ranks. is connected to Expected number of connected peer ranks is : 00 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 +[rank5]:[W830 06:31:26.834964518 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]:[W830 06:31:26.834965282 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. +[rank7]:[W830 06:31:26.834991806 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. +[rank3]:[W830 06:31:26.834997259 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]:[W830 06:31:26.835037304 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]:[W830 06:31:26.835101677 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. +[rank1]:[W830 06:31:26.835392786 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. +> compiling dataset index builder ... +make: Entering directory '/workspace/Megatron-MoE-Echo/megatron/core/datasets' +make: Nothing to be done for 'default'. +make: Leaving directory '/workspace/Megatron-MoE-Echo/megatron/core/datasets' +>>> done with dataset index builder. Compilation time: 0.036 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]:[W830 06:31:26.988264722 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: 3.818 seconds +time to initialize megatron (seconds): 7.366 +[after megatron is initialized] datetime: 2026-08-30 06:31:32 +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( + > number of parameters on (tensor, pipeline) model parallel rank (0, 0): 3906080768 +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.layer_norm_weight + module.decoder.layers.15.mlp.router.weight + module.decoder.layers.15.self_attention.linear_proj.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_proj.weight + module.decoder.layers.14.self_attention.linear_qkv.layer_norm_weight + module.decoder.layers.14.pre_mlp_layernorm.weight + module.decoder.layers.14.self_attention.linear_qkv.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.pre_mlp_layernorm.weight + module.decoder.layers.13.self_attention.linear_qkv.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.mlp.router.weight + module.decoder.layers.9.self_attention.linear_proj.weight + module.decoder.layers.9.self_attention.linear_qkv.layer_norm_weight + module.decoder.layers.9.self_attention.linear_qkv.weight + module.decoder.layers.9.pre_mlp_layernorm.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_proj.weight + module.decoder.layers.8.self_attention.linear_qkv.layer_norm_weight +Params for bucket 10 (42213376 elements, 42213376 padded size): + 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 +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.pre_mlp_layernorm.weight + module.decoder.layers.6.self_attention.linear_qkv.weight +Params for bucket 12 (42213376 elements, 42213376 padded size): + module.decoder.layers.5.mlp.router.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.self_attention.linear_qkv.layer_norm_weight +Params for bucket 13 (42213376 elements, 42213376 padded size): + module.decoder.layers.4.pre_mlp_layernorm.weight + module.decoder.layers.4.self_attention.linear_qkv.weight + 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 +Params for bucket 14 (42213376 elements, 42213376 padded size): + module.decoder.layers.3.mlp.router.weight + module.decoder.layers.3.self_attention.linear_qkv.layer_norm_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.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 +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 + 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 (42213376 elements, 42213376 padded size): + module.decoder.layers.0.self_attention.linear_proj.weight + module.decoder.layers.0.mlp.router.weight + module.decoder.layers.0.pre_mlp_layernorm.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: 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 +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 + module.decoder.layers.15.mlp.experts.linear_fc1.weight7 + module.decoder.layers.15.mlp.experts.linear_fc1.weight6 +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.weight7 + module.decoder.layers.14.mlp.experts.linear_fc2.weight5 + module.decoder.layers.14.mlp.experts.linear_fc2.weight3 + 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 +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_fc1.weight6 + module.decoder.layers.14.mlp.experts.linear_fc2.weight0 + 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.weight3 + module.decoder.layers.14.mlp.experts.linear_fc1.weight4 +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.weight0 + module.decoder.layers.14.mlp.experts.linear_fc1.weight1 +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.weight6 + module.decoder.layers.13.mlp.experts.linear_fc2.weight4 + 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_fc1.weight6 + module.decoder.layers.13.mlp.experts.linear_fc1.weight7 + module.decoder.layers.13.mlp.experts.linear_fc2.weight0 +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.weight4 + module.decoder.layers.13.mlp.experts.linear_fc1.weight3 +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.weight3 + module.decoder.layers.12.mlp.experts.linear_fc2.weight5 +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_fc1.weight7 + 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.weight3 + module.decoder.layers.12.mlp.experts.linear_fc1.weight4 +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 + module.decoder.layers.12.mlp.experts.linear_fc1.weight1 +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.weight3 + module.decoder.layers.11.mlp.experts.linear_fc2.weight5 + 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_fc2.weight0 + module.decoder.layers.11.mlp.experts.linear_fc1.weight6 + module.decoder.layers.11.mlp.experts.linear_fc1.weight7 +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.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.weight7 + module.decoder.layers.10.mlp.experts.linear_fc1.weight6 + module.decoder.layers.10.mlp.experts.linear_fc2.weight0 +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.weight6 + module.decoder.layers.9.mlp.experts.linear_fc2.weight4 + 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.weight6 + module.decoder.layers.9.mlp.experts.linear_fc1.weight7 +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.weight7 + module.decoder.layers.8.mlp.experts.linear_fc2.weight3 + module.decoder.layers.8.mlp.experts.linear_fc2.weight5 + 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 +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.weight6 + module.decoder.layers.8.mlp.experts.linear_fc2.weight0 + module.decoder.layers.8.mlp.experts.linear_fc1.weight7 +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.weight3 + module.decoder.layers.8.mlp.experts.linear_fc1.weight4 +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.weight3 + module.decoder.layers.7.mlp.experts.linear_fc2.weight5 +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_fc1.weight7 + 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.weight3 + module.decoder.layers.7.mlp.experts.linear_fc1.weight4 +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.weight6 + module.decoder.layers.6.mlp.experts.linear_fc2.weight4 + 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.weight3 + module.decoder.layers.6.mlp.experts.linear_fc2.weight5 +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.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.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.weight4 + module.decoder.layers.5.mlp.experts.linear_fc2.weight6 + 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 +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_fc1.weight7 + module.decoder.layers.5.mlp.experts.linear_fc2.weight0 +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 +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.weight5 + module.decoder.layers.4.mlp.experts.linear_fc2.weight3 + module.decoder.layers.4.mlp.experts.linear_fc2.weight6 + module.decoder.layers.4.mlp.experts.linear_fc2.weight4 + 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_fc1.weight7 + module.decoder.layers.4.mlp.experts.linear_fc1.weight6 + module.decoder.layers.4.mlp.experts.linear_fc2.weight0 +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.weight4 + module.decoder.layers.4.mlp.experts.linear_fc1.weight3 +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.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.weight6 + module.decoder.layers.3.mlp.experts.linear_fc1.weight7 +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.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.weight3 + module.decoder.layers.2.mlp.experts.linear_fc2.weight5 + 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_fc2.weight0 + module.decoder.layers.2.mlp.experts.linear_fc1.weight6 + module.decoder.layers.2.mlp.experts.linear_fc1.weight7 +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.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.weight4 + module.decoder.layers.1.mlp.experts.linear_fc2.weight6 + module.decoder.layers.1.mlp.experts.linear_fc2.weight2 + module.decoder.layers.1.mlp.experts.linear_fc2.weight7 + module.decoder.layers.1.mlp.experts.linear_fc2.weight3 + module.decoder.layers.1.mlp.experts.linear_fc2.weight5 +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.weight6 + module.decoder.layers.1.mlp.experts.linear_fc1.weight7 + module.decoder.layers.1.mlp.experts.linear_fc2.weight0 +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.weight7 + module.decoder.layers.0.mlp.experts.linear_fc2.weight5 + module.decoder.layers.0.mlp.experts.linear_fc2.weight3 + 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 +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_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 + module.decoder.layers.0.mlp.experts.linear_fc1.weight3 + module.decoder.layers.0.mlp.experts.linear_fc1.weight4 +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: 2026-08-30 06:31:33 +> building train, validation, and test datasets ... + > datasets target sizes (minimum size): + train: 192 + 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=(192, 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 44ed1ee5bcb9e5b030387b8d27a6c406-GPTDataset-train-document_index.npy +INFO:megatron.core.datasets.gpt_dataset: Load the sample index from 44ed1ee5bcb9e5b030387b8d27a6c406-GPTDataset-train-sample_index.npy +INFO:megatron.core.datasets.gpt_dataset: Load the shuffle index from 44ed1ee5bcb9e5b030387b8d27a6c406-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: 2026-08-30 06:31:33 +done with setup ... +training ... +(min, max) time across ranks (ms): + model-and-optimizer-setup ......................: (466.10, 472.76) + train/valid/test-data-iterators-setup ..........: (63.53, 72.24) +Overwriting rerun_state_machine.current_iteration from -1 to 0... +[before the start of training step] datetime: 2026-08-30 06:31:33 +/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: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: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: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: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.") + [2026-08-30 06:32:11] iteration 1/ 12 | consumed samples: 16 | elapsed time per iteration (ms): 38382.2 | throughput per GPU (TFLOP/s/GPU): 8.7 | learning rate: 2.000000E-05 | global batch size: 16 | lm loss: 1.103551E+01 | loss scale: 1.0 | grad norm: 44.037 | number of skipped iterations: 0 | number of nan iterations: 0 |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: 58851.314453125 | max allocated: 68669.056640625 | reserved: 72692.0 | max reserved: 72692.0 + [2026-08-30 06:32:12] iteration 2/ 12 | consumed samples: 32 | elapsed time per iteration (ms): 581.7 | throughput per GPU (TFLOP/s/GPU): 573.1 | learning rate: 4.000000E-05 | global batch size: 16 | lm loss: 1.102915E+01 | loss scale: 1.0 | grad norm: 45.936 | number of skipped iterations: 0 | number of nan iterations: 0 | + [2026-08-30 06:32:12] iteration 3/ 12 | consumed samples: 48 | elapsed time per iteration (ms): 501.6 | throughput per GPU (TFLOP/s/GPU): 664.6 | learning rate: 6.000000E-05 | global batch size: 16 | lm loss: 1.045600E+01 | loss scale: 1.0 | grad norm: 206.584 | number of skipped iterations: 0 | number of nan iterations: 0 | +Capture CUDA graph for training!!! +CUDA graph capture done!!! + [2026-08-30 06:32:14] iteration 4/ 12 | consumed samples: 64 | elapsed time per iteration (ms): 2263.5 | throughput per GPU (TFLOP/s/GPU): 147.3 | learning rate: 8.000000E-05 | global batch size: 16 | lm loss: 1.318230E+01 | loss scale: 1.0 | grad norm: 36.193 | number of skipped iterations: 0 | number of nan iterations: 0 | + [2026-08-30 06:32:15] iteration 5/ 12 | consumed samples: 80 | elapsed time per iteration (ms): 414.1 | throughput per GPU (TFLOP/s/GPU): 805.0 | learning rate: 1.000000E-04 | global batch size: 16 | lm loss: 1.039625E+01 | loss scale: 1.0 | grad norm: 16.068 | number of skipped iterations: 0 | number of nan iterations: 0 | + [2026-08-30 06:32:15] iteration 6/ 12 | consumed samples: 96 | elapsed time per iteration (ms): 412.9 | throughput per GPU (TFLOP/s/GPU): 807.4 | learning rate: 7.750000E-05 | global batch size: 16 | lm loss: 1.042547E+01 | loss scale: 1.0 | grad norm: 36.682 | number of skipped iterations: 0 | number of nan iterations: 0 | + [2026-08-30 06:32:16] iteration 7/ 12 | consumed samples: 112 | elapsed time per iteration (ms): 412.9 | throughput per GPU (TFLOP/s/GPU): 807.3 | learning rate: 3.250000E-05 | global batch size: 16 | lm loss: 1.002835E+01 | loss scale: 1.0 | grad norm: 7.871 | number of skipped iterations: 0 | number of nan iterations: 0 | + [2026-08-30 06:32:16] iteration 8/ 12 | consumed samples: 128 | elapsed time per iteration (ms): 410.8 | throughput per GPU (TFLOP/s/GPU): 811.5 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 9.531461E+00 | loss scale: 1.0 | grad norm: 5.786 | number of skipped iterations: 0 | number of nan iterations: 0 | + [2026-08-30 06:32:16] iteration 9/ 12 | consumed samples: 144 | elapsed time per iteration (ms): 412.9 | throughput per GPU (TFLOP/s/GPU): 807.5 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 9.298289E+00 | loss scale: 1.0 | grad norm: 5.642 | number of skipped iterations: 0 | number of nan iterations: 0 | + [2026-08-30 06:32:17] iteration 10/ 12 | consumed samples: 160 | elapsed time per iteration (ms): 412.7 | throughput per GPU (TFLOP/s/GPU): 807.8 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 9.141107E+00 | loss scale: 1.0 | grad norm: 5.465 | number of skipped iterations: 0 | number of nan iterations: 0 | + [2026-08-30 06:32:17] iteration 11/ 12 | consumed samples: 176 | elapsed time per iteration (ms): 412.2 | throughput per GPU (TFLOP/s/GPU): 808.9 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 9.021684E+00 | loss scale: 1.0 | grad norm: 5.268 | number of skipped iterations: 0 | number of nan iterations: 0 | + [2026-08-30 06:32:18] iteration 12/ 12 | consumed samples: 192 | elapsed time per iteration (ms): 413.2 | throughput per GPU (TFLOP/s/GPU): 806.7 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 8.896466E+00 | loss scale: 1.0 | grad norm: 5.345 | number of skipped iterations: 0 | number of nan iterations: 0 | +[after training is done] datetime: 2026-08-30 06:32:18 +[rank0]:[W830 06:32:19.544970062 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()) diff --git a/tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/G4__echo__one__seq4096__gbs16__cap1p095703125/rep1/mem_samples.csv b/tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/G4__echo__one__seq4096__gbs16__cap1p095703125/rep1/mem_samples.csv new file mode 100644 index 0000000000000000000000000000000000000000..95b158ea0f595f5f977a4ecb5fd961d7b8b8ac85 --- /dev/null +++ b/tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/G4__echo__one__seq4096__gbs16__cap1p095703125/rep1/mem_samples.csv @@ -0,0 +1,264 @@ +0, 0 +1, 0 +2, 0 +3, 0 +4, 0 +5, 0 +6, 0 +7, 0 +0, 0 +1, 0 +2, 0 +3, 0 +4, 0 +5, 0 +6, 0 +7, 0 +0, 4 +1, 2 +2, 1 +3, 1 +4, 1 +5, 1 +6, 1 +7, 1 +0, 4 +1, 4 +2, 4 +3, 4 +4, 4 +5, 1 +6, 1 +7, 1 +0, 4 +1, 4 +2, 4 +3, 4 +4, 4 +5, 4 +6, 4 +7, 4 +0, 4 +1, 4 +2, 4 +3, 4 +4, 4 +5, 4 +6, 4 +7, 4 +0, 821 +1, 821 +2, 821 +3, 821 +4, 821 +5, 821 +6, 821 +7, 821 +0, 1895 +1, 1895 +2, 1895 +3, 1895 +4, 1895 +5, 1895 +6, 1895 +7, 1895 +0, 38231 +1, 38679 +2, 38679 +3, 38679 +4, 38231 +5, 38679 +6, 38679 +7, 38679 +0, 38929 +1, 39377 +2, 39377 +3, 39377 +4, 38929 +5, 39377 +6, 39377 +7, 39377 +0, 39577 +1, 40025 +2, 40025 +3, 40025 +4, 39577 +5, 40025 +6, 40025 +7, 40025 +0, 41635 +1, 42083 +2, 42083 +3, 42083 +4, 41635 +5, 42083 +6, 42083 +7, 42083 +0, 41651 +1, 42099 +2, 42099 +3, 42099 +4, 41651 +5, 42099 +6, 42099 +7, 42099 +0, 41743 +1, 42189 +2, 42187 +3, 42195 +4, 41779 +5, 42217 +6, 42229 +7, 42213 +0, 42581 +1, 43029 +2, 43029 +3, 43029 +4, 42581 +5, 43029 +6, 43029 +7, 43029 +0, 43703 +1, 44151 +2, 44151 +3, 44151 +4, 43703 +5, 44151 +6, 44151 +7, 44151 +0, 45319 +1, 45767 +2, 45767 +3, 45767 +4, 45319 +5, 45767 +6, 45767 +7, 45767 +0, 46445 +1, 46893 +2, 46893 +3, 46893 +4, 46445 +5, 46893 +6, 46893 +7, 46893 +0, 73369 +1, 73341 +2, 73341 +3, 73341 +4, 73389 +5, 73341 +6, 73341 +7, 73341 +0, 74493 +1, 74465 +2, 74465 +3, 74465 +4, 74513 +5, 74465 +6, 74465 +7, 74465 +0, 74493 +1, 74465 +2, 74465 +3, 74465 +4, 74513 +5, 74465 +6, 74465 +7, 74465 +0, 74493 +1, 74465 +2, 74465 +3, 74465 +4, 74513 +5, 74465 +6, 74465 +7, 74465 +0, 83049 +1, 82775 +2, 83013 +3, 82819 +4, 83069 +5, 82733 +6, 83045 +7, 82799 +0, 83853 +1, 83541 +2, 83781 +3, 83559 +4, 83835 +5, 83475 +6, 83803 +7, 83607 +0, 84539 +1, 84285 +2, 84509 +3, 84285 +4, 84559 +5, 84397 +6, 84509 +7, 84285 +0, 85107 +1, 84929 +2, 85129 +3, 84883 +4, 85215 +5, 85079 +6, 85163 +7, 84941 +0, 92235 +1, 90315 +2, 105067 +3, 88503 +4, 93975 +5, 87181 +6, 90321 +7, 88901 +0, 126687 +1, 126673 +2, 126673 +3, 126675 +4, 126705 +5, 127009 +6, 126675 +7, 126895 +0, 126687 +1, 126673 +2, 126673 +3, 126675 +4, 126705 +5, 127009 +6, 126675 +7, 126895 +0, 125907 +1, 126389 +2, 126177 +3, 126391 +4, 126115 +5, 126123 +6, 125821 +7, 125983 +0, 125757 +1, 125743 +2, 3114 +3, 1834 +4, 1834 +5, 1834 +6, 1834 +7, 2090 +0, 0 +1, 0 +2, 0 +3, 0 +4, 0 +5, 0 +6, 0 +7, 0 +0, 0 +1, 0 +2, 0 +3, 0 +4, 0 +5, 0 +6, 0 +7, 0 diff --git a/tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/G4__echo__one__seq4096__gbs16__cap1p095703125/rep1/run.env b/tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/G4__echo__one__seq4096__gbs16__cap1p095703125/rep1/run.env new file mode 100644 index 0000000000000000000000000000000000000000..9a57e0eee2ddeb1ead96f25351d51cb4ef7cfbd2 --- /dev/null +++ b/tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/G4__echo__one__seq4096__gbs16__cap1p095703125/rep1/run.env @@ -0,0 +1,70 @@ +backend=echo +config=G4 +family=tierA +repo=/workspace/Megatron-MoE-Echo +branch=bracing-regions-through-kernels +commit=9a236793 +num_layers=16 +hidden=4096 +ffn_hidden=14336 +moe_ffn_hidden=896 +num_experts=128 +topk=32 +experts_per_rank=16 +expert_params=11010048 +heads=32 +kv_groups=8 +kv_channels= +qk_layernorm=0 +shared_ffn=0 +shared_expert_overlap=0 +moe_token_dispatcher=flex +mla=0 +dense_first_layers=0 +moe_layers=16 +rotary_base=1000000 +rotary_scaling= +qkv_bias=0 +router_balancing=aux_loss +aux_loss_coeff=1e-2 +router_pre_softmax=0 +router_fusion=0 +router_dtype=fp32 +device_initiated_grouped_gemm=1 +norm_epsilon=1e-5 +seq_length=4096 +micro_batch_size=1 +global_batch_size=16 +num_microbatches=2 +assignments_per_rank=131072 +mean_tokens_per_expert=8192 +slot_policy=one +slots_per_rank=1 +capacity_mode=static +capacity=1.095703125 +cuda_graph=1 +probe_mode=0 +async_grad_reduce=0 +branch_role=timing +memory_dump_mode=none +memory_snapshot_path= +cuda_graph_warmup_steps=3 +recompute=none +force_balance=0 +train_iters=12 +ep_size=8 +echo_dispatch_overlap=1 +echo_dispatch_recompute=0 +ultraep_quota_min_tokens= +ultraep_weight_sync_join= +profile_enable=0 +kernel_trace=0 +fp8_mode=mxfp8 +fp8_recipe=mxfp8 +training_seed=1234 +learning_rate=1e-4 +min_learning_rate=1.0e-5 +lr_warmup_iters=5 +state_dump_enable=0 +state_dump_dir= +moonep_assert_no_pending_reduce=0 diff --git a/tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/G4__echo__one__seq4096__gbs16__cap1p095703125/rep1/train.log b/tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/G4__echo__one__seq4096__gbs16__cap1p095703125/rep1/train.log new file mode 100644 index 0000000000000000000000000000000000000000..8804e3b0ee4254881d8793b6e04360f46354036c --- /dev/null +++ b/tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/G4__echo__one__seq4096__gbs16__cap1p095703125/rep1/train.log @@ -0,0 +1,1502 @@ +============================================================== + G4 / echo (slots/rank=1 policy=one) + E=128 K=32 H'=896 m=16 + L=16 seq=4096 gbs=16 (2 microbatches) + cap_mode=static cap=1.095703125 cg=1 recompute=none + memory_dump=none snapshots=/workspace/tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/G4__echo__one__seq4096__gbs16__cap1p095703125/rep1/memory_snapshots/snapshot.pickle + repo=/workspace/Megatron-MoE-Echo @ 9a236793 +============================================================== +W0830 06:32:29.253000 619339 torch/distributed/run.py:766] +W0830 06:32:29.253000 619339 torch/distributed/run.py:766] ***************************************** +W0830 06:32:29.253000 619339 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. +W0830 06:32:29.253000 619339 torch/distributed/run.py:766] ***************************************** +/workspace/Megatron-MoE-Echo/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-MoE-Echo/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-MoE-Echo/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-MoE-Echo/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-MoE-Echo/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-MoE-Echo/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_deepep_num_sms .............................. 20 + moe_echo_enable_random_offloading ............... False + moe_echo_expert_dispatch_overlap ................ True + moe_echo_expert_dispatcher_type ................. hybridep + moe_echo_recompute_expert_dispatch .............. False + moe_enable_deepep ............................... False + moe_enable_echo ................................. True + moe_enable_hybridep ............................. True + moe_expert_capacity_factor ...................... None + moe_extended_tp ................................. False + moe_ffn_hidden_size ............................. 896 + 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 ............................ 8 + moe_pad_expert_input_to_capacity ................ False + moe_per_layer_logging ........................... False + moe_permute_fusion .............................. False + moe_received_token_capacity ..................... 1.095703125 + 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 ................................. 32 + 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_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 ..................................... 128 + 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 ......................................... False + profile_ranks ................................... [0] + profile_step_end ................................ 12 + profile_step_start .............................. 10 + 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 ..................................... 12 + 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-MoE-Echo/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-MoE-Echo/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 +[Gloo] Rank 2 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 [Gloo] Rank 34 is connected to is connected to 7 peer ranks. 7Expected number of connected peer ranks is : peer ranks. 7Expected 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 1 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 [Gloo] Rank 1 is connected to 07 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 2 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 4[Gloo] Rank 7 is connected to 6 is connected to 7 is connected to 7 peer ranks. 7 peer ranks. Expected number of connected peer ranks is : peer ranks. Expected number of connected peer ranks is : 7Expected number of connected peer ranks is : 7 +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 +[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 +[rank2]:[W830 06:32:47.263099030 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]:[W830 06:32:47.263144014 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]:[W830 06:32:47.263448795 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. +[rank3]:[W830 06:32:47.263477418 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]:[W830 06:32:47.263484647 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]:[W830 06:32:47.265103749 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. +[rank4]:[W830 06:32:47.265127322 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-MoE-Echo/megatron/core/datasets' +make: Nothing to be done for 'default'. +make: Leaving directory '/workspace/Megatron-MoE-Echo/megatron/core/datasets' +>>> done with dataset index builder. Compilation time: 0.037 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]:[W830 06:32:47.427207219 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: 3.839 seconds +time to initialize megatron (seconds): 6.973 +[after megatron is initialized] datetime: 2026-08-30 06:32:53 +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( + > number of parameters on (tensor, pipeline) model parallel rank (0, 0): 3910275072 +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 (42479616 elements, 42479616 padded size): + module.decoder.layers.15.self_attention.linear_qkv.layer_norm_weight + module.decoder.layers.15.mlp.router.weight + module.decoder.final_layernorm.weight + module.decoder.layers.15.pre_mlp_layernorm.weight + module.decoder.layers.15.self_attention.linear_qkv.weight + module.decoder.layers.15.self_attention.linear_proj.weight +Params for bucket 3 (42475520 elements, 42475520 padded size): + module.decoder.layers.14.self_attention.linear_qkv.weight + module.decoder.layers.14.pre_mlp_layernorm.weight + module.decoder.layers.14.self_attention.linear_proj.weight + module.decoder.layers.14.mlp.router.weight + module.decoder.layers.14.self_attention.linear_qkv.layer_norm_weight +Params for bucket 4 (42475520 elements, 42475520 padded size): + module.decoder.layers.13.mlp.router.weight + module.decoder.layers.13.self_attention.linear_qkv.layer_norm_weight + module.decoder.layers.13.pre_mlp_layernorm.weight + module.decoder.layers.13.self_attention.linear_proj.weight + module.decoder.layers.13.self_attention.linear_qkv.weight +Params for bucket 5 (42475520 elements, 42475520 padded size): + module.decoder.layers.12.self_attention.linear_proj.weight + 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_qkv.layer_norm_weight +Params for bucket 6 (42475520 elements, 42475520 padded size): + module.decoder.layers.11.mlp.router.weight + module.decoder.layers.11.self_attention.linear_qkv.layer_norm_weight + module.decoder.layers.11.self_attention.linear_proj.weight + module.decoder.layers.11.self_attention.linear_qkv.weight + module.decoder.layers.11.pre_mlp_layernorm.weight +Params for bucket 7 (42475520 elements, 42475520 padded size): + module.decoder.layers.10.pre_mlp_layernorm.weight + module.decoder.layers.10.self_attention.linear_qkv.weight + module.decoder.layers.10.self_attention.linear_proj.weight + module.decoder.layers.10.mlp.router.weight + module.decoder.layers.10.self_attention.linear_qkv.layer_norm_weight +Params for bucket 8 (42475520 elements, 42475520 padded size): + module.decoder.layers.9.pre_mlp_layernorm.weight + module.decoder.layers.9.self_attention.linear_proj.weight + module.decoder.layers.9.self_attention.linear_qkv.weight + module.decoder.layers.9.mlp.router.weight + module.decoder.layers.9.self_attention.linear_qkv.layer_norm_weight +Params for bucket 9 (42475520 elements, 42475520 padded size): + module.decoder.layers.8.self_attention.linear_proj.weight + module.decoder.layers.8.pre_mlp_layernorm.weight + module.decoder.layers.8.self_attention.linear_qkv.weight + module.decoder.layers.8.mlp.router.weight + module.decoder.layers.8.self_attention.linear_qkv.layer_norm_weight +Params for bucket 10 (42475520 elements, 42475520 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_qkv.weight + module.decoder.layers.7.pre_mlp_layernorm.weight + module.decoder.layers.7.self_attention.linear_proj.weight +Params for bucket 11 (42475520 elements, 42475520 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.pre_mlp_layernorm.weight + module.decoder.layers.6.self_attention.linear_qkv.weight +Params for bucket 12 (42475520 elements, 42475520 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.pre_mlp_layernorm.weight + module.decoder.layers.5.self_attention.linear_proj.weight +Params for bucket 13 (42475520 elements, 42475520 padded size): + module.decoder.layers.4.self_attention.linear_proj.weight + module.decoder.layers.4.pre_mlp_layernorm.weight + module.decoder.layers.4.self_attention.linear_qkv.weight + module.decoder.layers.4.mlp.router.weight + module.decoder.layers.4.self_attention.linear_qkv.layer_norm_weight +Params for bucket 14 (42475520 elements, 42475520 padded size): + 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.mlp.router.weight + module.decoder.layers.3.self_attention.linear_qkv.layer_norm_weight +Params for bucket 15 (42475520 elements, 42475520 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 (42475520 elements, 42475520 padded size): + module.decoder.layers.1.mlp.router.weight + module.decoder.layers.1.pre_mlp_layernorm.weight + module.decoder.layers.1.self_attention.linear_proj.weight + module.decoder.layers.1.self_attention.linear_qkv.weight + module.decoder.layers.1.self_attention.linear_qkv.layer_norm_weight +Params for bucket 17 (42475520 elements, 42475520 padded size): + module.decoder.layers.0.self_attention.linear_qkv.layer_norm_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_proj.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: 67 +Params for bucket 1 (40370176 elements, 40370176 padded size): + module.decoder.layers.15.mlp.experts.linear_fc2.weight12 + module.decoder.layers.15.mlp.experts.linear_fc2.weight14 + module.decoder.layers.15.mlp.experts.linear_fc2.weight10 + module.decoder.layers.15.mlp.experts.linear_fc2.weight8 + module.decoder.layers.15.mlp.experts.linear_fc2.weight6 + module.decoder.layers.15.mlp.experts.linear_fc2.weight15 + module.decoder.layers.15.mlp.experts.linear_fc2.weight13 + module.decoder.layers.15.mlp.experts.linear_fc2.weight11 + module.decoder.layers.15.mlp.experts.linear_fc2.weight9 + module.decoder.layers.15.mlp.experts.linear_fc2.weight7 + module.decoder.layers.15.mlp.experts.linear_fc2.weight5 +Params for bucket 2 (40370176 elements, 40370176 padded size): + 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.weight0 + module.decoder.layers.15.mlp.experts.linear_fc1.weight15 + module.decoder.layers.15.mlp.experts.linear_fc1.weight13 + 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_fc1.weight14 +Params for bucket 3 (44040192 elements, 44040192 padded size): + module.decoder.layers.15.mlp.experts.linear_fc1.weight11 + module.decoder.layers.15.mlp.experts.linear_fc1.weight9 + module.decoder.layers.15.mlp.experts.linear_fc1.weight7 + module.decoder.layers.15.mlp.experts.linear_fc1.weight12 + module.decoder.layers.15.mlp.experts.linear_fc1.weight8 + module.decoder.layers.15.mlp.experts.linear_fc1.weight10 +Params for bucket 4 (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.weight1 + module.decoder.layers.15.mlp.experts.linear_fc1.weight6 + module.decoder.layers.15.mlp.experts.linear_fc1.weight4 + module.decoder.layers.15.mlp.experts.linear_fc1.weight2 +Params for bucket 5 (40370176 elements, 40370176 padded size): + module.decoder.layers.14.mlp.experts.linear_fc2.weight15 + module.decoder.layers.14.mlp.experts.linear_fc2.weight13 + module.decoder.layers.14.mlp.experts.linear_fc2.weight11 + module.decoder.layers.14.mlp.experts.linear_fc2.weight9 + module.decoder.layers.14.mlp.experts.linear_fc2.weight7 + module.decoder.layers.15.mlp.experts.linear_fc1.weight0 + module.decoder.layers.14.mlp.experts.linear_fc2.weight12 + module.decoder.layers.14.mlp.experts.linear_fc2.weight14 + module.decoder.layers.14.mlp.experts.linear_fc2.weight10 + module.decoder.layers.14.mlp.experts.linear_fc2.weight8 +Params for bucket 6 (40370176 elements, 40370176 padded size): + module.decoder.layers.14.mlp.experts.linear_fc2.weight3 + module.decoder.layers.14.mlp.experts.linear_fc2.weight5 + module.decoder.layers.14.mlp.experts.linear_fc2.weight1 + module.decoder.layers.14.mlp.experts.linear_fc1.weight14 + module.decoder.layers.14.mlp.experts.linear_fc2.weight6 + module.decoder.layers.14.mlp.experts.linear_fc2.weight2 + module.decoder.layers.14.mlp.experts.linear_fc2.weight4 + module.decoder.layers.14.mlp.experts.linear_fc2.weight0 + module.decoder.layers.14.mlp.experts.linear_fc1.weight15 +Params for bucket 7 (44040192 elements, 44040192 padded size): + module.decoder.layers.14.mlp.experts.linear_fc1.weight12 + module.decoder.layers.14.mlp.experts.linear_fc1.weight10 + module.decoder.layers.14.mlp.experts.linear_fc1.weight8 + module.decoder.layers.14.mlp.experts.linear_fc1.weight13 + module.decoder.layers.14.mlp.experts.linear_fc1.weight9 + module.decoder.layers.14.mlp.experts.linear_fc1.weight11 +Params for bucket 8 (44040192 elements, 44040192 padded size): + module.decoder.layers.14.mlp.experts.linear_fc1.weight6 + module.decoder.layers.14.mlp.experts.linear_fc1.weight4 + module.decoder.layers.14.mlp.experts.linear_fc1.weight2 + module.decoder.layers.14.mlp.experts.linear_fc1.weight7 + module.decoder.layers.14.mlp.experts.linear_fc1.weight3 + module.decoder.layers.14.mlp.experts.linear_fc1.weight5 +Params for bucket 9 (40370176 elements, 40370176 padded size): + module.decoder.layers.13.mlp.experts.linear_fc2.weight14 + module.decoder.layers.14.mlp.experts.linear_fc1.weight0 + module.decoder.layers.13.mlp.experts.linear_fc2.weight12 + module.decoder.layers.13.mlp.experts.linear_fc2.weight10 + module.decoder.layers.14.mlp.experts.linear_fc1.weight1 + module.decoder.layers.13.mlp.experts.linear_fc2.weight15 + module.decoder.layers.13.mlp.experts.linear_fc2.weight13 + module.decoder.layers.13.mlp.experts.linear_fc2.weight11 + module.decoder.layers.13.mlp.experts.linear_fc2.weight9 +Params for bucket 10 (40370176 elements, 40370176 padded size): + module.decoder.layers.13.mlp.experts.linear_fc2.weight8 + module.decoder.layers.13.mlp.experts.linear_fc2.weight6 + module.decoder.layers.13.mlp.experts.linear_fc2.weight4 + module.decoder.layers.13.mlp.experts.linear_fc2.weight2 + module.decoder.layers.13.mlp.experts.linear_fc2.weight0 + module.decoder.layers.13.mlp.experts.linear_fc1.weight15 + 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.weight1 +Params for bucket 11 (44040192 elements, 44040192 padded size): + module.decoder.layers.13.mlp.experts.linear_fc1.weight13 + module.decoder.layers.13.mlp.experts.linear_fc1.weight11 + module.decoder.layers.13.mlp.experts.linear_fc1.weight9 + module.decoder.layers.13.mlp.experts.linear_fc1.weight14 + module.decoder.layers.13.mlp.experts.linear_fc1.weight10 + module.decoder.layers.13.mlp.experts.linear_fc1.weight12 +Params for bucket 12 (44040192 elements, 44040192 padded size): + module.decoder.layers.13.mlp.experts.linear_fc1.weight7 + 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.weight8 + module.decoder.layers.13.mlp.experts.linear_fc1.weight6 + module.decoder.layers.13.mlp.experts.linear_fc1.weight4 +Params for bucket 13 (40370176 elements, 40370176 padded size): + module.decoder.layers.12.mlp.experts.linear_fc2.weight15 + module.decoder.layers.13.mlp.experts.linear_fc1.weight1 + module.decoder.layers.12.mlp.experts.linear_fc2.weight13 + module.decoder.layers.12.mlp.experts.linear_fc2.weight11 + module.decoder.layers.13.mlp.experts.linear_fc1.weight2 + module.decoder.layers.13.mlp.experts.linear_fc1.weight0 + module.decoder.layers.12.mlp.experts.linear_fc2.weight14 + module.decoder.layers.12.mlp.experts.linear_fc2.weight12 +Params for bucket 14 (40370176 elements, 40370176 padded size): + module.decoder.layers.12.mlp.experts.linear_fc2.weight9 + 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.weight1 + module.decoder.layers.12.mlp.experts.linear_fc2.weight10 + module.decoder.layers.12.mlp.experts.linear_fc2.weight8 + 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.weight0 +Params for bucket 15 (44040192 elements, 44040192 padded size): + module.decoder.layers.12.mlp.experts.linear_fc1.weight14 + module.decoder.layers.12.mlp.experts.linear_fc1.weight12 + module.decoder.layers.12.mlp.experts.linear_fc1.weight10 + module.decoder.layers.12.mlp.experts.linear_fc1.weight15 + module.decoder.layers.12.mlp.experts.linear_fc1.weight11 + module.decoder.layers.12.mlp.experts.linear_fc1.weight13 +Params for bucket 16 (44040192 elements, 44040192 padded size): + module.decoder.layers.12.mlp.experts.linear_fc1.weight8 + module.decoder.layers.12.mlp.experts.linear_fc1.weight6 + module.decoder.layers.12.mlp.experts.linear_fc1.weight4 + module.decoder.layers.12.mlp.experts.linear_fc1.weight9 + module.decoder.layers.12.mlp.experts.linear_fc1.weight5 + module.decoder.layers.12.mlp.experts.linear_fc1.weight7 +Params for bucket 17 (40370176 elements, 40370176 padded size): + module.decoder.layers.11.mlp.experts.linear_fc2.weight15 + module.decoder.layers.12.mlp.experts.linear_fc1.weight2 + module.decoder.layers.12.mlp.experts.linear_fc1.weight0 + module.decoder.layers.11.mlp.experts.linear_fc2.weight13 + module.decoder.layers.12.mlp.experts.linear_fc1.weight3 + module.decoder.layers.12.mlp.experts.linear_fc1.weight1 + module.decoder.layers.11.mlp.experts.linear_fc2.weight14 +Params for bucket 18 (40370176 elements, 40370176 padded size): + module.decoder.layers.11.mlp.experts.linear_fc2.weight11 + module.decoder.layers.11.mlp.experts.linear_fc2.weight9 + 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.weight12 + module.decoder.layers.11.mlp.experts.linear_fc2.weight8 + module.decoder.layers.11.mlp.experts.linear_fc2.weight10 + 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 19 (44040192 elements, 44040192 padded size): + module.decoder.layers.11.mlp.experts.linear_fc2.weight1 + module.decoder.layers.11.mlp.experts.linear_fc1.weight14 + module.decoder.layers.11.mlp.experts.linear_fc1.weight12 + module.decoder.layers.11.mlp.experts.linear_fc1.weight15 + module.decoder.layers.11.mlp.experts.linear_fc2.weight0 + module.decoder.layers.11.mlp.experts.linear_fc1.weight13 + module.decoder.layers.11.mlp.experts.linear_fc1.weight11 +Params for bucket 20 (44040192 elements, 44040192 padded size): + module.decoder.layers.11.mlp.experts.linear_fc1.weight10 + module.decoder.layers.11.mlp.experts.linear_fc1.weight8 + module.decoder.layers.11.mlp.experts.linear_fc1.weight6 + module.decoder.layers.11.mlp.experts.linear_fc1.weight9 + module.decoder.layers.11.mlp.experts.linear_fc1.weight7 + module.decoder.layers.11.mlp.experts.linear_fc1.weight5 +Params for bucket 21 (40370176 elements, 40370176 padded size): + module.decoder.layers.11.mlp.experts.linear_fc1.weight4 + module.decoder.layers.11.mlp.experts.linear_fc1.weight2 + module.decoder.layers.11.mlp.experts.linear_fc1.weight0 + module.decoder.layers.10.mlp.experts.linear_fc2.weight15 + module.decoder.layers.11.mlp.experts.linear_fc1.weight3 + module.decoder.layers.11.mlp.experts.linear_fc1.weight1 +Params for bucket 22 (40370176 elements, 40370176 padded size): + module.decoder.layers.10.mlp.experts.linear_fc2.weight13 + module.decoder.layers.10.mlp.experts.linear_fc2.weight11 + module.decoder.layers.10.mlp.experts.linear_fc2.weight9 + 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.weight14 + module.decoder.layers.10.mlp.experts.linear_fc2.weight10 + module.decoder.layers.10.mlp.experts.linear_fc2.weight12 + module.decoder.layers.10.mlp.experts.linear_fc2.weight8 + module.decoder.layers.10.mlp.experts.linear_fc2.weight6 + module.decoder.layers.10.mlp.experts.linear_fc2.weight4 +Params for bucket 23 (44040192 elements, 44040192 padded size): + module.decoder.layers.10.mlp.experts.linear_fc2.weight3 + module.decoder.layers.10.mlp.experts.linear_fc2.weight1 + module.decoder.layers.10.mlp.experts.linear_fc1.weight14 + module.decoder.layers.10.mlp.experts.linear_fc1.weight12 + module.decoder.layers.10.mlp.experts.linear_fc1.weight15 + module.decoder.layers.10.mlp.experts.linear_fc2.weight0 + module.decoder.layers.10.mlp.experts.linear_fc2.weight2 + module.decoder.layers.10.mlp.experts.linear_fc1.weight13 +Params for bucket 24 (44040192 elements, 44040192 padded size): + module.decoder.layers.10.mlp.experts.linear_fc1.weight10 + module.decoder.layers.10.mlp.experts.linear_fc1.weight8 + module.decoder.layers.10.mlp.experts.linear_fc1.weight6 + module.decoder.layers.10.mlp.experts.linear_fc1.weight11 + module.decoder.layers.10.mlp.experts.linear_fc1.weight9 + module.decoder.layers.10.mlp.experts.linear_fc1.weight7 +Params for bucket 25 (44040192 elements, 44040192 padded size): + module.decoder.layers.10.mlp.experts.linear_fc1.weight2 + module.decoder.layers.10.mlp.experts.linear_fc1.weight4 + module.decoder.layers.10.mlp.experts.linear_fc1.weight0 + 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.weight1 +Params for bucket 26 (40370176 elements, 40370176 padded size): + module.decoder.layers.9.mlp.experts.linear_fc2.weight12 + module.decoder.layers.9.mlp.experts.linear_fc2.weight14 + module.decoder.layers.9.mlp.experts.linear_fc2.weight10 + module.decoder.layers.9.mlp.experts.linear_fc2.weight8 + module.decoder.layers.9.mlp.experts.linear_fc2.weight6 + module.decoder.layers.9.mlp.experts.linear_fc2.weight15 + module.decoder.layers.9.mlp.experts.linear_fc2.weight13 + module.decoder.layers.9.mlp.experts.linear_fc2.weight11 + module.decoder.layers.9.mlp.experts.linear_fc2.weight9 + module.decoder.layers.9.mlp.experts.linear_fc2.weight7 + module.decoder.layers.9.mlp.experts.linear_fc2.weight5 +Params for bucket 27 (40370176 elements, 40370176 padded size): + 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.weight0 + module.decoder.layers.9.mlp.experts.linear_fc1.weight15 + module.decoder.layers.9.mlp.experts.linear_fc1.weight13 + module.decoder.layers.9.mlp.experts.linear_fc2.weight3 + module.decoder.layers.9.mlp.experts.linear_fc2.weight1 + module.decoder.layers.9.mlp.experts.linear_fc1.weight14 +Params for bucket 28 (44040192 elements, 44040192 padded size): + module.decoder.layers.9.mlp.experts.linear_fc1.weight11 + module.decoder.layers.9.mlp.experts.linear_fc1.weight9 + module.decoder.layers.9.mlp.experts.linear_fc1.weight7 + module.decoder.layers.9.mlp.experts.linear_fc1.weight12 + module.decoder.layers.9.mlp.experts.linear_fc1.weight8 + module.decoder.layers.9.mlp.experts.linear_fc1.weight10 +Params for bucket 29 (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.weight1 + module.decoder.layers.9.mlp.experts.linear_fc1.weight6 + module.decoder.layers.9.mlp.experts.linear_fc1.weight4 + module.decoder.layers.9.mlp.experts.linear_fc1.weight2 +Params for bucket 30 (40370176 elements, 40370176 padded size): + module.decoder.layers.8.mlp.experts.linear_fc2.weight12 + module.decoder.layers.8.mlp.experts.linear_fc2.weight14 + module.decoder.layers.8.mlp.experts.linear_fc2.weight10 + module.decoder.layers.8.mlp.experts.linear_fc2.weight8 + module.decoder.layers.9.mlp.experts.linear_fc1.weight0 + module.decoder.layers.8.mlp.experts.linear_fc2.weight15 + module.decoder.layers.8.mlp.experts.linear_fc2.weight13 + module.decoder.layers.8.mlp.experts.linear_fc2.weight11 + module.decoder.layers.8.mlp.experts.linear_fc2.weight9 + module.decoder.layers.8.mlp.experts.linear_fc2.weight7 +Params for bucket 31 (40370176 elements, 40370176 padded size): + 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.weight4 + module.decoder.layers.8.mlp.experts.linear_fc2.weight0 + module.decoder.layers.8.mlp.experts.linear_fc1.weight14 + 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.weight1 + module.decoder.layers.8.mlp.experts.linear_fc1.weight15 +Params for bucket 32 (44040192 elements, 44040192 padded size): + module.decoder.layers.8.mlp.experts.linear_fc1.weight10 + module.decoder.layers.8.mlp.experts.linear_fc1.weight12 + module.decoder.layers.8.mlp.experts.linear_fc1.weight8 + module.decoder.layers.8.mlp.experts.linear_fc1.weight13 + module.decoder.layers.8.mlp.experts.linear_fc1.weight9 + module.decoder.layers.8.mlp.experts.linear_fc1.weight11 +Params for bucket 33 (44040192 elements, 44040192 padded size): + module.decoder.layers.8.mlp.experts.linear_fc1.weight4 + module.decoder.layers.8.mlp.experts.linear_fc1.weight6 + module.decoder.layers.8.mlp.experts.linear_fc1.weight2 + module.decoder.layers.8.mlp.experts.linear_fc1.weight7 + module.decoder.layers.8.mlp.experts.linear_fc1.weight5 + module.decoder.layers.8.mlp.experts.linear_fc1.weight3 +Params for bucket 34 (40370176 elements, 40370176 padded size): + module.decoder.layers.8.mlp.experts.linear_fc1.weight0 + module.decoder.layers.7.mlp.experts.linear_fc2.weight14 + module.decoder.layers.7.mlp.experts.linear_fc2.weight12 + module.decoder.layers.7.mlp.experts.linear_fc2.weight10 + module.decoder.layers.8.mlp.experts.linear_fc1.weight1 + module.decoder.layers.7.mlp.experts.linear_fc2.weight13 + module.decoder.layers.7.mlp.experts.linear_fc2.weight15 + module.decoder.layers.7.mlp.experts.linear_fc2.weight11 + module.decoder.layers.7.mlp.experts.linear_fc2.weight9 +Params for bucket 35 (40370176 elements, 40370176 padded size): + module.decoder.layers.7.mlp.experts.linear_fc2.weight8 + 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.weight0 + module.decoder.layers.7.mlp.experts.linear_fc2.weight5 + 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.weight1 + module.decoder.layers.7.mlp.experts.linear_fc1.weight15 +Params for bucket 36 (44040192 elements, 44040192 padded size): + module.decoder.layers.7.mlp.experts.linear_fc1.weight14 + module.decoder.layers.7.mlp.experts.linear_fc1.weight10 + module.decoder.layers.7.mlp.experts.linear_fc1.weight12 + module.decoder.layers.7.mlp.experts.linear_fc1.weight13 + module.decoder.layers.7.mlp.experts.linear_fc1.weight11 + module.decoder.layers.7.mlp.experts.linear_fc1.weight9 +Params for bucket 37 (44040192 elements, 44040192 padded size): + module.decoder.layers.7.mlp.experts.linear_fc1.weight8 + module.decoder.layers.7.mlp.experts.linear_fc1.weight6 + module.decoder.layers.7.mlp.experts.linear_fc1.weight4 + module.decoder.layers.7.mlp.experts.linear_fc1.weight7 + module.decoder.layers.7.mlp.experts.linear_fc1.weight5 + module.decoder.layers.7.mlp.experts.linear_fc1.weight3 +Params for bucket 38 (40370176 elements, 40370176 padded size): + module.decoder.layers.7.mlp.experts.linear_fc1.weight2 + module.decoder.layers.7.mlp.experts.linear_fc1.weight0 + module.decoder.layers.6.mlp.experts.linear_fc2.weight15 + module.decoder.layers.6.mlp.experts.linear_fc2.weight13 + module.decoder.layers.6.mlp.experts.linear_fc2.weight11 + module.decoder.layers.6.mlp.experts.linear_fc2.weight14 + module.decoder.layers.7.mlp.experts.linear_fc1.weight1 + module.decoder.layers.6.mlp.experts.linear_fc2.weight12 +Params for bucket 39 (40370176 elements, 40370176 padded size): + module.decoder.layers.6.mlp.experts.linear_fc2.weight7 + module.decoder.layers.6.mlp.experts.linear_fc2.weight9 + 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.weight1 + module.decoder.layers.6.mlp.experts.linear_fc2.weight10 + module.decoder.layers.6.mlp.experts.linear_fc2.weight8 + 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 + module.decoder.layers.6.mlp.experts.linear_fc2.weight0 +Params for bucket 40 (44040192 elements, 44040192 padded size): + module.decoder.layers.6.mlp.experts.linear_fc1.weight14 + module.decoder.layers.6.mlp.experts.linear_fc1.weight12 + module.decoder.layers.6.mlp.experts.linear_fc1.weight10 + module.decoder.layers.6.mlp.experts.linear_fc1.weight15 + module.decoder.layers.6.mlp.experts.linear_fc1.weight13 + module.decoder.layers.6.mlp.experts.linear_fc1.weight11 +Params for bucket 41 (44040192 elements, 44040192 padded size): + module.decoder.layers.6.mlp.experts.linear_fc1.weight8 + module.decoder.layers.6.mlp.experts.linear_fc1.weight6 + module.decoder.layers.6.mlp.experts.linear_fc1.weight4 + module.decoder.layers.6.mlp.experts.linear_fc1.weight9 + module.decoder.layers.6.mlp.experts.linear_fc1.weight7 + module.decoder.layers.6.mlp.experts.linear_fc1.weight5 +Params for bucket 42 (40370176 elements, 40370176 padded size): + module.decoder.layers.6.mlp.experts.linear_fc1.weight0 + module.decoder.layers.6.mlp.experts.linear_fc1.weight2 + module.decoder.layers.5.mlp.experts.linear_fc2.weight14 + module.decoder.layers.6.mlp.experts.linear_fc1.weight3 + module.decoder.layers.5.mlp.experts.linear_fc2.weight15 + module.decoder.layers.6.mlp.experts.linear_fc1.weight1 + module.decoder.layers.5.mlp.experts.linear_fc2.weight13 +Params for bucket 43 (40370176 elements, 40370176 padded size): + module.decoder.layers.5.mlp.experts.linear_fc2.weight12 + module.decoder.layers.5.mlp.experts.linear_fc2.weight10 + module.decoder.layers.5.mlp.experts.linear_fc2.weight8 + 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.weight11 + module.decoder.layers.5.mlp.experts.linear_fc2.weight9 + 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 +Params for bucket 44 (44040192 elements, 44040192 padded size): + module.decoder.layers.5.mlp.experts.linear_fc2.weight0 + module.decoder.layers.5.mlp.experts.linear_fc1.weight15 + module.decoder.layers.5.mlp.experts.linear_fc1.weight13 + module.decoder.layers.5.mlp.experts.linear_fc1.weight11 + module.decoder.layers.5.mlp.experts.linear_fc2.weight1 + module.decoder.layers.5.mlp.experts.linear_fc1.weight14 + module.decoder.layers.5.mlp.experts.linear_fc1.weight12 +Params for bucket 45 (44040192 elements, 44040192 padded size): + module.decoder.layers.5.mlp.experts.linear_fc1.weight9 + module.decoder.layers.5.mlp.experts.linear_fc1.weight7 + module.decoder.layers.5.mlp.experts.linear_fc1.weight5 + module.decoder.layers.5.mlp.experts.linear_fc1.weight10 + module.decoder.layers.5.mlp.experts.linear_fc1.weight6 + module.decoder.layers.5.mlp.experts.linear_fc1.weight8 +Params for bucket 46 (40370176 elements, 40370176 padded size): + module.decoder.layers.5.mlp.experts.linear_fc1.weight3 + module.decoder.layers.5.mlp.experts.linear_fc1.weight1 + module.decoder.layers.4.mlp.experts.linear_fc2.weight15 + module.decoder.layers.5.mlp.experts.linear_fc1.weight4 + module.decoder.layers.5.mlp.experts.linear_fc1.weight2 + module.decoder.layers.5.mlp.experts.linear_fc1.weight0 +Params for bucket 47 (40370176 elements, 40370176 padded size): + module.decoder.layers.4.mlp.experts.linear_fc2.weight13 + module.decoder.layers.4.mlp.experts.linear_fc2.weight11 + module.decoder.layers.4.mlp.experts.linear_fc2.weight9 + 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.weight14 + module.decoder.layers.4.mlp.experts.linear_fc2.weight10 + module.decoder.layers.4.mlp.experts.linear_fc2.weight12 + module.decoder.layers.4.mlp.experts.linear_fc2.weight8 + module.decoder.layers.4.mlp.experts.linear_fc2.weight6 + module.decoder.layers.4.mlp.experts.linear_fc2.weight4 +Params for bucket 48 (44040192 elements, 44040192 padded size): + module.decoder.layers.4.mlp.experts.linear_fc2.weight3 + module.decoder.layers.4.mlp.experts.linear_fc2.weight1 + module.decoder.layers.4.mlp.experts.linear_fc1.weight14 + module.decoder.layers.4.mlp.experts.linear_fc1.weight12 + module.decoder.layers.4.mlp.experts.linear_fc2.weight2 + module.decoder.layers.4.mlp.experts.linear_fc1.weight15 + module.decoder.layers.4.mlp.experts.linear_fc2.weight0 + module.decoder.layers.4.mlp.experts.linear_fc1.weight13 +Params for bucket 49 (44040192 elements, 44040192 padded size): + module.decoder.layers.4.mlp.experts.linear_fc1.weight10 + module.decoder.layers.4.mlp.experts.linear_fc1.weight8 + module.decoder.layers.4.mlp.experts.linear_fc1.weight6 + module.decoder.layers.4.mlp.experts.linear_fc1.weight11 + module.decoder.layers.4.mlp.experts.linear_fc1.weight7 + module.decoder.layers.4.mlp.experts.linear_fc1.weight9 +Params for bucket 50 (44040192 elements, 44040192 padded size): + module.decoder.layers.4.mlp.experts.linear_fc1.weight4 + 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.weight5 + module.decoder.layers.4.mlp.experts.linear_fc1.weight1 + module.decoder.layers.4.mlp.experts.linear_fc1.weight3 +Params for bucket 51 (40370176 elements, 40370176 padded size): + module.decoder.layers.3.mlp.experts.linear_fc2.weight15 + module.decoder.layers.3.mlp.experts.linear_fc2.weight11 + module.decoder.layers.3.mlp.experts.linear_fc2.weight13 + module.decoder.layers.3.mlp.experts.linear_fc2.weight9 + module.decoder.layers.3.mlp.experts.linear_fc2.weight7 + module.decoder.layers.3.mlp.experts.linear_fc2.weight5 + module.decoder.layers.3.mlp.experts.linear_fc2.weight14 + module.decoder.layers.3.mlp.experts.linear_fc2.weight12 + module.decoder.layers.3.mlp.experts.linear_fc2.weight10 + module.decoder.layers.3.mlp.experts.linear_fc2.weight8 + module.decoder.layers.3.mlp.experts.linear_fc2.weight6 +Params for bucket 52 (40370176 elements, 40370176 padded size): + module.decoder.layers.3.mlp.experts.linear_fc2.weight1 + module.decoder.layers.3.mlp.experts.linear_fc2.weight3 + module.decoder.layers.3.mlp.experts.linear_fc1.weight14 + 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.weight0 + module.decoder.layers.3.mlp.experts.linear_fc1.weight15 + module.decoder.layers.3.mlp.experts.linear_fc1.weight13 +Params for bucket 53 (44040192 elements, 44040192 padded size): + module.decoder.layers.3.mlp.experts.linear_fc1.weight12 + module.decoder.layers.3.mlp.experts.linear_fc1.weight10 + module.decoder.layers.3.mlp.experts.linear_fc1.weight8 + module.decoder.layers.3.mlp.experts.linear_fc1.weight11 + module.decoder.layers.3.mlp.experts.linear_fc1.weight9 + module.decoder.layers.3.mlp.experts.linear_fc1.weight7 +Params for bucket 54 (44040192 elements, 44040192 padded size): + module.decoder.layers.3.mlp.experts.linear_fc1.weight6 + module.decoder.layers.3.mlp.experts.linear_fc1.weight4 + module.decoder.layers.3.mlp.experts.linear_fc1.weight2 + module.decoder.layers.3.mlp.experts.linear_fc1.weight3 + module.decoder.layers.3.mlp.experts.linear_fc1.weight5 + module.decoder.layers.3.mlp.experts.linear_fc1.weight1 +Params for bucket 55 (40370176 elements, 40370176 padded size): + module.decoder.layers.3.mlp.experts.linear_fc1.weight0 + module.decoder.layers.2.mlp.experts.linear_fc2.weight12 + module.decoder.layers.2.mlp.experts.linear_fc2.weight14 + module.decoder.layers.2.mlp.experts.linear_fc2.weight10 + module.decoder.layers.2.mlp.experts.linear_fc2.weight8 + module.decoder.layers.2.mlp.experts.linear_fc2.weight15 + module.decoder.layers.2.mlp.experts.linear_fc2.weight13 + module.decoder.layers.2.mlp.experts.linear_fc2.weight11 + module.decoder.layers.2.mlp.experts.linear_fc2.weight9 + module.decoder.layers.2.mlp.experts.linear_fc2.weight7 +Params for bucket 56 (40370176 elements, 40370176 padded size): + module.decoder.layers.2.mlp.experts.linear_fc2.weight6 + module.decoder.layers.2.mlp.experts.linear_fc2.weight2 + module.decoder.layers.2.mlp.experts.linear_fc2.weight4 + module.decoder.layers.2.mlp.experts.linear_fc2.weight0 + module.decoder.layers.2.mlp.experts.linear_fc1.weight15 + 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.weight1 + module.decoder.layers.2.mlp.experts.linear_fc1.weight14 +Params for bucket 57 (44040192 elements, 44040192 padded size): + module.decoder.layers.2.mlp.experts.linear_fc1.weight13 + module.decoder.layers.2.mlp.experts.linear_fc1.weight11 + module.decoder.layers.2.mlp.experts.linear_fc1.weight9 + module.decoder.layers.2.mlp.experts.linear_fc1.weight12 + module.decoder.layers.2.mlp.experts.linear_fc1.weight10 + module.decoder.layers.2.mlp.experts.linear_fc1.weight8 +Params for bucket 58 (44040192 elements, 44040192 padded size): + module.decoder.layers.2.mlp.experts.linear_fc1.weight7 + module.decoder.layers.2.mlp.experts.linear_fc1.weight5 + module.decoder.layers.2.mlp.experts.linear_fc1.weight3 + module.decoder.layers.2.mlp.experts.linear_fc1.weight6 + module.decoder.layers.2.mlp.experts.linear_fc1.weight4 + module.decoder.layers.2.mlp.experts.linear_fc1.weight2 +Params for bucket 59 (40370176 elements, 40370176 padded size): + module.decoder.layers.2.mlp.experts.linear_fc1.weight1 + module.decoder.layers.1.mlp.experts.linear_fc2.weight13 + module.decoder.layers.1.mlp.experts.linear_fc2.weight15 + module.decoder.layers.1.mlp.experts.linear_fc2.weight11 + module.decoder.layers.1.mlp.experts.linear_fc2.weight9 + module.decoder.layers.1.mlp.experts.linear_fc2.weight14 + module.decoder.layers.2.mlp.experts.linear_fc1.weight0 + module.decoder.layers.1.mlp.experts.linear_fc2.weight12 + module.decoder.layers.1.mlp.experts.linear_fc2.weight10 +Params for bucket 60 (40370176 elements, 40370176 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.weight1 + module.decoder.layers.1.mlp.experts.linear_fc1.weight15 + module.decoder.layers.1.mlp.experts.linear_fc2.weight8 + 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 + module.decoder.layers.1.mlp.experts.linear_fc2.weight0 +Params for bucket 61 (44040192 elements, 44040192 padded size): + module.decoder.layers.1.mlp.experts.linear_fc1.weight11 + module.decoder.layers.1.mlp.experts.linear_fc1.weight13 + module.decoder.layers.1.mlp.experts.linear_fc1.weight9 + module.decoder.layers.1.mlp.experts.linear_fc1.weight14 + module.decoder.layers.1.mlp.experts.linear_fc1.weight10 + module.decoder.layers.1.mlp.experts.linear_fc1.weight12 +Params for bucket 62 (44040192 elements, 44040192 padded size): + module.decoder.layers.1.mlp.experts.linear_fc1.weight7 + 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.weight8 + module.decoder.layers.1.mlp.experts.linear_fc1.weight4 + module.decoder.layers.1.mlp.experts.linear_fc1.weight6 +Params for bucket 63 (40370176 elements, 40370176 padded size): + module.decoder.layers.1.mlp.experts.linear_fc1.weight1 + module.decoder.layers.0.mlp.experts.linear_fc2.weight15 + module.decoder.layers.0.mlp.experts.linear_fc2.weight13 + module.decoder.layers.0.mlp.experts.linear_fc2.weight11 + module.decoder.layers.1.mlp.experts.linear_fc1.weight2 + module.decoder.layers.0.mlp.experts.linear_fc2.weight14 + module.decoder.layers.1.mlp.experts.linear_fc1.weight0 + module.decoder.layers.0.mlp.experts.linear_fc2.weight12 +Params for bucket 64 (40370176 elements, 40370176 padded size): + module.decoder.layers.0.mlp.experts.linear_fc2.weight8 + 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.weight1 + module.decoder.layers.0.mlp.experts.linear_fc2.weight10 + module.decoder.layers.0.mlp.experts.linear_fc2.weight9 + 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 + module.decoder.layers.0.mlp.experts.linear_fc2.weight0 +Params for bucket 65 (44040192 elements, 44040192 padded size): + module.decoder.layers.0.mlp.experts.linear_fc1.weight15 + module.decoder.layers.0.mlp.experts.linear_fc1.weight13 + module.decoder.layers.0.mlp.experts.linear_fc1.weight11 + module.decoder.layers.0.mlp.experts.linear_fc1.weight14 + module.decoder.layers.0.mlp.experts.linear_fc1.weight12 + module.decoder.layers.0.mlp.experts.linear_fc1.weight10 +Params for bucket 66 (44040192 elements, 44040192 padded size): + module.decoder.layers.0.mlp.experts.linear_fc1.weight8 + module.decoder.layers.0.mlp.experts.linear_fc1.weight7 + module.decoder.layers.0.mlp.experts.linear_fc1.weight4 + module.decoder.layers.0.mlp.experts.linear_fc1.weight9 + module.decoder.layers.0.mlp.experts.linear_fc1.weight6 + module.decoder.layers.0.mlp.experts.linear_fc1.weight5 +Params for bucket 67 (29360128 elements, 29360128 padded size): + module.decoder.layers.0.mlp.experts.linear_fc1.weight3 + 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: 2026-08-30 06:32:53 +> building train, validation, and test datasets ... + > datasets target sizes (minimum size): + train: 192 + 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=(192, 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 44ed1ee5bcb9e5b030387b8d27a6c406-GPTDataset-train-document_index.npy +INFO:megatron.core.datasets.gpt_dataset: Load the sample index from 44ed1ee5bcb9e5b030387b8d27a6c406-GPTDataset-train-sample_index.npy +INFO:megatron.core.datasets.gpt_dataset: Load the shuffle index from 44ed1ee5bcb9e5b030387b8d27a6c406-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: 2026-08-30 06:32:53 +done with setup ... +training ... +(min, max) time across ranks (ms): + model-and-optimizer-setup ......................: (685.73, 698.08) + train/valid/test-data-iterators-setup ..........: (66.44, 70.54) +Overwriting rerun_state_machine.current_iteration from -1 to 0... +[before the start of training step] datetime: 2026-08-30 06:32:53 +/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: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: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: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: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: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.") + [2026-08-30 06:33:32] iteration 1/ 12 | consumed samples: 16 | elapsed time per iteration (ms): 38282.4 | throughput per GPU (TFLOP/s/GPU): 8.7 | learning rate: 2.000000E-05 | global batch size: 16 | lm loss: 1.102834E+01 | loss scale: 1.0 | grad norm: 44.250 | number of skipped iterations: 0 | number of nan iterations: 0 |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: 58498.755859375 | max allocated: 73513.7607421875 | reserved: 77116.0 | max reserved: 77116.0 + [2026-08-30 06:33:33] iteration 2/ 12 | consumed samples: 32 | elapsed time per iteration (ms): 860.9 | throughput per GPU (TFLOP/s/GPU): 387.2 | learning rate: 4.000000E-05 | global batch size: 16 | lm loss: 1.102085E+01 | loss scale: 1.0 | grad norm: 45.950 | number of skipped iterations: 0 | number of nan iterations: 0 | + [2026-08-30 06:33:33] iteration 3/ 12 | consumed samples: 48 | elapsed time per iteration (ms): 548.1 | throughput per GPU (TFLOP/s/GPU): 608.2 | learning rate: 6.000000E-05 | global batch size: 16 | lm loss: 1.057767E+01 | loss scale: 1.0 | grad norm: 218.195 | number of skipped iterations: 0 | number of nan iterations: 0 | +Capture CUDA graph for training!!! +CUDA graph capture done!!! + [2026-08-30 06:33:35] iteration 4/ 12 | consumed samples: 64 | elapsed time per iteration (ms): 1820.6 | throughput per GPU (TFLOP/s/GPU): 183.1 | learning rate: 8.000000E-05 | global batch size: 16 | lm loss: 1.300301E+01 | loss scale: 1.0 | grad norm: 45.897 | number of skipped iterations: 0 | number of nan iterations: 0 | + [2026-08-30 06:33:35] iteration 5/ 12 | consumed samples: 80 | elapsed time per iteration (ms): 464.7 | throughput per GPU (TFLOP/s/GPU): 717.5 | learning rate: 1.000000E-04 | global batch size: 16 | lm loss: 1.034157E+01 | loss scale: 1.0 | grad norm: 18.004 | number of skipped iterations: 0 | number of nan iterations: 0 | + [2026-08-30 06:33:36] iteration 6/ 12 | consumed samples: 96 | elapsed time per iteration (ms): 460.2 | throughput per GPU (TFLOP/s/GPU): 724.4 | learning rate: 7.750000E-05 | global batch size: 16 | lm loss: 1.009405E+01 | loss scale: 1.0 | grad norm: 10.004 | number of skipped iterations: 0 | number of nan iterations: 0 | + [2026-08-30 06:33:36] iteration 7/ 12 | consumed samples: 112 | elapsed time per iteration (ms): 459.9 | throughput per GPU (TFLOP/s/GPU): 724.9 | learning rate: 3.250000E-05 | global batch size: 16 | lm loss: 9.658836E+00 | loss scale: 1.0 | grad norm: 6.161 | number of skipped iterations: 0 | number of nan iterations: 0 | + [2026-08-30 06:33:37] iteration 8/ 12 | consumed samples: 128 | elapsed time per iteration (ms): 458.7 | throughput per GPU (TFLOP/s/GPU): 726.8 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 9.024432E+00 | loss scale: 1.0 | grad norm: 5.422 | number of skipped iterations: 0 | number of nan iterations: 0 | + [2026-08-30 06:33:37] iteration 9/ 12 | consumed samples: 144 | elapsed time per iteration (ms): 459.0 | throughput per GPU (TFLOP/s/GPU): 726.3 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 8.765420E+00 | loss scale: 1.0 | grad norm: 5.068 | number of skipped iterations: 0 | number of nan iterations: 0 | + [2026-08-30 06:33:38] iteration 10/ 12 | consumed samples: 160 | elapsed time per iteration (ms): 460.2 | throughput per GPU (TFLOP/s/GPU): 724.5 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 8.629656E+00 | loss scale: 1.0 | grad norm: 4.476 | number of skipped iterations: 0 | number of nan iterations: 0 | + [2026-08-30 06:33:38] iteration 11/ 12 | consumed samples: 176 | elapsed time per iteration (ms): 460.3 | throughput per GPU (TFLOP/s/GPU): 724.3 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 8.553773E+00 | loss scale: 1.0 | grad norm: 3.975 | number of skipped iterations: 0 | number of nan iterations: 0 | + [2026-08-30 06:33:39] iteration 12/ 12 | consumed samples: 192 | elapsed time per iteration (ms): 460.3 | throughput per GPU (TFLOP/s/GPU): 724.3 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 8.467318E+00 | loss scale: 1.0 | grad norm: 4.646 | number of skipped iterations: 0 | number of nan iterations: 0 | +[after training is done] datetime: 2026-08-30 06:33:39 +[rank0]:[W830 06:33:40.443365606 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()) diff --git a/tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/G4__echo__one__seq4096__gbs16__cap1p095703125/rep2/mem_samples.csv b/tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/G4__echo__one__seq4096__gbs16__cap1p095703125/rep2/mem_samples.csv new file mode 100644 index 0000000000000000000000000000000000000000..1e6cd0503e1c7d98613949e899a550c58143267f --- /dev/null +++ b/tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/G4__echo__one__seq4096__gbs16__cap1p095703125/rep2/mem_samples.csv @@ -0,0 +1,256 @@ +0, 0 +1, 0 +2, 0 +3, 0 +4, 0 +5, 0 +6, 0 +7, 0 +0, 0 +1, 0 +2, 0 +3, 0 +4, 0 +5, 0 +6, 0 +7, 0 +0, 2 +1, 1 +2, 1 +3, 1 +4, 1 +5, 1 +6, 1 +7, 1 +0, 4 +1, 4 +2, 4 +3, 4 +4, 1 +5, 1 +6, 1 +7, 1 +0, 4 +1, 4 +2, 4 +3, 4 +4, 4 +5, 4 +6, 4 +7, 4 +0, 4 +1, 4 +2, 4 +3, 4 +4, 4 +5, 4 +6, 4 +7, 4 +0, 821 +1, 821 +2, 821 +3, 821 +4, 821 +5, 821 +6, 821 +7, 821 +0, 1895 +1, 1895 +2, 1895 +3, 1895 +4, 1895 +5, 1895 +6, 1895 +7, 1895 +0, 32513 +1, 32513 +2, 32513 +3, 32513 +4, 32513 +5, 32513 +6, 32513 +7, 32513 +0, 38897 +1, 39345 +2, 39341 +3, 39345 +4, 38889 +5, 39337 +6, 39345 +7, 39345 +0, 39493 +1, 39941 +2, 39923 +3, 39967 +4, 39513 +5, 39941 +6, 39969 +7, 39951 +0, 41635 +1, 42083 +2, 42083 +3, 42083 +4, 41635 +5, 42083 +6, 42083 +7, 42083 +0, 41651 +1, 42099 +2, 42099 +3, 42099 +4, 41651 +5, 42099 +6, 42099 +7, 42099 +0, 41727 +1, 42177 +2, 42181 +3, 42169 +4, 41733 +5, 42183 +6, 42195 +7, 42187 +0, 42581 +1, 43029 +2, 43029 +3, 43029 +4, 42581 +5, 43029 +6, 43029 +7, 43029 +0, 43703 +1, 44151 +2, 44151 +3, 44151 +4, 43703 +5, 44151 +6, 44151 +7, 44151 +0, 45319 +1, 45767 +2, 45767 +3, 45767 +4, 45319 +5, 45767 +6, 45767 +7, 45767 +0, 46445 +1, 46893 +2, 46893 +3, 46893 +4, 46445 +5, 46893 +6, 46893 +7, 46893 +0, 73369 +1, 73341 +2, 73341 +3, 73341 +4, 73389 +5, 73341 +6, 73341 +7, 73341 +0, 74493 +1, 74465 +2, 74465 +3, 74465 +4, 74513 +5, 74465 +6, 74465 +7, 74465 +0, 74493 +1, 74465 +2, 74465 +3, 74465 +4, 74513 +5, 74465 +6, 74465 +7, 74465 +0, 74493 +1, 74465 +2, 74465 +3, 74465 +4, 74513 +5, 74465 +6, 74465 +7, 74465 +0, 82681 +1, 82673 +2, 82629 +3, 82631 +4, 82681 +5, 82971 +6, 82977 +7, 82645 +0, 83501 +1, 83427 +2, 83439 +3, 83449 +4, 83509 +5, 83851 +6, 83845 +7, 83507 +0, 84427 +1, 84397 +2, 84397 +3, 84397 +4, 84447 +5, 84509 +6, 84509 +7, 84397 +0, 84987 +1, 84973 +2, 84981 +3, 85017 +4, 85037 +5, 85077 +6, 85105 +7, 85041 +0, 76807 +1, 76577 +2, 76651 +3, 76587 +4, 76719 +5, 76419 +6, 76557 +7, 76587 +0, 127023 +1, 126895 +2, 126897 +3, 126897 +4, 127043 +5, 126671 +6, 126785 +7, 126895 +0, 127023 +1, 126895 +2, 126897 +3, 126897 +4, 127043 +5, 126671 +6, 126785 +7, 126895 +0, 126739 +1, 126611 +2, 126613 +3, 126613 +4, 126759 +5, 126387 +6, 126501 +7, 126611 +0, 1833 +1, 1834 +2, 1834 +3, 1834 +4, 1834 +5, 2090 +6, 125855 +7, 3114 +0, 0 +1, 0 +2, 0 +3, 0 +4, 0 +5, 0 +6, 0 +7, 0 diff --git a/tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/G4__echo__one__seq4096__gbs16__cap1p095703125/rep2/run.env b/tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/G4__echo__one__seq4096__gbs16__cap1p095703125/rep2/run.env new file mode 100644 index 0000000000000000000000000000000000000000..9a57e0eee2ddeb1ead96f25351d51cb4ef7cfbd2 --- /dev/null +++ b/tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/G4__echo__one__seq4096__gbs16__cap1p095703125/rep2/run.env @@ -0,0 +1,70 @@ +backend=echo +config=G4 +family=tierA +repo=/workspace/Megatron-MoE-Echo +branch=bracing-regions-through-kernels +commit=9a236793 +num_layers=16 +hidden=4096 +ffn_hidden=14336 +moe_ffn_hidden=896 +num_experts=128 +topk=32 +experts_per_rank=16 +expert_params=11010048 +heads=32 +kv_groups=8 +kv_channels= +qk_layernorm=0 +shared_ffn=0 +shared_expert_overlap=0 +moe_token_dispatcher=flex +mla=0 +dense_first_layers=0 +moe_layers=16 +rotary_base=1000000 +rotary_scaling= +qkv_bias=0 +router_balancing=aux_loss +aux_loss_coeff=1e-2 +router_pre_softmax=0 +router_fusion=0 +router_dtype=fp32 +device_initiated_grouped_gemm=1 +norm_epsilon=1e-5 +seq_length=4096 +micro_batch_size=1 +global_batch_size=16 +num_microbatches=2 +assignments_per_rank=131072 +mean_tokens_per_expert=8192 +slot_policy=one +slots_per_rank=1 +capacity_mode=static +capacity=1.095703125 +cuda_graph=1 +probe_mode=0 +async_grad_reduce=0 +branch_role=timing +memory_dump_mode=none +memory_snapshot_path= +cuda_graph_warmup_steps=3 +recompute=none +force_balance=0 +train_iters=12 +ep_size=8 +echo_dispatch_overlap=1 +echo_dispatch_recompute=0 +ultraep_quota_min_tokens= +ultraep_weight_sync_join= +profile_enable=0 +kernel_trace=0 +fp8_mode=mxfp8 +fp8_recipe=mxfp8 +training_seed=1234 +learning_rate=1e-4 +min_learning_rate=1.0e-5 +lr_warmup_iters=5 +state_dump_enable=0 +state_dump_dir= +moonep_assert_no_pending_reduce=0 diff --git a/tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/G4__echo__one__seq4096__gbs16__cap1p095703125/rep2/train.log b/tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/G4__echo__one__seq4096__gbs16__cap1p095703125/rep2/train.log new file mode 100644 index 0000000000000000000000000000000000000000..eaf0d770f89f364181114d16ffd96efe0e78d6ee --- /dev/null +++ b/tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/G4__echo__one__seq4096__gbs16__cap1p095703125/rep2/train.log @@ -0,0 +1,1502 @@ +============================================================== + G4 / echo (slots/rank=1 policy=one) + E=128 K=32 H'=896 m=16 + L=16 seq=4096 gbs=16 (2 microbatches) + cap_mode=static cap=1.095703125 cg=1 recompute=none + memory_dump=none snapshots=/workspace/tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/G4__echo__one__seq4096__gbs16__cap1p095703125/rep2/memory_snapshots/snapshot.pickle + repo=/workspace/Megatron-MoE-Echo @ 9a236793 +============================================================== +W0830 06:33:49.507000 622813 torch/distributed/run.py:766] +W0830 06:33:49.507000 622813 torch/distributed/run.py:766] ***************************************** +W0830 06:33:49.507000 622813 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. +W0830 06:33:49.507000 622813 torch/distributed/run.py:766] ***************************************** +/workspace/Megatron-MoE-Echo/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-MoE-Echo/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-MoE-Echo/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-MoE-Echo/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-MoE-Echo/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-MoE-Echo/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_deepep_num_sms .............................. 20 + moe_echo_enable_random_offloading ............... False + moe_echo_expert_dispatch_overlap ................ True + moe_echo_expert_dispatcher_type ................. hybridep + moe_echo_recompute_expert_dispatch .............. False + moe_enable_deepep ............................... False + moe_enable_echo ................................. True + moe_enable_hybridep ............................. True + moe_expert_capacity_factor ...................... None + moe_extended_tp ................................. False + moe_ffn_hidden_size ............................. 896 + 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 ............................ 8 + moe_pad_expert_input_to_capacity ................ False + moe_per_layer_logging ........................... False + moe_permute_fusion .............................. False + moe_received_token_capacity ..................... 1.095703125 + 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 ................................. 32 + 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_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 ..................................... 128 + 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 ......................................... False + profile_ranks ................................... [0] + profile_step_end ................................ 12 + profile_step_start .............................. 10 + 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 ..................................... 12 + 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 ... +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 + > 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 ... +/workspace/Megatron-MoE-Echo/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-MoE-Echo/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 5 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 43 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 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 4[Gloo] Rank [Gloo] Rank [Gloo] Rank [Gloo] Rank is connected to 25776 is connected to is connected to is connected to peer ranks. is connected to 777Expected number of connected peer ranks is : 7 peer ranks. peer ranks. peer ranks. 7 peer ranks. Expected number of connected peer ranks is : Expected number of connected peer ranks is : Expected number of connected peer ranks is : +Expected number of connected peer ranks is : 777 + +7 + +3 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 [Gloo] Rank 0 is connected to 00 is connected to peer ranks. 0Expected number of connected peer ranks is : peer ranks. 0Expected 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 +[rank6]:[W830 06:34:07.453462577 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. +[rank1]:[W830 06:34:07.453798117 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. +[rank2]:[W830 06:34:07.453974894 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. +[rank3]:[W830 06:34:07.454156336 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. +[rank4]:[W830 06:34:07.460984777 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. +[rank5]:[W830 06:34:07.461008393 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. +[rank7]:[W830 06:34:07.461049053 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. +> compiling dataset index builder ... +make: Entering directory '/workspace/Megatron-MoE-Echo/megatron/core/datasets' +make: Nothing to be done for 'default'. +make: Leaving directory '/workspace/Megatron-MoE-Echo/megatron/core/datasets' +>>> done with dataset index builder. Compilation time: 0.036 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]:[W830 06:34:07.623415649 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: 3.989 seconds +time to initialize megatron (seconds): 7.181 +[after megatron is initialized] datetime: 2026-08-30 06:34:13 +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( + > number of parameters on (tensor, pipeline) model parallel rank (0, 0): 3910275072 +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 (42479616 elements, 42479616 padded size): + module.decoder.layers.15.pre_mlp_layernorm.weight + module.decoder.layers.15.self_attention.linear_qkv.weight + module.decoder.layers.15.self_attention.linear_proj.weight + module.decoder.final_layernorm.weight + module.decoder.layers.15.mlp.router.weight + module.decoder.layers.15.self_attention.linear_qkv.layer_norm_weight +Params for bucket 3 (42475520 elements, 42475520 padded size): + module.decoder.layers.14.pre_mlp_layernorm.weight + module.decoder.layers.14.self_attention.linear_qkv.weight + module.decoder.layers.14.self_attention.linear_proj.weight + module.decoder.layers.14.mlp.router.weight + module.decoder.layers.14.self_attention.linear_qkv.layer_norm_weight +Params for bucket 4 (42475520 elements, 42475520 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 (42475520 elements, 42475520 padded size): + module.decoder.layers.12.self_attention.linear_proj.weight + module.decoder.layers.12.pre_mlp_layernorm.weight + module.decoder.layers.12.self_attention.linear_qkv.weight + module.decoder.layers.12.mlp.router.weight + module.decoder.layers.12.self_attention.linear_qkv.layer_norm_weight +Params for bucket 6 (42475520 elements, 42475520 padded size): + module.decoder.layers.11.pre_mlp_layernorm.weight + module.decoder.layers.11.self_attention.linear_proj.weight + module.decoder.layers.11.self_attention.linear_qkv.weight + module.decoder.layers.11.mlp.router.weight + module.decoder.layers.11.self_attention.linear_qkv.layer_norm_weight +Params for bucket 7 (42475520 elements, 42475520 padded size): + module.decoder.layers.10.mlp.router.weight + module.decoder.layers.10.self_attention.linear_qkv.layer_norm_weight + module.decoder.layers.10.self_attention.linear_proj.weight + module.decoder.layers.10.self_attention.linear_qkv.weight + module.decoder.layers.10.pre_mlp_layernorm.weight +Params for bucket 8 (42475520 elements, 42475520 padded size): + 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 + module.decoder.layers.9.self_attention.linear_qkv.weight + module.decoder.layers.9.pre_mlp_layernorm.weight +Params for bucket 9 (42475520 elements, 42475520 padded size): + 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 + module.decoder.layers.8.self_attention.linear_qkv.weight + module.decoder.layers.8.pre_mlp_layernorm.weight +Params for bucket 10 (42475520 elements, 42475520 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 (42475520 elements, 42475520 padded size): + module.decoder.layers.6.self_attention.linear_proj.weight + module.decoder.layers.6.pre_mlp_layernorm.weight + module.decoder.layers.6.self_attention.linear_qkv.weight + module.decoder.layers.6.mlp.router.weight + module.decoder.layers.6.self_attention.linear_qkv.layer_norm_weight +Params for bucket 12 (42475520 elements, 42475520 padded size): + module.decoder.layers.5.self_attention.linear_qkv.weight + module.decoder.layers.5.pre_mlp_layernorm.weight + module.decoder.layers.5.mlp.router.weight + module.decoder.layers.5.self_attention.linear_proj.weight + module.decoder.layers.5.self_attention.linear_qkv.layer_norm_weight +Params for bucket 13 (42475520 elements, 42475520 padded size): + module.decoder.layers.4.self_attention.linear_qkv.weight + module.decoder.layers.4.pre_mlp_layernorm.weight + module.decoder.layers.4.self_attention.linear_proj.weight + module.decoder.layers.4.mlp.router.weight + module.decoder.layers.4.self_attention.linear_qkv.layer_norm_weight +Params for bucket 14 (42475520 elements, 42475520 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.weight + module.decoder.layers.3.pre_mlp_layernorm.weight + module.decoder.layers.3.self_attention.linear_qkv.layer_norm_weight +Params for bucket 15 (42475520 elements, 42475520 padded size): + 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.self_attention.linear_qkv.weight + module.decoder.layers.2.pre_mlp_layernorm.weight +Params for bucket 16 (42475520 elements, 42475520 padded size): + module.decoder.layers.1.self_attention.linear_qkv.layer_norm_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.self_attention.linear_proj.weight +Params for bucket 17 (42475520 elements, 42475520 padded size): + module.decoder.layers.0.self_attention.linear_qkv.weight + module.decoder.layers.0.self_attention.linear_proj.weight + module.decoder.layers.0.self_attention.linear_qkv.layer_norm_weight + module.decoder.layers.0.pre_mlp_layernorm.weight + module.decoder.layers.0.mlp.router.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: 67 +Params for bucket 1 (40370176 elements, 40370176 padded size): + module.decoder.layers.15.mlp.experts.linear_fc2.weight12 + module.decoder.layers.15.mlp.experts.linear_fc2.weight14 + module.decoder.layers.15.mlp.experts.linear_fc2.weight10 + module.decoder.layers.15.mlp.experts.linear_fc2.weight8 + module.decoder.layers.15.mlp.experts.linear_fc2.weight6 + module.decoder.layers.15.mlp.experts.linear_fc2.weight15 + module.decoder.layers.15.mlp.experts.linear_fc2.weight13 + module.decoder.layers.15.mlp.experts.linear_fc2.weight11 + module.decoder.layers.15.mlp.experts.linear_fc2.weight9 + module.decoder.layers.15.mlp.experts.linear_fc2.weight7 + module.decoder.layers.15.mlp.experts.linear_fc2.weight5 +Params for bucket 2 (40370176 elements, 40370176 padded size): + 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.weight0 + module.decoder.layers.15.mlp.experts.linear_fc1.weight15 + module.decoder.layers.15.mlp.experts.linear_fc1.weight13 + 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_fc1.weight14 +Params for bucket 3 (44040192 elements, 44040192 padded size): + module.decoder.layers.15.mlp.experts.linear_fc1.weight11 + module.decoder.layers.15.mlp.experts.linear_fc1.weight9 + module.decoder.layers.15.mlp.experts.linear_fc1.weight7 + module.decoder.layers.15.mlp.experts.linear_fc1.weight12 + module.decoder.layers.15.mlp.experts.linear_fc1.weight8 + module.decoder.layers.15.mlp.experts.linear_fc1.weight10 +Params for bucket 4 (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.weight1 + module.decoder.layers.15.mlp.experts.linear_fc1.weight6 + module.decoder.layers.15.mlp.experts.linear_fc1.weight4 + module.decoder.layers.15.mlp.experts.linear_fc1.weight2 +Params for bucket 5 (40370176 elements, 40370176 padded size): + module.decoder.layers.14.mlp.experts.linear_fc2.weight12 + module.decoder.layers.14.mlp.experts.linear_fc2.weight14 + module.decoder.layers.14.mlp.experts.linear_fc2.weight10 + module.decoder.layers.14.mlp.experts.linear_fc2.weight8 + module.decoder.layers.15.mlp.experts.linear_fc1.weight0 + module.decoder.layers.14.mlp.experts.linear_fc2.weight15 + module.decoder.layers.14.mlp.experts.linear_fc2.weight13 + module.decoder.layers.14.mlp.experts.linear_fc2.weight11 + module.decoder.layers.14.mlp.experts.linear_fc2.weight9 + module.decoder.layers.14.mlp.experts.linear_fc2.weight7 +Params for bucket 6 (40370176 elements, 40370176 padded size): + module.decoder.layers.14.mlp.experts.linear_fc2.weight6 + module.decoder.layers.14.mlp.experts.linear_fc2.weight2 + module.decoder.layers.14.mlp.experts.linear_fc2.weight4 + module.decoder.layers.14.mlp.experts.linear_fc2.weight0 + module.decoder.layers.14.mlp.experts.linear_fc1.weight15 + module.decoder.layers.14.mlp.experts.linear_fc2.weight5 + module.decoder.layers.14.mlp.experts.linear_fc2.weight3 + module.decoder.layers.14.mlp.experts.linear_fc2.weight1 + module.decoder.layers.14.mlp.experts.linear_fc1.weight14 +Params for bucket 7 (44040192 elements, 44040192 padded size): + module.decoder.layers.14.mlp.experts.linear_fc1.weight13 + module.decoder.layers.14.mlp.experts.linear_fc1.weight11 + module.decoder.layers.14.mlp.experts.linear_fc1.weight9 + module.decoder.layers.14.mlp.experts.linear_fc1.weight12 + module.decoder.layers.14.mlp.experts.linear_fc1.weight10 + module.decoder.layers.14.mlp.experts.linear_fc1.weight8 +Params for bucket 8 (44040192 elements, 44040192 padded size): + module.decoder.layers.14.mlp.experts.linear_fc1.weight7 + module.decoder.layers.14.mlp.experts.linear_fc1.weight5 + module.decoder.layers.14.mlp.experts.linear_fc1.weight3 + module.decoder.layers.14.mlp.experts.linear_fc1.weight6 + module.decoder.layers.14.mlp.experts.linear_fc1.weight4 + module.decoder.layers.14.mlp.experts.linear_fc1.weight2 +Params for bucket 9 (40370176 elements, 40370176 padded size): + module.decoder.layers.14.mlp.experts.linear_fc1.weight1 + module.decoder.layers.13.mlp.experts.linear_fc2.weight14 + module.decoder.layers.13.mlp.experts.linear_fc2.weight12 + module.decoder.layers.13.mlp.experts.linear_fc2.weight10 + module.decoder.layers.13.mlp.experts.linear_fc2.weight15 + module.decoder.layers.13.mlp.experts.linear_fc2.weight13 + module.decoder.layers.14.mlp.experts.linear_fc1.weight0 + module.decoder.layers.13.mlp.experts.linear_fc2.weight11 + module.decoder.layers.13.mlp.experts.linear_fc2.weight9 +Params for bucket 10 (40370176 elements, 40370176 padded size): + module.decoder.layers.13.mlp.experts.linear_fc2.weight8 + module.decoder.layers.13.mlp.experts.linear_fc2.weight6 + module.decoder.layers.13.mlp.experts.linear_fc2.weight4 + module.decoder.layers.13.mlp.experts.linear_fc2.weight2 + module.decoder.layers.13.mlp.experts.linear_fc2.weight0 + module.decoder.layers.13.mlp.experts.linear_fc1.weight15 + module.decoder.layers.13.mlp.experts.linear_fc2.weight5 + module.decoder.layers.13.mlp.experts.linear_fc2.weight7 + module.decoder.layers.13.mlp.experts.linear_fc2.weight3 + module.decoder.layers.13.mlp.experts.linear_fc2.weight1 +Params for bucket 11 (44040192 elements, 44040192 padded size): + module.decoder.layers.13.mlp.experts.linear_fc1.weight11 + module.decoder.layers.13.mlp.experts.linear_fc1.weight13 + module.decoder.layers.13.mlp.experts.linear_fc1.weight9 + module.decoder.layers.13.mlp.experts.linear_fc1.weight14 + module.decoder.layers.13.mlp.experts.linear_fc1.weight10 + module.decoder.layers.13.mlp.experts.linear_fc1.weight12 +Params for bucket 12 (44040192 elements, 44040192 padded size): + module.decoder.layers.13.mlp.experts.linear_fc1.weight7 + 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.weight8 + module.decoder.layers.13.mlp.experts.linear_fc1.weight4 + module.decoder.layers.13.mlp.experts.linear_fc1.weight6 +Params for bucket 13 (40370176 elements, 40370176 padded size): + module.decoder.layers.13.mlp.experts.linear_fc1.weight1 + module.decoder.layers.12.mlp.experts.linear_fc2.weight15 + module.decoder.layers.12.mlp.experts.linear_fc2.weight13 + module.decoder.layers.12.mlp.experts.linear_fc2.weight11 + module.decoder.layers.13.mlp.experts.linear_fc1.weight2 + module.decoder.layers.12.mlp.experts.linear_fc2.weight14 + module.decoder.layers.13.mlp.experts.linear_fc1.weight0 + module.decoder.layers.12.mlp.experts.linear_fc2.weight12 +Params for bucket 14 (40370176 elements, 40370176 padded size): + module.decoder.layers.12.mlp.experts.linear_fc2.weight7 + module.decoder.layers.12.mlp.experts.linear_fc2.weight9 + 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.weight1 + module.decoder.layers.12.mlp.experts.linear_fc2.weight10 + module.decoder.layers.12.mlp.experts.linear_fc2.weight8 + 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.weight0 +Params for bucket 15 (44040192 elements, 44040192 padded size): + module.decoder.layers.12.mlp.experts.linear_fc1.weight14 + module.decoder.layers.12.mlp.experts.linear_fc1.weight12 + module.decoder.layers.12.mlp.experts.linear_fc1.weight10 + module.decoder.layers.12.mlp.experts.linear_fc1.weight15 + module.decoder.layers.12.mlp.experts.linear_fc1.weight13 + module.decoder.layers.12.mlp.experts.linear_fc1.weight11 +Params for bucket 16 (44040192 elements, 44040192 padded size): + module.decoder.layers.12.mlp.experts.linear_fc1.weight8 + module.decoder.layers.12.mlp.experts.linear_fc1.weight6 + module.decoder.layers.12.mlp.experts.linear_fc1.weight4 + module.decoder.layers.12.mlp.experts.linear_fc1.weight9 + module.decoder.layers.12.mlp.experts.linear_fc1.weight7 + module.decoder.layers.12.mlp.experts.linear_fc1.weight5 +Params for bucket 17 (40370176 elements, 40370176 padded size): + module.decoder.layers.12.mlp.experts.linear_fc1.weight0 + module.decoder.layers.12.mlp.experts.linear_fc1.weight2 + module.decoder.layers.11.mlp.experts.linear_fc2.weight14 + module.decoder.layers.12.mlp.experts.linear_fc1.weight3 + module.decoder.layers.12.mlp.experts.linear_fc1.weight1 + module.decoder.layers.11.mlp.experts.linear_fc2.weight15 + module.decoder.layers.11.mlp.experts.linear_fc2.weight13 +Params for bucket 18 (40370176 elements, 40370176 padded size): + module.decoder.layers.11.mlp.experts.linear_fc2.weight12 + module.decoder.layers.11.mlp.experts.linear_fc2.weight8 + module.decoder.layers.11.mlp.experts.linear_fc2.weight10 + 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 + module.decoder.layers.11.mlp.experts.linear_fc2.weight9 + module.decoder.layers.11.mlp.experts.linear_fc2.weight11 + 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 +Params for bucket 19 (44040192 elements, 44040192 padded size): + module.decoder.layers.11.mlp.experts.linear_fc2.weight0 + module.decoder.layers.11.mlp.experts.linear_fc1.weight14 + module.decoder.layers.11.mlp.experts.linear_fc1.weight12 + module.decoder.layers.11.mlp.experts.linear_fc2.weight1 + module.decoder.layers.11.mlp.experts.linear_fc1.weight13 + module.decoder.layers.11.mlp.experts.linear_fc1.weight15 + module.decoder.layers.11.mlp.experts.linear_fc1.weight11 +Params for bucket 20 (44040192 elements, 44040192 padded size): + module.decoder.layers.11.mlp.experts.linear_fc1.weight10 + module.decoder.layers.11.mlp.experts.linear_fc1.weight6 + module.decoder.layers.11.mlp.experts.linear_fc1.weight8 + module.decoder.layers.11.mlp.experts.linear_fc1.weight7 + module.decoder.layers.11.mlp.experts.linear_fc1.weight9 + module.decoder.layers.11.mlp.experts.linear_fc1.weight5 +Params for bucket 21 (40370176 elements, 40370176 padded size): + module.decoder.layers.11.mlp.experts.linear_fc1.weight4 + module.decoder.layers.11.mlp.experts.linear_fc1.weight0 + module.decoder.layers.11.mlp.experts.linear_fc1.weight2 + module.decoder.layers.11.mlp.experts.linear_fc1.weight3 + module.decoder.layers.11.mlp.experts.linear_fc1.weight1 + module.decoder.layers.10.mlp.experts.linear_fc2.weight15 +Params for bucket 22 (40370176 elements, 40370176 padded size): + module.decoder.layers.10.mlp.experts.linear_fc2.weight14 + module.decoder.layers.10.mlp.experts.linear_fc2.weight12 + module.decoder.layers.10.mlp.experts.linear_fc2.weight10 + module.decoder.layers.10.mlp.experts.linear_fc2.weight8 + 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.weight11 + module.decoder.layers.10.mlp.experts.linear_fc2.weight13 + module.decoder.layers.10.mlp.experts.linear_fc2.weight9 + module.decoder.layers.10.mlp.experts.linear_fc2.weight7 + module.decoder.layers.10.mlp.experts.linear_fc2.weight5 +Params for bucket 23 (44040192 elements, 44040192 padded size): + module.decoder.layers.10.mlp.experts.linear_fc2.weight2 + module.decoder.layers.10.mlp.experts.linear_fc1.weight15 + module.decoder.layers.10.mlp.experts.linear_fc2.weight0 + module.decoder.layers.10.mlp.experts.linear_fc1.weight13 + module.decoder.layers.10.mlp.experts.linear_fc2.weight3 + module.decoder.layers.10.mlp.experts.linear_fc2.weight1 + module.decoder.layers.10.mlp.experts.linear_fc1.weight14 + module.decoder.layers.10.mlp.experts.linear_fc1.weight12 +Params for bucket 24 (44040192 elements, 44040192 padded size): + module.decoder.layers.10.mlp.experts.linear_fc1.weight11 + module.decoder.layers.10.mlp.experts.linear_fc1.weight7 + module.decoder.layers.10.mlp.experts.linear_fc1.weight9 + module.decoder.layers.10.mlp.experts.linear_fc1.weight8 + module.decoder.layers.10.mlp.experts.linear_fc1.weight10 + module.decoder.layers.10.mlp.experts.linear_fc1.weight6 +Params for bucket 25 (44040192 elements, 44040192 padded size): + module.decoder.layers.10.mlp.experts.linear_fc1.weight5 + module.decoder.layers.10.mlp.experts.linear_fc1.weight1 + module.decoder.layers.10.mlp.experts.linear_fc1.weight3 + module.decoder.layers.10.mlp.experts.linear_fc1.weight2 + module.decoder.layers.10.mlp.experts.linear_fc1.weight4 + module.decoder.layers.10.mlp.experts.linear_fc1.weight0 +Params for bucket 26 (40370176 elements, 40370176 padded size): + module.decoder.layers.9.mlp.experts.linear_fc2.weight12 + module.decoder.layers.9.mlp.experts.linear_fc2.weight14 + module.decoder.layers.9.mlp.experts.linear_fc2.weight10 + module.decoder.layers.9.mlp.experts.linear_fc2.weight8 + module.decoder.layers.9.mlp.experts.linear_fc2.weight6 + module.decoder.layers.9.mlp.experts.linear_fc2.weight15 + module.decoder.layers.9.mlp.experts.linear_fc2.weight13 + module.decoder.layers.9.mlp.experts.linear_fc2.weight11 + module.decoder.layers.9.mlp.experts.linear_fc2.weight9 + module.decoder.layers.9.mlp.experts.linear_fc2.weight7 + module.decoder.layers.9.mlp.experts.linear_fc2.weight5 +Params for bucket 27 (40370176 elements, 40370176 padded size): + 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.weight0 + module.decoder.layers.9.mlp.experts.linear_fc1.weight15 + module.decoder.layers.9.mlp.experts.linear_fc1.weight13 + module.decoder.layers.9.mlp.experts.linear_fc2.weight3 + module.decoder.layers.9.mlp.experts.linear_fc2.weight1 + module.decoder.layers.9.mlp.experts.linear_fc1.weight14 +Params for bucket 28 (44040192 elements, 44040192 padded size): + module.decoder.layers.9.mlp.experts.linear_fc1.weight11 + module.decoder.layers.9.mlp.experts.linear_fc1.weight9 + module.decoder.layers.9.mlp.experts.linear_fc1.weight7 + module.decoder.layers.9.mlp.experts.linear_fc1.weight12 + module.decoder.layers.9.mlp.experts.linear_fc1.weight8 + module.decoder.layers.9.mlp.experts.linear_fc1.weight10 +Params for bucket 29 (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.weight1 + module.decoder.layers.9.mlp.experts.linear_fc1.weight6 + module.decoder.layers.9.mlp.experts.linear_fc1.weight4 + module.decoder.layers.9.mlp.experts.linear_fc1.weight2 +Params for bucket 30 (40370176 elements, 40370176 padded size): + module.decoder.layers.8.mlp.experts.linear_fc2.weight12 + module.decoder.layers.8.mlp.experts.linear_fc2.weight14 + module.decoder.layers.8.mlp.experts.linear_fc2.weight10 + module.decoder.layers.8.mlp.experts.linear_fc2.weight8 + module.decoder.layers.9.mlp.experts.linear_fc1.weight0 + module.decoder.layers.8.mlp.experts.linear_fc2.weight15 + module.decoder.layers.8.mlp.experts.linear_fc2.weight13 + module.decoder.layers.8.mlp.experts.linear_fc2.weight11 + module.decoder.layers.8.mlp.experts.linear_fc2.weight9 + module.decoder.layers.8.mlp.experts.linear_fc2.weight7 +Params for bucket 31 (40370176 elements, 40370176 padded size): + 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.weight4 + module.decoder.layers.8.mlp.experts.linear_fc2.weight0 + module.decoder.layers.8.mlp.experts.linear_fc1.weight15 + 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.weight1 + module.decoder.layers.8.mlp.experts.linear_fc1.weight14 +Params for bucket 32 (44040192 elements, 44040192 padded size): + module.decoder.layers.8.mlp.experts.linear_fc1.weight13 + module.decoder.layers.8.mlp.experts.linear_fc1.weight11 + module.decoder.layers.8.mlp.experts.linear_fc1.weight9 + module.decoder.layers.8.mlp.experts.linear_fc1.weight12 + module.decoder.layers.8.mlp.experts.linear_fc1.weight10 + module.decoder.layers.8.mlp.experts.linear_fc1.weight8 +Params for bucket 33 (44040192 elements, 44040192 padded size): + module.decoder.layers.8.mlp.experts.linear_fc1.weight7 + module.decoder.layers.8.mlp.experts.linear_fc1.weight5 + module.decoder.layers.8.mlp.experts.linear_fc1.weight3 + module.decoder.layers.8.mlp.experts.linear_fc1.weight6 + module.decoder.layers.8.mlp.experts.linear_fc1.weight4 + module.decoder.layers.8.mlp.experts.linear_fc1.weight2 +Params for bucket 34 (40370176 elements, 40370176 padded size): + module.decoder.layers.8.mlp.experts.linear_fc1.weight1 + module.decoder.layers.7.mlp.experts.linear_fc2.weight13 + module.decoder.layers.7.mlp.experts.linear_fc2.weight15 + module.decoder.layers.7.mlp.experts.linear_fc2.weight11 + module.decoder.layers.7.mlp.experts.linear_fc2.weight9 + module.decoder.layers.7.mlp.experts.linear_fc2.weight14 + module.decoder.layers.8.mlp.experts.linear_fc1.weight0 + module.decoder.layers.7.mlp.experts.linear_fc2.weight12 + module.decoder.layers.7.mlp.experts.linear_fc2.weight10 +Params for bucket 35 (40370176 elements, 40370176 padded size): + 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 + module.decoder.layers.7.mlp.experts.linear_fc2.weight1 + module.decoder.layers.7.mlp.experts.linear_fc2.weight8 + 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.weight0 + module.decoder.layers.7.mlp.experts.linear_fc1.weight15 +Params for bucket 36 (44040192 elements, 44040192 padded size): + module.decoder.layers.7.mlp.experts.linear_fc1.weight14 + module.decoder.layers.7.mlp.experts.linear_fc1.weight12 + module.decoder.layers.7.mlp.experts.linear_fc1.weight10 + module.decoder.layers.7.mlp.experts.linear_fc1.weight13 + module.decoder.layers.7.mlp.experts.linear_fc1.weight11 + module.decoder.layers.7.mlp.experts.linear_fc1.weight9 +Params for bucket 37 (44040192 elements, 44040192 padded size): + module.decoder.layers.7.mlp.experts.linear_fc1.weight8 + module.decoder.layers.7.mlp.experts.linear_fc1.weight6 + module.decoder.layers.7.mlp.experts.linear_fc1.weight4 + module.decoder.layers.7.mlp.experts.linear_fc1.weight5 + module.decoder.layers.7.mlp.experts.linear_fc1.weight7 + module.decoder.layers.7.mlp.experts.linear_fc1.weight3 +Params for bucket 38 (40370176 elements, 40370176 padded size): + module.decoder.layers.7.mlp.experts.linear_fc1.weight2 + module.decoder.layers.6.mlp.experts.linear_fc2.weight15 + module.decoder.layers.7.mlp.experts.linear_fc1.weight0 + module.decoder.layers.6.mlp.experts.linear_fc2.weight13 + module.decoder.layers.6.mlp.experts.linear_fc2.weight11 + module.decoder.layers.6.mlp.experts.linear_fc2.weight14 + module.decoder.layers.7.mlp.experts.linear_fc1.weight1 + module.decoder.layers.6.mlp.experts.linear_fc2.weight12 +Params for bucket 39 (40370176 elements, 40370176 padded size): + module.decoder.layers.6.mlp.experts.linear_fc2.weight9 + 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.weight1 + module.decoder.layers.6.mlp.experts.linear_fc2.weight10 + module.decoder.layers.6.mlp.experts.linear_fc2.weight8 + 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 + module.decoder.layers.6.mlp.experts.linear_fc2.weight0 +Params for bucket 40 (44040192 elements, 44040192 padded size): + module.decoder.layers.6.mlp.experts.linear_fc1.weight14 + module.decoder.layers.6.mlp.experts.linear_fc1.weight12 + module.decoder.layers.6.mlp.experts.linear_fc1.weight10 + module.decoder.layers.6.mlp.experts.linear_fc1.weight15 + module.decoder.layers.6.mlp.experts.linear_fc1.weight13 + module.decoder.layers.6.mlp.experts.linear_fc1.weight11 +Params for bucket 41 (44040192 elements, 44040192 padded size): + module.decoder.layers.6.mlp.experts.linear_fc1.weight8 + module.decoder.layers.6.mlp.experts.linear_fc1.weight6 + module.decoder.layers.6.mlp.experts.linear_fc1.weight4 + module.decoder.layers.6.mlp.experts.linear_fc1.weight9 + module.decoder.layers.6.mlp.experts.linear_fc1.weight7 + module.decoder.layers.6.mlp.experts.linear_fc1.weight5 +Params for bucket 42 (40370176 elements, 40370176 padded size): + module.decoder.layers.6.mlp.experts.linear_fc1.weight0 + module.decoder.layers.6.mlp.experts.linear_fc1.weight2 + module.decoder.layers.5.mlp.experts.linear_fc2.weight14 + module.decoder.layers.6.mlp.experts.linear_fc1.weight3 + module.decoder.layers.6.mlp.experts.linear_fc1.weight1 + module.decoder.layers.5.mlp.experts.linear_fc2.weight15 + module.decoder.layers.5.mlp.experts.linear_fc2.weight13 +Params for bucket 43 (40370176 elements, 40370176 padded size): + module.decoder.layers.5.mlp.experts.linear_fc2.weight12 + module.decoder.layers.5.mlp.experts.linear_fc2.weight8 + module.decoder.layers.5.mlp.experts.linear_fc2.weight10 + 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.weight9 + module.decoder.layers.5.mlp.experts.linear_fc2.weight11 + 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 +Params for bucket 44 (44040192 elements, 44040192 padded size): + module.decoder.layers.5.mlp.experts.linear_fc2.weight0 + module.decoder.layers.5.mlp.experts.linear_fc1.weight15 + module.decoder.layers.5.mlp.experts.linear_fc1.weight13 + module.decoder.layers.5.mlp.experts.linear_fc1.weight11 + module.decoder.layers.5.mlp.experts.linear_fc2.weight1 + module.decoder.layers.5.mlp.experts.linear_fc1.weight14 + module.decoder.layers.5.mlp.experts.linear_fc1.weight12 +Params for bucket 45 (44040192 elements, 44040192 padded size): + module.decoder.layers.5.mlp.experts.linear_fc1.weight7 + module.decoder.layers.5.mlp.experts.linear_fc1.weight9 + module.decoder.layers.5.mlp.experts.linear_fc1.weight5 + module.decoder.layers.5.mlp.experts.linear_fc1.weight10 + module.decoder.layers.5.mlp.experts.linear_fc1.weight8 + module.decoder.layers.5.mlp.experts.linear_fc1.weight6 +Params for bucket 46 (40370176 elements, 40370176 padded size): + module.decoder.layers.5.mlp.experts.linear_fc1.weight1 + module.decoder.layers.5.mlp.experts.linear_fc1.weight3 + module.decoder.layers.4.mlp.experts.linear_fc2.weight15 + module.decoder.layers.5.mlp.experts.linear_fc1.weight4 + module.decoder.layers.5.mlp.experts.linear_fc1.weight0 + module.decoder.layers.5.mlp.experts.linear_fc1.weight2 +Params for bucket 47 (40370176 elements, 40370176 padded size): + module.decoder.layers.4.mlp.experts.linear_fc2.weight11 + module.decoder.layers.4.mlp.experts.linear_fc2.weight13 + module.decoder.layers.4.mlp.experts.linear_fc2.weight9 + 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.weight14 + module.decoder.layers.4.mlp.experts.linear_fc2.weight12 + module.decoder.layers.4.mlp.experts.linear_fc2.weight10 + module.decoder.layers.4.mlp.experts.linear_fc2.weight8 + module.decoder.layers.4.mlp.experts.linear_fc2.weight6 + module.decoder.layers.4.mlp.experts.linear_fc2.weight4 +Params for bucket 48 (44040192 elements, 44040192 padded size): + module.decoder.layers.4.mlp.experts.linear_fc2.weight3 + module.decoder.layers.4.mlp.experts.linear_fc2.weight1 + module.decoder.layers.4.mlp.experts.linear_fc1.weight14 + module.decoder.layers.4.mlp.experts.linear_fc1.weight12 + module.decoder.layers.4.mlp.experts.linear_fc2.weight2 + module.decoder.layers.4.mlp.experts.linear_fc1.weight15 + module.decoder.layers.4.mlp.experts.linear_fc2.weight0 + module.decoder.layers.4.mlp.experts.linear_fc1.weight13 +Params for bucket 49 (44040192 elements, 44040192 padded size): + module.decoder.layers.4.mlp.experts.linear_fc1.weight8 + module.decoder.layers.4.mlp.experts.linear_fc1.weight10 + module.decoder.layers.4.mlp.experts.linear_fc1.weight6 + module.decoder.layers.4.mlp.experts.linear_fc1.weight11 + module.decoder.layers.4.mlp.experts.linear_fc1.weight9 + module.decoder.layers.4.mlp.experts.linear_fc1.weight7 +Params for bucket 50 (44040192 elements, 44040192 padded size): + module.decoder.layers.4.mlp.experts.linear_fc1.weight4 + 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.weight5 + module.decoder.layers.4.mlp.experts.linear_fc1.weight1 + module.decoder.layers.4.mlp.experts.linear_fc1.weight3 +Params for bucket 51 (40370176 elements, 40370176 padded size): + module.decoder.layers.3.mlp.experts.linear_fc2.weight12 + module.decoder.layers.3.mlp.experts.linear_fc2.weight14 + module.decoder.layers.3.mlp.experts.linear_fc2.weight10 + module.decoder.layers.3.mlp.experts.linear_fc2.weight8 + module.decoder.layers.3.mlp.experts.linear_fc2.weight6 + module.decoder.layers.3.mlp.experts.linear_fc2.weight15 + module.decoder.layers.3.mlp.experts.linear_fc2.weight13 + module.decoder.layers.3.mlp.experts.linear_fc2.weight11 + module.decoder.layers.3.mlp.experts.linear_fc2.weight9 + module.decoder.layers.3.mlp.experts.linear_fc2.weight7 + module.decoder.layers.3.mlp.experts.linear_fc2.weight5 +Params for bucket 52 (40370176 elements, 40370176 padded size): + module.decoder.layers.3.mlp.experts.linear_fc2.weight4 + module.decoder.layers.3.mlp.experts.linear_fc2.weight0 + module.decoder.layers.3.mlp.experts.linear_fc2.weight2 + module.decoder.layers.3.mlp.experts.linear_fc1.weight15 + module.decoder.layers.3.mlp.experts.linear_fc1.weight13 + 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_fc1.weight14 +Params for bucket 53 (44040192 elements, 44040192 padded size): + module.decoder.layers.3.mlp.experts.linear_fc1.weight11 + module.decoder.layers.3.mlp.experts.linear_fc1.weight9 + module.decoder.layers.3.mlp.experts.linear_fc1.weight7 + module.decoder.layers.3.mlp.experts.linear_fc1.weight12 + module.decoder.layers.3.mlp.experts.linear_fc1.weight8 + module.decoder.layers.3.mlp.experts.linear_fc1.weight10 +Params for bucket 54 (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.weight1 + module.decoder.layers.3.mlp.experts.linear_fc1.weight6 + module.decoder.layers.3.mlp.experts.linear_fc1.weight2 + module.decoder.layers.3.mlp.experts.linear_fc1.weight4 +Params for bucket 55 (40370176 elements, 40370176 padded size): + module.decoder.layers.2.mlp.experts.linear_fc2.weight12 + module.decoder.layers.2.mlp.experts.linear_fc2.weight14 + module.decoder.layers.2.mlp.experts.linear_fc2.weight10 + module.decoder.layers.2.mlp.experts.linear_fc2.weight8 + module.decoder.layers.3.mlp.experts.linear_fc1.weight0 + module.decoder.layers.2.mlp.experts.linear_fc2.weight15 + module.decoder.layers.2.mlp.experts.linear_fc2.weight13 + module.decoder.layers.2.mlp.experts.linear_fc2.weight11 + module.decoder.layers.2.mlp.experts.linear_fc2.weight9 + module.decoder.layers.2.mlp.experts.linear_fc2.weight7 +Params for bucket 56 (40370176 elements, 40370176 padded size): + module.decoder.layers.2.mlp.experts.linear_fc2.weight6 + module.decoder.layers.2.mlp.experts.linear_fc2.weight2 + module.decoder.layers.2.mlp.experts.linear_fc2.weight4 + module.decoder.layers.2.mlp.experts.linear_fc2.weight0 + module.decoder.layers.2.mlp.experts.linear_fc1.weight15 + module.decoder.layers.2.mlp.experts.linear_fc2.weight3 + module.decoder.layers.2.mlp.experts.linear_fc2.weight5 + module.decoder.layers.2.mlp.experts.linear_fc2.weight1 + module.decoder.layers.2.mlp.experts.linear_fc1.weight14 +Params for bucket 57 (44040192 elements, 44040192 padded size): + module.decoder.layers.2.mlp.experts.linear_fc1.weight13 + module.decoder.layers.2.mlp.experts.linear_fc1.weight9 + module.decoder.layers.2.mlp.experts.linear_fc1.weight11 + module.decoder.layers.2.mlp.experts.linear_fc1.weight12 + module.decoder.layers.2.mlp.experts.linear_fc1.weight10 + module.decoder.layers.2.mlp.experts.linear_fc1.weight8 +Params for bucket 58 (44040192 elements, 44040192 padded size): + module.decoder.layers.2.mlp.experts.linear_fc1.weight7 + module.decoder.layers.2.mlp.experts.linear_fc1.weight5 + module.decoder.layers.2.mlp.experts.linear_fc1.weight3 + module.decoder.layers.2.mlp.experts.linear_fc1.weight6 + module.decoder.layers.2.mlp.experts.linear_fc1.weight4 + module.decoder.layers.2.mlp.experts.linear_fc1.weight2 +Params for bucket 59 (40370176 elements, 40370176 padded size): + module.decoder.layers.2.mlp.experts.linear_fc1.weight1 + module.decoder.layers.1.mlp.experts.linear_fc2.weight13 + module.decoder.layers.1.mlp.experts.linear_fc2.weight15 + module.decoder.layers.1.mlp.experts.linear_fc2.weight11 + module.decoder.layers.1.mlp.experts.linear_fc2.weight9 + module.decoder.layers.1.mlp.experts.linear_fc2.weight14 + module.decoder.layers.2.mlp.experts.linear_fc1.weight0 + module.decoder.layers.1.mlp.experts.linear_fc2.weight12 + module.decoder.layers.1.mlp.experts.linear_fc2.weight10 +Params for bucket 60 (40370176 elements, 40370176 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.weight1 + module.decoder.layers.1.mlp.experts.linear_fc2.weight8 + 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 + module.decoder.layers.1.mlp.experts.linear_fc2.weight0 + module.decoder.layers.1.mlp.experts.linear_fc1.weight15 +Params for bucket 61 (44040192 elements, 44040192 padded size): + module.decoder.layers.1.mlp.experts.linear_fc1.weight14 + module.decoder.layers.1.mlp.experts.linear_fc1.weight12 + module.decoder.layers.1.mlp.experts.linear_fc1.weight10 + module.decoder.layers.1.mlp.experts.linear_fc1.weight13 + module.decoder.layers.1.mlp.experts.linear_fc1.weight11 + module.decoder.layers.1.mlp.experts.linear_fc1.weight9 +Params for bucket 62 (44040192 elements, 44040192 padded size): + module.decoder.layers.1.mlp.experts.linear_fc1.weight8 + module.decoder.layers.1.mlp.experts.linear_fc1.weight6 + module.decoder.layers.1.mlp.experts.linear_fc1.weight4 + module.decoder.layers.1.mlp.experts.linear_fc1.weight7 + module.decoder.layers.1.mlp.experts.linear_fc1.weight5 + module.decoder.layers.1.mlp.experts.linear_fc1.weight3 +Params for bucket 63 (40370176 elements, 40370176 padded size): + module.decoder.layers.1.mlp.experts.linear_fc1.weight2 + module.decoder.layers.1.mlp.experts.linear_fc1.weight0 + module.decoder.layers.0.mlp.experts.linear_fc2.weight15 + module.decoder.layers.0.mlp.experts.linear_fc2.weight13 + module.decoder.layers.0.mlp.experts.linear_fc2.weight11 + module.decoder.layers.0.mlp.experts.linear_fc2.weight14 + module.decoder.layers.1.mlp.experts.linear_fc1.weight1 + module.decoder.layers.0.mlp.experts.linear_fc2.weight12 +Params for bucket 64 (40370176 elements, 40370176 padded size): + module.decoder.layers.0.mlp.experts.linear_fc2.weight7 + module.decoder.layers.0.mlp.experts.linear_fc2.weight3 + module.decoder.layers.0.mlp.experts.linear_fc2.weight1 + module.decoder.layers.0.mlp.experts.linear_fc2.weight10 + module.decoder.layers.0.mlp.experts.linear_fc2.weight9 + module.decoder.layers.0.mlp.experts.linear_fc2.weight8 + module.decoder.layers.0.mlp.experts.linear_fc2.weight6 + module.decoder.layers.0.mlp.experts.linear_fc2.weight5 + 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.weight0 +Params for bucket 65 (44040192 elements, 44040192 padded size): + module.decoder.layers.0.mlp.experts.linear_fc1.weight15 + module.decoder.layers.0.mlp.experts.linear_fc1.weight13 + module.decoder.layers.0.mlp.experts.linear_fc1.weight11 + module.decoder.layers.0.mlp.experts.linear_fc1.weight14 + module.decoder.layers.0.mlp.experts.linear_fc1.weight12 + module.decoder.layers.0.mlp.experts.linear_fc1.weight10 +Params for bucket 66 (44040192 elements, 44040192 padded size): + module.decoder.layers.0.mlp.experts.linear_fc1.weight9 + module.decoder.layers.0.mlp.experts.linear_fc1.weight6 + module.decoder.layers.0.mlp.experts.linear_fc1.weight7 + module.decoder.layers.0.mlp.experts.linear_fc1.weight4 + module.decoder.layers.0.mlp.experts.linear_fc1.weight5 + module.decoder.layers.0.mlp.experts.linear_fc1.weight8 +Params for bucket 67 (29360128 elements, 29360128 padded size): + module.decoder.layers.0.mlp.experts.linear_fc1.weight3 + 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: 2026-08-30 06:34:14 +> building train, validation, and test datasets ... + > datasets target sizes (minimum size): + train: 192 + 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=(192, 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 44ed1ee5bcb9e5b030387b8d27a6c406-GPTDataset-train-document_index.npy +INFO:megatron.core.datasets.gpt_dataset: Load the sample index from 44ed1ee5bcb9e5b030387b8d27a6c406-GPTDataset-train-sample_index.npy +INFO:megatron.core.datasets.gpt_dataset: Load the shuffle index from 44ed1ee5bcb9e5b030387b8d27a6c406-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: 2026-08-30 06:34:14 +done with setup ... +training ... +(min, max) time across ranks (ms): + model-and-optimizer-setup ......................: (763.17, 770.91) + train/valid/test-data-iterators-setup ..........: (66.51, 77.02) +Overwriting rerun_state_machine.current_iteration from -1 to 0... +[before the start of training step] datetime: 2026-08-30 06:34:14 +/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.") +/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: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. + 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: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.") + [2026-08-30 06:34:52] iteration 1/ 12 | consumed samples: 16 | elapsed time per iteration (ms): 38370.6 | throughput per GPU (TFLOP/s/GPU): 8.7 | learning rate: 2.000000E-05 | global batch size: 16 | lm loss: 1.102834E+01 | loss scale: 1.0 | grad norm: 44.250 | number of skipped iterations: 0 | number of nan iterations: 0 | +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: 58498.755859375 | max allocated: 73513.7607421875 | reserved: 77004.0 | max reserved: 77004.0 + [2026-08-30 06:34:53] iteration 2/ 12 | consumed samples: 32 | elapsed time per iteration (ms): 981.6 | throughput per GPU (TFLOP/s/GPU): 339.6 | learning rate: 4.000000E-05 | global batch size: 16 | lm loss: 1.102085E+01 | loss scale: 1.0 | grad norm: 45.950 | number of skipped iterations: 0 | number of nan iterations: 0 | + [2026-08-30 06:34:54] iteration 3/ 12 | consumed samples: 48 | elapsed time per iteration (ms): 550.7 | throughput per GPU (TFLOP/s/GPU): 605.4 | learning rate: 6.000000E-05 | global batch size: 16 | lm loss: 1.057874E+01 | loss scale: 1.0 | grad norm: 218.301 | number of skipped iterations: 0 | number of nan iterations: 0 | +Capture CUDA graph for training!!! +CUDA graph capture done!!! + [2026-08-30 06:34:56] iteration 4/ 12 | consumed samples: 64 | elapsed time per iteration (ms): 1929.7 | throughput per GPU (TFLOP/s/GPU): 172.8 | learning rate: 8.000000E-05 | global batch size: 16 | lm loss: 1.300300E+01 | loss scale: 1.0 | grad norm: 45.905 | number of skipped iterations: 0 | number of nan iterations: 0 | + [2026-08-30 06:34:56] iteration 5/ 12 | consumed samples: 80 | elapsed time per iteration (ms): 462.1 | throughput per GPU (TFLOP/s/GPU): 721.5 | learning rate: 1.000000E-04 | global batch size: 16 | lm loss: 1.034165E+01 | loss scale: 1.0 | grad norm: 18.005 | number of skipped iterations: 0 | number of nan iterations: 0 | + [2026-08-30 06:34:57] iteration 6/ 12 | consumed samples: 96 | elapsed time per iteration (ms): 463.5 | throughput per GPU (TFLOP/s/GPU): 719.3 | learning rate: 7.750000E-05 | global batch size: 16 | lm loss: 1.009386E+01 | loss scale: 1.0 | grad norm: 9.966 | number of skipped iterations: 0 | number of nan iterations: 0 | + [2026-08-30 06:34:57] iteration 7/ 12 | consumed samples: 112 | elapsed time per iteration (ms): 462.2 | throughput per GPU (TFLOP/s/GPU): 721.4 | learning rate: 3.250000E-05 | global batch size: 16 | lm loss: 9.657560E+00 | loss scale: 1.0 | grad norm: 6.161 | number of skipped iterations: 0 | number of nan iterations: 0 | + [2026-08-30 06:34:57] iteration 8/ 12 | consumed samples: 128 | elapsed time per iteration (ms): 468.6 | throughput per GPU (TFLOP/s/GPU): 711.5 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 9.023036E+00 | loss scale: 1.0 | grad norm: 5.425 | number of skipped iterations: 0 | number of nan iterations: 0 | + [2026-08-30 06:34:58] iteration 9/ 12 | consumed samples: 144 | elapsed time per iteration (ms): 459.9 | throughput per GPU (TFLOP/s/GPU): 724.8 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 8.763896E+00 | loss scale: 1.0 | grad norm: 5.064 | number of skipped iterations: 0 | number of nan iterations: 0 | + [2026-08-30 06:34:58] iteration 10/ 12 | consumed samples: 160 | elapsed time per iteration (ms): 460.1 | throughput per GPU (TFLOP/s/GPU): 724.7 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 8.627990E+00 | loss scale: 1.0 | grad norm: 4.462 | number of skipped iterations: 0 | number of nan iterations: 0 | + [2026-08-30 06:34:59] iteration 11/ 12 | consumed samples: 176 | elapsed time per iteration (ms): 460.4 | throughput per GPU (TFLOP/s/GPU): 724.0 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 8.552834E+00 | loss scale: 1.0 | grad norm: 3.979 | number of skipped iterations: 0 | number of nan iterations: 0 | + [2026-08-30 06:34:59] iteration 12/ 12 | consumed samples: 192 | elapsed time per iteration (ms): 462.7 | throughput per GPU (TFLOP/s/GPU): 720.5 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 8.466097E+00 | loss scale: 1.0 | grad norm: 4.669 | number of skipped iterations: 0 | number of nan iterations: 0 | +[after training is done] datetime: 2026-08-30 06:34:59 +[rank0]:[W830 06:35:00.170058409 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()) diff --git a/tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/G4__echo__one__seq4096__gbs16__cap1p095703125/rep3/mem_samples.csv b/tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/G4__echo__one__seq4096__gbs16__cap1p095703125/rep3/mem_samples.csv new file mode 100644 index 0000000000000000000000000000000000000000..1cf4b5d536eb3db82b7a55ac9ef6ca2beafd032c --- /dev/null +++ b/tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/G4__echo__one__seq4096__gbs16__cap1p095703125/rep3/mem_samples.csv @@ -0,0 +1,256 @@ +0, 0 +1, 0 +2, 0 +3, 0 +4, 0 +5, 0 +6, 0 +7, 0 +0, 0 +1, 0 +2, 0 +3, 0 +4, 0 +5, 0 +6, 0 +7, 0 +0, 2 +1, 1 +2, 1 +3, 1 +4, 1 +5, 1 +6, 1 +7, 1 +0, 4 +1, 4 +2, 4 +3, 4 +4, 1 +5, 1 +6, 1 +7, 1 +0, 4 +1, 4 +2, 4 +3, 4 +4, 4 +5, 4 +6, 4 +7, 4 +0, 821 +1, 819 +2, 819 +3, 819 +4, 819 +5, 821 +6, 819 +7, 819 +0, 1537 +1, 1537 +2, 1537 +3, 1537 +4, 1537 +5, 1537 +6, 1537 +7, 1537 +0, 2231 +1, 2231 +2, 2231 +3, 2231 +4, 2231 +5, 2231 +6, 2231 +7, 2231 +0, 38511 +1, 39003 +2, 38981 +3, 39005 +4, 38553 +5, 39001 +6, 39013 +7, 39039 +0, 39167 +1, 39619 +2, 39627 +3, 39605 +4, 39171 +5, 39603 +6, 39637 +7, 39635 +0, 41635 +1, 42083 +2, 42083 +3, 42083 +4, 41635 +5, 42083 +6, 42083 +7, 42083 +0, 41651 +1, 42099 +2, 42099 +3, 42099 +4, 41651 +5, 42099 +6, 42099 +7, 42099 +0, 41651 +1, 42099 +2, 42099 +3, 42099 +4, 41651 +5, 42099 +6, 42099 +7, 42099 +0, 42581 +1, 43029 +2, 43029 +3, 43029 +4, 42581 +5, 43029 +6, 43029 +7, 43029 +0, 43703 +1, 44151 +2, 44151 +3, 44151 +4, 43703 +5, 44151 +6, 44151 +7, 44151 +0, 45319 +1, 45767 +2, 45767 +3, 45767 +4, 45319 +5, 45767 +6, 45767 +7, 44151 +0, 46445 +1, 46893 +2, 46893 +3, 46893 +4, 46445 +5, 46893 +6, 46893 +7, 46893 +0, 73369 +1, 73341 +2, 73341 +3, 73341 +4, 73389 +5, 73341 +6, 73341 +7, 73341 +0, 74493 +1, 74465 +2, 74465 +3, 74465 +4, 74513 +5, 74465 +6, 74465 +7, 74465 +0, 74493 +1, 74465 +2, 74465 +3, 74465 +4, 74513 +5, 74465 +6, 74465 +7, 74465 +0, 74493 +1, 74465 +2, 74465 +3, 74465 +4, 74513 +5, 74465 +6, 74465 +7, 74465 +0, 82957 +1, 82917 +2, 82947 +3, 82609 +4, 82999 +5, 82617 +6, 82921 +7, 82951 +0, 83821 +1, 83821 +2, 83821 +3, 83463 +4, 83845 +5, 83479 +6, 83839 +7, 83847 +0, 84539 +1, 84509 +2, 84509 +3, 84397 +4, 84559 +5, 84397 +6, 84509 +7, 84509 +0, 85125 +1, 85127 +2, 85111 +3, 85023 +4, 85159 +5, 85035 +6, 85139 +7, 85123 +0, 84903 +1, 95753 +2, 85567 +3, 85901 +4, 84915 +5, 85111 +6, 94065 +7, 88503 +0, 126687 +1, 126675 +2, 126671 +3, 127009 +4, 126705 +5, 126895 +6, 126787 +7, 126673 +0, 126687 +1, 126675 +2, 126671 +3, 127009 +4, 126705 +5, 126895 +6, 126787 +7, 126673 +0, 126019 +1, 126065 +2, 125805 +3, 126725 +4, 126357 +5, 126349 +6, 126503 +7, 126127 +0, 1834 +1, 1834 +2, 2090 +3, 126079 +4, 3114 +5, 2090 +6, 125857 +7, 3114 +0, 0 +1, 0 +2, 0 +3, 0 +4, 0 +5, 0 +6, 0 +7, 0 +0, 0 +1, 0 +2, 0 +3, 0 +4, 0 +5, 0 +6, 0 +7, 0 diff --git a/tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/G4__echo__one__seq4096__gbs16__cap1p095703125/rep3/run.env b/tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/G4__echo__one__seq4096__gbs16__cap1p095703125/rep3/run.env new file mode 100644 index 0000000000000000000000000000000000000000..9a57e0eee2ddeb1ead96f25351d51cb4ef7cfbd2 --- /dev/null +++ b/tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/G4__echo__one__seq4096__gbs16__cap1p095703125/rep3/run.env @@ -0,0 +1,70 @@ +backend=echo +config=G4 +family=tierA +repo=/workspace/Megatron-MoE-Echo +branch=bracing-regions-through-kernels +commit=9a236793 +num_layers=16 +hidden=4096 +ffn_hidden=14336 +moe_ffn_hidden=896 +num_experts=128 +topk=32 +experts_per_rank=16 +expert_params=11010048 +heads=32 +kv_groups=8 +kv_channels= +qk_layernorm=0 +shared_ffn=0 +shared_expert_overlap=0 +moe_token_dispatcher=flex +mla=0 +dense_first_layers=0 +moe_layers=16 +rotary_base=1000000 +rotary_scaling= +qkv_bias=0 +router_balancing=aux_loss +aux_loss_coeff=1e-2 +router_pre_softmax=0 +router_fusion=0 +router_dtype=fp32 +device_initiated_grouped_gemm=1 +norm_epsilon=1e-5 +seq_length=4096 +micro_batch_size=1 +global_batch_size=16 +num_microbatches=2 +assignments_per_rank=131072 +mean_tokens_per_expert=8192 +slot_policy=one +slots_per_rank=1 +capacity_mode=static +capacity=1.095703125 +cuda_graph=1 +probe_mode=0 +async_grad_reduce=0 +branch_role=timing +memory_dump_mode=none +memory_snapshot_path= +cuda_graph_warmup_steps=3 +recompute=none +force_balance=0 +train_iters=12 +ep_size=8 +echo_dispatch_overlap=1 +echo_dispatch_recompute=0 +ultraep_quota_min_tokens= +ultraep_weight_sync_join= +profile_enable=0 +kernel_trace=0 +fp8_mode=mxfp8 +fp8_recipe=mxfp8 +training_seed=1234 +learning_rate=1e-4 +min_learning_rate=1.0e-5 +lr_warmup_iters=5 +state_dump_enable=0 +state_dump_dir= +moonep_assert_no_pending_reduce=0 diff --git a/tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/G4__echo__one__seq4096__gbs16__cap1p095703125/rep3/train.log b/tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/G4__echo__one__seq4096__gbs16__cap1p095703125/rep3/train.log new file mode 100644 index 0000000000000000000000000000000000000000..25e1b9ab61ff46a719f451231cab474197a63874 --- /dev/null +++ b/tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/G4__echo__one__seq4096__gbs16__cap1p095703125/rep3/train.log @@ -0,0 +1,1502 @@ +============================================================== + G4 / echo (slots/rank=1 policy=one) + E=128 K=32 H'=896 m=16 + L=16 seq=4096 gbs=16 (2 microbatches) + cap_mode=static cap=1.095703125 cg=1 recompute=none + memory_dump=none snapshots=/workspace/tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/G4__echo__one__seq4096__gbs16__cap1p095703125/rep3/memory_snapshots/snapshot.pickle + repo=/workspace/Megatron-MoE-Echo @ 9a236793 +============================================================== +W0830 06:35:09.705000 626231 torch/distributed/run.py:766] +W0830 06:35:09.705000 626231 torch/distributed/run.py:766] ***************************************** +W0830 06:35:09.705000 626231 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. +W0830 06:35:09.705000 626231 torch/distributed/run.py:766] ***************************************** +/workspace/Megatron-MoE-Echo/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-MoE-Echo/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-MoE-Echo/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-MoE-Echo/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-MoE-Echo/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_deepep_num_sms .............................. 20 + moe_echo_enable_random_offloading ............... False + moe_echo_expert_dispatch_overlap ................ True + moe_echo_expert_dispatcher_type ................. hybridep + moe_echo_recompute_expert_dispatch .............. False + moe_enable_deepep ............................... False + moe_enable_echo ................................. True + moe_enable_hybridep ............................. True + moe_expert_capacity_factor ...................... None + moe_extended_tp ................................. False + moe_ffn_hidden_size ............................. 896 + 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 ............................ 8 + moe_pad_expert_input_to_capacity ................ False + moe_per_layer_logging ........................... False + moe_permute_fusion .............................. False + moe_received_token_capacity ..................... 1.095703125 + 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 ................................. 32 + 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_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 ..................................... 128 + 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 ......................................... False + profile_ranks ................................... [0] + profile_step_end ................................ 12 + profile_step_start .............................. 10 + 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 ..................................... 12 + 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 ... + > 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 ... +/workspace/Megatron-MoE-Echo/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-MoE-Echo/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-MoE-Echo/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 +[Gloo] Rank 4 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 6 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 [Gloo] Rank 357 is connected to is connected to is connected to 777 peer ranks. peer ranks. peer ranks. Expected number of connected peer ranks is : Expected number of connected peer ranks is : Expected number of connected peer ranks is : 77 +7 + +[Gloo] Rank 4 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 7 is connected to 7 peer ranks. Expected number of connected peer ranks is : 7 +[Gloo] Rank [Gloo] Rank 56 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 [Gloo] Rank 3 is connected to 27 is connected to peer ranks. 7Expected number of connected peer ranks is : peer ranks. 7Expected 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 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[Gloo] Rank +[Gloo] Rank 00 is connected to is connected to 00 peer ranks. peer ranks. Expected number of connected peer ranks is : Expected number of connected peer ranks is : 00 + +[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 +[rank5]:[W830 06:35:27.473972034 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. +[rank7]:[W830 06:35:27.473984138 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]:[W830 06:35:27.474029291 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. +[rank3]:[W830 06:35:27.474046934 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]:[W830 06:35:27.474078295 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. +[rank1]:[W830 06:35:27.474090014 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. +[rank2]:[W830 06:35:27.478590198 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. +> compiling dataset index builder ... +make: Entering directory '/workspace/Megatron-MoE-Echo/megatron/core/datasets' +make: Nothing to be done for 'default'. +make: Leaving directory '/workspace/Megatron-MoE-Echo/megatron/core/datasets' +>>> done with dataset index builder. Compilation time: 0.034 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]:[W830 06:35:27.628222428 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: 3.826 seconds +time to initialize megatron (seconds): 7.190 +[after megatron is initialized] datetime: 2026-08-30 06:35:33 +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( + > number of parameters on (tensor, pipeline) model parallel rank (0, 0): 3910275072 +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 (42479616 elements, 42479616 padded size): + module.decoder.layers.15.mlp.router.weight + module.decoder.layers.15.self_attention.linear_qkv.weight + module.decoder.layers.15.self_attention.linear_proj.weight + module.decoder.final_layernorm.weight + module.decoder.layers.15.pre_mlp_layernorm.weight + module.decoder.layers.15.self_attention.linear_qkv.layer_norm_weight +Params for bucket 3 (42475520 elements, 42475520 padded size): + 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 + module.decoder.layers.14.self_attention.linear_qkv.weight + module.decoder.layers.14.pre_mlp_layernorm.weight +Params for bucket 4 (42475520 elements, 42475520 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_proj.weight + module.decoder.layers.13.self_attention.linear_qkv.layer_norm_weight +Params for bucket 5 (42475520 elements, 42475520 padded size): + module.decoder.layers.12.pre_mlp_layernorm.weight + module.decoder.layers.12.self_attention.linear_proj.weight + module.decoder.layers.12.self_attention.linear_qkv.weight + module.decoder.layers.12.mlp.router.weight + module.decoder.layers.12.self_attention.linear_qkv.layer_norm_weight +Params for bucket 6 (42475520 elements, 42475520 padded size): + module.decoder.layers.11.self_attention.linear_proj.weight + module.decoder.layers.11.self_attention.linear_qkv.weight + module.decoder.layers.11.pre_mlp_layernorm.weight + module.decoder.layers.11.mlp.router.weight + module.decoder.layers.11.self_attention.linear_qkv.layer_norm_weight +Params for bucket 7 (42475520 elements, 42475520 padded size): + module.decoder.layers.10.mlp.router.weight + module.decoder.layers.10.self_attention.linear_qkv.weight + module.decoder.layers.10.self_attention.linear_proj.weight + module.decoder.layers.10.pre_mlp_layernorm.weight + module.decoder.layers.10.self_attention.linear_qkv.layer_norm_weight +Params for bucket 8 (42475520 elements, 42475520 padded size): + module.decoder.layers.9.self_attention.linear_qkv.weight + module.decoder.layers.9.self_attention.linear_proj.weight + module.decoder.layers.9.pre_mlp_layernorm.weight + module.decoder.layers.9.mlp.router.weight + module.decoder.layers.9.self_attention.linear_qkv.layer_norm_weight +Params for bucket 9 (42475520 elements, 42475520 padded size): + 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 + module.decoder.layers.8.pre_mlp_layernorm.weight + module.decoder.layers.8.self_attention.linear_qkv.weight +Params for bucket 10 (42475520 elements, 42475520 padded size): + 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 + module.decoder.layers.7.mlp.router.weight + module.decoder.layers.7.self_attention.linear_qkv.layer_norm_weight +Params for bucket 11 (42475520 elements, 42475520 padded size): + module.decoder.layers.6.mlp.router.weight + module.decoder.layers.6.self_attention.linear_qkv.weight + module.decoder.layers.6.pre_mlp_layernorm.weight + module.decoder.layers.6.self_attention.linear_proj.weight + module.decoder.layers.6.self_attention.linear_qkv.layer_norm_weight +Params for bucket 12 (42475520 elements, 42475520 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 +Params for bucket 13 (42475520 elements, 42475520 padded size): + 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_qkv.layer_norm_weight +Params for bucket 14 (42475520 elements, 42475520 padded size): + module.decoder.layers.3.self_attention.linear_qkv.layer_norm_weight + module.decoder.layers.3.pre_mlp_layernorm.weight + module.decoder.layers.3.mlp.router.weight + module.decoder.layers.3.self_attention.linear_qkv.weight + module.decoder.layers.3.self_attention.linear_proj.weight +Params for bucket 15 (42475520 elements, 42475520 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 (42475520 elements, 42475520 padded size): + 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 + module.decoder.layers.1.self_attention.linear_qkv.weight + module.decoder.layers.1.pre_mlp_layernorm.weight +Params for bucket 17 (42475520 elements, 42475520 padded size): + 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 + module.decoder.layers.0.self_attention.linear_proj.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: 67 +Params for bucket 1 (40370176 elements, 40370176 padded size): + module.decoder.layers.15.mlp.experts.linear_fc2.weight14 + module.decoder.layers.15.mlp.experts.linear_fc2.weight12 + module.decoder.layers.15.mlp.experts.linear_fc2.weight10 + module.decoder.layers.15.mlp.experts.linear_fc2.weight8 + module.decoder.layers.15.mlp.experts.linear_fc2.weight6 + module.decoder.layers.15.mlp.experts.linear_fc2.weight15 + module.decoder.layers.15.mlp.experts.linear_fc2.weight13 + module.decoder.layers.15.mlp.experts.linear_fc2.weight11 + module.decoder.layers.15.mlp.experts.linear_fc2.weight9 + module.decoder.layers.15.mlp.experts.linear_fc2.weight7 + module.decoder.layers.15.mlp.experts.linear_fc2.weight5 +Params for bucket 2 (40370176 elements, 40370176 padded size): + module.decoder.layers.15.mlp.experts.linear_fc2.weight4 + module.decoder.layers.15.mlp.experts.linear_fc2.weight0 + module.decoder.layers.15.mlp.experts.linear_fc2.weight2 + module.decoder.layers.15.mlp.experts.linear_fc1.weight15 + module.decoder.layers.15.mlp.experts.linear_fc1.weight13 + module.decoder.layers.15.mlp.experts.linear_fc2.weight1 + module.decoder.layers.15.mlp.experts.linear_fc2.weight3 + module.decoder.layers.15.mlp.experts.linear_fc1.weight14 +Params for bucket 3 (44040192 elements, 44040192 padded size): + module.decoder.layers.15.mlp.experts.linear_fc1.weight9 + module.decoder.layers.15.mlp.experts.linear_fc1.weight11 + module.decoder.layers.15.mlp.experts.linear_fc1.weight7 + module.decoder.layers.15.mlp.experts.linear_fc1.weight12 + module.decoder.layers.15.mlp.experts.linear_fc1.weight8 + module.decoder.layers.15.mlp.experts.linear_fc1.weight10 +Params for bucket 4 (44040192 elements, 44040192 padded size): + module.decoder.layers.15.mlp.experts.linear_fc1.weight3 + module.decoder.layers.15.mlp.experts.linear_fc1.weight5 + module.decoder.layers.15.mlp.experts.linear_fc1.weight1 + module.decoder.layers.15.mlp.experts.linear_fc1.weight6 + module.decoder.layers.15.mlp.experts.linear_fc1.weight2 + module.decoder.layers.15.mlp.experts.linear_fc1.weight4 +Params for bucket 5 (40370176 elements, 40370176 padded size): + module.decoder.layers.14.mlp.experts.linear_fc2.weight15 + module.decoder.layers.14.mlp.experts.linear_fc2.weight13 + module.decoder.layers.14.mlp.experts.linear_fc2.weight11 + module.decoder.layers.14.mlp.experts.linear_fc2.weight9 + module.decoder.layers.14.mlp.experts.linear_fc2.weight7 + module.decoder.layers.15.mlp.experts.linear_fc1.weight0 + module.decoder.layers.14.mlp.experts.linear_fc2.weight14 + module.decoder.layers.14.mlp.experts.linear_fc2.weight12 + module.decoder.layers.14.mlp.experts.linear_fc2.weight10 + module.decoder.layers.14.mlp.experts.linear_fc2.weight8 +Params for bucket 6 (40370176 elements, 40370176 padded size): + module.decoder.layers.14.mlp.experts.linear_fc2.weight5 + module.decoder.layers.14.mlp.experts.linear_fc2.weight3 + module.decoder.layers.14.mlp.experts.linear_fc2.weight1 + module.decoder.layers.14.mlp.experts.linear_fc1.weight15 + module.decoder.layers.14.mlp.experts.linear_fc2.weight6 + module.decoder.layers.14.mlp.experts.linear_fc2.weight2 + module.decoder.layers.14.mlp.experts.linear_fc2.weight4 + module.decoder.layers.14.mlp.experts.linear_fc2.weight0 + module.decoder.layers.14.mlp.experts.linear_fc1.weight14 +Params for bucket 7 (44040192 elements, 44040192 padded size): + module.decoder.layers.14.mlp.experts.linear_fc1.weight13 + module.decoder.layers.14.mlp.experts.linear_fc1.weight11 + module.decoder.layers.14.mlp.experts.linear_fc1.weight9 + module.decoder.layers.14.mlp.experts.linear_fc1.weight12 + module.decoder.layers.14.mlp.experts.linear_fc1.weight10 + module.decoder.layers.14.mlp.experts.linear_fc1.weight8 +Params for bucket 8 (44040192 elements, 44040192 padded size): + module.decoder.layers.14.mlp.experts.linear_fc1.weight7 + module.decoder.layers.14.mlp.experts.linear_fc1.weight5 + module.decoder.layers.14.mlp.experts.linear_fc1.weight3 + module.decoder.layers.14.mlp.experts.linear_fc1.weight4 + module.decoder.layers.14.mlp.experts.linear_fc1.weight6 + module.decoder.layers.14.mlp.experts.linear_fc1.weight2 +Params for bucket 9 (40370176 elements, 40370176 padded size): + module.decoder.layers.14.mlp.experts.linear_fc1.weight1 + module.decoder.layers.13.mlp.experts.linear_fc2.weight14 + module.decoder.layers.13.mlp.experts.linear_fc2.weight12 + module.decoder.layers.13.mlp.experts.linear_fc2.weight10 + module.decoder.layers.13.mlp.experts.linear_fc2.weight15 + module.decoder.layers.13.mlp.experts.linear_fc2.weight13 + module.decoder.layers.14.mlp.experts.linear_fc1.weight0 + module.decoder.layers.13.mlp.experts.linear_fc2.weight11 + module.decoder.layers.13.mlp.experts.linear_fc2.weight9 +Params for bucket 10 (40370176 elements, 40370176 padded size): + module.decoder.layers.13.mlp.experts.linear_fc2.weight8 + module.decoder.layers.13.mlp.experts.linear_fc2.weight6 + module.decoder.layers.13.mlp.experts.linear_fc2.weight4 + module.decoder.layers.13.mlp.experts.linear_fc2.weight2 + module.decoder.layers.13.mlp.experts.linear_fc2.weight0 + module.decoder.layers.13.mlp.experts.linear_fc1.weight15 + module.decoder.layers.13.mlp.experts.linear_fc2.weight5 + module.decoder.layers.13.mlp.experts.linear_fc2.weight7 + module.decoder.layers.13.mlp.experts.linear_fc2.weight3 + module.decoder.layers.13.mlp.experts.linear_fc2.weight1 +Params for bucket 11 (44040192 elements, 44040192 padded size): + module.decoder.layers.13.mlp.experts.linear_fc1.weight13 + module.decoder.layers.13.mlp.experts.linear_fc1.weight11 + module.decoder.layers.13.mlp.experts.linear_fc1.weight9 + module.decoder.layers.13.mlp.experts.linear_fc1.weight14 + module.decoder.layers.13.mlp.experts.linear_fc1.weight10 + module.decoder.layers.13.mlp.experts.linear_fc1.weight12 +Params for bucket 12 (44040192 elements, 44040192 padded size): + module.decoder.layers.13.mlp.experts.linear_fc1.weight7 + 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.weight8 + module.decoder.layers.13.mlp.experts.linear_fc1.weight4 + module.decoder.layers.13.mlp.experts.linear_fc1.weight6 +Params for bucket 13 (40370176 elements, 40370176 padded size): + module.decoder.layers.13.mlp.experts.linear_fc1.weight1 + module.decoder.layers.12.mlp.experts.linear_fc2.weight14 + module.decoder.layers.12.mlp.experts.linear_fc2.weight12 + module.decoder.layers.13.mlp.experts.linear_fc1.weight2 + module.decoder.layers.12.mlp.experts.linear_fc2.weight15 + module.decoder.layers.13.mlp.experts.linear_fc1.weight0 + module.decoder.layers.12.mlp.experts.linear_fc2.weight13 + module.decoder.layers.12.mlp.experts.linear_fc2.weight11 +Params for bucket 14 (40370176 elements, 40370176 padded size): + module.decoder.layers.12.mlp.experts.linear_fc2.weight10 + module.decoder.layers.12.mlp.experts.linear_fc2.weight8 + 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.weight0 + module.decoder.layers.12.mlp.experts.linear_fc2.weight9 + 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.weight1 +Params for bucket 15 (44040192 elements, 44040192 padded size): + module.decoder.layers.12.mlp.experts.linear_fc1.weight15 + module.decoder.layers.12.mlp.experts.linear_fc1.weight11 + module.decoder.layers.12.mlp.experts.linear_fc1.weight13 + module.decoder.layers.12.mlp.experts.linear_fc1.weight14 + module.decoder.layers.12.mlp.experts.linear_fc1.weight12 + module.decoder.layers.12.mlp.experts.linear_fc1.weight10 +Params for bucket 16 (44040192 elements, 44040192 padded size): + module.decoder.layers.12.mlp.experts.linear_fc1.weight9 + module.decoder.layers.12.mlp.experts.linear_fc1.weight7 + module.decoder.layers.12.mlp.experts.linear_fc1.weight5 + module.decoder.layers.12.mlp.experts.linear_fc1.weight8 + module.decoder.layers.12.mlp.experts.linear_fc1.weight6 + module.decoder.layers.12.mlp.experts.linear_fc1.weight4 +Params for bucket 17 (40370176 elements, 40370176 padded size): + module.decoder.layers.12.mlp.experts.linear_fc1.weight3 + module.decoder.layers.12.mlp.experts.linear_fc1.weight1 + module.decoder.layers.11.mlp.experts.linear_fc2.weight14 + module.decoder.layers.12.mlp.experts.linear_fc1.weight2 + module.decoder.layers.11.mlp.experts.linear_fc2.weight15 + module.decoder.layers.12.mlp.experts.linear_fc1.weight0 + module.decoder.layers.11.mlp.experts.linear_fc2.weight13 +Params for bucket 18 (40370176 elements, 40370176 padded size): + module.decoder.layers.11.mlp.experts.linear_fc2.weight12 + module.decoder.layers.11.mlp.experts.linear_fc2.weight10 + module.decoder.layers.11.mlp.experts.linear_fc2.weight8 + 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 + module.decoder.layers.11.mlp.experts.linear_fc2.weight9 + module.decoder.layers.11.mlp.experts.linear_fc2.weight11 + 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 +Params for bucket 19 (44040192 elements, 44040192 padded size): + module.decoder.layers.11.mlp.experts.linear_fc2.weight0 + module.decoder.layers.11.mlp.experts.linear_fc1.weight15 + module.decoder.layers.11.mlp.experts.linear_fc1.weight13 + module.decoder.layers.11.mlp.experts.linear_fc1.weight11 + module.decoder.layers.11.mlp.experts.linear_fc2.weight1 + module.decoder.layers.11.mlp.experts.linear_fc1.weight14 + module.decoder.layers.11.mlp.experts.linear_fc1.weight12 +Params for bucket 20 (44040192 elements, 44040192 padded size): + module.decoder.layers.11.mlp.experts.linear_fc1.weight9 + module.decoder.layers.11.mlp.experts.linear_fc1.weight7 + module.decoder.layers.11.mlp.experts.linear_fc1.weight5 + module.decoder.layers.11.mlp.experts.linear_fc1.weight10 + module.decoder.layers.11.mlp.experts.linear_fc1.weight8 + module.decoder.layers.11.mlp.experts.linear_fc1.weight6 +Params for bucket 21 (40370176 elements, 40370176 padded size): + module.decoder.layers.11.mlp.experts.linear_fc1.weight3 + module.decoder.layers.11.mlp.experts.linear_fc1.weight1 + module.decoder.layers.11.mlp.experts.linear_fc1.weight4 + module.decoder.layers.11.mlp.experts.linear_fc1.weight2 + module.decoder.layers.11.mlp.experts.linear_fc1.weight0 + module.decoder.layers.10.mlp.experts.linear_fc2.weight15 +Params for bucket 22 (40370176 elements, 40370176 padded size): + module.decoder.layers.10.mlp.experts.linear_fc2.weight14 + module.decoder.layers.10.mlp.experts.linear_fc2.weight12 + module.decoder.layers.10.mlp.experts.linear_fc2.weight10 + module.decoder.layers.10.mlp.experts.linear_fc2.weight8 + 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.weight13 + module.decoder.layers.10.mlp.experts.linear_fc2.weight11 + module.decoder.layers.10.mlp.experts.linear_fc2.weight9 + module.decoder.layers.10.mlp.experts.linear_fc2.weight7 + module.decoder.layers.10.mlp.experts.linear_fc2.weight5 +Params for bucket 23 (44040192 elements, 44040192 padded size): + module.decoder.layers.10.mlp.experts.linear_fc2.weight2 + module.decoder.layers.10.mlp.experts.linear_fc1.weight15 + module.decoder.layers.10.mlp.experts.linear_fc2.weight0 + module.decoder.layers.10.mlp.experts.linear_fc1.weight13 + module.decoder.layers.10.mlp.experts.linear_fc2.weight3 + module.decoder.layers.10.mlp.experts.linear_fc2.weight1 + module.decoder.layers.10.mlp.experts.linear_fc1.weight14 + module.decoder.layers.10.mlp.experts.linear_fc1.weight12 +Params for bucket 24 (44040192 elements, 44040192 padded size): + module.decoder.layers.10.mlp.experts.linear_fc1.weight11 + module.decoder.layers.10.mlp.experts.linear_fc1.weight9 + module.decoder.layers.10.mlp.experts.linear_fc1.weight7 + module.decoder.layers.10.mlp.experts.linear_fc1.weight10 + module.decoder.layers.10.mlp.experts.linear_fc1.weight8 + module.decoder.layers.10.mlp.experts.linear_fc1.weight6 +Params for bucket 25 (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.weight1 + module.decoder.layers.10.mlp.experts.linear_fc1.weight2 + module.decoder.layers.10.mlp.experts.linear_fc1.weight4 + module.decoder.layers.10.mlp.experts.linear_fc1.weight0 +Params for bucket 26 (40370176 elements, 40370176 padded size): + module.decoder.layers.9.mlp.experts.linear_fc2.weight15 + module.decoder.layers.9.mlp.experts.linear_fc2.weight13 + module.decoder.layers.9.mlp.experts.linear_fc2.weight11 + module.decoder.layers.9.mlp.experts.linear_fc2.weight9 + module.decoder.layers.9.mlp.experts.linear_fc2.weight7 + module.decoder.layers.9.mlp.experts.linear_fc2.weight5 + module.decoder.layers.9.mlp.experts.linear_fc2.weight14 + module.decoder.layers.9.mlp.experts.linear_fc2.weight12 + module.decoder.layers.9.mlp.experts.linear_fc2.weight10 + module.decoder.layers.9.mlp.experts.linear_fc2.weight8 + module.decoder.layers.9.mlp.experts.linear_fc2.weight6 +Params for bucket 27 (40370176 elements, 40370176 padded size): + module.decoder.layers.9.mlp.experts.linear_fc2.weight3 + module.decoder.layers.9.mlp.experts.linear_fc2.weight1 + module.decoder.layers.9.mlp.experts.linear_fc1.weight14 + 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.weight0 + module.decoder.layers.9.mlp.experts.linear_fc1.weight15 + module.decoder.layers.9.mlp.experts.linear_fc1.weight13 +Params for bucket 28 (44040192 elements, 44040192 padded size): + module.decoder.layers.9.mlp.experts.linear_fc1.weight12 + module.decoder.layers.9.mlp.experts.linear_fc1.weight8 + module.decoder.layers.9.mlp.experts.linear_fc1.weight10 + module.decoder.layers.9.mlp.experts.linear_fc1.weight9 + module.decoder.layers.9.mlp.experts.linear_fc1.weight11 + module.decoder.layers.9.mlp.experts.linear_fc1.weight7 +Params for bucket 29 (44040192 elements, 44040192 padded size): + module.decoder.layers.9.mlp.experts.linear_fc1.weight6 + module.decoder.layers.9.mlp.experts.linear_fc1.weight2 + module.decoder.layers.9.mlp.experts.linear_fc1.weight4 + 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.weight1 +Params for bucket 30 (40370176 elements, 40370176 padded size): + module.decoder.layers.9.mlp.experts.linear_fc1.weight0 + module.decoder.layers.8.mlp.experts.linear_fc2.weight14 + module.decoder.layers.8.mlp.experts.linear_fc2.weight12 + module.decoder.layers.8.mlp.experts.linear_fc2.weight10 + module.decoder.layers.8.mlp.experts.linear_fc2.weight8 + module.decoder.layers.8.mlp.experts.linear_fc2.weight13 + module.decoder.layers.8.mlp.experts.linear_fc2.weight15 + module.decoder.layers.8.mlp.experts.linear_fc2.weight11 + module.decoder.layers.8.mlp.experts.linear_fc2.weight9 + module.decoder.layers.8.mlp.experts.linear_fc2.weight7 +Params for bucket 31 (40370176 elements, 40370176 padded size): + 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.weight2 + module.decoder.layers.8.mlp.experts.linear_fc2.weight0 + module.decoder.layers.8.mlp.experts.linear_fc1.weight15 + 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.weight1 + module.decoder.layers.8.mlp.experts.linear_fc1.weight14 +Params for bucket 32 (44040192 elements, 44040192 padded size): + module.decoder.layers.8.mlp.experts.linear_fc1.weight13 + module.decoder.layers.8.mlp.experts.linear_fc1.weight9 + module.decoder.layers.8.mlp.experts.linear_fc1.weight11 + module.decoder.layers.8.mlp.experts.linear_fc1.weight10 + module.decoder.layers.8.mlp.experts.linear_fc1.weight12 + module.decoder.layers.8.mlp.experts.linear_fc1.weight8 +Params for bucket 33 (44040192 elements, 44040192 padded size): + module.decoder.layers.8.mlp.experts.linear_fc1.weight7 + module.decoder.layers.8.mlp.experts.linear_fc1.weight3 + 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.weight6 + module.decoder.layers.8.mlp.experts.linear_fc1.weight2 +Params for bucket 34 (40370176 elements, 40370176 padded size): + module.decoder.layers.8.mlp.experts.linear_fc1.weight1 + module.decoder.layers.7.mlp.experts.linear_fc2.weight15 + module.decoder.layers.7.mlp.experts.linear_fc2.weight13 + module.decoder.layers.7.mlp.experts.linear_fc2.weight11 + module.decoder.layers.7.mlp.experts.linear_fc2.weight9 + module.decoder.layers.8.mlp.experts.linear_fc1.weight0 + module.decoder.layers.7.mlp.experts.linear_fc2.weight14 + module.decoder.layers.7.mlp.experts.linear_fc2.weight12 + module.decoder.layers.7.mlp.experts.linear_fc2.weight10 +Params for bucket 35 (40370176 elements, 40370176 padded size): + module.decoder.layers.7.mlp.experts.linear_fc2.weight5 + 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.weight1 + module.decoder.layers.7.mlp.experts.linear_fc2.weight8 + 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.weight0 + module.decoder.layers.7.mlp.experts.linear_fc1.weight15 +Params for bucket 36 (44040192 elements, 44040192 padded size): + module.decoder.layers.7.mlp.experts.linear_fc1.weight14 + module.decoder.layers.7.mlp.experts.linear_fc1.weight10 + module.decoder.layers.7.mlp.experts.linear_fc1.weight12 + module.decoder.layers.7.mlp.experts.linear_fc1.weight11 + module.decoder.layers.7.mlp.experts.linear_fc1.weight13 + module.decoder.layers.7.mlp.experts.linear_fc1.weight9 +Params for bucket 37 (44040192 elements, 44040192 padded size): + module.decoder.layers.7.mlp.experts.linear_fc1.weight8 + module.decoder.layers.7.mlp.experts.linear_fc1.weight4 + module.decoder.layers.7.mlp.experts.linear_fc1.weight6 + module.decoder.layers.7.mlp.experts.linear_fc1.weight7 + module.decoder.layers.7.mlp.experts.linear_fc1.weight5 + module.decoder.layers.7.mlp.experts.linear_fc1.weight3 +Params for bucket 38 (40370176 elements, 40370176 padded size): + module.decoder.layers.7.mlp.experts.linear_fc1.weight2 + module.decoder.layers.6.mlp.experts.linear_fc2.weight14 + module.decoder.layers.7.mlp.experts.linear_fc1.weight0 + module.decoder.layers.6.mlp.experts.linear_fc2.weight12 + module.decoder.layers.7.mlp.experts.linear_fc1.weight1 + module.decoder.layers.6.mlp.experts.linear_fc2.weight15 + module.decoder.layers.6.mlp.experts.linear_fc2.weight13 + module.decoder.layers.6.mlp.experts.linear_fc2.weight11 +Params for bucket 39 (40370176 elements, 40370176 padded size): + module.decoder.layers.6.mlp.experts.linear_fc2.weight10 + module.decoder.layers.6.mlp.experts.linear_fc2.weight6 + module.decoder.layers.6.mlp.experts.linear_fc2.weight8 + module.decoder.layers.6.mlp.experts.linear_fc2.weight4 + module.decoder.layers.6.mlp.experts.linear_fc2.weight2 + module.decoder.layers.6.mlp.experts.linear_fc2.weight0 + module.decoder.layers.6.mlp.experts.linear_fc2.weight9 + 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.weight1 +Params for bucket 40 (44040192 elements, 44040192 padded size): + module.decoder.layers.6.mlp.experts.linear_fc1.weight15 + module.decoder.layers.6.mlp.experts.linear_fc1.weight11 + module.decoder.layers.6.mlp.experts.linear_fc1.weight13 + module.decoder.layers.6.mlp.experts.linear_fc1.weight12 + module.decoder.layers.6.mlp.experts.linear_fc1.weight14 + module.decoder.layers.6.mlp.experts.linear_fc1.weight10 +Params for bucket 41 (44040192 elements, 44040192 padded size): + module.decoder.layers.6.mlp.experts.linear_fc1.weight9 + module.decoder.layers.6.mlp.experts.linear_fc1.weight5 + 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_fc1.weight8 + module.decoder.layers.6.mlp.experts.linear_fc1.weight4 +Params for bucket 42 (40370176 elements, 40370176 padded size): + module.decoder.layers.6.mlp.experts.linear_fc1.weight3 + module.decoder.layers.6.mlp.experts.linear_fc1.weight1 + module.decoder.layers.5.mlp.experts.linear_fc2.weight14 + module.decoder.layers.6.mlp.experts.linear_fc1.weight2 + module.decoder.layers.6.mlp.experts.linear_fc1.weight0 + module.decoder.layers.5.mlp.experts.linear_fc2.weight15 + module.decoder.layers.5.mlp.experts.linear_fc2.weight13 +Params for bucket 43 (40370176 elements, 40370176 padded size): + module.decoder.layers.5.mlp.experts.linear_fc2.weight12 + module.decoder.layers.5.mlp.experts.linear_fc2.weight10 + module.decoder.layers.5.mlp.experts.linear_fc2.weight8 + 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.weight9 + module.decoder.layers.5.mlp.experts.linear_fc2.weight11 + 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 +Params for bucket 44 (44040192 elements, 44040192 padded size): + module.decoder.layers.5.mlp.experts.linear_fc2.weight0 + module.decoder.layers.5.mlp.experts.linear_fc1.weight15 + module.decoder.layers.5.mlp.experts.linear_fc1.weight13 + module.decoder.layers.5.mlp.experts.linear_fc1.weight11 + module.decoder.layers.5.mlp.experts.linear_fc2.weight1 + module.decoder.layers.5.mlp.experts.linear_fc1.weight14 + module.decoder.layers.5.mlp.experts.linear_fc1.weight12 +Params for bucket 45 (44040192 elements, 44040192 padded size): + module.decoder.layers.5.mlp.experts.linear_fc1.weight9 + module.decoder.layers.5.mlp.experts.linear_fc1.weight7 + module.decoder.layers.5.mlp.experts.linear_fc1.weight5 + module.decoder.layers.5.mlp.experts.linear_fc1.weight10 + module.decoder.layers.5.mlp.experts.linear_fc1.weight8 + module.decoder.layers.5.mlp.experts.linear_fc1.weight6 +Params for bucket 46 (40370176 elements, 40370176 padded size): + module.decoder.layers.5.mlp.experts.linear_fc1.weight3 + module.decoder.layers.5.mlp.experts.linear_fc1.weight1 + module.decoder.layers.4.mlp.experts.linear_fc2.weight15 + module.decoder.layers.5.mlp.experts.linear_fc1.weight4 + module.decoder.layers.5.mlp.experts.linear_fc1.weight2 + module.decoder.layers.5.mlp.experts.linear_fc1.weight0 +Params for bucket 47 (40370176 elements, 40370176 padded size): + module.decoder.layers.4.mlp.experts.linear_fc2.weight13 + module.decoder.layers.4.mlp.experts.linear_fc2.weight11 + module.decoder.layers.4.mlp.experts.linear_fc2.weight9 + 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.weight14 + module.decoder.layers.4.mlp.experts.linear_fc2.weight12 + module.decoder.layers.4.mlp.experts.linear_fc2.weight10 + module.decoder.layers.4.mlp.experts.linear_fc2.weight8 + module.decoder.layers.4.mlp.experts.linear_fc2.weight6 + module.decoder.layers.4.mlp.experts.linear_fc2.weight4 +Params for bucket 48 (44040192 elements, 44040192 padded size): + module.decoder.layers.4.mlp.experts.linear_fc2.weight3 + module.decoder.layers.4.mlp.experts.linear_fc2.weight1 + module.decoder.layers.4.mlp.experts.linear_fc1.weight14 + module.decoder.layers.4.mlp.experts.linear_fc1.weight12 + module.decoder.layers.4.mlp.experts.linear_fc2.weight2 + module.decoder.layers.4.mlp.experts.linear_fc1.weight15 + module.decoder.layers.4.mlp.experts.linear_fc2.weight0 + module.decoder.layers.4.mlp.experts.linear_fc1.weight13 +Params for bucket 49 (44040192 elements, 44040192 padded size): + module.decoder.layers.4.mlp.experts.linear_fc1.weight8 + module.decoder.layers.4.mlp.experts.linear_fc1.weight10 + module.decoder.layers.4.mlp.experts.linear_fc1.weight6 + module.decoder.layers.4.mlp.experts.linear_fc1.weight11 + module.decoder.layers.4.mlp.experts.linear_fc1.weight9 + module.decoder.layers.4.mlp.experts.linear_fc1.weight7 +Params for bucket 50 (44040192 elements, 44040192 padded size): + module.decoder.layers.4.mlp.experts.linear_fc1.weight2 + module.decoder.layers.4.mlp.experts.linear_fc1.weight4 + module.decoder.layers.4.mlp.experts.linear_fc1.weight0 + module.decoder.layers.4.mlp.experts.linear_fc1.weight5 + module.decoder.layers.4.mlp.experts.linear_fc1.weight1 + module.decoder.layers.4.mlp.experts.linear_fc1.weight3 +Params for bucket 51 (40370176 elements, 40370176 padded size): + module.decoder.layers.3.mlp.experts.linear_fc2.weight15 + module.decoder.layers.3.mlp.experts.linear_fc2.weight13 + module.decoder.layers.3.mlp.experts.linear_fc2.weight11 + module.decoder.layers.3.mlp.experts.linear_fc2.weight9 + module.decoder.layers.3.mlp.experts.linear_fc2.weight7 + module.decoder.layers.3.mlp.experts.linear_fc2.weight5 + module.decoder.layers.3.mlp.experts.linear_fc2.weight12 + module.decoder.layers.3.mlp.experts.linear_fc2.weight14 + module.decoder.layers.3.mlp.experts.linear_fc2.weight10 + module.decoder.layers.3.mlp.experts.linear_fc2.weight8 + module.decoder.layers.3.mlp.experts.linear_fc2.weight6 +Params for bucket 52 (40370176 elements, 40370176 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_fc1.weight14 + module.decoder.layers.3.mlp.experts.linear_fc2.weight4 + module.decoder.layers.3.mlp.experts.linear_fc2.weight0 + module.decoder.layers.3.mlp.experts.linear_fc2.weight2 + module.decoder.layers.3.mlp.experts.linear_fc1.weight15 + module.decoder.layers.3.mlp.experts.linear_fc1.weight13 +Params for bucket 53 (44040192 elements, 44040192 padded size): + module.decoder.layers.3.mlp.experts.linear_fc1.weight12 + module.decoder.layers.3.mlp.experts.linear_fc1.weight8 + module.decoder.layers.3.mlp.experts.linear_fc1.weight10 + module.decoder.layers.3.mlp.experts.linear_fc1.weight11 + module.decoder.layers.3.mlp.experts.linear_fc1.weight9 + module.decoder.layers.3.mlp.experts.linear_fc1.weight7 +Params for bucket 54 (44040192 elements, 44040192 padded size): + module.decoder.layers.3.mlp.experts.linear_fc1.weight6 + module.decoder.layers.3.mlp.experts.linear_fc1.weight2 + module.decoder.layers.3.mlp.experts.linear_fc1.weight4 + 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.weight1 +Params for bucket 55 (40370176 elements, 40370176 padded size): + module.decoder.layers.3.mlp.experts.linear_fc1.weight0 + module.decoder.layers.2.mlp.experts.linear_fc2.weight15 + module.decoder.layers.2.mlp.experts.linear_fc2.weight13 + module.decoder.layers.2.mlp.experts.linear_fc2.weight11 + module.decoder.layers.2.mlp.experts.linear_fc2.weight9 + module.decoder.layers.2.mlp.experts.linear_fc2.weight7 + module.decoder.layers.2.mlp.experts.linear_fc2.weight14 + module.decoder.layers.2.mlp.experts.linear_fc2.weight12 + module.decoder.layers.2.mlp.experts.linear_fc2.weight10 + module.decoder.layers.2.mlp.experts.linear_fc2.weight8 +Params for bucket 56 (40370176 elements, 40370176 padded size): + 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.weight1 + module.decoder.layers.2.mlp.experts.linear_fc1.weight14 + module.decoder.layers.2.mlp.experts.linear_fc2.weight6 + module.decoder.layers.2.mlp.experts.linear_fc2.weight2 + module.decoder.layers.2.mlp.experts.linear_fc2.weight4 + module.decoder.layers.2.mlp.experts.linear_fc2.weight0 + module.decoder.layers.2.mlp.experts.linear_fc1.weight15 +Params for bucket 57 (44040192 elements, 44040192 padded size): + module.decoder.layers.2.mlp.experts.linear_fc1.weight12 + module.decoder.layers.2.mlp.experts.linear_fc1.weight10 + module.decoder.layers.2.mlp.experts.linear_fc1.weight8 + module.decoder.layers.2.mlp.experts.linear_fc1.weight13 + module.decoder.layers.2.mlp.experts.linear_fc1.weight9 + module.decoder.layers.2.mlp.experts.linear_fc1.weight11 +Params for bucket 58 (44040192 elements, 44040192 padded size): + module.decoder.layers.2.mlp.experts.linear_fc1.weight6 + module.decoder.layers.2.mlp.experts.linear_fc1.weight4 + module.decoder.layers.2.mlp.experts.linear_fc1.weight2 + module.decoder.layers.2.mlp.experts.linear_fc1.weight7 + module.decoder.layers.2.mlp.experts.linear_fc1.weight5 + module.decoder.layers.2.mlp.experts.linear_fc1.weight3 +Params for bucket 59 (40370176 elements, 40370176 padded size): + module.decoder.layers.1.mlp.experts.linear_fc2.weight14 + module.decoder.layers.2.mlp.experts.linear_fc1.weight0 + module.decoder.layers.1.mlp.experts.linear_fc2.weight12 + module.decoder.layers.1.mlp.experts.linear_fc2.weight10 + module.decoder.layers.2.mlp.experts.linear_fc1.weight1 + module.decoder.layers.1.mlp.experts.linear_fc2.weight15 + module.decoder.layers.1.mlp.experts.linear_fc2.weight13 + module.decoder.layers.1.mlp.experts.linear_fc2.weight11 + module.decoder.layers.1.mlp.experts.linear_fc2.weight9 +Params for bucket 60 (40370176 elements, 40370176 padded size): + module.decoder.layers.1.mlp.experts.linear_fc2.weight8 + 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 + module.decoder.layers.1.mlp.experts.linear_fc2.weight0 + module.decoder.layers.1.mlp.experts.linear_fc1.weight15 + module.decoder.layers.1.mlp.experts.linear_fc2.weight5 + module.decoder.layers.1.mlp.experts.linear_fc2.weight7 + module.decoder.layers.1.mlp.experts.linear_fc2.weight3 + module.decoder.layers.1.mlp.experts.linear_fc2.weight1 +Params for bucket 61 (44040192 elements, 44040192 padded size): + module.decoder.layers.1.mlp.experts.linear_fc1.weight13 + module.decoder.layers.1.mlp.experts.linear_fc1.weight11 + module.decoder.layers.1.mlp.experts.linear_fc1.weight9 + module.decoder.layers.1.mlp.experts.linear_fc1.weight14 + module.decoder.layers.1.mlp.experts.linear_fc1.weight10 + module.decoder.layers.1.mlp.experts.linear_fc1.weight12 +Params for bucket 62 (44040192 elements, 44040192 padded size): + module.decoder.layers.1.mlp.experts.linear_fc1.weight7 + 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.weight8 + module.decoder.layers.1.mlp.experts.linear_fc1.weight4 + module.decoder.layers.1.mlp.experts.linear_fc1.weight6 +Params for bucket 63 (40370176 elements, 40370176 padded size): + module.decoder.layers.0.mlp.experts.linear_fc2.weight15 + module.decoder.layers.1.mlp.experts.linear_fc1.weight1 + module.decoder.layers.0.mlp.experts.linear_fc2.weight13 + module.decoder.layers.0.mlp.experts.linear_fc2.weight11 + module.decoder.layers.1.mlp.experts.linear_fc1.weight2 + module.decoder.layers.1.mlp.experts.linear_fc1.weight0 + module.decoder.layers.0.mlp.experts.linear_fc2.weight14 + module.decoder.layers.0.mlp.experts.linear_fc2.weight12 +Params for bucket 64 (40370176 elements, 40370176 padded size): + module.decoder.layers.0.mlp.experts.linear_fc2.weight7 + module.decoder.layers.0.mlp.experts.linear_fc2.weight9 + module.decoder.layers.0.mlp.experts.linear_fc2.weight4 + module.decoder.layers.0.mlp.experts.linear_fc2.weight3 + module.decoder.layers.0.mlp.experts.linear_fc2.weight1 + module.decoder.layers.0.mlp.experts.linear_fc2.weight10 + module.decoder.layers.0.mlp.experts.linear_fc2.weight8 + module.decoder.layers.0.mlp.experts.linear_fc2.weight6 + module.decoder.layers.0.mlp.experts.linear_fc2.weight5 + module.decoder.layers.0.mlp.experts.linear_fc2.weight2 + module.decoder.layers.0.mlp.experts.linear_fc2.weight0 +Params for bucket 65 (44040192 elements, 44040192 padded size): + module.decoder.layers.0.mlp.experts.linear_fc1.weight15 + module.decoder.layers.0.mlp.experts.linear_fc1.weight13 + module.decoder.layers.0.mlp.experts.linear_fc1.weight11 + module.decoder.layers.0.mlp.experts.linear_fc1.weight12 + module.decoder.layers.0.mlp.experts.linear_fc1.weight14 + module.decoder.layers.0.mlp.experts.linear_fc1.weight10 +Params for bucket 66 (44040192 elements, 44040192 padded size): + module.decoder.layers.0.mlp.experts.linear_fc1.weight7 + module.decoder.layers.0.mlp.experts.linear_fc1.weight8 + 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.weight9 + module.decoder.layers.0.mlp.experts.linear_fc1.weight6 +Params for bucket 67 (29360128 elements, 29360128 padded size): + module.decoder.layers.0.mlp.experts.linear_fc1.weight3 + module.decoder.layers.0.mlp.experts.linear_fc1.weight1 + module.decoder.layers.0.mlp.experts.linear_fc1.weight0 + module.decoder.layers.0.mlp.experts.linear_fc1.weight2 +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: 2026-08-30 06:35:33 +> building train, validation, and test datasets ... + > datasets target sizes (minimum size): + train: 192 + 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=(192, 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 44ed1ee5bcb9e5b030387b8d27a6c406-GPTDataset-train-document_index.npy +INFO:megatron.core.datasets.gpt_dataset: Load the sample index from 44ed1ee5bcb9e5b030387b8d27a6c406-GPTDataset-train-sample_index.npy +INFO:megatron.core.datasets.gpt_dataset: Load the shuffle index from 44ed1ee5bcb9e5b030387b8d27a6c406-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: 2026-08-30 06:35:33 +done with setup ... +(min, max) time across ranks (ms): + model-and-optimizer-setup ......................: (631.79, 658.08) + train/valid/test-data-iterators-setup ..........: (62.19, 70.72) +training ... +Overwriting rerun_state_machine.current_iteration from -1 to 0... +[before the start of training step] datetime: 2026-08-30 06:35:33 +/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: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: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: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: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.") +Number of parameters in transformer block in billions: 23.22 [2026-08-30 06:36:12] iteration 1/ 12 | consumed samples: 16 | elapsed time per iteration (ms): 38586.5 | throughput per GPU (TFLOP/s/GPU): 8.6 | learning rate: 2.000000E-05 | global batch size: 16 | lm loss: 1.102834E+01 | loss scale: 1.0 | grad norm: 44.251 | number of skipped iterations: 0 | number of nan iterations: 0 | + +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: 58498.755859375 | max allocated: 73513.7607421875 | reserved: 77116.0 | max reserved: 77116.0 + [2026-08-30 06:36:13] iteration 2/ 12 | consumed samples: 32 | elapsed time per iteration (ms): 884.1 | throughput per GPU (TFLOP/s/GPU): 377.1 | learning rate: 4.000000E-05 | global batch size: 16 | lm loss: 1.102085E+01 | loss scale: 1.0 | grad norm: 45.950 | number of skipped iterations: 0 | number of nan iterations: 0 | + [2026-08-30 06:36:14] iteration 3/ 12 | consumed samples: 48 | elapsed time per iteration (ms): 552.7 | throughput per GPU (TFLOP/s/GPU): 603.2 | learning rate: 6.000000E-05 | global batch size: 16 | lm loss: 1.057726E+01 | loss scale: 1.0 | grad norm: 218.166 | number of skipped iterations: 0 | number of nan iterations: 0 | +Capture CUDA graph for training!!! +CUDA graph capture done!!! + [2026-08-30 06:36:15] iteration 4/ 12 | consumed samples: 64 | elapsed time per iteration (ms): 1818.4 | throughput per GPU (TFLOP/s/GPU): 183.3 | learning rate: 8.000000E-05 | global batch size: 16 | lm loss: 1.300323E+01 | loss scale: 1.0 | grad norm: 45.868 | number of skipped iterations: 0 | number of nan iterations: 0 | + [2026-08-30 06:36:16] iteration 5/ 12 | consumed samples: 80 | elapsed time per iteration (ms): 464.3 | throughput per GPU (TFLOP/s/GPU): 718.0 | learning rate: 1.000000E-04 | global batch size: 16 | lm loss: 1.034158E+01 | loss scale: 1.0 | grad norm: 18.008 | number of skipped iterations: 0 | number of nan iterations: 0 | + [2026-08-30 06:36:16] iteration 6/ 12 | consumed samples: 96 | elapsed time per iteration (ms): 462.2 | throughput per GPU (TFLOP/s/GPU): 721.2 | learning rate: 7.750000E-05 | global batch size: 16 | lm loss: 1.009415E+01 | loss scale: 1.0 | grad norm: 10.021 | number of skipped iterations: 0 | number of nan iterations: 0 | + [2026-08-30 06:36:17] iteration 7/ 12 | consumed samples: 112 | elapsed time per iteration (ms): 462.9 | throughput per GPU (TFLOP/s/GPU): 720.2 | learning rate: 3.250000E-05 | global batch size: 16 | lm loss: 9.659429E+00 | loss scale: 1.0 | grad norm: 6.158 | number of skipped iterations: 0 | number of nan iterations: 0 | + [2026-08-30 06:36:17] iteration 8/ 12 | consumed samples: 128 | elapsed time per iteration (ms): 459.7 | throughput per GPU (TFLOP/s/GPU): 725.2 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 9.025377E+00 | loss scale: 1.0 | grad norm: 5.421 | number of skipped iterations: 0 | number of nan iterations: 0 | + [2026-08-30 06:36:18] iteration 9/ 12 | consumed samples: 144 | elapsed time per iteration (ms): 458.9 | throughput per GPU (TFLOP/s/GPU): 726.5 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 8.766354E+00 | loss scale: 1.0 | grad norm: 5.073 | number of skipped iterations: 0 | number of nan iterations: 0 | + [2026-08-30 06:36:18] iteration 10/ 12 | consumed samples: 160 | elapsed time per iteration (ms): 460.5 | throughput per GPU (TFLOP/s/GPU): 723.9 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 8.629994E+00 | loss scale: 1.0 | grad norm: 4.476 | number of skipped iterations: 0 | number of nan iterations: 0 | + [2026-08-30 06:36:19] iteration 11/ 12 | consumed samples: 176 | elapsed time per iteration (ms): 462.1 | throughput per GPU (TFLOP/s/GPU): 721.4 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 8.554044E+00 | loss scale: 1.0 | grad norm: 3.970 | number of skipped iterations: 0 | number of nan iterations: 0 | + [2026-08-30 06:36:19] iteration 12/ 12 | consumed samples: 192 | elapsed time per iteration (ms): 459.2 | throughput per GPU (TFLOP/s/GPU): 725.9 | learning rate: 1.000000E-05 | global batch size: 16 | lm loss: 8.467697E+00 | loss scale: 1.0 | grad norm: 4.658 | number of skipped iterations: 0 | number of nan iterations: 0 | +[after training is done] datetime: 2026-08-30 06:36:19 +[rank0]:[W830 06:36:20.989750322 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()) diff --git a/tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/results.tsv b/tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/results.tsv index aeee0a8836a184b7a1950bb45d60683111ccdbfe..4c2dd95e13b0a3b7d7acfe08227a73e2cdf2053f 100644 --- a/tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/results.tsv +++ b/tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/results.tsv @@ -6,3 +6,8 @@ G2 echo one 1 4096 16 1.08203125 none 1 OK 394.21 392.20 120227 /workspace/tiera G2 echo one 1 4096 16 1.08203125 none 2 OK 394.20 392.20 120225 /workspace/tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/G2__echo__one__seq4096__gbs16__cap1p08203125/rep2 G2 echo one 1 4096 16 1.08203125 none 3 OK 394.28 392.20 120225 /workspace/tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/G2__echo__one__seq4096__gbs16__cap1p08203125/rep3 G3 echo one 1 4096 16 1.072265625 none 1 OK 412.23 410.50 116311 /workspace/tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/G3__echo__one__seq4096__gbs16__cap1p072265625/rep1 +G3 echo one 1 4096 16 1.072265625 none 2 OK 412.99 410.50 116423 /workspace/tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/G3__echo__one__seq4096__gbs16__cap1p072265625/rep2 +G3 echo one 1 4096 16 1.072265625 none 3 OK 412.83 410.50 116531 /workspace/tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/G3__echo__one__seq4096__gbs16__cap1p072265625/rep3 +G4 echo one 1 4096 16 1.095703125 none 1 OK 459.80 458.70 127009 /workspace/tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/G4__echo__one__seq4096__gbs16__cap1p095703125/rep1 +G4 echo one 1 4096 16 1.095703125 none 2 OK 461.14 458.70 127043 /workspace/tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/G4__echo__one__seq4096__gbs16__cap1p095703125/rep2 +G4 echo one 1 4096 16 1.095703125 none 3 OK 461.02 458.70 127009 /workspace/tiera-rerun-fixed-moonep/echo/a1/a1_oneslot_gbs16/G4__echo__one__seq4096__gbs16__cap1p095703125/rep3