Compare: 05_fp8 — MaxView

‹ 04_int8Case: 05_fp806_grad_accum ›

Metrics: `main (8a17c3d19)` vs `test/pipeline-scan-nnx (5b21b66fc)`

Metric	main `8a17c3d19`	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	85.783	85.854	+0.071
Tok/s	12930.1	12940.9	+10.785
Avg s/step	1.679	1.667	-0.012
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_18(process=5,(2,4,0,0))	0.45	31.25	1.4%
TPU_19(process=5,(3,4,0,0))	0.45	31.25	1.4%
TPU_22(process=5,(2,5,0,0))	0.45	31.25	1.4%
TPU_23(process=5,(3,5,0,0))	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	0.884	0.889	+0.0050
1	10.874	10.874	8.082	8.215	+0.1330
2	10.816	10.816	377.263	442.016	+64.7530
3	10.431	10.431	31.349	32.386	+1.0370
4	9.992	9.992	82.728	82.647	-0.0810
5	9.549	9.549	85.994	85.878	-0.1160
6	9.111	9.111	42.771	42.587	-0.1840
7	8.685	8.685	0.553	0.557	+0.0040
8	8.281	8.281	83.153	83.087	-0.0660
9	7.908	7.908	85.854	85.581	-0.2730
10	7.568	7.568	85.161	85.570	+0.4090
11	7.288	7.288	5.309	5.348	+0.0390
12	7.032	7.032	2063.972	2108.820	+44.8480
13	6.815	6.815	24.216	24.384	+0.1680
14	6.635	6.635	83.238	83.175	-0.0630
15	6.492	6.492	85.813	85.854	+0.0410
16	6.380	6.380	85.888	85.843	-0.0450
17	6.293	6.293	85.670	85.613	-0.0570
18	6.223	6.223	85.869	85.804	-0.0650
19	6.167	6.167	85.783	85.854	+0.0710

main · 8a17c3d19 · main_20260422_071422 · full log

XPK Start: Wed Apr 22 07:26:18 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. 
2026-04-22 07:26:43.851759: 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)
I0422 07:26:44.063495 135333372073792 max_utils.py:273] Attempting to initialize the jax distributed system...
I0422 07:26:53.104436 135333372073792 distributed.py:149] Starting JAX distributed service on [::]:8482
I0422 07:26:53.106833 135333372073792 distributed.py:172] Connecting to JAX distributed service on mt-05-fp8-68gsk-slice-job-0-0.mt-05-fp8-68gsk:8482
I0422 07:26:55.272066 135333372073792 max_utils.py:284] Jax distributed system initialized!
I0422 07:27:01.547978 135333372073792 max_utils.py:800] System Information: Jax Version: 0.9.2
I0422 07:27:01.548087 135333372073792 max_utils.py:801] System Information: Jaxlib Version: 0.9.2
I0422 07:27:01.548127 135333372073792 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
I0422 07:27:01.548163 135333372073792 train_utils.py:361] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing.
I0422 07:27:02.240637 135333372073792 maxtext_utils.py:1604] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0422 07:27:02.240922 135333372073792 checkpointing.py:677] Setting up checkpoint logger...
I0422 07:27:02.240976 135333372073792 checkpointing.py:233] Creating checkpoint manager with ocdbt=True and zarr3=True
I0422 07:27:02.241021 135333372073792 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None
I0422 07:27:02.241388 135333372073792 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 0x7b1509d50620>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB)
I0422 07:27:05.239210 135333372073792 checkpointing.py:265] Enabling policy for fixed interval checkpointing.
I0422 07:27:05.239479 135333372073792 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 0x7b003437aa50>}, handler_registry=None
I0422 07:27:05.239731 135333372073792 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7b003437aa50>` 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`.
I0422 07:27:05.239784 135333372073792 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7b0108284f80>` 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`.
I0422 07:27:05.239820 135333372073792 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 0x7b003437aa50>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7b003437aa50>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7b0108284f80>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7b0108284f80>}).
I0422 07:27:05.240174 135333372073792 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34
I0422 07:27:05.240271 135333372073792 async_checkpointer.py:192] [process=6][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7b000858d6c0> timeout: 1200 secs and primary_host=0 for async checkpoint writes
I0422 07:27:06.368745 135333372073792 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_main_20260422_071422/linen_xpk_main_20260422_071422_05_fp8/checkpoints
I0422 07:27:07.222933 135333372073792 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_20260422_071422/linen_xpk_main_20260422_071422_05_fp8/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7b0108283050>
I0422 07:27:07.223107 135333372073792 checkpointing.py:301] Checkpoint manager created!
I0422 07:27:08.304450 135333372073792 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed').
I0422 07:27:08.304571 135333372073792 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0422 07:27:08.697253 135333372073792 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0422 07:27:08.697368 135333372073792 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0422 07:27:08.713508 135333372073792 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0422 07:27:08.713570 135333372073792 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0422 07:27:08.765725 135333372073792 attentions.py:1154] attentions/query Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0422 07:27:08.765810 135333372073792 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0422 07:27:08.782114 135333372073792 attentions.py:1155] attentions/key Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0422 07:27:08.782179 135333372073792 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0422 07:27:08.798487 135333372073792 attentions.py:1156] attentions/value Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0422 07:27:08.798551 135333372073792 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0422 07:27:08.823938 135333372073792 attentions.py:1198] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv').
I0422 07:27:08.824009 135333372073792 attentions.py:1198] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0422 07:27:08.875435 135333372073792 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp').
I0422 07:27:08.875514 135333372073792 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None).
I0422 07:27:09.312471 135333372073792 checkpointing.py:577] checkpoint manager exists so trying to load this run's existing checkpoint
I0422 07:27:09.312587 135333372073792 checkpointing.py:665] No existing checkpoints found, not restoring checkpoint.
fsdp: 32
I0422 07:27:11.251715 135333372073792 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:  ()
I0422 07:27:11.251842 135333372073792 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/mlp/wi_0/Fp8DirectDotGeneralOp_0/input_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0422 07:27:11.251891 135333372073792 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:  ()
I0422 07:27:11.251931 135333372073792 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/mlp/wi_0/Fp8DirectDotGeneralOp_0/kernel_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0422 07:27:11.251965 135333372073792 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:  ()
I0422 07:27:11.251998 135333372073792 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:  ()
I0422 07:27:11.252030 135333372073792 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:  ()
I0422 07:27:11.252062 135333372073792 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/mlp/wi_1/Fp8DirectDotGeneralOp_0/input_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0422 07:27:11.252092 135333372073792 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:  ()
I0422 07:27:11.252130 135333372073792 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/mlp/wi_1/Fp8DirectDotGeneralOp_0/kernel_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0422 07:27:11.252161 135333372073792 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:  ()
I0422 07:27:11.252191 135333372073792 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:  ()
I0422 07:27:11.252223 135333372073792 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/mlp/wo/Fp8DirectDotGeneralOp_0/input_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0422 07:27:11.252254 135333372073792 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/mlp/wo/Fp8DirectDotGeneralOp_0/input_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0422 07:27:11.252303 135333372073792 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/mlp/wo/Fp8DirectDotGeneralOp_0/kernel_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0422 07:27:11.252342 135333372073792 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/mlp/wo/Fp8DirectDotGeneralOp_0/kernel_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0422 07:27:11.252376 135333372073792 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:  ()
I0422 07:27:11.252406 135333372073792 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/mlp/wo/Fp8DirectDotGeneralOp_0/output_grad_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0422 07:27:11.252435 135333372073792 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:  ()
I0422 07:27:11.252463 135333372073792 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/self_attention/key/Fp8DirectDotGeneralOp_0/input_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0422 07:27:11.252491 135333372073792 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:  ()
I0422 07:27:11.252518 135333372073792 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/self_attention/key/Fp8DirectDotGeneralOp_0/kernel_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0422 07:27:11.252545 135333372073792 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:  ()
I0422 07:27:11.252573 135333372073792 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:  ()
I0422 07:27:11.252602 135333372073792 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:  ()
I0422 07:27:11.252629 135333372073792 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/self_attention/out/Fp8DirectDotGeneralOp_0/input_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0422 07:27:11.252656 135333372073792 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:  ()
I0422 07:27:11.252703 135333372073792 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/self_attention/out/Fp8DirectDotGeneralOp_0/kernel_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0422 07:27:11.252734 135333372073792 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:  ()
I0422 07:27:11.252761 135333372073792 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:  ()
I0422 07:27:11.252788 135333372073792 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:  ()
I0422 07:27:11.252816 135333372073792 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/self_attention/query/Fp8DirectDotGeneralOp_0/input_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0422 07:27:11.252843 135333372073792 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:  ()
I0422 07:27:11.252869 135333372073792 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/self_attention/query/Fp8DirectDotGeneralOp_0/kernel_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0422 07:27:11.252896 135333372073792 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:  ()
I0422 07:27:11.252922 135333372073792 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:  ()
I0422 07:27:11.252948 135333372073792 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:  ()
I0422 07:27:11.252974 135333372073792 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/self_attention/value/Fp8DirectDotGeneralOp_0/input_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0422 07:27:11.253001 135333372073792 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:  ()
I0422 07:27:11.253028 135333372073792 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/self_attention/value/Fp8DirectDotGeneralOp_0/kernel_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0422 07:27:11.253055 135333372073792 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:  ()
I0422 07:27:11.253083 135333372073792 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:  ()
I0422 07:27:11.253140 135333372073792 maxtext_utils.py:1707]  params/params/decoder/decoder_norm/scale
    Shape:     float32[2048]
    Logical:   P('norm',)
    Physical:  (None,)
I0422 07:27:11.253196 135333372073792 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)
I0422 07:27:11.253235 135333372073792 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)
I0422 07:27:11.253293 135333372073792 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')
I0422 07:27:11.253343 135333372073792 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)
I0422 07:27:11.253378 135333372073792 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)
I0422 07:27:11.253426 135333372073792 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)
I0422 07:27:11.253473 135333372073792 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')
I0422 07:27:11.253508 135333372073792 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)
I0422 07:27:11.253541 135333372073792 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)
I0422 07:27:11.253585 135333372073792 maxtext_utils.py:1707]  params/params/decoder/logits_dense/kernel
    Shape:     float32[2048,32000]
    Logical:   P('embed_vocab', 'vocab')
    Physical:  ('fsdp', None)
I0422 07:27:11.253628 135333372073792 maxtext_utils.py:1707]  params/params/token_embedder/embedding
    Shape:     float32[32000,2048]
    Logical:   P('vocab', 'embed_vocab')
    Physical:  (None, 'fsdp')

I0422 07:27:12.892688 135333372073792 train.py:155] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0422 07:27:12.892782 135333372073792 train.py:155] train/xent Physical: float32[32,2048]............................................ ('fsdp', None).
I0422 07:27:12.908140 135333372073792 train.py:162] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0422 07:27:12.908202 135333372073792 train.py:162] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None).
I0422 07:27:26.961371 135333372073792 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.
I0422 07:27:26.962322 135333372073792 metric_logger.py:301] number parameters: 1.105 billion
I0422 07:27:42.335369 135333372073792 checkpointing.py:772] Waiting for step 0 to finish before checkpoint...
I0422 07:27:42.502927 135333372073792 checkpointing.py:776] Waited 0.16753888130187988 seconds for step 0 to finish before starting checkpointing.
I0422 07:27:42.505397 135333372073792 checkpoint_manager.py:2009] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0422 07:27:42.507488 135333372073792 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 0
I0422 07:27:42.508924 135333372073792 event_tracking.py:70] [process=6] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260422_071422/linen_xpk_main_20260422_071422_05_fp8/checkpoints/0.
I0422 07:27:43.268095 135333372073792 signaling_client.py:364] Using JaxDistributedSignalingClient
I0422 07:27:43.269100 135333372073792 jax_array_handlers.py:360] Scheduling D2H of 81 prioritized jax.Array.
I0422 07:27:43.269168 135333372073792 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
I0422 07:27:43.695445 135333372073792 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.427470s
I0422 07:27:43.695613 135333372073792 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 3.538 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.4360814094543457 s) (per-host)
I0422 07:27:43.695667 135333372073792 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.436145s (batch_requests_ready=0.003845s, total_serialization_initiated=0.432228s, others=0.000071s)
I0422 07:27:43.695761 135333372073792 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.440322s (all_items=0.000018s, per_item={'items': '0.00001764'}, temp_paths=0.440304)
I0422 07:27:43.696556 135333372073792 event_tracking.py:125] [process=6] [async] Finished blocking save in 1.19 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260422_071422/linen_xpk_main_20260422_071422_05_fp8/checkpoints/0.
I0422 07:27:43.696902 135204415121152 async_checkpointer.py:76] [process=6][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-22 07:47:43.696863
I0422 07:27:43.698942 135333372073792 checkpoint_manager.py:1560] [process=6][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize
I0422 07:27:43.699244 135204386715392 async_checkpointer.py:280] [process=6][thread=save_finalize] Waiting for background save thread=async_save.
I0422 07:27:43.699424 135333372073792 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260422_071422/linen_xpk_main_20260422_071422_05_fp8/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776842862.50538, 'wait_for_prev_duration_secs': 6.0558319091796875e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776842862.5075285, 'checkpointer_blocking_duration_secs': 1.18953537940979, 'get_old_steps_start_time': 1776842863.6970887, 'get_old_steps_duration_secs': 3.218650817871094e-05, 'checkpoint_manager_blocking_start_time': 1776842862.5035858, 'checkpoint_manager_blocking_duration_secs': 1.1957981586456299}
I0422 07:27:43.699532 135333372073792 checkpointing.py:408] Started an asynchronous checkpoint save for step 0
I0422 07:27:43.699585 135333372073792 max_utils.py:750] 
Memstats: After params initialized:
I0422 07:27:43.699634 135333372073792 max_utils.py:756] 	Using (GB) 0.45 / 31.25 (1.440000%) on TPU_24(process=6,(0,6,0,0))
I0422 07:27:43.699670 135333372073792 max_utils.py:756] 	Using (GB) 0.45 / 31.25 (1.440000%) on TPU_25(process=6,(1,6,0,0))
I0422 07:27:43.699698 135333372073792 max_utils.py:756] 	Using (GB) 0.45 / 31.25 (1.440000%) on TPU_28(process=6,(0,7,0,0))
I0422 07:27:43.699724 135333372073792 max_utils.py:756] 	Using (GB) 0.45 / 31.25 (1.440000%) on TPU_29(process=6,(1,7,0,0))
I0422 07:27:44.018577 135333372073792 metric_logger.py:196] completed step: 0, seconds: 15.373, TFLOP/s/device: 0.884, Tokens/s/device: 133.221, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52776.805
I0422 07:27:44.212087 135333372073792 metric_logger.py:196] completed step: 1, seconds: 1.681, TFLOP/s/device: 8.082, Tokens/s/device: 1218.225, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52761.840
I0422 07:27:44.645320 135333372073792 metric_logger.py:196] completed step: 2, seconds: 0.036, TFLOP/s/device: 377.263, Tokens/s/device: 56865.195, total_weights: 65536, loss: 10.816, lm_loss: 10.816, perplexity: 49832.398
I0422 07:27:44.803481 135333372073792 metric_logger.py:196] completed step: 3, seconds: 0.433, TFLOP/s/device: 31.349, Tokens/s/device: 4725.329, total_weights: 65536, loss: 10.431, lm_loss: 10.431, perplexity: 33901.820
I0422 07:27:45.121147 135333372073792 metric_logger.py:196] completed step: 4, seconds: 0.164, TFLOP/s/device: 82.728, Tokens/s/device: 12469.633, total_weights: 65536, loss: 9.992, lm_loss: 9.992, perplexity: 21847.639
I0422 07:27:45.127038 135333372073792 metric_logger.py:196] completed step: 5, seconds: 0.158, TFLOP/s/device: 85.994, Tokens/s/device: 12962.025, total_weights: 65536, loss: 9.549, lm_loss: 9.549, perplexity: 14035.006
I0422 07:27:46.602230    2798 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com
I0422 07:27:50.053838 135204395108096 array_metadata_store.py:203] [process=6][thread=array_type_handler] Wrote 81 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260422_071422/linen_xpk_main_20260422_071422_05_fp8/checkpoints/0/items/array_metadatas/process_6
I0422 07:28:03.224945 135204415121152 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 79.125 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 19.965373039245605 s) (per-host)
I0422 07:28:03.225085 135204415121152 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 19.528069s.
I0422 07:28:09.831923 135333372073792 metric_logger.py:196] completed step: 6, seconds: 0.318, TFLOP/s/device: 42.771, Tokens/s/device: 6446.982, total_weights: 65536, loss: 9.111, lm_loss: 9.111, perplexity: 9058.688
I0422 07:28:09.990351 135333372073792 metric_logger.py:196] completed step: 7, seconds: 24.548, TFLOP/s/device: 0.553, Tokens/s/device: 83.429, total_weights: 65536, loss: 8.685, lm_loss: 8.685, perplexity: 5914.720
I0422 07:28:10.148869 135333372073792 metric_logger.py:196] completed step: 8, seconds: 0.163, TFLOP/s/device: 83.153, Tokens/s/device: 12533.660, total_weights: 65536, loss: 8.281, lm_loss: 8.281, perplexity: 3950.010
I0422 07:28:10.155040 135333372073792 checkpointing.py:772] Waiting for step 10 to finish before checkpoint...
I0422 07:28:10.465562 135333372073792 checkpointing.py:776] Waited 0.310497522354126 seconds for step 10 to finish before starting checkpointing.
I0422 07:28:10.468173 135333372073792 checkpoint_manager.py:2020] [process=6][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0422 07:28:11.938017 135204415121152 async_checkpointer.py:160] [process=6][thread=async_save] Background save thread done. Time taken: 28.240983s.
I0422 07:28:11.938325 135204386715392 async_checkpointer.py:288] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save.
I0422 07:28:11.938446 135204386715392 async_checkpointer.py:298] [process=6][thread=save_finalize] No errors found in background save thread=async_save.
I0422 07:28:11.938498 135204386715392 checkpoint_manager.py:2137] [process=6][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts...
I0422 07:28:11.940122 135204386715392 checkpoint_manager.py:2146] [process=6][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts.
I0422 07:28:11.940251 135333372073792 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.
W0422 07:28:11.940355 135333372073792 checkpoint_manager.py:1452] Waiting for previous save to complete took 1.472185 seconds. If this number is high, consider checkpointing less frequently.
I0422 07:28:11.942583 135333372073792 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 10
I0422 07:28:11.944607 135333372073792 event_tracking.py:70] [process=6] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260422_071422/linen_xpk_main_20260422_071422_05_fp8/checkpoints/10.
I0422 07:28:12.643826 135333372073792 jax_array_handlers.py:360] Scheduling D2H of 81 prioritized jax.Array.
I0422 07:28:12.643918 135333372073792 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
I0422 07:28:12.704690 135333372073792 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.061921s
I0422 07:28:12.704854 135333372073792 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 22.579 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.06832718849182129 s) (per-host)
I0422 07:28:12.704901 135333372073792 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.068384s (batch_requests_ready=0.003321s, total_serialization_initiated=0.064999s, others=0.000064s)
I0422 07:28:12.704998 135333372073792 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.072731s (all_items=0.000015s, per_item={'items': '0.00001526'}, temp_paths=0.072716)
I0422 07:28:12.705651 135333372073792 event_tracking.py:125] [process=6] [async] Finished blocking save in 0.76 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260422_071422/linen_xpk_main_20260422_071422_05_fp8/checkpoints/10.
I0422 07:28:12.706000 135204386715392 async_checkpointer.py:76] [process=6][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-22 07:48:12.705962
I0422 07:28:12.708037 135333372073792 checkpoint_manager.py:1560] [process=6][thread=MainThread][step=10] Starting CheckpointManager Save Finalize thread=save_finalize
I0422 07:28:12.708356 135201200006912 async_checkpointer.py:280] [process=6][thread=save_finalize] Waiting for background save thread=async_save.
I0422 07:28:12.708498 135333372073792 standard_logger.py:34] {'step': 10, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260422_071422/linen_xpk_main_20260422_071422_05_fp8/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776842890.4681423, 'wait_for_prev_duration_secs': 1.4721853733062744, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776842891.942622, 'checkpointer_blocking_duration_secs': 0.7635326385498047, 'get_old_steps_start_time': 1776842892.706178, 'get_old_steps_duration_secs': 3.1948089599609375e-05, 'checkpoint_manager_blocking_start_time': 1776842890.4664054, 'checkpoint_manager_blocking_duration_secs': 2.242058277130127}
I0422 07:28:12.708675 135333372073792 checkpointing.py:408] Started an asynchronous checkpoint save for step 10
I0422 07:28:12.709346 135333372073792 metric_logger.py:196] completed step: 9, seconds: 0.158, TFLOP/s/device: 85.854, Tokens/s/device: 12940.894, total_weights: 65536, loss: 7.908, lm_loss: 7.908, perplexity: 2720.116
I0422 07:28:12.716763 135333372073792 metric_logger.py:196] completed step: 10, seconds: 0.160, TFLOP/s/device: 85.161, Tokens/s/device: 12836.423, total_weights: 65536, loss: 7.568, lm_loss: 7.568, perplexity: 1936.198
I0422 07:28:12.874706 135333372073792 metric_logger.py:196] completed step: 11, seconds: 2.559, TFLOP/s/device: 5.309, Tokens/s/device: 800.195, total_weights: 65536, loss: 7.288, lm_loss: 7.288, perplexity: 1463.272
I0422 07:28:13.440210 135333372073792 metric_logger.py:196] completed step: 12, seconds: 0.007, TFLOP/s/device: 2063.972, Tokens/s/device: 311104.360, total_weights: 65536, loss: 7.032, lm_loss: 7.032, perplexity: 1132.061
I0422 07:28:13.598575 135333372073792 metric_logger.py:196] completed step: 13, seconds: 0.561, TFLOP/s/device: 24.216, Tokens/s/device: 3650.084, total_weights: 65536, loss: 6.815, lm_loss: 6.815, perplexity: 911.031
I0422 07:28:13.756779 135333372073792 metric_logger.py:196] completed step: 14, seconds: 0.163, TFLOP/s/device: 83.238, Tokens/s/device: 12546.559, total_weights: 65536, loss: 6.635, lm_loss: 6.635, perplexity: 761.080
I0422 07:28:13.915349 135333372073792 metric_logger.py:196] completed step: 15, seconds: 0.158, TFLOP/s/device: 85.813, Tokens/s/device: 12934.682, total_weights: 65536, loss: 6.492, lm_loss: 6.492, perplexity: 659.915
I0422 07:28:14.073688 135333372073792 metric_logger.py:196] completed step: 16, seconds: 0.158, TFLOP/s/device: 85.888, Tokens/s/device: 12945.966, total_weights: 65536, loss: 6.380, lm_loss: 6.380, perplexity: 589.948
I0422 07:28:14.231964 135333372073792 metric_logger.py:196] completed step: 17, seconds: 0.159, TFLOP/s/device: 85.670, Tokens/s/device: 12913.070, total_weights: 65536, loss: 6.293, lm_loss: 6.293, perplexity: 540.941
I0422 07:28:14.390337 135333372073792 metric_logger.py:196] completed step: 18, seconds: 0.158, TFLOP/s/device: 85.869, Tokens/s/device: 12943.102, total_weights: 65536, loss: 6.223, lm_loss: 6.223, perplexity: 504.108
I0422 07:28:14.547967 135333372073792 checkpointing.py:772] Waiting for step 19 to finish before checkpoint...
I0422 07:28:14.548979 135333372073792 checkpointing.py:776] Waited 0.0010323524475097656 seconds for step 19 to finish before starting checkpointing.
I0422 07:28:14.551079 135333372073792 checkpoint_manager.py:2020] [process=6][thread=MainThread][step=10][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0422 07:28:18.239221 135201174828800 array_metadata_store.py:203] [process=6][thread=array_type_handler] Wrote 81 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260422_071422/linen_xpk_main_20260422_071422_05_fp8/checkpoints/10/items/array_metadatas/process_6
I0422 07:28:54.934067 135204386715392 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 37.349 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 42.2974808216095 s) (per-host)
I0422 07:28:54.934198 135204386715392 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 42.228076s.
I0422 07:29:02.679600 135204386715392 async_checkpointer.py:160] [process=6][thread=async_save] Background save thread done. Time taken: 49.973461s.
I0422 07:29:02.679893 135201200006912 async_checkpointer.py:288] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save.
I0422 07:29:02.680017 135201200006912 async_checkpointer.py:298] [process=6][thread=save_finalize] No errors found in background save thread=async_save.
I0422 07:29:02.680068 135201200006912 checkpoint_manager.py:2137] [process=6][thread=save_finalize][step=10] CheckpointManager Save Finalize is syncing with other hosts...
I0422 07:29:02.681487 135201200006912 checkpoint_manager.py:2146] [process=6][thread=save_finalize][step=10] CheckpointManager Save Finalize is done on all hosts.
I0422 07:29:02.681666 135333372073792 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.
W0422 07:29:02.681789 135333372073792 checkpoint_manager.py:1452] Waiting for previous save to complete took 48.130715 seconds. If this number is high, consider checkpointing less frequently.
I0422 07:29:02.684276 135333372073792 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 19
I0422 07:29:02.686276 135333372073792 event_tracking.py:70] [process=6] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260422_071422/linen_xpk_main_20260422_071422_05_fp8/checkpoints/19.
I0422 07:29:03.430472 135333372073792 jax_array_handlers.py:360] Scheduling D2H of 81 prioritized jax.Array.
I0422 07:29:03.430567 135333372073792 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
I0422 07:29:03.490147 135333372073792 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.060668s
I0422 07:29:03.490324 135333372073792 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 22.946 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.06723475456237793 s) (per-host)
I0422 07:29:03.490372 135333372073792 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.067294s (batch_requests_ready=0.003303s, total_serialization_initiated=0.063927s, others=0.000064s)
I0422 07:29:03.490470 135333372073792 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.071550s (all_items=0.000012s, per_item={'items': '0.00001168'}, temp_paths=0.071539)
I0422 07:29:03.491123 135333372073792 event_tracking.py:125] [process=6] [async] Finished blocking save in 0.81 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260422_071422/linen_xpk_main_20260422_071422_05_fp8/checkpoints/19.
I0422 07:29:03.491473 135201200006912 async_checkpointer.py:76] [process=6][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-22 07:49:03.491433
I0422 07:29:03.493467 135333372073792 checkpoint_manager.py:1560] [process=6][thread=MainThread][step=19] Starting CheckpointManager Save Finalize thread=save_finalize
I0422 07:29:03.493692 135201174828800 async_checkpointer.py:280] [process=6][thread=save_finalize] Waiting for background save thread=async_save.
I0422 07:29:03.493792 135333372073792 standard_logger.py:34] {'step': 19, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260422_071422/linen_xpk_main_20260422_071422_05_fp8/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776842894.5510504, 'wait_for_prev_duration_secs': 48.13071537017822, 'time_between_consecutive_saves_sec': 2.610888957977295, 'checkpointer_blocking_start_time': 1776842942.6843297, 'checkpointer_blocking_duration_secs': 0.8072867393493652, 'get_old_steps_start_time': 1776842943.4916358, 'get_old_steps_duration_secs': 2.6226043701171875e-05, 'checkpoint_manager_blocking_start_time': 1776842894.5492687, 'checkpoint_manager_blocking_duration_secs': 48.94449162483215}
I0422 07:29:03.493960 135333372073792 checkpointing.py:408] Started an asynchronous checkpoint save for step 19
I0422 07:29:03.494005 135333372073792 checkpoint_manager.py:2020] [process=6][thread=MainThread][step=19][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0422 07:29:08.636494 135201703307008 array_metadata_store.py:203] [process=6][thread=array_type_handler] Wrote 81 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260422_071422/linen_xpk_main_20260422_071422_05_fp8/checkpoints/19/items/array_metadatas/process_6
I0422 07:29:45.772080 135201200006912 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 37.303 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 42.348944425582886 s) (per-host)
I0422 07:29:45.772202 135201200006912 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 42.280615s.
I0422 07:29:52.540896 135201200006912 async_checkpointer.py:160] [process=6][thread=async_save] Background save thread done. Time taken: 49.049291s.
I0422 07:29:52.541145 135201174828800 async_checkpointer.py:288] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save.
I0422 07:29:52.541203 135201174828800 async_checkpointer.py:298] [process=6][thread=save_finalize] No errors found in background save thread=async_save.
I0422 07:29:52.541252 135201174828800 checkpoint_manager.py:2137] [process=6][thread=save_finalize][step=19] CheckpointManager Save Finalize is syncing with other hosts...
I0422 07:29:52.542946 135201174828800 checkpoint_manager.py:2146] [process=6][thread=save_finalize][step=19] CheckpointManager Save Finalize is done on all hosts.
I0422 07:29:52.543098 135333372073792 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.
I0422 07:29:52.543248 135333372073792 checkpoint_manager.py:2009] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0422 07:29:52.544183 135333372073792 metric_logger.py:196] completed step: 19, seconds: 0.158, TFLOP/s/device: 85.783, Tokens/s/device: 12930.109, 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: Wed Apr 22 07:30:03 UTC 2026
EXIT_CODE=0

test/pipeline-scan-nnx · 5b21b66fc · test_pipeline_scan_nnx_20260422_212603 · full log

XPK Start: Wed Apr 22 21:38:40 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. 
2026-04-22 21:39:05.330680: 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)
I0422 21:39:05.540894 138957515740992 max_utils.py:273] Attempting to initialize the jax distributed system...
I0422 21:39:14.581381 138957515740992 distributed.py:149] Starting JAX distributed service on [::]:8482
I0422 21:39:14.583801 138957515740992 distributed.py:172] Connecting to JAX distributed service on mt-05-fp8-chefp-slice-job-0-0.mt-05-fp8-chefp:8482
I0422 21:39:16.376136 138957515740992 max_utils.py:284] Jax distributed system initialized!
I0422 21:39:22.355074 138957515740992 max_utils.py:800] System Information: Jax Version: 0.9.2
I0422 21:39:22.355179 138957515740992 max_utils.py:801] System Information: Jaxlib Version: 0.9.2
I0422 21:39:22.355221 138957515740992 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
I0422 21:39:22.355256 138957515740992 train_utils.py:361] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing.
I0422 21:39:23.098489 138957515740992 maxtext_utils.py:1565] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0422 21:39:23.098786 138957515740992 checkpointing.py:677] Setting up checkpoint logger...
I0422 21:39:23.098842 138957515740992 checkpointing.py:233] Creating checkpoint manager with ocdbt=True and zarr3=True
I0422 21:39:23.098887 138957515740992 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None
I0422 21:39:23.099242 138957515740992 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 0x7e60d9a500e0>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB)
I0422 21:39:25.972617 138957515740992 checkpointing.py:265] Enabling policy for fixed interval checkpointing.
I0422 21:39:25.972866 138957515740992 checkpoint_manager.py:708] [process=5][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=('items',), item_handlers={'items': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7e4c90252510>}, handler_registry=None
I0422 21:39:25.973103 138957515740992 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7e4c90252510>` 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`.
I0422 21:39:25.973152 138957515740992 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7e4c902570e0>` 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`.
I0422 21:39:25.973191 138957515740992 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 0x7e4c90252510>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7e4c90252510>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7e4c902570e0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7e4c902570e0>}).
I0422 21:39:25.973520 138957515740992 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34
I0422 21:39:25.973590 138957515740992 async_checkpointer.py:192] [process=5][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7e4b9c7e72e0> timeout: 1200 secs and primary_host=0 for async checkpoint writes
I0422 21:39:27.088112 138957515740992 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_test_pipeline_scan_nnx_20260422_212603/linen_xpk_test_pipeline_scan_nnx_20260422_212603_05_fp8/checkpoints
I0422 21:39:27.539575 138957515740992 checkpoint_manager.py:929] [process=5][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_20260422_212603/linen_xpk_test_pipeline_scan_nnx_20260422_212603_05_fp8/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7e4c90253140>
I0422 21:39:27.539786 138957515740992 checkpointing.py:301] Checkpoint manager created!
I0422 21:39:28.617497 138957515740992 nnx_wrappers.py:453] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed').
I0422 21:39:28.617615 138957515740992 nnx_wrappers.py:453] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0422 21:39:29.000883 138957515740992 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0422 21:39:29.000975 138957515740992 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0422 21:39:29.017268 138957515740992 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0422 21:39:29.017323 138957515740992 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0422 21:39:29.069931 138957515740992 attentions.py:1154] attentions/query Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0422 21:39:29.070017 138957515740992 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0422 21:39:29.086499 138957515740992 attentions.py:1155] attentions/key Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0422 21:39:29.086564 138957515740992 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0422 21:39:29.103016 138957515740992 attentions.py:1156] attentions/value Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0422 21:39:29.103074 138957515740992 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0422 21:39:29.133706 138957515740992 attentions.py:1198] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv').
I0422 21:39:29.133806 138957515740992 attentions.py:1198] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0422 21:39:29.185219 138957515740992 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp').
I0422 21:39:29.185294 138957515740992 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None).
I0422 21:39:29.616585 138957515740992 checkpointing.py:577] checkpoint manager exists so trying to load this run's existing checkpoint
I0422 21:39:29.616724 138957515740992 checkpointing.py:665] No existing checkpoints found, not restoring checkpoint.
fsdp: 32
I0422 21:39:31.545114 138957515740992 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:  ()
I0422 21:39:31.545246 138957515740992 maxtext_utils.py:1668]  params/_overwrite_with_gradient/decoder/layers/mlp/wi_0/Fp8DirectDotGeneralOp_0/input_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0422 21:39:31.545295 138957515740992 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:  ()
I0422 21:39:31.545333 138957515740992 maxtext_utils.py:1668]  params/_overwrite_with_gradient/decoder/layers/mlp/wi_0/Fp8DirectDotGeneralOp_0/kernel_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0422 21:39:31.545375 138957515740992 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:  ()
I0422 21:39:31.545411 138957515740992 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:  ()
I0422 21:39:31.545444 138957515740992 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:  ()
I0422 21:39:31.545474 138957515740992 maxtext_utils.py:1668]  params/_overwrite_with_gradient/decoder/layers/mlp/wi_1/Fp8DirectDotGeneralOp_0/input_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0422 21:39:31.545505 138957515740992 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:  ()
I0422 21:39:31.545533 138957515740992 maxtext_utils.py:1668]  params/_overwrite_with_gradient/decoder/layers/mlp/wi_1/Fp8DirectDotGeneralOp_0/kernel_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0422 21:39:31.545562 138957515740992 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:  ()
I0422 21:39:31.545592 138957515740992 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:  ()
I0422 21:39:31.545624 138957515740992 maxtext_utils.py:1668]  params/_overwrite_with_gradient/decoder/layers/mlp/wo/Fp8DirectDotGeneralOp_0/input_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0422 21:39:31.545667 138957515740992 maxtext_utils.py:1668]  params/_overwrite_with_gradient/decoder/layers/mlp/wo/Fp8DirectDotGeneralOp_0/input_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0422 21:39:31.545700 138957515740992 maxtext_utils.py:1668]  params/_overwrite_with_gradient/decoder/layers/mlp/wo/Fp8DirectDotGeneralOp_0/kernel_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0422 21:39:31.545730 138957515740992 maxtext_utils.py:1668]  params/_overwrite_with_gradient/decoder/layers/mlp/wo/Fp8DirectDotGeneralOp_0/kernel_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0422 21:39:31.545758 138957515740992 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:  ()
I0422 21:39:31.545786 138957515740992 maxtext_utils.py:1668]  params/_overwrite_with_gradient/decoder/layers/mlp/wo/Fp8DirectDotGeneralOp_0/output_grad_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0422 21:39:31.545816 138957515740992 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:  ()
I0422 21:39:31.545844 138957515740992 maxtext_utils.py:1668]  params/_overwrite_with_gradient/decoder/layers/self_attention/key/Fp8DirectDotGeneralOp_0/input_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0422 21:39:31.545872 138957515740992 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:  ()
I0422 21:39:31.545898 138957515740992 maxtext_utils.py:1668]  params/_overwrite_with_gradient/decoder/layers/self_attention/key/Fp8DirectDotGeneralOp_0/kernel_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0422 21:39:31.545925 138957515740992 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:  ()
I0422 21:39:31.545952 138957515740992 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:  ()
I0422 21:39:31.545983 138957515740992 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:  ()
I0422 21:39:31.546012 138957515740992 maxtext_utils.py:1668]  params/_overwrite_with_gradient/decoder/layers/self_attention/out/Fp8DirectDotGeneralOp_0/input_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0422 21:39:31.546038 138957515740992 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:  ()
I0422 21:39:31.546064 138957515740992 maxtext_utils.py:1668]  params/_overwrite_with_gradient/decoder/layers/self_attention/out/Fp8DirectDotGeneralOp_0/kernel_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0422 21:39:31.546091 138957515740992 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:  ()
I0422 21:39:31.546117 138957515740992 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:  ()
I0422 21:39:31.546144 138957515740992 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:  ()
I0422 21:39:31.546170 138957515740992 maxtext_utils.py:1668]  params/_overwrite_with_gradient/decoder/layers/self_attention/query/Fp8DirectDotGeneralOp_0/input_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0422 21:39:31.546196 138957515740992 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:  ()
I0422 21:39:31.546222 138957515740992 maxtext_utils.py:1668]  params/_overwrite_with_gradient/decoder/layers/self_attention/query/Fp8DirectDotGeneralOp_0/kernel_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0422 21:39:31.546248 138957515740992 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:  ()
I0422 21:39:31.546274 138957515740992 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:  ()
I0422 21:39:31.546300 138957515740992 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:  ()
I0422 21:39:31.546326 138957515740992 maxtext_utils.py:1668]  params/_overwrite_with_gradient/decoder/layers/self_attention/value/Fp8DirectDotGeneralOp_0/input_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0422 21:39:31.546353 138957515740992 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:  ()
I0422 21:39:31.546379 138957515740992 maxtext_utils.py:1668]  params/_overwrite_with_gradient/decoder/layers/self_attention/value/Fp8DirectDotGeneralOp_0/kernel_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0422 21:39:31.546407 138957515740992 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:  ()
I0422 21:39:31.546433 138957515740992 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:  ()
I0422 21:39:31.546485 138957515740992 maxtext_utils.py:1668]  params/params/decoder/decoder_norm/scale
    Shape:     float32[2048]
    Logical:   P('norm',)
    Physical:  (None,)
I0422 21:39:31.546545 138957515740992 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)
I0422 21:39:31.546582 138957515740992 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)
I0422 21:39:31.546628 138957515740992 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')
I0422 21:39:31.546703 138957515740992 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)
I0422 21:39:31.546744 138957515740992 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)
I0422 21:39:31.546794 138957515740992 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)
I0422 21:39:31.546842 138957515740992 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')
I0422 21:39:31.546877 138957515740992 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)
I0422 21:39:31.546909 138957515740992 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)
I0422 21:39:31.546953 138957515740992 maxtext_utils.py:1668]  params/params/decoder/logits_dense/kernel
    Shape:     float32[2048,32000]
    Logical:   P('embed_vocab', 'vocab')
    Physical:  ('fsdp', None)

I0422 21:39:31.547002 138957515740992 maxtext_utils.py:1668]  params/params/token_embedder/embedding
    Shape:     float32[32000,2048]
    Logical:   P('vocab', 'embed_vocab')
    Physical:  (None, 'fsdp')
I0422 21:39:33.175905 138957515740992 train.py:155] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0422 21:39:33.176006 138957515740992 train.py:155] train/xent Physical: float32[32,2048]............................................ ('fsdp', None).
I0422 21:39:33.191454 138957515740992 train.py:162] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0422 21:39:33.191514 138957515740992 train.py:162] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None).
I0422 21:39:46.984049 138957515740992 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.
I0422 21:39:46.984930 138957515740992 metric_logger.py:301] number parameters: 1.105 billion
I0422 21:40:02.269160 138957515740992 checkpointing.py:772] Waiting for step 0 to finish before checkpoint...
I0422 21:40:02.440023 138957515740992 checkpointing.py:776] Waited 0.17083144187927246 seconds for step 0 to finish before starting checkpointing.
I0422 21:40:02.442326 138957515740992 checkpoint_manager.py:2009] [process=5][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0422 21:40:02.444288 138957515740992 checkpoint_manager.py:1512] [process=5] Saving checkpoint at step 0
I0422 21:40:02.445705 138957515740992 event_tracking.py:70] [process=5] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_test_pipeline_scan_nnx_20260422_212603/linen_xpk_test_pipeline_scan_nnx_20260422_212603_05_fp8/checkpoints/0.
I0422 21:40:03.180912 138957515740992 signaling_client.py:364] Using JaxDistributedSignalingClient
I0422 21:40:03.181857 138957515740992 jax_array_handlers.py:360] Scheduling D2H of 81 prioritized jax.Array.
I0422 21:40:03.181917 138957515740992 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
I0422 21:40:03.601113 138957515740992 base_pytree_checkpoint_handler.py:154] [process=5][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.420324s
I0422 21:40:03.601288 138957515740992 base_pytree_checkpoint_handler.py:130] [process=5] /jax/orbax/write/blocking_gbytes_per_sec: 3.597 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.4289381504058838 s) (per-host)
I0422 21:40:03.601342 138957515740992 base_pytree_checkpoint_handler.py:768] [process=5][thread=MainThread] Initiated Pytree async_save. Time taken: 0.429003s (batch_requests_ready=0.003860s, total_serialization_initiated=0.425070s, others=0.000073s)
I0422 21:40:03.601444 138957515740992 composite_checkpoint_handler.py:715] [process=5][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.433120s (all_items=0.000019s, per_item={'items': '0.00001860'}, temp_paths=0.433102)
I0422 21:40:03.602226 138957515740992 event_tracking.py:125] [process=5] [async] Finished blocking save in 1.16 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_test_pipeline_scan_nnx_20260422_212603/linen_xpk_test_pipeline_scan_nnx_20260422_212603_05_fp8/checkpoints/0.
I0422 21:40:03.602504 138827513071360 async_checkpointer.py:76] [process=5][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-22 22:00:03.602481
I0422 21:40:03.607945 138957515740992 checkpoint_manager.py:1560] [process=5][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize
I0422 21:40:03.608267 138825959470848 async_checkpointer.py:280] [process=5][thread=save_finalize] Waiting for background save thread=async_save.
I0422 21:40:03.608431 138957515740992 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_test_pipeline_scan_nnx_20260422_212603/linen_xpk_test_pipeline_scan_nnx_20260422_212603_05_fp8/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776894002.4423072, 'wait_for_prev_duration_secs': 6.0558319091796875e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776894002.444326, 'checkpointer_blocking_duration_secs': 1.1582872867584229, 'get_old_steps_start_time': 1776894003.6026366, 'get_old_steps_duration_secs': 4.291534423828125e-05, 'checkpoint_manager_blocking_start_time': 1776894002.44051, 'checkpoint_manager_blocking_duration_secs': 1.1678798198699951}
I0422 21:40:03.608548 138957515740992 checkpointing.py:408] Started an asynchronous checkpoint save for step 0
I0422 21:40:03.608601 138957515740992 max_utils.py:750] 
Memstats: After params initialized:
I0422 21:40:03.608667 138957515740992 max_utils.py:756] 	Using (GB) 0.45 / 31.25 (1.440000%) on TPU_18(process=5,(2,4,0,0))
I0422 21:40:03.608705 138957515740992 max_utils.py:756] 	Using (GB) 0.45 / 31.25 (1.440000%) on TPU_19(process=5,(3,4,0,0))
I0422 21:40:03.608735 138957515740992 max_utils.py:756] 	Using (GB) 0.45 / 31.25 (1.440000%) on TPU_22(process=5,(2,5,0,0))
I0422 21:40:03.608761 138957515740992 max_utils.py:756] 	Using (GB) 0.45 / 31.25 (1.440000%) on TPU_23(process=5,(3,5,0,0))
I0422 21:40:03.925884 138957515740992 metric_logger.py:196] completed step: 0, seconds: 15.284, TFLOP/s/device: 0.889, Tokens/s/device: 133.995, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52776.805
I0422 21:40:04.113258 138957515740992 metric_logger.py:196] completed step: 1, seconds: 1.654, TFLOP/s/device: 8.215, Tokens/s/device: 1238.282, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52761.840
I0422 21:40:04.532904 138957515740992 metric_logger.py:196] completed step: 2, seconds: 0.031, TFLOP/s/device: 442.016, Tokens/s/device: 66625.460, total_weights: 65536, loss: 10.816, lm_loss: 10.816, perplexity: 49832.398
I0422 21:40:04.691360 138957515740992 metric_logger.py:196] completed step: 3, seconds: 0.420, TFLOP/s/device: 32.386, Tokens/s/device: 4881.630, total_weights: 65536, loss: 10.431, lm_loss: 10.431, perplexity: 33901.820
I0422 21:40:05.009282 138957515740992 metric_logger.py:196] completed step: 4, seconds: 0.164, TFLOP/s/device: 82.647, Tokens/s/device: 12457.421, total_weights: 65536, loss: 9.992, lm_loss: 9.992, perplexity: 21847.639
I0422 21:40:05.016377 138957515740992 metric_logger.py:196] completed step: 5, seconds: 0.158, TFLOP/s/device: 85.878, Tokens/s/device: 12944.411, total_weights: 65536, loss: 9.549, lm_loss: 9.549, perplexity: 14035.006
I0422 21:40:06.578024    2837 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com
I0422 21:40:08.535791 138827014321920 array_metadata_store.py:203] [process=5][thread=array_type_handler] Wrote 81 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_test_pipeline_scan_nnx_20260422_212603/linen_xpk_test_pipeline_scan_nnx_20260422_212603_05_fp8/checkpoints/0/items/array_metadatas/process_5
I0422 21:40:23.845853 138827513071360 base_pytree_checkpoint_handler.py:130] [process=5] /jax/orbax/write/gbytes_per_sec: 76.415 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 20.67346501350403 s) (per-host)
I0422 21:40:23.845977 138827513071360 async_checkpointer.py:90] [process=5][thread=async_save] 3 Handler Commit operations completed. Time taken: 20.243398s.
I0422 21:40:29.568785 138957515740992 metric_logger.py:196] completed step: 6, seconds: 0.319, TFLOP/s/device: 42.587, Tokens/s/device: 6419.238, total_weights: 65536, loss: 9.111, lm_loss: 9.111, perplexity: 9058.688
I0422 21:40:29.727630 138957515740992 metric_logger.py:196] completed step: 7, seconds: 24.395, TFLOP/s/device: 0.557, Tokens/s/device: 83.951, total_weights: 65536, loss: 8.685, lm_loss: 8.685, perplexity: 5914.720
I0422 21:40:29.886546 138957515740992 metric_logger.py:196] completed step: 8, seconds: 0.164, TFLOP/s/device: 83.087, Tokens/s/device: 12523.773, total_weights: 65536, loss: 8.281, lm_loss: 8.281, perplexity: 3950.010
I0422 21:40:29.891551 138957515740992 checkpointing.py:772] Waiting for step 10 to finish before checkpoint...
I0422 21:40:30.203035 138957515740992 checkpointing.py:776] Waited 0.3114490509033203 seconds for step 10 to finish before starting checkpointing.
I0422 21:40:30.205682 138957515740992 checkpoint_manager.py:2020] [process=5][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0422 21:40:31.668529 138827513071360 async_checkpointer.py:160] [process=5][thread=async_save] Background save thread done. Time taken: 28.065934s.
I0422 21:40:31.668849 138825959470848 async_checkpointer.py:288] [process=5][thread=save_finalize] Done with waiting for background save thread=async_save.
I0422 21:40:31.668976 138825959470848 async_checkpointer.py:298] [process=5][thread=save_finalize] No errors found in background save thread=async_save.
I0422 21:40:31.669028 138825959470848 checkpoint_manager.py:2137] [process=5][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts...
I0422 21:40:31.670768 138825959470848 checkpoint_manager.py:2146] [process=5][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts.
I0422 21:40:31.670950 138957515740992 checkpoint_manager.py:2032] [process=5][thread=MainThread][step=0][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=0.
W0422 21:40:31.671078 138957515740992 checkpoint_manager.py:1452] Waiting for previous save to complete took 1.465410 seconds. If this number is high, consider checkpointing less frequently.
I0422 21:40:31.672851 138957515740992 checkpoint_manager.py:1512] [process=5] Saving checkpoint at step 10
I0422 21:40:31.674846 138957515740992 event_tracking.py:70] [process=5] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_test_pipeline_scan_nnx_20260422_212603/linen_xpk_test_pipeline_scan_nnx_20260422_212603_05_fp8/checkpoints/10.
I0422 21:40:32.382464 138957515740992 jax_array_handlers.py:360] Scheduling D2H of 81 prioritized jax.Array.
I0422 21:40:32.382557 138957515740992 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
I0422 21:40:32.420544 138957515740992 base_pytree_checkpoint_handler.py:154] [process=5][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.038947s
I0422 21:40:32.420695 138957515740992 base_pytree_checkpoint_handler.py:130] [process=5] /jax/orbax/write/blocking_gbytes_per_sec: 34.060 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.04529404640197754 s) (per-host)
I0422 21:40:32.420746 138957515740992 base_pytree_checkpoint_handler.py:768] [process=5][thread=MainThread] Initiated Pytree async_save. Time taken: 0.045355s (batch_requests_ready=0.003293s, total_serialization_initiated=0.041994s, others=0.000068s)
I0422 21:40:32.420831 138957515740992 composite_checkpoint_handler.py:715] [process=5][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.049517s (all_items=0.000016s, per_item={'items': '0.00001597'}, temp_paths=0.049501)
I0422 21:40:32.421545 138957515740992 event_tracking.py:125] [process=5] [async] Finished blocking save in 0.75 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_test_pipeline_scan_nnx_20260422_212603/linen_xpk_test_pipeline_scan_nnx_20260422_212603_05_fp8/checkpoints/10.
I0422 21:40:32.421893 138825959470848 async_checkpointer.py:76] [process=5][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-22 22:00:32.421856
I0422 21:40:32.425631 138957515740992 checkpoint_manager.py:1560] [process=5][thread=MainThread][step=10] Starting CheckpointManager Save Finalize thread=save_finalize
I0422 21:40:32.425936 138824832276224 async_checkpointer.py:280] [process=5][thread=save_finalize] Waiting for background save thread=async_save.
I0422 21:40:32.426079 138957515740992 standard_logger.py:34] {'step': 10, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_test_pipeline_scan_nnx_20260422_212603/linen_xpk_test_pipeline_scan_nnx_20260422_212603_05_fp8/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776894030.2056372, 'wait_for_prev_duration_secs': 1.4654102325439453, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776894031.6728888, 'checkpointer_blocking_duration_secs': 0.7491481304168701, 'get_old_steps_start_time': 1776894032.4220588, 'get_old_steps_duration_secs': 3.147125244140625e-05, 'checkpoint_manager_blocking_start_time': 1776894030.203811, 'checkpoint_manager_blocking_duration_secs': 2.2222349643707275}
I0422 21:40:32.426247 138957515740992 checkpointing.py:408] Started an asynchronous checkpoint save for step 10
I0422 21:40:32.426976 138957515740992 metric_logger.py:196] completed step: 9, seconds: 0.159, TFLOP/s/device: 85.581, Tokens/s/device: 12899.650, total_weights: 65536, loss: 7.908, lm_loss: 7.908, perplexity: 2720.116
I0422 21:40:32.434510 138957515740992 metric_logger.py:196] completed step: 10, seconds: 0.159, TFLOP/s/device: 85.570, Tokens/s/device: 12898.025, total_weights: 65536, loss: 7.568, lm_loss: 7.568, perplexity: 1936.198
I0422 21:40:32.591918 138957515740992 metric_logger.py:196] completed step: 11, seconds: 2.541, TFLOP/s/device: 5.348, Tokens/s/device: 806.099, total_weights: 65536, loss: 7.288, lm_loss: 7.288, perplexity: 1463.272
I0422 21:40:33.154016 138957515740992 metric_logger.py:196] completed step: 12, seconds: 0.006, TFLOP/s/device: 2108.820, Tokens/s/device: 317864.349, total_weights: 65536, loss: 7.032, lm_loss: 7.032, perplexity: 1132.061
I0422 21:40:33.312323 138957515740992 metric_logger.py:196] completed step: 13, seconds: 0.557, TFLOP/s/device: 24.384, Tokens/s/device: 3675.448, total_weights: 65536, loss: 6.815, lm_loss: 6.815, perplexity: 911.031
I0422 21:40:33.470796 138957515740992 metric_logger.py:196] completed step: 14, seconds: 0.163, TFLOP/s/device: 83.175, Tokens/s/device: 12537.036, total_weights: 65536, loss: 6.635, lm_loss: 6.635, perplexity: 761.080
I0422 21:40:33.629552 138957515740992 metric_logger.py:196] completed step: 15, seconds: 0.158, TFLOP/s/device: 85.854, Tokens/s/device: 12940.812, total_weights: 65536, loss: 6.492, lm_loss: 6.492, perplexity: 659.915
I0422 21:40:33.787833 138957515740992 metric_logger.py:196] completed step: 16, seconds: 0.158, TFLOP/s/device: 85.843, Tokens/s/device: 12939.177, total_weights: 65536, loss: 6.380, lm_loss: 6.380, perplexity: 589.948
I0422 21:40:33.946089 138957515740992 metric_logger.py:196] completed step: 17, seconds: 0.159, TFLOP/s/device: 85.613, Tokens/s/device: 12904.527, total_weights: 65536, loss: 6.293, lm_loss: 6.293, perplexity: 540.941
I0422 21:40:34.104264 138957515740992 metric_logger.py:196] completed step: 18, seconds: 0.158, TFLOP/s/device: 85.804, Tokens/s/device: 12933.294, total_weights: 65536, loss: 6.223, lm_loss: 6.223, perplexity: 504.108
I0422 21:40:34.261949 138957515740992 checkpointing.py:772] Waiting for step 19 to finish before checkpoint...
I0422 21:40:34.262984 138957515740992 checkpointing.py:776] Waited 0.0010612010955810547 seconds for step 19 to finish before starting checkpointing.
I0422 21:40:34.265113 138957515740992 checkpoint_manager.py:2020] [process=5][thread=MainThread][step=10][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0422 21:40:38.462736 138823786809088 array_metadata_store.py:203] [process=5][thread=array_type_handler] Wrote 81 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_test_pipeline_scan_nnx_20260422_212603/linen_xpk_test_pipeline_scan_nnx_20260422_212603_05_fp8/checkpoints/10/items/array_metadatas/process_5
I0422 21:41:15.684344 138825959470848 base_pytree_checkpoint_handler.py:130] [process=5] /jax/orbax/write/gbytes_per_sec: 36.477 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 43.30888509750366 s) (per-host)
I0422 21:41:15.684470 138825959470848 async_checkpointer.py:90] [process=5][thread=async_save] 3 Handler Commit operations completed. Time taken: 43.262462s.
I0422 21:41:21.968394 138825959470848 async_checkpointer.py:160] [process=5][thread=async_save] Background save thread done. Time taken: 49.546371s.
I0422 21:41:21.968746 138824832276224 async_checkpointer.py:288] [process=5][thread=save_finalize] Done with waiting for background save thread=async_save.
I0422 21:41:21.968868 138824832276224 async_checkpointer.py:298] [process=5][thread=save_finalize] No errors found in background save thread=async_save.
I0422 21:41:21.968918 138824832276224 checkpoint_manager.py:2137] [process=5][thread=save_finalize][step=10] CheckpointManager Save Finalize is syncing with other hosts...
I0422 21:41:21.970463 138824832276224 checkpoint_manager.py:2146] [process=5][thread=save_finalize][step=10] CheckpointManager Save Finalize is done on all hosts.
I0422 21:41:21.970678 138957515740992 checkpoint_manager.py:2032] [process=5][thread=MainThread][step=10][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=10.
W0422 21:41:21.970808 138957515740992 checkpoint_manager.py:1452] Waiting for previous save to complete took 47.705699 seconds. If this number is high, consider checkpointing less frequently.
I0422 21:41:21.972391 138957515740992 checkpoint_manager.py:1512] [process=5] Saving checkpoint at step 19
I0422 21:41:21.974685 138957515740992 event_tracking.py:70] [process=5] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_test_pipeline_scan_nnx_20260422_212603/linen_xpk_test_pipeline_scan_nnx_20260422_212603_05_fp8/checkpoints/19.
I0422 21:41:22.675162 138957515740992 jax_array_handlers.py:360] Scheduling D2H of 81 prioritized jax.Array.
I0422 21:41:22.675264 138957515740992 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
I0422 21:41:22.712500 138957515740992 base_pytree_checkpoint_handler.py:154] [process=5][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.038276s
I0422 21:41:22.712693 138957515740992 base_pytree_checkpoint_handler.py:130] [process=5] /jax/orbax/write/blocking_gbytes_per_sec: 34.536 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.044670820236206055 s) (per-host)
I0422 21:41:22.712749 138957515740992 base_pytree_checkpoint_handler.py:768] [process=5][thread=MainThread] Initiated Pytree async_save. Time taken: 0.044739s (batch_requests_ready=0.003250s, total_serialization_initiated=0.041415s, others=0.000074s)
I0422 21:41:22.712881 138957515740992 composite_checkpoint_handler.py:715] [process=5][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.049057s (all_items=0.000010s, per_item={'items': '0.00001049'}, temp_paths=0.049046)
I0422 21:41:22.713594 138957515740992 event_tracking.py:125] [process=5] [async] Finished blocking save in 0.74 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_test_pipeline_scan_nnx_20260422_212603/linen_xpk_test_pipeline_scan_nnx_20260422_212603_05_fp8/checkpoints/19.
I0422 21:41:22.713925 138824832276224 async_checkpointer.py:76] [process=5][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-22 22:01:22.713888
I0422 21:41:22.718164 138957515740992 checkpoint_manager.py:1560] [process=5][thread=MainThread][step=19] Starting CheckpointManager Save Finalize thread=save_finalize
I0422 21:41:22.718471 138822721459968 async_checkpointer.py:280] [process=5][thread=save_finalize] Waiting for background save thread=async_save.
I0422 21:41:22.718629 138957515740992 standard_logger.py:34] {'step': 19, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_test_pipeline_scan_nnx_20260422_212603/linen_xpk_test_pipeline_scan_nnx_20260422_212603_05_fp8/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776894034.2650847, 'wait_for_prev_duration_secs': 47.7056987285614, 'time_between_consecutive_saves_sec': 2.5942773818969727, 'checkpointer_blocking_start_time': 1776894081.97243, 'checkpointer_blocking_duration_secs': 0.7416467666625977, 'get_old_steps_start_time': 1776894082.7140946, 'get_old_steps_duration_secs': 2.574920654296875e-05, 'checkpoint_manager_blocking_start_time': 1776894034.2632616, 'checkpoint_manager_blocking_duration_secs': 48.45533585548401}
I0422 21:41:22.718750 138957515740992 checkpointing.py:408] Started an asynchronous checkpoint save for step 19
I0422 21:41:22.718792 138957515740992 checkpoint_manager.py:2020] [process=5][thread=MainThread][step=19][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0422 21:41:28.260414 138823786809088 array_metadata_store.py:203] [process=5][thread=array_type_handler] Wrote 81 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_test_pipeline_scan_nnx_20260422_212603/linen_xpk_test_pipeline_scan_nnx_20260422_212603_05_fp8/checkpoints/19/items/array_metadatas/process_5
I0422 21:42:04.350531 138824832276224 base_pytree_checkpoint_handler.py:130] [process=5] /jax/orbax/write/gbytes_per_sec: 37.900 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 41.682472467422485 s) (per-host)
I0422 21:42:04.350648 138824832276224 async_checkpointer.py:90] [process=5][thread=async_save] 3 Handler Commit operations completed. Time taken: 41.636612s.
I0422 21:42:13.089827 138824832276224 async_checkpointer.py:160] [process=5][thread=async_save] Background save thread done. Time taken: 50.375775s.
I0422 21:42:13.090124 138822721459968 async_checkpointer.py:288] [process=5][thread=save_finalize] Done with waiting for background save thread=async_save.
I0422 21:42:13.090256 138822721459968 async_checkpointer.py:298] [process=5][thread=save_finalize] No errors found in background save thread=async_save.
I0422 21:42:13.090306 138822721459968 checkpoint_manager.py:2137] [process=5][thread=save_finalize][step=19] CheckpointManager Save Finalize is syncing with other hosts...
I0422 21:42:13.092910 138822721459968 checkpoint_manager.py:2146] [process=5][thread=save_finalize][step=19] CheckpointManager Save Finalize is done on all hosts.
I0422 21:42:13.093080 138957515740992 checkpoint_manager.py:2032] [process=5][thread=MainThread][step=19][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=19.
I0422 21:42:13.093228 138957515740992 checkpoint_manager.py:2009] [process=5][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0422 21:42:13.094208 138957515740992 metric_logger.py:196] completed step: 19, seconds: 0.158, TFLOP/s/device: 85.854, Tokens/s/device: 12940.894, 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: Wed Apr 22 21:42:25 UTC 2026
EXIT_CODE=0

Metrics: main (8a17c3d19) vs test/pipeline-scan-nnx (5b21b66fc)

Metrics: `main (8a17c3d19)` vs `test/pipeline-scan-nnx (5b21b66fc)`