Compare: 05_fp8 — MaxView

‹ 04_int8Case: 05_fp806_grad_accum ›

Metrics: `main (59e0f1759)` vs `test/pipeline-scan-nnx (5b21b66fc)`

Metric	main `59e0f1759`	test/pipeline-scan-nnx `5b21b66fc`	Diff (test/pipeline-scan-nnx − main)
Parameters	1.105 billion	1.105 billion	—
Final loss	6.1670	6.1670	0
TFLOP/s	85.760	85.889	+0.129
Tok/s	12926.7	12946.1	+19.448
Avg s/step	1.599	1.675	+0.076
Memory %	1.44	1.44	0
JAX	0.9.2	0.9.2	—

Diff = branch value − main value. Green = branch improved. Red = branch regressed.

Field	main	test/pipeline-scan-nnx
JAX	0.9.2	0.9.2
Jaxlib	0.9.2	0.9.2
Hardware	PJRT C API / TFRT TPU v6 lite	PJRT C API / TFRT TPU v6 lite
Devices	32	32

Memory Breakdown

Field	main	test/pipeline-scan-nnx
Total	1.5 GB	1.5 GB
Output	0.4 GB	0.4 GB
Temp	1.1 GB	1.1 GB
Argument	0.4 GB	0.4 GB

main

Device	Used (GB)	Total (GB)	%
TPU_24(process=6,(0,6,0,0))	0.45	31.25	1.4%
TPU_25(process=6,(1,6,0,0))	0.45	31.25	1.4%
TPU_28(process=6,(0,7,0,0))	0.45	31.25	1.4%
TPU_29(process=6,(1,7,0,0))	0.45	31.25	1.4%

test/pipeline-scan-nnx

Device	Used (GB)	Total (GB)	%
TPU_16(process=4,(0,4,0,0))	0.45	31.25	1.4%
TPU_17(process=4,(1,4,0,0))	0.45	31.25	1.4%
TPU_20(process=4,(0,5,0,0))	0.45	31.25	1.4%
TPU_21(process=4,(1,5,0,0))	0.45	31.25	1.4%

main TFLOP/s

test/pipeline-scan-nnx TFLOP/s

Step	main Loss	test/pipeline-scan-nnx Loss	main TFLOP/s	test/pipeline-scan-nnx TFLOP/s	TFLOP/s Δ
0	10.874	10.874	0.888	0.889	+0.0010
1	10.874	10.874	8.750	8.651	-0.0990
2	10.816	10.816	117.438	380.240	+262.8020
3	10.431	10.431	33.176	33.248	+0.0720
4	9.992	9.992	82.529	82.640	+0.1110
5	9.549	9.549	85.700	85.948	+0.2480
6	9.111	9.111	42.599	42.749	+0.1500
7	8.685	8.685	0.570	0.576	+0.0060
8	8.281	8.281	83.138	83.033	-0.1050
9	7.908	7.908	85.888	85.801	-0.0870
10	7.568	7.568	85.574	85.599	+0.0250
11	7.288	7.288	8.118	3.836	-4.2820
12	7.032	7.032	24.568	2176.727	+2152.1590
13	6.815	6.815	83.317	23.641	-59.6760
14	6.635	6.635	85.892	83.156	-2.7360
15	6.492	6.492	85.847	85.937	+0.0900
16	6.380	6.380	85.651	85.829	+0.1780
17	6.293	6.293	85.824	85.790	-0.0340
18	6.223	6.223	85.931	85.873	-0.0580
19	6.167	6.167	85.760	85.889	+0.1290

main · 59e0f1759 · main_20260425_071456 · full log

XPK Start: Sat Apr 25 07:24:28 UTC 2026
PyTorch was not found. Models won't be available and only tokenizers, configuration and file/data utilities can be used.
`rope_parameters`'s factor field must be a float >= 1, got 40
`rope_parameters`'s beta_fast field must be a float, got 32
`rope_parameters`'s beta_slow field must be a float, got 1
DeepseekV32Config got `key=rope_scaling` in kwargs but hasn't set it as attribute. For RoPE standardization you need to set `self.rope_parameters` in model's config. 
2026-04-25 07:24:53.519576: E external/local_xla/xla/stream_executor/cuda/cuda_platform.cc:51] failed call to cuInit: INTERNAL: CUDA error: Failed call to cuInit: UNKNOWN ERROR (303)
I0425 07:24:53.726617 136390415513408 max_utils.py:273] Attempting to initialize the jax distributed system...
I0425 07:25:02.768357 136390415513408 distributed.py:149] Starting JAX distributed service on [::]:8482
I0425 07:25:02.770697 136390415513408 distributed.py:172] Connecting to JAX distributed service on mt-05-fp8-peg96-slice-job-0-0.mt-05-fp8-peg96:8482
I0425 07:25:04.453753 136390415513408 max_utils.py:284] Jax distributed system initialized!
I0425 07:25:10.762574 136390415513408 max_utils.py:800] System Information: Jax Version: 0.9.2
I0425 07:25:10.762679 136390415513408 max_utils.py:801] System Information: Jaxlib Version: 0.9.2
I0425 07:25:10.762721 136390415513408 max_utils.py:802] System Information: Jax Backend: PJRT C API
TFRT TPU v6 lite
Built on Mar 4 2026 11:32:08 (1772652728) cl/878335365
I0425 07:25:10.762756 136390415513408 train_utils.py:361] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing.
I0425 07:25:11.456088 136390415513408 maxtext_utils.py:1604] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1)
I0425 07:25:11.456395 136390415513408 checkpointing.py:677] Setting up checkpoint logger...
I0425 07:25:11.456452 136390415513408 checkpointing.py:233] Creating checkpoint manager with ocdbt=True and zarr3=True
I0425 07:25:11.456497 136390415513408 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None
I0425 07:25:11.456837 136390415513408 base_pytree_checkpoint_handler.py:441] Created BasePyTreeCheckpointHandler: use_ocdbt=True, use_zarr3=True, pytree_metadata_options=PyTreeMetadataOptions(support_rich_types=False), array_metadata_store=<orbax.checkpoint._src.metadata.array_metadata_store.Store object at 0x7c0b26b3f650>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB)
I0425 07:25:14.349413 136390415513408 checkpointing.py:265] Enabling policy for fixed interval checkpointing.
I0425 07:25:14.349650 136390415513408 checkpoint_manager.py:708] [process=6][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=('items',), item_handlers={'items': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7bf7087c08c0>}, handler_registry=None
I0425 07:25:14.349889 136390415513408 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7bf7087c08c0>` for item "items" and save args `<class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeSaveArgs'>` and restore args `<class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>` to `_handler_registry`.
I0425 07:25:14.349937 136390415513408 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7bf7087be630>` for item "metrics" and save args `<class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>` and restore args `<class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>` to `_handler_registry`.
I0425 07:25:14.349973 136390415513408 composite_checkpoint_handler.py:505] Initialized registry DefaultCheckpointHandlerRegistry({('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeSaveArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7bf7087c08c0>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7bf7087c08c0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7bf7087be630>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7bf7087be630>}).
I0425 07:25:14.350314 136390415513408 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34
I0425 07:25:14.350384 136390415513408 async_checkpointer.py:192] [process=6][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7bf708195940> timeout: 1200 secs and primary_host=0 for async checkpoint writes
I0425 07:25:15.091848 136390415513408 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_main_20260425_071456/linen_xpk_main_20260425_071456_05_fp8/checkpoints
I0425 07:25:15.097051 136390415513408 checkpoint_manager.py:929] [process=6][thread=MainThread] CheckpointManager created,  primary_host=0, CheckpointManagerOptions=CheckpointManagerOptions(save_interval_steps=1, max_to_keep=None, keep_time_interval=None, keep_period=None, should_keep_fn=None, best_fn=None, best_mode='max', keep_checkpoints_without_metrics=True, step_prefix=None, step_format_fixed_length=None, step_name_format=None, create=True, cleanup_tmp_directories=False, save_on_steps=frozenset(), single_host_load_and_broadcast=False, todelete_subdir=None, todelete_full_path=None, enable_background_delete=False, read_only=False, enable_async_checkpointing=True, async_options=None, multiprocessing_options=MultiprocessingOptions(primary_host=0, active_processes=None, barrier_sync_key_prefix=None), should_save_fn=None, file_options=FileOptions(path_permission_mode=None), save_root_metadata=True, temporary_path_class=None, save_decision_policy=FixedIntervalPolicy(interval=10), preservation_policy=LatestN(n=None), prevent_write_metrics=False, enable_should_save_is_saving_in_progress_check=True, enable_per_process_directory_creation=False, lightweight_initialize=False), root_directory=gs://lance-maxtext/linen_ckpt_xpk_main_20260425_071456/linen_xpk_main_20260425_071456_05_fp8/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7bf7087c2780>
I0425 07:25:15.097178 136390415513408 checkpointing.py:301] Checkpoint manager created!
I0425 07:25:16.181551 136390415513408 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed').
I0425 07:25:16.181676 136390415513408 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0425 07:25:16.568191 136390415513408 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch_attn', 'activation_length_attn', 'activation_embed_attn').
I0425 07:25:16.568288 136390415513408 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0425 07:25:16.584976 136390415513408 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch_attn', 'activation_length_attn', 'activation_embed_attn').
I0425 07:25:16.585039 136390415513408 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0425 07:25:16.638058 136390415513408 attentions.py:1154] attentions/query Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_length_attn', 'activation_kv_heads', 'activation_kv_head_dim').
I0425 07:25:16.638156 136390415513408 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0425 07:25:16.654814 136390415513408 attentions.py:1155] attentions/key Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_length_attn', 'activation_kv_heads', 'activation_kv_head_dim').
I0425 07:25:16.654877 136390415513408 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0425 07:25:16.671667 136390415513408 attentions.py:1156] attentions/value Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_length_attn', 'activation_kv_heads', 'activation_kv_head_dim').
I0425 07:25:16.671730 136390415513408 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0425 07:25:16.696830 136390415513408 attentions.py:1198] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch_attn', 'activation_length_attn', 'activation_heads', 'activation_kv').
I0425 07:25:16.696899 136390415513408 attentions.py:1198] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0425 07:25:16.749223 136390415513408 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp').
I0425 07:25:16.749303 136390415513408 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None).
I0425 07:25:17.182524 136390415513408 checkpointing.py:577] checkpoint manager exists so trying to load this run's existing checkpoint
I0425 07:25:17.182638 136390415513408 checkpointing.py:665] No existing checkpoints found, not restoring checkpoint.
fsdp: 32
I0425 07:25:19.117067 136390415513408 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/mlp/wi_0/Fp8DirectDotGeneralOp_0/input_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0425 07:25:19.117223 136390415513408 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/mlp/wi_0/Fp8DirectDotGeneralOp_0/input_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0425 07:25:19.117297 136390415513408 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/mlp/wi_0/Fp8DirectDotGeneralOp_0/kernel_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0425 07:25:19.117358 136390415513408 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/mlp/wi_0/Fp8DirectDotGeneralOp_0/kernel_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0425 07:25:19.117416 136390415513408 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/mlp/wi_0/Fp8DirectDotGeneralOp_0/output_grad_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0425 07:25:19.117473 136390415513408 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/mlp/wi_0/Fp8DirectDotGeneralOp_0/output_grad_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0425 07:25:19.117527 136390415513408 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/mlp/wi_1/Fp8DirectDotGeneralOp_0/input_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0425 07:25:19.117578 136390415513408 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/mlp/wi_1/Fp8DirectDotGeneralOp_0/input_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0425 07:25:19.117627 136390415513408 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/mlp/wi_1/Fp8DirectDotGeneralOp_0/kernel_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0425 07:25:19.117674 136390415513408 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/mlp/wi_1/Fp8DirectDotGeneralOp_0/kernel_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0425 07:25:19.117722 136390415513408 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/mlp/wi_1/Fp8DirectDotGeneralOp_0/output_grad_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0425 07:25:19.117769 136390415513408 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/mlp/wi_1/Fp8DirectDotGeneralOp_0/output_grad_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0425 07:25:19.117818 136390415513408 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/mlp/wo/Fp8DirectDotGeneralOp_0/input_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0425 07:25:19.117864 136390415513408 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/mlp/wo/Fp8DirectDotGeneralOp_0/input_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0425 07:25:19.117910 136390415513408 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/mlp/wo/Fp8DirectDotGeneralOp_0/kernel_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0425 07:25:19.117955 136390415513408 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/mlp/wo/Fp8DirectDotGeneralOp_0/kernel_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0425 07:25:19.118000 136390415513408 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/mlp/wo/Fp8DirectDotGeneralOp_0/output_grad_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0425 07:25:19.118045 136390415513408 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/mlp/wo/Fp8DirectDotGeneralOp_0/output_grad_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0425 07:25:19.118091 136390415513408 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/self_attention/key/Fp8DirectDotGeneralOp_0/input_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0425 07:25:19.118155 136390415513408 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/self_attention/key/Fp8DirectDotGeneralOp_0/input_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0425 07:25:19.118201 136390415513408 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/self_attention/key/Fp8DirectDotGeneralOp_0/kernel_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0425 07:25:19.118244 136390415513408 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/self_attention/key/Fp8DirectDotGeneralOp_0/kernel_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0425 07:25:19.118293 136390415513408 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/self_attention/key/Fp8DirectDotGeneralOp_0/output_grad_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0425 07:25:19.118336 136390415513408 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/self_attention/key/Fp8DirectDotGeneralOp_0/output_grad_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0425 07:25:19.118380 136390415513408 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/self_attention/out/Fp8DirectDotGeneralOp_0/input_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0425 07:25:19.118422 136390415513408 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/self_attention/out/Fp8DirectDotGeneralOp_0/input_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0425 07:25:19.118465 136390415513408 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/self_attention/out/Fp8DirectDotGeneralOp_0/kernel_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0425 07:25:19.118509 136390415513408 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/self_attention/out/Fp8DirectDotGeneralOp_0/kernel_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0425 07:25:19.118553 136390415513408 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/self_attention/out/Fp8DirectDotGeneralOp_0/output_grad_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0425 07:25:19.118596 136390415513408 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/self_attention/out/Fp8DirectDotGeneralOp_0/output_grad_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0425 07:25:19.118638 136390415513408 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/self_attention/query/Fp8DirectDotGeneralOp_0/input_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0425 07:25:19.118682 136390415513408 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/self_attention/query/Fp8DirectDotGeneralOp_0/input_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0425 07:25:19.118725 136390415513408 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/self_attention/query/Fp8DirectDotGeneralOp_0/kernel_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0425 07:25:19.118774 136390415513408 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/self_attention/query/Fp8DirectDotGeneralOp_0/kernel_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0425 07:25:19.118816 136390415513408 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/self_attention/query/Fp8DirectDotGeneralOp_0/output_grad_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0425 07:25:19.118845 136390415513408 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/self_attention/query/Fp8DirectDotGeneralOp_0/output_grad_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0425 07:25:19.118873 136390415513408 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/self_attention/value/Fp8DirectDotGeneralOp_0/input_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0425 07:25:19.118899 136390415513408 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/self_attention/value/Fp8DirectDotGeneralOp_0/input_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0425 07:25:19.118926 136390415513408 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/self_attention/value/Fp8DirectDotGeneralOp_0/kernel_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0425 07:25:19.118953 136390415513408 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/self_attention/value/Fp8DirectDotGeneralOp_0/kernel_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0425 07:25:19.118980 136390415513408 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/self_attention/value/Fp8DirectDotGeneralOp_0/output_grad_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0425 07:25:19.119006 136390415513408 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/self_attention/value/Fp8DirectDotGeneralOp_0/output_grad_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0425 07:25:19.119058 136390415513408 maxtext_utils.py:1707]  params/params/decoder/decoder_norm/scale
    Shape:     float32[2048]
    Logical:   P('norm',)
    Physical:  (None,)
I0425 07:25:19.119144 136390415513408 maxtext_utils.py:1707]  params/params/decoder/layers/mlp/wi_0/kernel
    Shape:     float32[2048,16,7168]
    Logical:   P('embed', 'layers', 'mlp')
    Physical:  ('fsdp', None, None)
I0425 07:25:19.119186 136390415513408 maxtext_utils.py:1707]  params/params/decoder/layers/mlp/wi_1/kernel
    Shape:     float32[2048,16,7168]
    Logical:   P('embed', 'layers', 'mlp')
    Physical:  ('fsdp', None, None)
I0425 07:25:19.119235 136390415513408 maxtext_utils.py:1707]  params/params/decoder/layers/mlp/wo/kernel
    Shape:     float32[7168,16,2048]
    Logical:   P('mlp', 'layers', 'embed')
    Physical:  (None, None, 'fsdp')
I0425 07:25:19.119287 136390415513408 maxtext_utils.py:1707]  params/params/decoder/layers/post_self_attention_layer_norm/scale
    Shape:     float32[2048,16]
    Logical:   P('norm', 'layers')
    Physical:  (None, None)
I0425 07:25:19.119320 136390415513408 maxtext_utils.py:1707]  params/params/decoder/layers/pre_self_attention_layer_norm/scale
    Shape:     float32[2048,16]
    Logical:   P('norm', 'layers')
    Physical:  (None, None)
I0425 07:25:19.119366 136390415513408 maxtext_utils.py:1707]  params/params/decoder/layers/self_attention/key/kernel
    Shape:     float32[2048,16,16,128]
    Logical:   P('embed', 'layers', 'kv_heads', 'kv_head_dim')
    Physical:  ('fsdp', None, None, None)
I0425 07:25:19.119413 136390415513408 maxtext_utils.py:1707]  params/params/decoder/layers/self_attention/out/kernel
    Shape:     float32[16,16,128,2048]
    Logical:   P('heads', 'layers', 'kv', 'embed')
    Physical:  (None, None, None, 'fsdp')
I0425 07:25:19.119448 136390415513408 maxtext_utils.py:1707]  params/params/decoder/layers/self_attention/query/kernel
    Shape:     float32[2048,16,16,128]
    Logical:   P('embed', 'layers', 'q_heads', 'kv')
    Physical:  ('fsdp', None, None, None)
I0425 07:25:19.119481 136390415513408 maxtext_utils.py:1707]  params/params/decoder/layers/self_attention/value/kernel
    Shape:     float32[2048,16,16,128]
    Logical:   P('embed', 'layers', 'kv_heads', 'kv_head_dim')
    Physical:  ('fsdp', None, None, None)
I0425 07:25:19.119525 136390415513408 maxtext_utils.py:1707]  params/params/decoder/logits_dense/kernel
    Shape:     float32[2048,32000]
    Logical:   P('embed_vocab', 'vocab')
    Physical:  ('fsdp', None)
I0425 07:25:19.119568 136390415513408 maxtext_utils.py:1707]  params/params/token_embedder/embedding
    Shape:     float32[32000,2048]
    Logical:   P('vocab', 'embed_vocab')
    Physical:  (None, 'fsdp')

I0425 07:25:20.752288 136390415513408 train.py:155] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0425 07:25:20.752384 136390415513408 train.py:155] train/xent Physical: float32[32,2048]............................................ ('fsdp', None).
I0425 07:25:20.768080 136390415513408 train.py:162] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0425 07:25:20.768154 136390415513408 train.py:162] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None).
I0425 07:25:34.605504 136390415513408 max_utils.py:791] Total memory size: 1.5 GB, Output size: 0.4 GB, Temp size: 1.1 GB, Argument size: 0.4 GB, Host temp size: 0.0 GB.
I0425 07:25:34.606443 136390415513408 metric_logger.py:301] number parameters: 1.105 billion
I0425 07:25:49.899983 136390415513408 checkpointing.py:772] Waiting for step 0 to finish before checkpoint...
I0425 07:25:50.069348 136390415513408 checkpointing.py:776] Waited 0.16933631896972656 seconds for step 0 to finish before starting checkpointing.
I0425 07:25:50.071800 136390415513408 checkpoint_manager.py:2009] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0425 07:25:50.073707 136390415513408 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 0
I0425 07:25:50.075090 136390415513408 event_tracking.py:70] [process=6] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260425_071456/linen_xpk_main_20260425_071456_05_fp8/checkpoints/0.
I0425 07:25:50.723269 136390415513408 signaling_client.py:364] Using JaxDistributedSignalingClient
I0425 07:25:50.724310 136390415513408 jax_array_handlers.py:360] Scheduling D2H of 81 prioritized jax.Array.
I0425 07:25:50.724373 136390415513408 replica_slices.py:424] Transferring arrays to host memory with options: use_replica_parallel=True, min_slice_bytes_for_replica_parallel=None, max_replicas_for_replica_parallel=None, enable_pinned_host_transfer=False
I0425 07:25:51.134683 136390415513408 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.411533s
I0425 07:25:51.134854 136390415513408 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 3.675 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.4198460578918457 s) (per-host)
I0425 07:25:51.134907 136390415513408 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.419909s (batch_requests_ready=0.003619s, total_serialization_initiated=0.416219s, others=0.000072s)
I0425 07:25:51.135011 136390415513408 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.424046s (all_items=0.000018s, per_item={'items': '0.00001764'}, temp_paths=0.424029)
I0425 07:25:51.135826 136390415513408 event_tracking.py:125] [process=6] [async] Finished blocking save in 1.06 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260425_071456/linen_xpk_main_20260425_071456_05_fp8/checkpoints/0.
I0425 07:25:51.136180 136260723144448 async_checkpointer.py:76] [process=6][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-25 07:45:51.136140
I0425 07:25:51.139879 136390415513408 checkpoint_manager.py:1560] [process=6][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize
I0425 07:25:51.140120 136260665087744 async_checkpointer.py:280] [process=6][thread=save_finalize] Waiting for background save thread=async_save.
I0425 07:25:51.140239 136390415513408 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260425_071456/linen_xpk_main_20260425_071456_05_fp8/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1777101950.0717812, 'wait_for_prev_duration_secs': 6.246566772460938e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1777101950.0737474, 'checkpointer_blocking_duration_secs': 1.0625362396240234, 'get_old_steps_start_time': 1777101951.1363106, 'get_old_steps_duration_secs': 2.9802322387695312e-05, 'checkpoint_manager_blocking_start_time': 1777101950.0699418, 'checkpoint_manager_blocking_duration_secs': 1.0702595710754395}
I0425 07:25:51.140352 136390415513408 checkpointing.py:408] Started an asynchronous checkpoint save for step 0
I0425 07:25:51.140402 136390415513408 max_utils.py:750] 
Memstats: After params initialized:
I0425 07:25:51.140450 136390415513408 max_utils.py:756] 	Using (GB) 0.45 / 31.25 (1.440000%) on TPU_24(process=6,(0,6,0,0))
I0425 07:25:51.140489 136390415513408 max_utils.py:756] 	Using (GB) 0.45 / 31.25 (1.440000%) on TPU_25(process=6,(1,6,0,0))
I0425 07:25:51.140517 136390415513408 max_utils.py:756] 	Using (GB) 0.45 / 31.25 (1.440000%) on TPU_28(process=6,(0,7,0,0))
I0425 07:25:51.140543 136390415513408 max_utils.py:756] 	Using (GB) 0.45 / 31.25 (1.440000%) on TPU_29(process=6,(1,7,0,0))
I0425 07:25:51.454913 136390415513408 metric_logger.py:196] completed step: 0, seconds: 15.293, TFLOP/s/device: 0.888, Tokens/s/device: 133.914, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52776.805
I0425 07:25:51.727996 136390415513408 metric_logger.py:196] completed step: 1, seconds: 1.553, TFLOP/s/device: 8.750, Tokens/s/device: 1318.864, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52761.840
I0425 07:25:52.137902 136390415513408 metric_logger.py:196] completed step: 2, seconds: 0.116, TFLOP/s/device: 117.438, Tokens/s/device: 17701.563, total_weights: 65536, loss: 10.816, lm_loss: 10.816, perplexity: 49832.398
I0425 07:25:52.296509 136390415513408 metric_logger.py:196] completed step: 3, seconds: 0.410, TFLOP/s/device: 33.176, Tokens/s/device: 5000.659, total_weights: 65536, loss: 10.431, lm_loss: 10.431, perplexity: 33901.820
I0425 07:25:52.614904 136390415513408 metric_logger.py:196] completed step: 4, seconds: 0.165, TFLOP/s/device: 82.529, Tokens/s/device: 12439.715, total_weights: 65536, loss: 9.992, lm_loss: 9.992, perplexity: 21847.639
I0425 07:25:52.621479 136390415513408 metric_logger.py:196] completed step: 5, seconds: 0.159, TFLOP/s/device: 85.700, Tokens/s/device: 12917.631, total_weights: 65536, loss: 9.549, lm_loss: 9.549, perplexity: 14035.006
I0425 07:25:54.182582    2815 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com
I0425 07:25:56.225918 136260706359040 array_metadata_store.py:203] [process=6][thread=array_type_handler] Wrote 81 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260425_071456/linen_xpk_main_20260425_071456_05_fp8/checkpoints/0/items/array_metadatas/process_6
I0425 07:26:11.245802 136260723144448 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 76.946 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 20.530755281448364 s) (per-host)
I0425 07:26:11.245925 136260723144448 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 20.109086s.
I0425 07:26:16.625491 136390415513408 metric_logger.py:196] completed step: 6, seconds: 0.319, TFLOP/s/device: 42.599, Tokens/s/device: 6421.009, total_weights: 65536, loss: 9.111, lm_loss: 9.111, perplexity: 9058.688
I0425 07:26:16.783863 136390415513408 metric_logger.py:196] completed step: 7, seconds: 23.847, TFLOP/s/device: 0.570, Tokens/s/device: 85.882, total_weights: 65536, loss: 8.685, lm_loss: 8.685, perplexity: 5914.720
I0425 07:26:16.942298 136390415513408 metric_logger.py:196] completed step: 8, seconds: 0.163, TFLOP/s/device: 83.138, Tokens/s/device: 12531.512, total_weights: 65536, loss: 8.281, lm_loss: 8.281, perplexity: 3950.010
I0425 07:26:16.947610 136390415513408 checkpointing.py:772] Waiting for step 10 to finish before checkpoint...
I0425 07:26:17.259315 136390415513408 checkpointing.py:776] Waited 0.31166744232177734 seconds for step 10 to finish before starting checkpointing.
I0425 07:26:17.261840 136390415513408 checkpoint_manager.py:2020] [process=6][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0425 07:26:18.270532 136260723144448 async_checkpointer.py:160] [process=6][thread=async_save] Background save thread done. Time taken: 27.133677s.
I0425 07:26:18.270800 136260665087744 async_checkpointer.py:288] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save.
I0425 07:26:18.270868 136260665087744 async_checkpointer.py:298] [process=6][thread=save_finalize] No errors found in background save thread=async_save.
I0425 07:26:18.270924 136260665087744 checkpoint_manager.py:2137] [process=6][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts...
I0425 07:26:18.272624 136260665087744 checkpoint_manager.py:2146] [process=6][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts.
I0425 07:26:18.272781 136390415513408 checkpoint_manager.py:2032] [process=6][thread=MainThread][step=0][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=0.
W0425 07:26:18.272913 136390415513408 checkpoint_manager.py:1452] Waiting for previous save to complete took 1.011072 seconds. If this number is high, consider checkpointing less frequently.
I0425 07:26:18.275038 136390415513408 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 10
I0425 07:26:18.277140 136390415513408 event_tracking.py:70] [process=6] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260425_071456/linen_xpk_main_20260425_071456_05_fp8/checkpoints/10.
I0425 07:26:18.559673 136390415513408 jax_array_handlers.py:360] Scheduling D2H of 81 prioritized jax.Array.
I0425 07:26:18.559770 136390415513408 replica_slices.py:424] Transferring arrays to host memory with options: use_replica_parallel=True, min_slice_bytes_for_replica_parallel=None, max_replicas_for_replica_parallel=None, enable_pinned_host_transfer=False
I0425 07:26:18.612209 136390415513408 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.053529s
I0425 07:26:18.612378 136390415513408 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 25.885 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.059600114822387695 s) (per-host)
I0425 07:26:18.612428 136390415513408 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.059660s (batch_requests_ready=0.003022s, total_serialization_initiated=0.056570s, others=0.000068s)
I0425 07:26:18.612531 136390415513408 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.064102s (all_items=0.000015s, per_item={'items': '0.00001454'}, temp_paths=0.064088)
I0425 07:26:18.613256 136390415513408 event_tracking.py:125] [process=6] [async] Finished blocking save in 0.34 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260425_071456/linen_xpk_main_20260425_071456_05_fp8/checkpoints/10.
I0425 07:26:18.613604 136260665087744 async_checkpointer.py:76] [process=6][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-25 07:46:18.613567
I0425 07:26:18.615598 136390415513408 checkpoint_manager.py:1560] [process=6][thread=MainThread][step=10] Starting CheckpointManager Save Finalize thread=save_finalize
I0425 07:26:18.615838 136259615831808 async_checkpointer.py:280] [process=6][thread=save_finalize] Waiting for background save thread=async_save.
I0425 07:26:18.615995 136390415513408 standard_logger.py:34] {'step': 10, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260425_071456/linen_xpk_main_20260425_071456_05_fp8/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1777101977.2618113, 'wait_for_prev_duration_secs': 1.0110721588134766, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1777101978.275079, 'checkpointer_blocking_duration_secs': 0.3386685848236084, 'get_old_steps_start_time': 1777101978.613771, 'get_old_steps_duration_secs': 3.361701965332031e-05, 'checkpoint_manager_blocking_start_time': 1777101977.2600794, 'checkpoint_manager_blocking_duration_secs': 1.355881690979004}
I0425 07:26:18.616124 136390415513408 checkpointing.py:408] Started an asynchronous checkpoint save for step 10
I0425 07:26:18.616721 136390415513408 metric_logger.py:196] completed step: 9, seconds: 0.158, TFLOP/s/device: 85.888, Tokens/s/device: 12946.048, total_weights: 65536, loss: 7.908, lm_loss: 7.908, perplexity: 2720.116
I0425 07:26:18.623249 136390415513408 metric_logger.py:196] completed step: 10, seconds: 0.159, TFLOP/s/device: 85.574, Tokens/s/device: 12898.675, total_weights: 65536, loss: 7.568, lm_loss: 7.568, perplexity: 1936.198
I0425 07:26:19.332669 136390415513408 metric_logger.py:196] completed step: 11, seconds: 1.674, TFLOP/s/device: 8.118, Tokens/s/device: 1223.619, total_weights: 65536, loss: 7.288, lm_loss: 7.288, perplexity: 1463.272
I0425 07:26:19.490967 136390415513408 metric_logger.py:196] completed step: 12, seconds: 0.553, TFLOP/s/device: 24.568, Tokens/s/device: 3703.195, total_weights: 65536, loss: 7.032, lm_loss: 7.032, perplexity: 1132.061
I0425 07:26:19.649172 136390415513408 metric_logger.py:196] completed step: 13, seconds: 0.163, TFLOP/s/device: 83.317, Tokens/s/device: 12558.408, total_weights: 65536, loss: 6.815, lm_loss: 6.815, perplexity: 911.031
I0425 07:26:19.807804 136390415513408 metric_logger.py:196] completed step: 14, seconds: 0.158, TFLOP/s/device: 85.892, Tokens/s/device: 12946.539, total_weights: 65536, loss: 6.635, lm_loss: 6.635, perplexity: 761.080
I0425 07:26:19.966045 136390415513408 metric_logger.py:196] completed step: 15, seconds: 0.158, TFLOP/s/device: 85.847, Tokens/s/device: 12939.749, total_weights: 65536, loss: 6.492, lm_loss: 6.492, perplexity: 659.915
I0425 07:26:20.124291 136390415513408 metric_logger.py:196] completed step: 16, seconds: 0.159, TFLOP/s/device: 85.651, Tokens/s/device: 12910.302, total_weights: 65536, loss: 6.380, lm_loss: 6.380, perplexity: 589.948
I0425 07:26:20.282649 136390415513408 metric_logger.py:196] completed step: 17, seconds: 0.158, TFLOP/s/device: 85.824, Tokens/s/device: 12936.316, total_weights: 65536, loss: 6.293, lm_loss: 6.293, perplexity: 540.941
I0425 07:26:20.440768 136390415513408 metric_logger.py:196] completed step: 18, seconds: 0.158, TFLOP/s/device: 85.931, Tokens/s/device: 12952.434, total_weights: 65536, loss: 6.223, lm_loss: 6.223, perplexity: 504.108
I0425 07:26:20.598376 136390415513408 checkpointing.py:772] Waiting for step 19 to finish before checkpoint...
I0425 07:26:20.599289 136390415513408 checkpointing.py:776] Waited 0.0009329319000244141 seconds for step 19 to finish before starting checkpointing.
I0425 07:26:20.601536 136390415513408 checkpoint_manager.py:2020] [process=6][thread=MainThread][step=10][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0425 07:26:24.178619 136259641009920 array_metadata_store.py:203] [process=6][thread=array_type_handler] Wrote 81 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260425_071456/linen_xpk_main_20260425_071456_05_fp8/checkpoints/10/items/array_metadatas/process_6
I0425 07:27:01.180286 136260665087744 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 37.060 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 42.62746524810791 s) (per-host)
I0425 07:27:01.180416 136260665087744 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 42.566700s.
I0425 07:27:08.500976 136260665087744 async_checkpointer.py:160] [process=6][thread=async_save] Background save thread done. Time taken: 49.887244s.
I0425 07:27:08.501283 136259615831808 async_checkpointer.py:288] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save.
I0425 07:27:08.501411 136259615831808 async_checkpointer.py:298] [process=6][thread=save_finalize] No errors found in background save thread=async_save.
I0425 07:27:08.501469 136259615831808 checkpoint_manager.py:2137] [process=6][thread=save_finalize][step=10] CheckpointManager Save Finalize is syncing with other hosts...
I0425 07:27:08.503132 136259615831808 checkpoint_manager.py:2146] [process=6][thread=save_finalize][step=10] CheckpointManager Save Finalize is done on all hosts.
I0425 07:27:08.503320 136390415513408 checkpoint_manager.py:2032] [process=6][thread=MainThread][step=10][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=10.
W0425 07:27:08.503450 136390415513408 checkpoint_manager.py:1452] Waiting for previous save to complete took 47.901927 seconds. If this number is high, consider checkpointing less frequently.
I0425 07:27:08.505011 136390415513408 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 19
I0425 07:27:08.507140 136390415513408 event_tracking.py:70] [process=6] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260425_071456/linen_xpk_main_20260425_071456_05_fp8/checkpoints/19.
I0425 07:27:09.214677 136390415513408 jax_array_handlers.py:360] Scheduling D2H of 81 prioritized jax.Array.
I0425 07:27:09.214774 136390415513408 replica_slices.py:424] Transferring arrays to host memory with options: use_replica_parallel=True, min_slice_bytes_for_replica_parallel=None, max_replicas_for_replica_parallel=None, enable_pinned_host_transfer=False
I0425 07:27:09.268934 136390415513408 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.055197s
I0425 07:27:09.269122 136390415513408 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 25.192 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.06123828887939453 s) (per-host)
I0425 07:27:09.269171 136390415513408 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.061321s (batch_requests_ready=0.002976s, total_serialization_initiated=0.058257s, others=0.000088s)
I0425 07:27:09.269269 136390415513408 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.065483s (all_items=0.000010s, per_item={'items': '0.00000954'}, temp_paths=0.065474)
I0425 07:27:09.269985 136390415513408 event_tracking.py:125] [process=6] [async] Finished blocking save in 0.76 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260425_071456/linen_xpk_main_20260425_071456_05_fp8/checkpoints/19.
I0425 07:27:09.270366 136259615831808 async_checkpointer.py:76] [process=6][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-25 07:47:09.270332
I0425 07:27:09.272365 136390415513408 checkpoint_manager.py:1560] [process=6][thread=MainThread][step=19] Starting CheckpointManager Save Finalize thread=save_finalize
I0425 07:27:09.272573 136259065964288 async_checkpointer.py:280] [process=6][thread=save_finalize] Waiting for background save thread=async_save.
I0425 07:27:09.272686 136390415513408 standard_logger.py:34] {'step': 19, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260425_071456/linen_xpk_main_20260425_071456_05_fp8/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1777101980.6014988, 'wait_for_prev_duration_secs': 47.90192747116089, 'time_between_consecutive_saves_sec': 2.3288350105285645, 'checkpointer_blocking_start_time': 1777102028.5050497, 'checkpointer_blocking_duration_secs': 0.7654757499694824, 'get_old_steps_start_time': 1777102029.2705438, 'get_old_steps_duration_secs': 2.5987625122070312e-05, 'checkpoint_manager_blocking_start_time': 1777101980.5995584, 'checkpoint_manager_blocking_duration_secs': 48.67309784889221}
I0425 07:27:09.272795 136390415513408 checkpointing.py:408] Started an asynchronous checkpoint save for step 19
I0425 07:27:09.272836 136390415513408 checkpoint_manager.py:2020] [process=6][thread=MainThread][step=19][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0425 07:27:15.403674 136256925325056 array_metadata_store.py:203] [process=6][thread=array_type_handler] Wrote 81 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260425_071456/linen_xpk_main_20260425_071456_05_fp8/checkpoints/19/items/array_metadatas/process_6
I0425 07:27:52.232533 136259615831808 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 36.718 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 43.02461338043213 s) (per-host)
I0425 07:27:52.232652 136259615831808 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 42.962170s.
I0425 07:28:00.802175 136259615831808 async_checkpointer.py:160] [process=6][thread=async_save] Background save thread done. Time taken: 51.531678s.
I0425 07:28:00.802396 136259065964288 async_checkpointer.py:288] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save.
I0425 07:28:00.802463 136259065964288 async_checkpointer.py:298] [process=6][thread=save_finalize] No errors found in background save thread=async_save.
I0425 07:28:00.802509 136259065964288 checkpoint_manager.py:2137] [process=6][thread=save_finalize][step=19] CheckpointManager Save Finalize is syncing with other hosts...
I0425 07:28:00.804114 136259065964288 checkpoint_manager.py:2146] [process=6][thread=save_finalize][step=19] CheckpointManager Save Finalize is done on all hosts.
I0425 07:28:00.804279 136390415513408 checkpoint_manager.py:2032] [process=6][thread=MainThread][step=19][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=19.
I0425 07:28:00.804436 136390415513408 checkpoint_manager.py:2009] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0425 07:28:00.805332 136390415513408 metric_logger.py:196] completed step: 19, seconds: 0.158, TFLOP/s/device: 85.760, Tokens/s/device: 12926.681, total_weights: 65536, loss: 6.167, lm_loss: 6.167, perplexity: 476.947
Per train step:
 Total TFLOPs: 13.59 
 split as 93.93% learnable weight flops and 6.07% attention flops
XPK End: Sat Apr 25 07:28:13 UTC 2026
EXIT_CODE=0

test/pipeline-scan-nnx · 5b21b66fc · test_pipeline_scan_nnx_20260425_201227 · full log

XPK Start: Sat Apr 25 20:21:39 UTC 2026
PyTorch was not found. Models won't be available and only tokenizers, configuration and file/data utilities can be used.
`rope_parameters`'s factor field must be a float >= 1, got 40
`rope_parameters`'s beta_fast field must be a float, got 32
`rope_parameters`'s beta_slow field must be a float, got 1
DeepseekV32Config got `key=rope_scaling` in kwargs but hasn't set it as attribute. For RoPE standardization you need to set `self.rope_parameters` in model's config. 
2026-04-25 20:22:04.474612: E external/local_xla/xla/stream_executor/cuda/cuda_platform.cc:51] failed call to cuInit: INTERNAL: CUDA error: Failed call to cuInit: UNKNOWN ERROR (303)
I0425 20:22:04.683605 136344560469824 max_utils.py:273] Attempting to initialize the jax distributed system...
I0425 20:22:13.725615 136344560469824 distributed.py:149] Starting JAX distributed service on [::]:8482
I0425 20:22:13.727932 136344560469824 distributed.py:172] Connecting to JAX distributed service on mt-05-fp8-prd8l-slice-job-0-0.mt-05-fp8-prd8l:8482
I0425 20:22:15.183812 136344560469824 max_utils.py:284] Jax distributed system initialized!
I0425 20:22:21.187173 136344560469824 max_utils.py:800] System Information: Jax Version: 0.9.2
I0425 20:22:21.187281 136344560469824 max_utils.py:801] System Information: Jaxlib Version: 0.9.2
I0425 20:22:21.187324 136344560469824 max_utils.py:802] System Information: Jax Backend: PJRT C API
TFRT TPU v6 lite
Built on Mar 4 2026 11:32:08 (1772652728) cl/878335365
I0425 20:22:21.187364 136344560469824 train_utils.py:361] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing.
I0425 20:22:21.878225 136344560469824 maxtext_utils.py:1565] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0425 20:22:21.878513 136344560469824 checkpointing.py:677] Setting up checkpoint logger...
I0425 20:22:21.878569 136344560469824 checkpointing.py:233] Creating checkpoint manager with ocdbt=True and zarr3=True
I0425 20:22:21.878614 136344560469824 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None
I0425 20:22:21.878955 136344560469824 base_pytree_checkpoint_handler.py:441] Created BasePyTreeCheckpointHandler: use_ocdbt=True, use_zarr3=True, pytree_metadata_options=PyTreeMetadataOptions(support_rich_types=False), array_metadata_store=<orbax.checkpoint._src.metadata.array_metadata_store.Store object at 0x7c00983eb5f0>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB)
I0425 20:22:24.715064 136344560469824 checkpointing.py:265] Enabling policy for fixed interval checkpointing.
I0425 20:22:24.715301 136344560469824 checkpoint_manager.py:708] [process=4][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=('items',), item_handlers={'items': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7bed006f38f0>}, handler_registry=None
I0425 20:22:24.715570 136344560469824 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7bed006f38f0>` for item "items" and save args `<class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeSaveArgs'>` and restore args `<class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>` to `_handler_registry`.
I0425 20:22:24.715622 136344560469824 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7bed004fc440>` for item "metrics" and save args `<class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>` and restore args `<class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>` to `_handler_registry`.
I0425 20:22:24.715658 136344560469824 composite_checkpoint_handler.py:505] Initialized registry DefaultCheckpointHandlerRegistry({('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeSaveArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7bed006f38f0>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7bed006f38f0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7bed004fc440>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7bed004fc440>}).
I0425 20:22:24.715984 136344560469824 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34
I0425 20:22:24.716056 136344560469824 async_checkpointer.py:192] [process=4][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7bed002772e0> timeout: 1200 secs and primary_host=0 for async checkpoint writes
I0425 20:22:26.316956 136344560469824 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_test_pipeline_scan_nnx_20260425_201227/linen_xpk_test_pipeline_scan_nnx_20260425_201227_05_fp8/checkpoints
I0425 20:22:26.319190 136344560469824 checkpoint_manager.py:929] [process=4][thread=MainThread] CheckpointManager created,  primary_host=0, CheckpointManagerOptions=CheckpointManagerOptions(save_interval_steps=1, max_to_keep=None, keep_time_interval=None, keep_period=None, should_keep_fn=None, best_fn=None, best_mode='max', keep_checkpoints_without_metrics=True, step_prefix=None, step_format_fixed_length=None, step_name_format=None, create=True, cleanup_tmp_directories=False, save_on_steps=frozenset(), single_host_load_and_broadcast=False, todelete_subdir=None, todelete_full_path=None, enable_background_delete=False, read_only=False, enable_async_checkpointing=True, async_options=None, multiprocessing_options=MultiprocessingOptions(primary_host=0, active_processes=None, barrier_sync_key_prefix=None), should_save_fn=None, file_options=FileOptions(path_permission_mode=None), save_root_metadata=True, temporary_path_class=None, save_decision_policy=FixedIntervalPolicy(interval=10), preservation_policy=LatestN(n=None), prevent_write_metrics=False, enable_should_save_is_saving_in_progress_check=True, enable_per_process_directory_creation=False, lightweight_initialize=False), root_directory=gs://lance-maxtext/linen_ckpt_xpk_test_pipeline_scan_nnx_20260425_201227/linen_xpk_test_pipeline_scan_nnx_20260425_201227_05_fp8/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7bed006f2630>
I0425 20:22:26.319303 136344560469824 checkpointing.py:301] Checkpoint manager created!
I0425 20:22:27.401155 136344560469824 nnx_wrappers.py:453] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed').
I0425 20:22:27.401282 136344560469824 nnx_wrappers.py:453] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0425 20:22:27.784620 136344560469824 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0425 20:22:27.784714 136344560469824 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0425 20:22:27.801196 136344560469824 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0425 20:22:27.801267 136344560469824 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0425 20:22:27.853724 136344560469824 attentions.py:1154] attentions/query Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0425 20:22:27.853804 136344560469824 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0425 20:22:27.870256 136344560469824 attentions.py:1155] attentions/key Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0425 20:22:27.870320 136344560469824 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0425 20:22:27.886826 136344560469824 attentions.py:1156] attentions/value Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0425 20:22:27.886903 136344560469824 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0425 20:22:27.912192 136344560469824 attentions.py:1198] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv').
I0425 20:22:27.912265 136344560469824 attentions.py:1198] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0425 20:22:27.963438 136344560469824 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp').
I0425 20:22:27.963519 136344560469824 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None).
I0425 20:22:28.393753 136344560469824 checkpointing.py:577] checkpoint manager exists so trying to load this run's existing checkpoint
I0425 20:22:28.393866 136344560469824 checkpointing.py:665] No existing checkpoints found, not restoring checkpoint.
fsdp: 32
I0425 20:22:30.297282 136344560469824 maxtext_utils.py:1668]  params/_overwrite_with_gradient/decoder/layers/mlp/wi_0/Fp8DirectDotGeneralOp_0/input_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0425 20:22:30.297404 136344560469824 maxtext_utils.py:1668]  params/_overwrite_with_gradient/decoder/layers/mlp/wi_0/Fp8DirectDotGeneralOp_0/input_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0425 20:22:30.297452 136344560469824 maxtext_utils.py:1668]  params/_overwrite_with_gradient/decoder/layers/mlp/wi_0/Fp8DirectDotGeneralOp_0/kernel_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0425 20:22:30.297490 136344560469824 maxtext_utils.py:1668]  params/_overwrite_with_gradient/decoder/layers/mlp/wi_0/Fp8DirectDotGeneralOp_0/kernel_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0425 20:22:30.297530 136344560469824 maxtext_utils.py:1668]  params/_overwrite_with_gradient/decoder/layers/mlp/wi_0/Fp8DirectDotGeneralOp_0/output_grad_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0425 20:22:30.297564 136344560469824 maxtext_utils.py:1668]  params/_overwrite_with_gradient/decoder/layers/mlp/wi_0/Fp8DirectDotGeneralOp_0/output_grad_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0425 20:22:30.297596 136344560469824 maxtext_utils.py:1668]  params/_overwrite_with_gradient/decoder/layers/mlp/wi_1/Fp8DirectDotGeneralOp_0/input_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0425 20:22:30.297627 136344560469824 maxtext_utils.py:1668]  params/_overwrite_with_gradient/decoder/layers/mlp/wi_1/Fp8DirectDotGeneralOp_0/input_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0425 20:22:30.297659 136344560469824 maxtext_utils.py:1668]  params/_overwrite_with_gradient/decoder/layers/mlp/wi_1/Fp8DirectDotGeneralOp_0/kernel_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0425 20:22:30.297688 136344560469824 maxtext_utils.py:1668]  params/_overwrite_with_gradient/decoder/layers/mlp/wi_1/Fp8DirectDotGeneralOp_0/kernel_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0425 20:22:30.297716 136344560469824 maxtext_utils.py:1668]  params/_overwrite_with_gradient/decoder/layers/mlp/wi_1/Fp8DirectDotGeneralOp_0/output_grad_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0425 20:22:30.297745 136344560469824 maxtext_utils.py:1668]  params/_overwrite_with_gradient/decoder/layers/mlp/wi_1/Fp8DirectDotGeneralOp_0/output_grad_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0425 20:22:30.297778 136344560469824 maxtext_utils.py:1668]  params/_overwrite_with_gradient/decoder/layers/mlp/wo/Fp8DirectDotGeneralOp_0/input_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0425 20:22:30.297810 136344560469824 maxtext_utils.py:1668]  params/_overwrite_with_gradient/decoder/layers/mlp/wo/Fp8DirectDotGeneralOp_0/input_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0425 20:22:30.297839 136344560469824 maxtext_utils.py:1668]  params/_overwrite_with_gradient/decoder/layers/mlp/wo/Fp8DirectDotGeneralOp_0/kernel_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0425 20:22:30.297881 136344560469824 maxtext_utils.py:1668]  params/_overwrite_with_gradient/decoder/layers/mlp/wo/Fp8DirectDotGeneralOp_0/kernel_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0425 20:22:30.297913 136344560469824 maxtext_utils.py:1668]  params/_overwrite_with_gradient/decoder/layers/mlp/wo/Fp8DirectDotGeneralOp_0/output_grad_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0425 20:22:30.297944 136344560469824 maxtext_utils.py:1668]  params/_overwrite_with_gradient/decoder/layers/mlp/wo/Fp8DirectDotGeneralOp_0/output_grad_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0425 20:22:30.297974 136344560469824 maxtext_utils.py:1668]  params/_overwrite_with_gradient/decoder/layers/self_attention/key/Fp8DirectDotGeneralOp_0/input_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0425 20:22:30.298002 136344560469824 maxtext_utils.py:1668]  params/_overwrite_with_gradient/decoder/layers/self_attention/key/Fp8DirectDotGeneralOp_0/input_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0425 20:22:30.298030 136344560469824 maxtext_utils.py:1668]  params/_overwrite_with_gradient/decoder/layers/self_attention/key/Fp8DirectDotGeneralOp_0/kernel_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0425 20:22:30.298058 136344560469824 maxtext_utils.py:1668]  params/_overwrite_with_gradient/decoder/layers/self_attention/key/Fp8DirectDotGeneralOp_0/kernel_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0425 20:22:30.298087 136344560469824 maxtext_utils.py:1668]  params/_overwrite_with_gradient/decoder/layers/self_attention/key/Fp8DirectDotGeneralOp_0/output_grad_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0425 20:22:30.298114 136344560469824 maxtext_utils.py:1668]  params/_overwrite_with_gradient/decoder/layers/self_attention/key/Fp8DirectDotGeneralOp_0/output_grad_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0425 20:22:30.298143 136344560469824 maxtext_utils.py:1668]  params/_overwrite_with_gradient/decoder/layers/self_attention/out/Fp8DirectDotGeneralOp_0/input_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0425 20:22:30.298172 136344560469824 maxtext_utils.py:1668]  params/_overwrite_with_gradient/decoder/layers/self_attention/out/Fp8DirectDotGeneralOp_0/input_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0425 20:22:30.298200 136344560469824 maxtext_utils.py:1668]  params/_overwrite_with_gradient/decoder/layers/self_attention/out/Fp8DirectDotGeneralOp_0/kernel_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0425 20:22:30.298228 136344560469824 maxtext_utils.py:1668]  params/_overwrite_with_gradient/decoder/layers/self_attention/out/Fp8DirectDotGeneralOp_0/kernel_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0425 20:22:30.298260 136344560469824 maxtext_utils.py:1668]  params/_overwrite_with_gradient/decoder/layers/self_attention/out/Fp8DirectDotGeneralOp_0/output_grad_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0425 20:22:30.298288 136344560469824 maxtext_utils.py:1668]  params/_overwrite_with_gradient/decoder/layers/self_attention/out/Fp8DirectDotGeneralOp_0/output_grad_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0425 20:22:30.298316 136344560469824 maxtext_utils.py:1668]  params/_overwrite_with_gradient/decoder/layers/self_attention/query/Fp8DirectDotGeneralOp_0/input_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0425 20:22:30.298342 136344560469824 maxtext_utils.py:1668]  params/_overwrite_with_gradient/decoder/layers/self_attention/query/Fp8DirectDotGeneralOp_0/input_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0425 20:22:30.298370 136344560469824 maxtext_utils.py:1668]  params/_overwrite_with_gradient/decoder/layers/self_attention/query/Fp8DirectDotGeneralOp_0/kernel_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0425 20:22:30.298397 136344560469824 maxtext_utils.py:1668]  params/_overwrite_with_gradient/decoder/layers/self_attention/query/Fp8DirectDotGeneralOp_0/kernel_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0425 20:22:30.298423 136344560469824 maxtext_utils.py:1668]  params/_overwrite_with_gradient/decoder/layers/self_attention/query/Fp8DirectDotGeneralOp_0/output_grad_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0425 20:22:30.298449 136344560469824 maxtext_utils.py:1668]  params/_overwrite_with_gradient/decoder/layers/self_attention/query/Fp8DirectDotGeneralOp_0/output_grad_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0425 20:22:30.298475 136344560469824 maxtext_utils.py:1668]  params/_overwrite_with_gradient/decoder/layers/self_attention/value/Fp8DirectDotGeneralOp_0/input_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0425 20:22:30.298502 136344560469824 maxtext_utils.py:1668]  params/_overwrite_with_gradient/decoder/layers/self_attention/value/Fp8DirectDotGeneralOp_0/input_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0425 20:22:30.298528 136344560469824 maxtext_utils.py:1668]  params/_overwrite_with_gradient/decoder/layers/self_attention/value/Fp8DirectDotGeneralOp_0/kernel_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0425 20:22:30.298555 136344560469824 maxtext_utils.py:1668]  params/_overwrite_with_gradient/decoder/layers/self_attention/value/Fp8DirectDotGeneralOp_0/kernel_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0425 20:22:30.298583 136344560469824 maxtext_utils.py:1668]  params/_overwrite_with_gradient/decoder/layers/self_attention/value/Fp8DirectDotGeneralOp_0/output_grad_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0425 20:22:30.298611 136344560469824 maxtext_utils.py:1668]  params/_overwrite_with_gradient/decoder/layers/self_attention/value/Fp8DirectDotGeneralOp_0/output_grad_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0425 20:22:30.298664 136344560469824 maxtext_utils.py:1668]  params/params/decoder/decoder_norm/scale
    Shape:     float32[2048]
    Logical:   P('norm',)
    Physical:  (None,)
I0425 20:22:30.298718 136344560469824 maxtext_utils.py:1668]  params/params/decoder/layers/mlp/wi_0/kernel
    Shape:     float32[2048,16,7168]
    Logical:   P('embed', 'layers', 'mlp')
    Physical:  ('fsdp', None, None)
I0425 20:22:30.298757 136344560469824 maxtext_utils.py:1668]  params/params/decoder/layers/mlp/wi_1/kernel
    Shape:     float32[2048,16,7168]
    Logical:   P('embed', 'layers', 'mlp')
    Physical:  ('fsdp', None, None)
I0425 20:22:30.298805 136344560469824 maxtext_utils.py:1668]  params/params/decoder/layers/mlp/wo/kernel
    Shape:     float32[7168,16,2048]
    Logical:   P('mlp', 'layers', 'embed')
    Physical:  (None, None, 'fsdp')
I0425 20:22:30.298852 136344560469824 maxtext_utils.py:1668]  params/params/decoder/layers/post_self_attention_layer_norm/scale
    Shape:     float32[2048,16]
    Logical:   P('norm', 'layers')
    Physical:  (None, None)
I0425 20:22:30.298915 136344560469824 maxtext_utils.py:1668]  params/params/decoder/layers/pre_self_attention_layer_norm/scale
    Shape:     float32[2048,16]
    Logical:   P('norm', 'layers')
    Physical:  (None, None)
I0425 20:22:30.298970 136344560469824 maxtext_utils.py:1668]  params/params/decoder/layers/self_attention/key/kernel
    Shape:     float32[2048,16,16,128]
    Logical:   P('embed', 'layers', 'kv_heads', 'kv_head_dim')
    Physical:  ('fsdp', None, None, None)
I0425 20:22:30.299020 136344560469824 maxtext_utils.py:1668]  params/params/decoder/layers/self_attention/out/kernel
    Shape:     float32[16,16,128,2048]
    Logical:   P('heads', 'layers', 'kv', 'embed')
    Physical:  (None, None, None, 'fsdp')
I0425 20:22:30.299057 136344560469824 maxtext_utils.py:1668]  params/params/decoder/layers/self_attention/query/kernel
    Shape:     float32[2048,16,16,128]
    Logical:   P('embed', 'layers', 'q_heads', 'kv')
    Physical:  ('fsdp', None, None, None)
I0425 20:22:30.299090 136344560469824 maxtext_utils.py:1668]  params/params/decoder/layers/self_attention/value/kernel
    Shape:     float32[2048,16,16,128]
    Logical:   P('embed', 'layers', 'kv_heads', 'kv_head_dim')
    Physical:  ('fsdp', None, None, None)
I0425 20:22:30.299134 136344560469824 maxtext_utils.py:1668]  params/params/decoder/logits_dense/kernel
    Shape:     float32[2048,32000]
    Logical:   P('embed_vocab', 'vocab')
    Physical:  ('fsdp', None)
I0425 20:22:30.299178 136344560469824 maxtext_utils.py:1668]  params/params/token_embedder/embedding
    Shape:     float32[32000,2048]
    Logical:   P('vocab', 'embed_vocab')
    Physical:  (None, 'fsdp')

I0425 20:22:31.920017 136344560469824 train.py:155] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0425 20:22:31.920112 136344560469824 train.py:155] train/xent Physical: float32[32,2048]............................................ ('fsdp', None).
I0425 20:22:31.935495 136344560469824 train.py:162] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0425 20:22:31.935556 136344560469824 train.py:162] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None).
I0425 20:22:45.347977 136344560469824 max_utils.py:791] Total memory size: 1.5 GB, Output size: 0.4 GB, Temp size: 1.1 GB, Argument size: 0.4 GB, Host temp size: 0.0 GB.
I0425 20:22:45.348781 136344560469824 metric_logger.py:301] number parameters: 1.105 billion
I0425 20:23:00.634273 136344560469824 checkpointing.py:772] Waiting for step 0 to finish before checkpoint...
I0425 20:23:00.834247 136344560469824 checkpointing.py:776] Waited 0.1999516487121582 seconds for step 0 to finish before starting checkpointing.
I0425 20:23:00.836641 136344560469824 checkpoint_manager.py:2009] [process=4][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0425 20:23:00.838657 136344560469824 checkpoint_manager.py:1512] [process=4] Saving checkpoint at step 0
I0425 20:23:00.840148 136344560469824 event_tracking.py:70] [process=4] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_test_pipeline_scan_nnx_20260425_201227/linen_xpk_test_pipeline_scan_nnx_20260425_201227_05_fp8/checkpoints/0.
I0425 20:23:01.459163 136344560469824 signaling_client.py:364] Using JaxDistributedSignalingClient
I0425 20:23:01.460179 136344560469824 jax_array_handlers.py:360] Scheduling D2H of 81 prioritized jax.Array.
I0425 20:23:01.460239 136344560469824 replica_slices.py:424] Transferring arrays to host memory with options: use_replica_parallel=True, min_slice_bytes_for_replica_parallel=None, max_replicas_for_replica_parallel=None, enable_pinned_host_transfer=False
I0425 20:23:01.879333 136344560469824 base_pytree_checkpoint_handler.py:154] [process=4][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.420290s
I0425 20:23:01.879504 136344560469824 base_pytree_checkpoint_handler.py:130] [process=4] /jax/orbax/write/blocking_gbytes_per_sec: 3.597 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.42884349822998047 s) (per-host)
I0425 20:23:01.879558 136344560469824 base_pytree_checkpoint_handler.py:768] [process=4][thread=MainThread] Initiated Pytree async_save. Time taken: 0.428907s (batch_requests_ready=0.003861s, total_serialization_initiated=0.424975s, others=0.000071s)
I0425 20:23:01.879659 136344560469824 composite_checkpoint_handler.py:715] [process=4][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.433041s (all_items=0.000019s, per_item={'items': '0.00001860'}, temp_paths=0.433022)
I0425 20:23:01.880509 136344560469824 event_tracking.py:125] [process=4] [async] Finished blocking save in 1.04 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_test_pipeline_scan_nnx_20260425_201227/linen_xpk_test_pipeline_scan_nnx_20260425_201227_05_fp8/checkpoints/0.
I0425 20:23:01.880820 136218015180544 async_checkpointer.py:76] [process=4][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-25 20:43:01.880787
I0425 20:23:01.890163 136344560469824 checkpoint_manager.py:1560] [process=4][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize
I0425 20:23:01.890487 136217990002432 async_checkpointer.py:280] [process=4][thread=save_finalize] Waiting for background save thread=async_save.
I0425 20:23:01.890640 136344560469824 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_test_pipeline_scan_nnx_20260425_201227/linen_xpk_test_pipeline_scan_nnx_20260425_201227_05_fp8/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1777148580.836622, 'wait_for_prev_duration_secs': 6.151199340820312e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1777148580.838697, 'checkpointer_blocking_duration_secs': 1.0422823429107666, 'get_old_steps_start_time': 1777148581.881004, 'get_old_steps_duration_secs': 3.0994415283203125e-05, 'checkpoint_manager_blocking_start_time': 1777148580.834795, 'checkpoint_manager_blocking_duration_secs': 1.0558075904846191}
I0425 20:23:01.890764 136344560469824 checkpointing.py:408] Started an asynchronous checkpoint save for step 0
I0425 20:23:01.890815 136344560469824 max_utils.py:750] 
Memstats: After params initialized:
I0425 20:23:01.890878 136344560469824 max_utils.py:756] 	Using (GB) 0.45 / 31.25 (1.440000%) on TPU_16(process=4,(0,4,0,0))
I0425 20:23:01.890914 136344560469824 max_utils.py:756] 	Using (GB) 0.45 / 31.25 (1.440000%) on TPU_17(process=4,(1,4,0,0))
I0425 20:23:01.890943 136344560469824 max_utils.py:756] 	Using (GB) 0.45 / 31.25 (1.440000%) on TPU_20(process=4,(0,5,0,0))
I0425 20:23:01.890968 136344560469824 max_utils.py:756] 	Using (GB) 0.45 / 31.25 (1.440000%) on TPU_21(process=4,(1,5,0,0))
I0425 20:23:02.206898 136344560469824 metric_logger.py:196] completed step: 0, seconds: 15.285, TFLOP/s/device: 0.889, Tokens/s/device: 133.984, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52776.805
I0425 20:23:02.400123 136344560469824 metric_logger.py:196] completed step: 1, seconds: 1.570, TFLOP/s/device: 8.651, Tokens/s/device: 1304.045, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52761.840
I0425 20:23:02.808504 136344560469824 metric_logger.py:196] completed step: 2, seconds: 0.036, TFLOP/s/device: 380.240, Tokens/s/device: 57313.967, total_weights: 65536, loss: 10.816, lm_loss: 10.816, perplexity: 49832.398
I0425 20:23:02.966718 136344560469824 metric_logger.py:196] completed step: 3, seconds: 0.409, TFLOP/s/device: 33.248, Tokens/s/device: 5011.501, total_weights: 65536, loss: 10.431, lm_loss: 10.431, perplexity: 33901.820
I0425 20:23:03.284109 136344560469824 metric_logger.py:196] completed step: 4, seconds: 0.164, TFLOP/s/device: 82.640, Tokens/s/device: 12456.360, total_weights: 65536, loss: 9.992, lm_loss: 9.992, perplexity: 21847.639
I0425 20:23:03.290879 136344560469824 metric_logger.py:196] completed step: 5, seconds: 0.158, TFLOP/s/device: 85.948, Tokens/s/device: 12955.056, total_weights: 65536, loss: 9.549, lm_loss: 9.549, perplexity: 14035.006
I0425 20:23:05.058263    2785 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com
I0425 20:23:07.082934 136217998395136 array_metadata_store.py:203] [process=4][thread=array_type_handler] Wrote 81 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_test_pipeline_scan_nnx_20260425_201227/linen_xpk_test_pipeline_scan_nnx_20260425_201227_05_fp8/checkpoints/0/items/array_metadatas/process_4
I0425 20:23:22.016442 136218015180544 base_pytree_checkpoint_handler.py:130] [process=4] /jax/orbax/write/gbytes_per_sec: 76.815 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 20.56575298309326 s) (per-host)
I0425 20:23:22.016551 136218015180544 async_checkpointer.py:90] [process=4][thread=async_save] 3 Handler Commit operations completed. Time taken: 20.135609s.
I0425 20:23:27.053656 136344560469824 metric_logger.py:196] completed step: 6, seconds: 0.318, TFLOP/s/device: 42.749, Tokens/s/device: 6443.554, total_weights: 65536, loss: 9.111, lm_loss: 9.111, perplexity: 9058.688
I0425 20:23:27.212047 136344560469824 metric_logger.py:196] completed step: 7, seconds: 23.606, TFLOP/s/device: 0.576, Tokens/s/device: 86.759, total_weights: 65536, loss: 8.685, lm_loss: 8.685, perplexity: 5914.720
I0425 20:23:27.370847 136344560469824 metric_logger.py:196] completed step: 8, seconds: 0.164, TFLOP/s/device: 83.033, Tokens/s/device: 12515.660, total_weights: 65536, loss: 8.281, lm_loss: 8.281, perplexity: 3950.010
I0425 20:23:27.376035 136344560469824 checkpointing.py:772] Waiting for step 10 to finish before checkpoint...
I0425 20:23:27.687250 136344560469824 checkpointing.py:776] Waited 0.3111765384674072 seconds for step 10 to finish before starting checkpointing.
I0425 20:23:27.689866 136344560469824 checkpoint_manager.py:2020] [process=4][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0425 20:23:30.558586 136218015180544 async_checkpointer.py:160] [process=4][thread=async_save] Background save thread done. Time taken: 28.677628s.
I0425 20:23:30.558887 136217990002432 async_checkpointer.py:288] [process=4][thread=save_finalize] Done with waiting for background save thread=async_save.
I0425 20:23:30.559016 136217990002432 async_checkpointer.py:298] [process=4][thread=save_finalize] No errors found in background save thread=async_save.
I0425 20:23:30.559069 136217990002432 checkpoint_manager.py:2137] [process=4][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts...
I0425 20:23:30.560640 136217990002432 checkpoint_manager.py:2146] [process=4][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts.
I0425 20:23:30.560846 136344560469824 checkpoint_manager.py:2032] [process=4][thread=MainThread][step=0][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=0.
W0425 20:23:30.560995 136344560469824 checkpoint_manager.py:1452] Waiting for previous save to complete took 2.871128 seconds. If this number is high, consider checkpointing less frequently.
I0425 20:23:30.562771 136344560469824 checkpoint_manager.py:1512] [process=4] Saving checkpoint at step 10
I0425 20:23:30.565248 136344560469824 event_tracking.py:70] [process=4] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_test_pipeline_scan_nnx_20260425_201227/linen_xpk_test_pipeline_scan_nnx_20260425_201227_05_fp8/checkpoints/10.
I0425 20:23:30.854366 136344560469824 jax_array_handlers.py:360] Scheduling D2H of 81 prioritized jax.Array.
I0425 20:23:30.854458 136344560469824 replica_slices.py:424] Transferring arrays to host memory with options: use_replica_parallel=True, min_slice_bytes_for_replica_parallel=None, max_replicas_for_replica_parallel=None, enable_pinned_host_transfer=False
I0425 20:23:30.908118 136344560469824 base_pytree_checkpoint_handler.py:154] [process=4][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.054684s
I0425 20:23:30.908285 136344560469824 base_pytree_checkpoint_handler.py:130] [process=4] /jax/orbax/write/blocking_gbytes_per_sec: 25.303 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.06097126007080078 s) (per-host)
I0425 20:23:30.908337 136344560469824 base_pytree_checkpoint_handler.py:768] [process=4][thread=MainThread] Initiated Pytree async_save. Time taken: 0.061034s (batch_requests_ready=0.003264s, total_serialization_initiated=0.057701s, others=0.000069s)
I0425 20:23:30.908433 136344560469824 composite_checkpoint_handler.py:715] [process=4][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.065094s (all_items=0.000015s, per_item={'items': '0.00001454'}, temp_paths=0.065079)
I0425 20:23:30.909177 136344560469824 event_tracking.py:125] [process=4] [async] Finished blocking save in 0.35 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_test_pipeline_scan_nnx_20260425_201227/linen_xpk_test_pipeline_scan_nnx_20260425_201227_05_fp8/checkpoints/10.
I0425 20:23:30.909479 136217990002432 async_checkpointer.py:76] [process=4][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-25 20:43:30.909446
I0425 20:23:30.911499 136344560469824 checkpoint_manager.py:1560] [process=4][thread=MainThread][step=10] Starting CheckpointManager Save Finalize thread=save_finalize
I0425 20:23:30.911728 136215323981568 async_checkpointer.py:280] [process=4][thread=save_finalize] Waiting for background save thread=async_save.
I0425 20:23:30.911852 136344560469824 standard_logger.py:34] {'step': 10, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_test_pipeline_scan_nnx_20260425_201227/linen_xpk_test_pipeline_scan_nnx_20260425_201227_05_fp8/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1777148607.6898353, 'wait_for_prev_duration_secs': 2.8711278438568115, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1777148610.56281, 'checkpointer_blocking_duration_secs': 0.3468148708343506, 'get_old_steps_start_time': 1777148610.9096515, 'get_old_steps_duration_secs': 3.147125244140625e-05, 'checkpoint_manager_blocking_start_time': 1777148607.6880958, 'checkpoint_manager_blocking_duration_secs': 3.2237229347229004}
I0425 20:23:30.911976 136344560469824 checkpointing.py:408] Started an asynchronous checkpoint save for step 10
I0425 20:23:30.912725 136344560469824 metric_logger.py:196] completed step: 9, seconds: 0.158, TFLOP/s/device: 85.801, Tokens/s/device: 12932.885, total_weights: 65536, loss: 7.908, lm_loss: 7.908, perplexity: 2720.116
I0425 20:23:30.919822 136344560469824 metric_logger.py:196] completed step: 10, seconds: 0.159, TFLOP/s/device: 85.599, Tokens/s/device: 12902.413, total_weights: 65536, loss: 7.568, lm_loss: 7.568, perplexity: 1936.198
I0425 20:23:31.078472 136344560469824 metric_logger.py:196] completed step: 11, seconds: 3.542, TFLOP/s/device: 3.836, Tokens/s/device: 578.279, total_weights: 65536, loss: 7.288, lm_loss: 7.288, perplexity: 1463.272
I0425 20:23:31.656994 136344560469824 metric_logger.py:196] completed step: 12, seconds: 0.006, TFLOP/s/device: 2176.727, Tokens/s/device: 328099.968, total_weights: 65536, loss: 7.032, lm_loss: 7.032, perplexity: 1132.061
I0425 20:23:31.815197 136344560469824 metric_logger.py:196] completed step: 13, seconds: 0.575, TFLOP/s/device: 23.641, Tokens/s/device: 3563.406, total_weights: 65536, loss: 6.815, lm_loss: 6.815, perplexity: 911.031
I0425 20:23:31.973484 136344560469824 metric_logger.py:196] completed step: 14, seconds: 0.163, TFLOP/s/device: 83.156, Tokens/s/device: 12534.197, total_weights: 65536, loss: 6.635, lm_loss: 6.635, perplexity: 761.080
I0425 20:23:32.131930 136344560469824 metric_logger.py:196] completed step: 15, seconds: 0.158, TFLOP/s/device: 85.937, Tokens/s/device: 12953.417, total_weights: 65536, loss: 6.492, lm_loss: 6.492, perplexity: 659.915
I0425 20:23:32.290091 136344560469824 metric_logger.py:196] completed step: 16, seconds: 0.158, TFLOP/s/device: 85.829, Tokens/s/device: 12937.052, total_weights: 65536, loss: 6.380, lm_loss: 6.380, perplexity: 589.948
I0425 20:23:32.448345 136344560469824 metric_logger.py:196] completed step: 17, seconds: 0.158, TFLOP/s/device: 85.790, Tokens/s/device: 12931.252, total_weights: 65536, loss: 6.293, lm_loss: 6.293, perplexity: 540.941
I0425 20:23:32.606714 136344560469824 metric_logger.py:196] completed step: 18, seconds: 0.158, TFLOP/s/device: 85.873, Tokens/s/device: 12943.757, total_weights: 65536, loss: 6.223, lm_loss: 6.223, perplexity: 504.108
I0425 20:23:32.764373 136344560469824 checkpointing.py:772] Waiting for step 19 to finish before checkpoint...
I0425 20:23:32.765342 136344560469824 checkpointing.py:776] Waited 0.000989675521850586 seconds for step 19 to finish before starting checkpointing.
I0425 20:23:32.767525 136344560469824 checkpoint_manager.py:2020] [process=4][thread=MainThread][step=10][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0425 20:23:36.652759 136215315588864 array_metadata_store.py:203] [process=4][thread=array_type_handler] Wrote 81 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_test_pipeline_scan_nnx_20260425_201227/linen_xpk_test_pipeline_scan_nnx_20260425_201227_05_fp8/checkpoints/10/items/array_metadatas/process_4
I0425 20:24:12.347169 136217990002432 base_pytree_checkpoint_handler.py:130] [process=4] /jax/orbax/write/gbytes_per_sec: 38.067 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 41.49980068206787 s) (per-host)
I0425 20:24:12.347292 136217990002432 async_checkpointer.py:90] [process=4][thread=async_save] 3 Handler Commit operations completed. Time taken: 41.437698s.
I0425 20:24:21.867263 136217990002432 async_checkpointer.py:160] [process=4][thread=async_save] Background save thread done. Time taken: 50.957653s.
I0425 20:24:21.867514 136215323981568 async_checkpointer.py:288] [process=4][thread=save_finalize] Done with waiting for background save thread=async_save.
I0425 20:24:21.867569 136215323981568 async_checkpointer.py:298] [process=4][thread=save_finalize] No errors found in background save thread=async_save.
I0425 20:24:21.867619 136215323981568 checkpoint_manager.py:2137] [process=4][thread=save_finalize][step=10] CheckpointManager Save Finalize is syncing with other hosts...
I0425 20:24:21.869661 136215323981568 checkpoint_manager.py:2146] [process=4][thread=save_finalize][step=10] CheckpointManager Save Finalize is done on all hosts.
I0425 20:24:21.869817 136344560469824 checkpoint_manager.py:2032] [process=4][thread=MainThread][step=10][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=10.
W0425 20:24:21.869957 136344560469824 checkpoint_manager.py:1452] Waiting for previous save to complete took 49.102434 seconds. If this number is high, consider checkpointing less frequently.
I0425 20:24:21.871495 136344560469824 checkpoint_manager.py:1512] [process=4] Saving checkpoint at step 19
I0425 20:24:21.873623 136344560469824 event_tracking.py:70] [process=4] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_test_pipeline_scan_nnx_20260425_201227/linen_xpk_test_pipeline_scan_nnx_20260425_201227_05_fp8/checkpoints/19.
I0425 20:24:22.153454 136344560469824 jax_array_handlers.py:360] Scheduling D2H of 81 prioritized jax.Array.
I0425 20:24:22.153552 136344560469824 replica_slices.py:424] Transferring arrays to host memory with options: use_replica_parallel=True, min_slice_bytes_for_replica_parallel=None, max_replicas_for_replica_parallel=None, enable_pinned_host_transfer=False
I0425 20:24:22.204609 136344560469824 base_pytree_checkpoint_handler.py:154] [process=4][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.052200s
I0425 20:24:22.204774 136344560469824 base_pytree_checkpoint_handler.py:130] [process=4] /jax/orbax/write/blocking_gbytes_per_sec: 26.398 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.058441877365112305 s) (per-host)
I0425 20:24:22.204823 136344560469824 base_pytree_checkpoint_handler.py:768] [process=4][thread=MainThread] Initiated Pytree async_save. Time taken: 0.058502s (batch_requests_ready=0.003213s, total_serialization_initiated=0.055224s, others=0.000065s)
I0425 20:24:22.204939 136344560469824 composite_checkpoint_handler.py:715] [process=4][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.062507s (all_items=0.000011s, per_item={'items': '0.00001144'}, temp_paths=0.062495)
I0425 20:24:22.205653 136344560469824 event_tracking.py:125] [process=4] [async] Finished blocking save in 0.33 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_test_pipeline_scan_nnx_20260425_201227/linen_xpk_test_pipeline_scan_nnx_20260425_201227_05_fp8/checkpoints/19.
I0425 20:24:22.205997 136215323981568 async_checkpointer.py:76] [process=4][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-25 20:44:22.205961
I0425 20:24:22.208129 136344560469824 checkpoint_manager.py:1560] [process=4][thread=MainThread][step=19] Starting CheckpointManager Save Finalize thread=save_finalize
I0425 20:24:22.208422 136215315588864 async_checkpointer.py:280] [process=4][thread=save_finalize] Waiting for background save thread=async_save.
I0425 20:24:22.208585 136344560469824 standard_logger.py:34] {'step': 19, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_test_pipeline_scan_nnx_20260425_201227/linen_xpk_test_pipeline_scan_nnx_20260425_201227_05_fp8/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1777148612.7674966, 'wait_for_prev_duration_secs': 49.102434158325195, 'time_between_consecutive_saves_sec': 2.2068159580230713, 'checkpointer_blocking_start_time': 1777148661.8715343, 'checkpointer_blocking_duration_secs': 0.33460283279418945, 'get_old_steps_start_time': 1777148662.2061563, 'get_old_steps_duration_secs': 2.574920654296875e-05, 'checkpoint_manager_blocking_start_time': 1777148612.7656143, 'checkpoint_manager_blocking_duration_secs': 49.44293761253357}
I0425 20:24:22.208714 136344560469824 checkpointing.py:408] Started an asynchronous checkpoint save for step 19
I0425 20:24:22.208761 136344560469824 checkpoint_manager.py:2020] [process=4][thread=MainThread][step=19][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0425 20:24:27.244697 136214778717952 array_metadata_store.py:203] [process=4][thread=array_type_handler] Wrote 81 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_test_pipeline_scan_nnx_20260425_201227/linen_xpk_test_pipeline_scan_nnx_20260425_201227_05_fp8/checkpoints/19/items/array_metadatas/process_4
I0425 20:25:03.705909 136215323981568 base_pytree_checkpoint_handler.py:130] [process=4] /jax/orbax/write/gbytes_per_sec: 38.012 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 41.55952477455139 s) (per-host)
I0425 20:25:03.706031 136215323981568 async_checkpointer.py:90] [process=4][thread=async_save] 3 Handler Commit operations completed. Time taken: 41.499925s.
I0425 20:25:10.633102 136215323981568 async_checkpointer.py:160] [process=4][thread=async_save] Background save thread done. Time taken: 48.426980s.
I0425 20:25:10.633451 136215315588864 async_checkpointer.py:288] [process=4][thread=save_finalize] Done with waiting for background save thread=async_save.
I0425 20:25:10.633572 136215315588864 async_checkpointer.py:298] [process=4][thread=save_finalize] No errors found in background save thread=async_save.
I0425 20:25:10.633613 136215315588864 checkpoint_manager.py:2137] [process=4][thread=save_finalize][step=19] CheckpointManager Save Finalize is syncing with other hosts...
I0425 20:25:10.636012 136215315588864 checkpoint_manager.py:2146] [process=4][thread=save_finalize][step=19] CheckpointManager Save Finalize is done on all hosts.
I0425 20:25:10.636178 136344560469824 checkpoint_manager.py:2032] [process=4][thread=MainThread][step=19][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=19.
I0425 20:25:10.636327 136344560469824 checkpoint_manager.py:2009] [process=4][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0425 20:25:10.637269 136344560469824 metric_logger.py:196] completed step: 19, seconds: 0.158, TFLOP/s/device: 85.889, Tokens/s/device: 12946.129, total_weights: 65536, loss: 6.167, lm_loss: 6.167, perplexity: 476.947
Per train step:
 Total TFLOPs: 13.59 
 split as 93.93% learnable weight flops and 6.07% attention flops
XPK End: Sat Apr 25 20:25:19 UTC 2026
EXIT_CODE=0

Metrics: main (59e0f1759) vs test/pipeline-scan-nnx (5b21b66fc)

Metrics: `main (59e0f1759)` vs `test/pipeline-scan-nnx (5b21b66fc)`