Compare: 05_fp8 — MaxView

‹ 04_int8Case: 05_fp806_grad_accum ›

Metrics: `main (574ad3fb9)` vs `feat/nnx-trainstate-and-training-loop (574ad3fb9)`

Metric	main `574ad3fb9`	feat/nnx-trainstate-and-training-loop `574ad3fb9`	Diff (feat/nnx-trainstate-and-training-loop − main)
Parameters	1.105 billion	1.105 billion	—
Final loss	6.1670	6.1670	0
TFLOP/s	85.720	85.745	+0.025
Tok/s	12920.6	12924.5	+3.832
Avg s/step	1.633	1.706	+0.073
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	feat/nnx-trainstate-and-training-loop
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	feat/nnx-trainstate-and-training-loop
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_8(process=2,(0,2,0,0))	0.45	31.25	1.4%
TPU_9(process=2,(1,2,0,0))	0.45	31.25	1.4%
TPU_12(process=2,(0,3,0,0))	0.45	31.25	1.4%
TPU_13(process=2,(1,3,0,0))	0.45	31.25	1.4%

feat/nnx-trainstate-and-training-loop

Device	Used (GB)	Total (GB)	%
TPU_2(process=1,(2,0,0,0))	0.45	31.25	1.4%
TPU_3(process=1,(3,0,0,0))	0.45	31.25	1.4%
TPU_6(process=1,(2,1,0,0))	0.45	31.25	1.4%
TPU_7(process=1,(3,1,0,0))	0.45	31.25	1.4%

main TFLOP/s

feat/nnx-trainstate-and-training-loop TFLOP/s

Step	main Loss	feat/nnx-trainstate-and-training-loop Loss	main TFLOP/s	feat/nnx-trainstate-and-training-loop TFLOP/s	TFLOP/s Δ
0	10.874	10.874	0.902	0.913	+0.0110
1	10.874	10.874	7.954	6.685	-1.2690
2	10.816	10.816	302.582	308.616	+6.0340
3	10.431	10.431	30.445	31.425	+0.9800
4	9.992	9.992	82.674	82.983	+0.3090
5	9.549	9.549	86.022	85.909	-0.1130
6	9.111	9.111	42.718	42.685	-0.0330
7	8.685	8.685	0.558	0.562	+0.0040
8	8.281	8.281	82.328	82.981	+0.6530
9	7.908	7.908	85.775	85.974	+0.1990
10	7.568	7.568	85.709	85.715	+0.0060
11	7.288	7.288	7.571	4.379	-3.1920
12	7.032	7.032	2243.581	2199.632	-43.9490
13	6.815	6.815	84.922	85.576	+0.6540
14	6.635	6.635	23.683	85.775	+62.0920
15	6.492	6.492	83.252	24.403	-58.8490
16	6.380	6.380	85.827	83.248	-2.5790
17	6.293	6.293	85.731	85.834	+0.1030
18	6.223	6.223	85.931	85.828	-0.1030
19	6.167	6.167	85.720	85.745	+0.0250

main · 574ad3fb9 · main_20260420_060028 · full log

XPK Start: Mon Apr 20 06:13:39 UTC 2026
PyTorch was not found. Models won't be available and only tokenizers, configuration and file/data utilities can be used.
2026-04-20 06:14:03.882403: 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)
I0420 06:14:04.416744 140037021194048 max_utils.py:273] Attempting to initialize the jax distributed system...
I0420 06:14:13.456727 140037021194048 distributed.py:149] Starting JAX distributed service on [::]:8482
I0420 06:14:13.459094 140037021194048 distributed.py:172] Connecting to JAX distributed service on mt-05-fp8-2sk9q-slice-job-0-0.mt-05-fp8-2sk9q:8482
I0420 06:14:15.161294 140037021194048 max_utils.py:284] Jax distributed system initialized!
I0420 06:14:21.294174 140037021194048 max_utils.py:800] System Information: Jax Version: 0.9.2
I0420 06:14:21.294277 140037021194048 max_utils.py:801] System Information: Jaxlib Version: 0.9.2
I0420 06:14:21.294317 140037021194048 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
I0420 06:14:21.294351 140037021194048 train_utils.py:348] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing.
I0420 06:14:21.993407 140037021194048 maxtext_utils.py:1551] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0420 06:14:21.993689 140037021194048 checkpointing.py:677] Setting up checkpoint logger...
I0420 06:14:21.993742 140037021194048 checkpointing.py:233] Creating checkpoint manager with ocdbt=True and zarr3=True
I0420 06:14:21.993786 140037021194048 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None
I0420 06:14:21.994133 140037021194048 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 0x7f5c52bb1520>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB)
I0420 06:14:25.010873 140037021194048 checkpointing.py:265] Enabling policy for fixed interval checkpointing.
I0420 06:14:25.011113 140037021194048 checkpoint_manager.py:708] [process=2][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=('items',), item_handlers={'items': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7f47f43773e0>}, handler_registry=None
I0420 06:14:25.011356 140037021194048 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7f47f43773e0>` 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`.
I0420 06:14:25.011408 140037021194048 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7f47f4378a10>` 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`.
I0420 06:14:25.011445 140037021194048 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 0x7f47f43773e0>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7f47f43773e0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7f47f4378a10>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7f47f4378a10>}).
I0420 06:14:25.011762 140037021194048 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34
I0420 06:14:25.011832 140037021194048 async_checkpointer.py:192] [process=2][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7f46ec687740> timeout: 1200 secs and primary_host=0 for async checkpoint writes
I0420 06:14:25.786688 140037021194048 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_main_20260420_060028/linen_xpk_main_20260420_060028_05_fp8/checkpoints
I0420 06:14:26.203381 140037021194048 checkpoint_manager.py:929] [process=2][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_20260420_060028/linen_xpk_main_20260420_060028_05_fp8/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7f47f4376c00>
I0420 06:14:26.203548 140037021194048 checkpointing.py:301] Checkpoint manager created!
I0420 06:14:27.278738 140037021194048 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed').
I0420 06:14:27.278853 140037021194048 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0420 06:14:27.666759 140037021194048 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0420 06:14:27.666848 140037021194048 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0420 06:14:27.683507 140037021194048 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0420 06:14:27.683566 140037021194048 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0420 06:14:27.735884 140037021194048 attentions.py:1154] attentions/query Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0420 06:14:27.735962 140037021194048 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0420 06:14:27.752640 140037021194048 attentions.py:1155] attentions/key Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0420 06:14:27.752699 140037021194048 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0420 06:14:27.769283 140037021194048 attentions.py:1156] attentions/value Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0420 06:14:27.769342 140037021194048 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0420 06:14:27.794201 140037021194048 attentions.py:1197] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv').
I0420 06:14:27.794271 140037021194048 attentions.py:1197] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0420 06:14:27.845529 140037021194048 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp').
I0420 06:14:27.845605 140037021194048 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None).
I0420 06:14:28.285381 140037021194048 checkpointing.py:577] checkpoint manager exists so trying to load this run's existing checkpoint
I0420 06:14:28.285491 140037021194048 checkpointing.py:665] No existing checkpoints found, not restoring checkpoint.
fsdp: 32
I0420 06:14:30.204820 140037021194048 maxtext_utils.py:1654]  params/_overwrite_with_gradient/decoder/layers/mlp/wi_0/Fp8DirectDotGeneralOp_0/input_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0420 06:14:30.204937 140037021194048 maxtext_utils.py:1654]  params/_overwrite_with_gradient/decoder/layers/mlp/wi_0/Fp8DirectDotGeneralOp_0/input_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0420 06:14:30.204985 140037021194048 maxtext_utils.py:1654]  params/_overwrite_with_gradient/decoder/layers/mlp/wi_0/Fp8DirectDotGeneralOp_0/kernel_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0420 06:14:30.205049 140037021194048 maxtext_utils.py:1654]  params/_overwrite_with_gradient/decoder/layers/mlp/wi_0/Fp8DirectDotGeneralOp_0/kernel_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0420 06:14:30.205097 140037021194048 maxtext_utils.py:1654]  params/_overwrite_with_gradient/decoder/layers/mlp/wi_0/Fp8DirectDotGeneralOp_0/output_grad_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0420 06:14:30.205136 140037021194048 maxtext_utils.py:1654]  params/_overwrite_with_gradient/decoder/layers/mlp/wi_0/Fp8DirectDotGeneralOp_0/output_grad_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0420 06:14:30.205168 140037021194048 maxtext_utils.py:1654]  params/_overwrite_with_gradient/decoder/layers/mlp/wi_1/Fp8DirectDotGeneralOp_0/input_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0420 06:14:30.205199 140037021194048 maxtext_utils.py:1654]  params/_overwrite_with_gradient/decoder/layers/mlp/wi_1/Fp8DirectDotGeneralOp_0/input_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0420 06:14:30.205229 140037021194048 maxtext_utils.py:1654]  params/_overwrite_with_gradient/decoder/layers/mlp/wi_1/Fp8DirectDotGeneralOp_0/kernel_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0420 06:14:30.205258 140037021194048 maxtext_utils.py:1654]  params/_overwrite_with_gradient/decoder/layers/mlp/wi_1/Fp8DirectDotGeneralOp_0/kernel_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0420 06:14:30.205287 140037021194048 maxtext_utils.py:1654]  params/_overwrite_with_gradient/decoder/layers/mlp/wi_1/Fp8DirectDotGeneralOp_0/output_grad_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0420 06:14:30.205318 140037021194048 maxtext_utils.py:1654]  params/_overwrite_with_gradient/decoder/layers/mlp/wi_1/Fp8DirectDotGeneralOp_0/output_grad_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0420 06:14:30.205347 140037021194048 maxtext_utils.py:1654]  params/_overwrite_with_gradient/decoder/layers/mlp/wo/Fp8DirectDotGeneralOp_0/input_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0420 06:14:30.205375 140037021194048 maxtext_utils.py:1654]  params/_overwrite_with_gradient/decoder/layers/mlp/wo/Fp8DirectDotGeneralOp_0/input_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0420 06:14:30.205403 140037021194048 maxtext_utils.py:1654]  params/_overwrite_with_gradient/decoder/layers/mlp/wo/Fp8DirectDotGeneralOp_0/kernel_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0420 06:14:30.205430 140037021194048 maxtext_utils.py:1654]  params/_overwrite_with_gradient/decoder/layers/mlp/wo/Fp8DirectDotGeneralOp_0/kernel_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0420 06:14:30.205459 140037021194048 maxtext_utils.py:1654]  params/_overwrite_with_gradient/decoder/layers/mlp/wo/Fp8DirectDotGeneralOp_0/output_grad_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0420 06:14:30.205488 140037021194048 maxtext_utils.py:1654]  params/_overwrite_with_gradient/decoder/layers/mlp/wo/Fp8DirectDotGeneralOp_0/output_grad_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0420 06:14:30.205516 140037021194048 maxtext_utils.py:1654]  params/_overwrite_with_gradient/decoder/layers/self_attention/key/Fp8DirectDotGeneralOp_0/input_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0420 06:14:30.205544 140037021194048 maxtext_utils.py:1654]  params/_overwrite_with_gradient/decoder/layers/self_attention/key/Fp8DirectDotGeneralOp_0/input_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0420 06:14:30.205572 140037021194048 maxtext_utils.py:1654]  params/_overwrite_with_gradient/decoder/layers/self_attention/key/Fp8DirectDotGeneralOp_0/kernel_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0420 06:14:30.205599 140037021194048 maxtext_utils.py:1654]  params/_overwrite_with_gradient/decoder/layers/self_attention/key/Fp8DirectDotGeneralOp_0/kernel_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0420 06:14:30.205626 140037021194048 maxtext_utils.py:1654]  params/_overwrite_with_gradient/decoder/layers/self_attention/key/Fp8DirectDotGeneralOp_0/output_grad_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0420 06:14:30.205653 140037021194048 maxtext_utils.py:1654]  params/_overwrite_with_gradient/decoder/layers/self_attention/key/Fp8DirectDotGeneralOp_0/output_grad_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0420 06:14:30.205680 140037021194048 maxtext_utils.py:1654]  params/_overwrite_with_gradient/decoder/layers/self_attention/out/Fp8DirectDotGeneralOp_0/input_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0420 06:14:30.205707 140037021194048 maxtext_utils.py:1654]  params/_overwrite_with_gradient/decoder/layers/self_attention/out/Fp8DirectDotGeneralOp_0/input_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0420 06:14:30.205734 140037021194048 maxtext_utils.py:1654]  params/_overwrite_with_gradient/decoder/layers/self_attention/out/Fp8DirectDotGeneralOp_0/kernel_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0420 06:14:30.205761 140037021194048 maxtext_utils.py:1654]  params/_overwrite_with_gradient/decoder/layers/self_attention/out/Fp8DirectDotGeneralOp_0/kernel_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0420 06:14:30.205787 140037021194048 maxtext_utils.py:1654]  params/_overwrite_with_gradient/decoder/layers/self_attention/out/Fp8DirectDotGeneralOp_0/output_grad_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0420 06:14:30.205814 140037021194048 maxtext_utils.py:1654]  params/_overwrite_with_gradient/decoder/layers/self_attention/out/Fp8DirectDotGeneralOp_0/output_grad_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0420 06:14:30.205840 140037021194048 maxtext_utils.py:1654]  params/_overwrite_with_gradient/decoder/layers/self_attention/query/Fp8DirectDotGeneralOp_0/input_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0420 06:14:30.205866 140037021194048 maxtext_utils.py:1654]  params/_overwrite_with_gradient/decoder/layers/self_attention/query/Fp8DirectDotGeneralOp_0/input_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0420 06:14:30.205892 140037021194048 maxtext_utils.py:1654]  params/_overwrite_with_gradient/decoder/layers/self_attention/query/Fp8DirectDotGeneralOp_0/kernel_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0420 06:14:30.205919 140037021194048 maxtext_utils.py:1654]  params/_overwrite_with_gradient/decoder/layers/self_attention/query/Fp8DirectDotGeneralOp_0/kernel_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0420 06:14:30.205945 140037021194048 maxtext_utils.py:1654]  params/_overwrite_with_gradient/decoder/layers/self_attention/query/Fp8DirectDotGeneralOp_0/output_grad_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0420 06:14:30.205971 140037021194048 maxtext_utils.py:1654]  params/_overwrite_with_gradient/decoder/layers/self_attention/query/Fp8DirectDotGeneralOp_0/output_grad_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0420 06:14:30.205997 140037021194048 maxtext_utils.py:1654]  params/_overwrite_with_gradient/decoder/layers/self_attention/value/Fp8DirectDotGeneralOp_0/input_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0420 06:14:30.206054 140037021194048 maxtext_utils.py:1654]  params/_overwrite_with_gradient/decoder/layers/self_attention/value/Fp8DirectDotGeneralOp_0/input_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0420 06:14:30.206087 140037021194048 maxtext_utils.py:1654]  params/_overwrite_with_gradient/decoder/layers/self_attention/value/Fp8DirectDotGeneralOp_0/kernel_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0420 06:14:30.206119 140037021194048 maxtext_utils.py:1654]  params/_overwrite_with_gradient/decoder/layers/self_attention/value/Fp8DirectDotGeneralOp_0/kernel_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0420 06:14:30.206146 140037021194048 maxtext_utils.py:1654]  params/_overwrite_with_gradient/decoder/layers/self_attention/value/Fp8DirectDotGeneralOp_0/output_grad_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0420 06:14:30.206175 140037021194048 maxtext_utils.py:1654]  params/_overwrite_with_gradient/decoder/layers/self_attention/value/Fp8DirectDotGeneralOp_0/output_grad_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0420 06:14:30.206225 140037021194048 maxtext_utils.py:1654]  params/params/decoder/decoder_norm/scale
    Shape:     float32[2048]
    Logical:   P('norm',)
    Physical:  (None,)
I0420 06:14:30.206278 140037021194048 maxtext_utils.py:1654]  params/params/decoder/layers/mlp/wi_0/kernel
    Shape:     float32[2048,16,7168]
    Logical:   P('embed', 'layers', 'mlp')
    Physical:  ('fsdp', None, None)
I0420 06:14:30.206315 140037021194048 maxtext_utils.py:1654]  params/params/decoder/layers/mlp/wi_1/kernel
    Shape:     float32[2048,16,7168]
    Logical:   P('embed', 'layers', 'mlp')
    Physical:  ('fsdp', None, None)
I0420 06:14:30.206361 140037021194048 maxtext_utils.py:1654]  params/params/decoder/layers/mlp/wo/kernel
    Shape:     float32[7168,16,2048]
    Logical:   P('mlp', 'layers', 'embed')
    Physical:  (None, None, 'fsdp')
I0420 06:14:30.206406 140037021194048 maxtext_utils.py:1654]  params/params/decoder/layers/post_self_attention_layer_norm/scale
    Shape:     float32[2048,16]
    Logical:   P('norm', 'layers')
    Physical:  (None, None)
I0420 06:14:30.206438 140037021194048 maxtext_utils.py:1654]  params/params/decoder/layers/pre_self_attention_layer_norm/scale
    Shape:     float32[2048,16]
    Logical:   P('norm', 'layers')
    Physical:  (None, None)
I0420 06:14:30.206484 140037021194048 maxtext_utils.py:1654]  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)
I0420 06:14:30.206529 140037021194048 maxtext_utils.py:1654]  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')
I0420 06:14:30.206564 140037021194048 maxtext_utils.py:1654]  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)
I0420 06:14:30.206596 140037021194048 maxtext_utils.py:1654]  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)
I0420 06:14:30.206640 140037021194048 maxtext_utils.py:1654]  params/params/decoder/logits_dense/kernel
    Shape:     float32[2048,32000]
    Logical:   P('embed_vocab', 'vocab')
    Physical:  ('fsdp', None)
I0420 06:14:30.206683 140037021194048 maxtext_utils.py:1654]  params/params/token_embedder/embedding
    Shape:     float32[32000,2048]
    Logical:   P('vocab', 'embed_vocab')
    Physical:  (None, 'fsdp')

I0420 06:14:31.839607 140037021194048 train.py:155] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0420 06:14:31.839704 140037021194048 train.py:155] train/xent Physical: float32[32,2048]............................................ ('fsdp', None).
I0420 06:14:31.855167 140037021194048 train.py:162] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0420 06:14:31.855230 140037021194048 train.py:162] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None).
I0420 06:14:45.589514 140037021194048 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.
I0420 06:14:45.590441 140037021194048 metric_logger.py:301] number parameters: 1.105 billion
I0420 06:15:00.658443 140037021194048 checkpointing.py:772] Waiting for step 0 to finish before checkpoint...
I0420 06:15:00.829529 140037021194048 checkpointing.py:776] Waited 0.17106914520263672 seconds for step 0 to finish before starting checkpointing.
I0420 06:15:00.831979 140037021194048 checkpoint_manager.py:2009] [process=2][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0420 06:15:00.833924 140037021194048 checkpoint_manager.py:1512] [process=2] Saving checkpoint at step 0
I0420 06:15:00.835392 140037021194048 event_tracking.py:70] [process=2] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260420_060028/linen_xpk_main_20260420_060028_05_fp8/checkpoints/0.
I0420 06:15:01.625127 140037021194048 signaling_client.py:364] Using JaxDistributedSignalingClient
I0420 06:15:01.626078 140037021194048 jax_array_handlers.py:360] Scheduling D2H of 81 prioritized jax.Array.
I0420 06:15:01.626137 140037021194048 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
I0420 06:15:02.052099 140037021194048 base_pytree_checkpoint_handler.py:154] [process=2][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.427112s
I0420 06:15:02.052274 140037021194048 base_pytree_checkpoint_handler.py:130] [process=2] /jax/orbax/write/blocking_gbytes_per_sec: 3.541 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.435657262802124 s) (per-host)
I0420 06:15:02.052325 140037021194048 base_pytree_checkpoint_handler.py:768] [process=2][thread=MainThread] Initiated Pytree async_save. Time taken: 0.435719s (batch_requests_ready=0.003662s, total_serialization_initiated=0.431988s, others=0.000069s)
I0420 06:15:02.052433 140037021194048 composite_checkpoint_handler.py:715] [process=2][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.439816s (all_items=0.000017s, per_item={'items': '0.00001669'}, temp_paths=0.439799)
I0420 06:15:02.053315 140037021194048 event_tracking.py:125] [process=2] [async] Finished blocking save in 1.22 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260420_060028/linen_xpk_main_20260420_060028_05_fp8/checkpoints/0.
I0420 06:15:02.053607 139907302815488 async_checkpointer.py:76] [process=2][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-20 06:35:02.053581
I0420 06:15:02.055664 140037021194048 checkpoint_manager.py:1560] [process=2][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize
I0420 06:15:02.055979 139906814834432 async_checkpointer.py:280] [process=2][thread=save_finalize] Waiting for background save thread=async_save.
I0420 06:15:02.056122 140037021194048 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260420_060028/linen_xpk_main_20260420_060028_05_fp8/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776665700.8319607, 'wait_for_prev_duration_secs': 7.390975952148438e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776665700.8339627, 'checkpointer_blocking_duration_secs': 1.2197585105895996, 'get_old_steps_start_time': 1776665702.0537472, 'get_old_steps_duration_secs': 3.1948089599609375e-05, 'checkpoint_manager_blocking_start_time': 1776665700.8301563, 'checkpoint_manager_blocking_duration_secs': 1.2259268760681152}
I0420 06:15:02.056288 140037021194048 checkpointing.py:408] Started an asynchronous checkpoint save for step 0
I0420 06:15:02.056339 140037021194048 max_utils.py:750] 
Memstats: After params initialized:
I0420 06:15:02.056388 140037021194048 max_utils.py:756] 	Using (GB) 0.45 / 31.25 (1.440000%) on TPU_8(process=2,(0,2,0,0))
I0420 06:15:02.056422 140037021194048 max_utils.py:756] 	Using (GB) 0.45 / 31.25 (1.440000%) on TPU_9(process=2,(1,2,0,0))
I0420 06:15:02.056448 140037021194048 max_utils.py:756] 	Using (GB) 0.45 / 31.25 (1.440000%) on TPU_12(process=2,(0,3,0,0))
I0420 06:15:02.056472 140037021194048 max_utils.py:756] 	Using (GB) 0.45 / 31.25 (1.440000%) on TPU_13(process=2,(1,3,0,0))
I0420 06:15:02.368779 140037021194048 metric_logger.py:196] completed step: 0, seconds: 15.068, TFLOP/s/device: 0.902, Tokens/s/device: 135.918, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52776.805
I0420 06:15:02.570897 140037021194048 metric_logger.py:196] completed step: 1, seconds: 1.708, TFLOP/s/device: 7.954, Tokens/s/device: 1198.894, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52761.840
I0420 06:15:03.016955 140037021194048 metric_logger.py:196] completed step: 2, seconds: 0.045, TFLOP/s/device: 302.582, Tokens/s/device: 45608.409, total_weights: 65536, loss: 10.816, lm_loss: 10.816, perplexity: 49832.398
I0420 06:15:03.175469 140037021194048 metric_logger.py:196] completed step: 3, seconds: 0.446, TFLOP/s/device: 30.445, Tokens/s/device: 4589.027, total_weights: 65536, loss: 10.431, lm_loss: 10.431, perplexity: 33901.820
I0420 06:15:03.493142 140037021194048 metric_logger.py:196] completed step: 4, seconds: 0.164, TFLOP/s/device: 82.674, Tokens/s/device: 12461.590, total_weights: 65536, loss: 9.992, lm_loss: 9.992, perplexity: 21847.639
I0420 06:15:03.499568 140037021194048 metric_logger.py:196] completed step: 5, seconds: 0.158, TFLOP/s/device: 86.022, Tokens/s/device: 12966.211, total_weights: 65536, loss: 9.549, lm_loss: 9.549, perplexity: 14035.006
I0420 06:15:04.866975    2851 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com
I0420 06:15:07.250153 139912156043008 array_metadata_store.py:203] [process=2][thread=array_type_handler] Wrote 81 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260420_060028/linen_xpk_main_20260420_060028_05_fp8/checkpoints/0/items/array_metadatas/process_2
I0420 06:15:21.943785 139907302815488 base_pytree_checkpoint_handler.py:130] [process=2] /jax/orbax/write/gbytes_per_sec: 77.717 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 20.32711935043335 s) (per-host)
I0420 06:15:21.943903 139907302815488 async_checkpointer.py:90] [process=2][thread=async_save] 3 Handler Commit operations completed. Time taken: 19.890217s.
I0420 06:15:28.028983 140037021194048 metric_logger.py:196] completed step: 6, seconds: 0.318, TFLOP/s/device: 42.718, Tokens/s/device: 6438.976, total_weights: 65536, loss: 9.111, lm_loss: 9.111, perplexity: 9058.688
I0420 06:15:28.187443 140037021194048 metric_logger.py:196] completed step: 7, seconds: 24.371, TFLOP/s/device: 0.558, Tokens/s/device: 84.036, total_weights: 65536, loss: 8.685, lm_loss: 8.685, perplexity: 5914.720
I0420 06:15:28.345853 140037021194048 metric_logger.py:196] completed step: 8, seconds: 0.165, TFLOP/s/device: 82.328, Tokens/s/device: 12409.414, total_weights: 65536, loss: 8.281, lm_loss: 8.281, perplexity: 3950.010
I0420 06:15:28.350981 140037021194048 checkpointing.py:772] Waiting for step 10 to finish before checkpoint...
I0420 06:15:28.662614 140037021194048 checkpointing.py:776] Waited 0.3115963935852051 seconds for step 10 to finish before starting checkpointing.
I0420 06:15:28.665349 140037021194048 checkpoint_manager.py:2020] [process=2][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0420 06:15:29.798048 139907302815488 async_checkpointer.py:160] [process=2][thread=async_save] Background save thread done. Time taken: 27.744323s.
I0420 06:15:29.798338 139906814834432 async_checkpointer.py:288] [process=2][thread=save_finalize] Done with waiting for background save thread=async_save.
I0420 06:15:29.798454 139906814834432 async_checkpointer.py:298] [process=2][thread=save_finalize] No errors found in background save thread=async_save.
I0420 06:15:29.798504 139906814834432 checkpoint_manager.py:2137] [process=2][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts...
I0420 06:15:29.799930 139906814834432 checkpoint_manager.py:2146] [process=2][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts.
I0420 06:15:29.800114 140037021194048 checkpoint_manager.py:2032] [process=2][thread=MainThread][step=0][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=0.
W0420 06:15:29.800262 140037021194048 checkpoint_manager.py:1452] Waiting for previous save to complete took 1.134912 seconds. If this number is high, consider checkpointing less frequently.
I0420 06:15:29.801921 140037021194048 checkpoint_manager.py:1512] [process=2] Saving checkpoint at step 10
I0420 06:15:29.804118 140037021194048 event_tracking.py:70] [process=2] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260420_060028/linen_xpk_main_20260420_060028_05_fp8/checkpoints/10.
I0420 06:15:30.097207 140037021194048 jax_array_handlers.py:360] Scheduling D2H of 81 prioritized jax.Array.
I0420 06:15:30.097298 140037021194048 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
I0420 06:15:30.133331 140037021194048 base_pytree_checkpoint_handler.py:154] [process=2][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.037198s
I0420 06:15:30.133498 140037021194048 base_pytree_checkpoint_handler.py:130] [process=2] /jax/orbax/write/blocking_gbytes_per_sec: 35.583 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.04335618019104004 s) (per-host)
I0420 06:15:30.133548 140037021194048 base_pytree_checkpoint_handler.py:768] [process=2][thread=MainThread] Initiated Pytree async_save. Time taken: 0.043416s (batch_requests_ready=0.003028s, total_serialization_initiated=0.040322s, others=0.000066s)
I0420 06:15:30.133652 140037021194048 composite_checkpoint_handler.py:715] [process=2][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.047442s (all_items=0.000015s, per_item={'items': '0.00001502'}, temp_paths=0.047427)
I0420 06:15:30.134229 140037021194048 event_tracking.py:125] [process=2] [async] Finished blocking save in 0.33 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260420_060028/linen_xpk_main_20260420_060028_05_fp8/checkpoints/10.
I0420 06:15:30.134478 139906814834432 async_checkpointer.py:76] [process=2][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-20 06:35:30.134454
I0420 06:15:30.139755 140037021194048 checkpoint_manager.py:1560] [process=2][thread=MainThread][step=10] Starting CheckpointManager Save Finalize thread=save_finalize
I0420 06:15:30.139975 139904113694464 async_checkpointer.py:280] [process=2][thread=save_finalize] Waiting for background save thread=async_save.
I0420 06:15:30.140088 140037021194048 standard_logger.py:34] {'step': 10, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260420_060028/linen_xpk_main_20260420_060028_05_fp8/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776665728.6653187, 'wait_for_prev_duration_secs': 1.1349117755889893, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776665729.8019607, 'checkpointer_blocking_duration_secs': 0.33261919021606445, 'get_old_steps_start_time': 1776665730.1346004, 'get_old_steps_duration_secs': 2.574920654296875e-05, 'checkpoint_manager_blocking_start_time': 1776665728.6635242, 'checkpoint_manager_blocking_duration_secs': 1.4765315055847168}
I0420 06:15:30.140248 140037021194048 checkpointing.py:408] Started an asynchronous checkpoint save for step 10
I0420 06:15:30.140841 140037021194048 metric_logger.py:196] completed step: 9, seconds: 0.158, TFLOP/s/device: 85.775, Tokens/s/device: 12928.966, total_weights: 65536, loss: 7.908, lm_loss: 7.908, perplexity: 2720.116
I0420 06:15:30.147691 140037021194048 metric_logger.py:196] completed step: 10, seconds: 0.159, TFLOP/s/device: 85.709, Tokens/s/device: 12918.935, total_weights: 65536, loss: 7.568, lm_loss: 7.568, perplexity: 1936.198
I0420 06:15:30.306720 140037021194048 metric_logger.py:196] completed step: 11, seconds: 1.795, TFLOP/s/device: 7.571, Tokens/s/device: 1141.162, total_weights: 65536, loss: 7.288, lm_loss: 7.288, perplexity: 1463.272
I0420 06:15:30.464992 140037021194048 metric_logger.py:196] completed step: 12, seconds: 0.006, TFLOP/s/device: 2243.581, Tokens/s/device: 338177.015, total_weights: 65536, loss: 7.032, lm_loss: 7.032, perplexity: 1132.061
I0420 06:15:31.043534 140037021194048 metric_logger.py:196] completed step: 13, seconds: 0.160, TFLOP/s/device: 84.922, Tokens/s/device: 12800.320, total_weights: 65536, loss: 6.815, lm_loss: 6.815, perplexity: 911.031
I0420 06:15:31.201839 140037021194048 metric_logger.py:196] completed step: 14, seconds: 0.574, TFLOP/s/device: 23.683, Tokens/s/device: 3569.798, total_weights: 65536, loss: 6.635, lm_loss: 6.635, perplexity: 761.080
I0420 06:15:31.360439 140037021194048 metric_logger.py:196] completed step: 15, seconds: 0.163, TFLOP/s/device: 83.252, Tokens/s/device: 12548.712, total_weights: 65536, loss: 6.492, lm_loss: 6.492, perplexity: 659.915
I0420 06:15:31.518600 140037021194048 metric_logger.py:196] completed step: 16, seconds: 0.158, TFLOP/s/device: 85.827, Tokens/s/device: 12936.725, total_weights: 65536, loss: 6.380, lm_loss: 6.380, perplexity: 589.948
I0420 06:15:31.676953 140037021194048 metric_logger.py:196] completed step: 17, seconds: 0.158, TFLOP/s/device: 85.731, Tokens/s/device: 12922.277, total_weights: 65536, loss: 6.293, lm_loss: 6.293, perplexity: 540.941
I0420 06:15:31.835141 140037021194048 metric_logger.py:196] completed step: 18, seconds: 0.158, TFLOP/s/device: 85.931, Tokens/s/device: 12952.434, total_weights: 65536, loss: 6.223, lm_loss: 6.223, perplexity: 504.108
I0420 06:15:31.992768 140037021194048 checkpointing.py:772] Waiting for step 19 to finish before checkpoint...
I0420 06:15:31.993788 140037021194048 checkpointing.py:776] Waited 0.0010402202606201172 seconds for step 19 to finish before starting checkpointing.
I0420 06:15:31.995893 140037021194048 checkpoint_manager.py:2020] [process=2][thread=MainThread][step=10][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0420 06:15:36.468224 139904658958080 array_metadata_store.py:203] [process=2][thread=array_type_handler] Wrote 81 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260420_060028/linen_xpk_main_20260420_060028_05_fp8/checkpoints/10/items/array_metadatas/process_2
I0420 06:16:11.290865 139906814834432 base_pytree_checkpoint_handler.py:130] [process=2] /jax/orbax/write/gbytes_per_sec: 38.343 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 41.20066976547241 s) (per-host)
I0420 06:16:11.290987 139906814834432 async_checkpointer.py:90] [process=2][thread=async_save] 3 Handler Commit operations completed. Time taken: 41.156429s.
I0420 06:16:21.454650 139906814834432 async_checkpointer.py:160] [process=2][thread=async_save] Background save thread done. Time taken: 51.320077s.
I0420 06:16:21.454880 139904113694464 async_checkpointer.py:288] [process=2][thread=save_finalize] Done with waiting for background save thread=async_save.
I0420 06:16:21.454946 139904113694464 async_checkpointer.py:298] [process=2][thread=save_finalize] No errors found in background save thread=async_save.
I0420 06:16:21.454995 139904113694464 checkpoint_manager.py:2137] [process=2][thread=save_finalize][step=10] CheckpointManager Save Finalize is syncing with other hosts...
I0420 06:16:21.456567 139904113694464 checkpoint_manager.py:2146] [process=2][thread=save_finalize][step=10] CheckpointManager Save Finalize is done on all hosts.
I0420 06:16:21.456797 140037021194048 checkpoint_manager.py:2032] [process=2][thread=MainThread][step=10][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=10.
W0420 06:16:21.456935 140037021194048 checkpoint_manager.py:1452] Waiting for previous save to complete took 49.461046 seconds. If this number is high, consider checkpointing less frequently.
I0420 06:16:21.458297 140037021194048 checkpoint_manager.py:1512] [process=2] Saving checkpoint at step 19
I0420 06:16:21.460462 140037021194048 event_tracking.py:70] [process=2] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260420_060028/linen_xpk_main_20260420_060028_05_fp8/checkpoints/19.
I0420 06:16:21.754616 140037021194048 jax_array_handlers.py:360] Scheduling D2H of 81 prioritized jax.Array.
I0420 06:16:21.754703 140037021194048 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
I0420 06:16:21.792761 140037021194048 base_pytree_checkpoint_handler.py:154] [process=2][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.039031s
I0420 06:16:21.792926 140037021194048 base_pytree_checkpoint_handler.py:130] [process=2] /jax/orbax/write/blocking_gbytes_per_sec: 34.150 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.045174598693847656 s) (per-host)
I0420 06:16:21.792976 140037021194048 base_pytree_checkpoint_handler.py:768] [process=2][thread=MainThread] Initiated Pytree async_save. Time taken: 0.045234s (batch_requests_ready=0.003015s, total_serialization_initiated=0.042155s, others=0.000064s)
I0420 06:16:21.793107 140037021194048 composite_checkpoint_handler.py:715] [process=2][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.049412s (all_items=0.000012s, per_item={'items': '0.00001216'}, temp_paths=0.049400)
I0420 06:16:21.793838 140037021194048 event_tracking.py:125] [process=2] [async] Finished blocking save in 0.34 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260420_060028/linen_xpk_main_20260420_060028_05_fp8/checkpoints/19.
I0420 06:16:21.794126 139904113694464 async_checkpointer.py:76] [process=2][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-20 06:36:21.794093
I0420 06:16:21.797944 140037021194048 checkpoint_manager.py:1560] [process=2][thread=MainThread][step=19] Starting CheckpointManager Save Finalize thread=save_finalize
I0420 06:16:21.798261 139907302815488 async_checkpointer.py:280] [process=2][thread=save_finalize] Waiting for background save thread=async_save.
I0420 06:16:21.798426 140037021194048 standard_logger.py:34] {'step': 19, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260420_060028/linen_xpk_main_20260420_060028_05_fp8/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776665731.995865, 'wait_for_prev_duration_secs': 49.46104574203491, 'time_between_consecutive_saves_sec': 2.195894956588745, 'checkpointer_blocking_start_time': 1776665781.4583354, 'checkpointer_blocking_duration_secs': 0.3359408378601074, 'get_old_steps_start_time': 1776665781.7942963, 'get_old_steps_duration_secs': 2.4080276489257812e-05, 'checkpoint_manager_blocking_start_time': 1776665731.994081, 'checkpoint_manager_blocking_duration_secs': 49.80431127548218}
I0420 06:16:21.798557 140037021194048 checkpointing.py:408] Started an asynchronous checkpoint save for step 19
I0420 06:16:21.798599 140037021194048 checkpoint_manager.py:2020] [process=2][thread=MainThread][step=19][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0420 06:16:26.897582 139904658958080 array_metadata_store.py:203] [process=2][thread=array_type_handler] Wrote 81 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260420_060028/linen_xpk_main_20260420_060028_05_fp8/checkpoints/19/items/array_metadatas/process_2
I0420 06:17:03.216748 139904113694464 base_pytree_checkpoint_handler.py:130] [process=2] /jax/orbax/write/gbytes_per_sec: 38.095 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 41.46896767616272 s) (per-host)
I0420 06:17:03.216854 139904113694464 async_checkpointer.py:90] [process=2][thread=async_save] 3 Handler Commit operations completed. Time taken: 41.422607s.
I0420 06:17:12.819980 139904113694464 async_checkpointer.py:160] [process=2][thread=async_save] Background save thread done. Time taken: 51.025716s.
I0420 06:17:12.820342 139907302815488 async_checkpointer.py:288] [process=2][thread=save_finalize] Done with waiting for background save thread=async_save.
I0420 06:17:12.820502 139907302815488 async_checkpointer.py:298] [process=2][thread=save_finalize] No errors found in background save thread=async_save.
I0420 06:17:12.820544 139907302815488 checkpoint_manager.py:2137] [process=2][thread=save_finalize][step=19] CheckpointManager Save Finalize is syncing with other hosts...
I0420 06:17:12.821902 139907302815488 checkpoint_manager.py:2146] [process=2][thread=save_finalize][step=19] CheckpointManager Save Finalize is done on all hosts.
I0420 06:17:12.822093 140037021194048 checkpoint_manager.py:2032] [process=2][thread=MainThread][step=19][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=19.
I0420 06:17:12.822251 140037021194048 checkpoint_manager.py:2009] [process=2][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0420 06:17:12.823183 140037021194048 metric_logger.py:196] completed step: 19, seconds: 0.159, TFLOP/s/device: 85.720, Tokens/s/device: 12920.647, 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: Mon Apr 20 06:17:25 UTC 2026
EXIT_CODE=0

feat/nnx-trainstate-and-training-loop · 574ad3fb9 · feat_nnx_trainstate_and_training_loop_20260420_081533 · full log

XPK Start: Mon Apr 20 08:25:06 UTC 2026
PyTorch was not found. Models won't be available and only tokenizers, configuration and file/data utilities can be used.
2026-04-20 08:25:30.780483: 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)
I0420 08:25:30.962353 136472689899328 max_utils.py:273] Attempting to initialize the jax distributed system...
I0420 08:25:40.002446 136472689899328 distributed.py:149] Starting JAX distributed service on [::]:8482
I0420 08:25:40.004869 136472689899328 distributed.py:172] Connecting to JAX distributed service on mt-05-fp8-jd8bf-slice-job-0-0.mt-05-fp8-jd8bf:8482
I0420 08:25:41.315674 136472689899328 max_utils.py:284] Jax distributed system initialized!
I0420 08:25:47.454397 136472689899328 max_utils.py:800] System Information: Jax Version: 0.9.2
I0420 08:25:47.454499 136472689899328 max_utils.py:801] System Information: Jaxlib Version: 0.9.2
I0420 08:25:47.454539 136472689899328 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
I0420 08:25:47.454572 136472689899328 train_utils.py:378] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing.
I0420 08:25:48.150251 136472689899328 maxtext_utils.py:1718] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0420 08:25:48.150850 136472689899328 maxtext_utils.py:1718] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0420 08:25:48.151034 136472689899328 checkpointing.py:688] Setting up checkpoint logger...
I0420 08:25:48.151084 136472689899328 checkpointing.py:234] Creating checkpoint manager with ocdbt=True and zarr3=True
I0420 08:25:48.151126 136472689899328 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None
I0420 08:25:48.151450 136472689899328 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 0x7c1e4e648b90>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB)
I0420 08:25:51.022633 136472689899328 checkpointing.py:266] Enabling policy for fixed interval checkpointing.
I0420 08:25:51.022888 136472689899328 checkpoint_manager.py:708] [process=1][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=('items',), item_handlers={'items': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7c0a001321b0>}, handler_registry=None
I0420 08:25:51.023128 136472689899328 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7c0a001321b0>` 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`.
I0420 08:25:51.023176 136472689899328 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7c0a001d2a80>` 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`.
I0420 08:25:51.023213 136472689899328 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 0x7c0a001321b0>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7c0a001321b0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7c0a001d2a80>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7c0a001d2a80>}).
I0420 08:25:51.023526 136472689899328 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34
I0420 08:25:51.023597 136472689899328 async_checkpointer.py:192] [process=1][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7c09d03702c0> timeout: 1200 secs and primary_host=0 for async checkpoint writes
I0420 08:25:52.150454 136472689899328 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260420_081533/linen_xpk_feat_nnx_trainstate_and_training_loop_20260420_081533_05_fp8/checkpoints
I0420 08:25:52.597804 136472689899328 checkpoint_manager.py:929] [process=1][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_feat_nnx_trainstate_and_training_loop_20260420_081533/linen_xpk_feat_nnx_trainstate_and_training_loop_20260420_081533_05_fp8/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7c0a0012e570>
I0420 08:25:52.597975 136472689899328 checkpointing.py:302] Checkpoint manager created!
I0420 08:25:53.695462 136472689899328 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed').
I0420 08:25:53.695570 136472689899328 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0420 08:25:54.086839 136472689899328 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0420 08:25:54.086932 136472689899328 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0420 08:25:54.103026 136472689899328 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0420 08:25:54.103085 136472689899328 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0420 08:25:54.155392 136472689899328 attentions.py:1154] attentions/query Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0420 08:25:54.155469 136472689899328 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0420 08:25:54.171634 136472689899328 attentions.py:1155] attentions/key Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0420 08:25:54.171697 136472689899328 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0420 08:25:54.187833 136472689899328 attentions.py:1156] attentions/value Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0420 08:25:54.187892 136472689899328 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0420 08:25:54.212901 136472689899328 attentions.py:1197] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv').
I0420 08:25:54.212974 136472689899328 attentions.py:1197] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0420 08:25:54.264387 136472689899328 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp').
I0420 08:25:54.264464 136472689899328 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None).
I0420 08:25:54.706819 136472689899328 checkpointing.py:578] checkpoint manager exists so trying to load this run's existing checkpoint
I0420 08:25:54.706934 136472689899328 checkpointing.py:676] No existing checkpoints found, not restoring checkpoint.
fsdp: 32
I0420 08:25:56.636900 136472689899328 maxtext_utils.py:1821]  params/_overwrite_with_gradient/decoder/layers/mlp/wi_0/Fp8DirectDotGeneralOp_0/input_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0420 08:25:56.637023 136472689899328 maxtext_utils.py:1821]  params/_overwrite_with_gradient/decoder/layers/mlp/wi_0/Fp8DirectDotGeneralOp_0/input_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0420 08:25:56.637355 136472689899328 maxtext_utils.py:1821]  params/_overwrite_with_gradient/decoder/layers/mlp/wi_0/Fp8DirectDotGeneralOp_0/kernel_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0420 08:25:56.637441 136472689899328 maxtext_utils.py:1821]  params/_overwrite_with_gradient/decoder/layers/mlp/wi_0/Fp8DirectDotGeneralOp_0/kernel_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0420 08:25:56.637511 136472689899328 maxtext_utils.py:1821]  params/_overwrite_with_gradient/decoder/layers/mlp/wi_0/Fp8DirectDotGeneralOp_0/output_grad_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0420 08:25:56.637571 136472689899328 maxtext_utils.py:1821]  params/_overwrite_with_gradient/decoder/layers/mlp/wi_0/Fp8DirectDotGeneralOp_0/output_grad_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0420 08:25:56.637626 136472689899328 maxtext_utils.py:1821]  params/_overwrite_with_gradient/decoder/layers/mlp/wi_1/Fp8DirectDotGeneralOp_0/input_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0420 08:25:56.637679 136472689899328 maxtext_utils.py:1821]  params/_overwrite_with_gradient/decoder/layers/mlp/wi_1/Fp8DirectDotGeneralOp_0/input_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0420 08:25:56.637831 136472689899328 maxtext_utils.py:1821]  params/_overwrite_with_gradient/decoder/layers/mlp/wi_1/Fp8DirectDotGeneralOp_0/kernel_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0420 08:25:56.637889 136472689899328 maxtext_utils.py:1821]  params/_overwrite_with_gradient/decoder/layers/mlp/wi_1/Fp8DirectDotGeneralOp_0/kernel_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0420 08:25:56.637946 136472689899328 maxtext_utils.py:1821]  params/_overwrite_with_gradient/decoder/layers/mlp/wi_1/Fp8DirectDotGeneralOp_0/output_grad_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0420 08:25:56.638125 136472689899328 maxtext_utils.py:1821]  params/_overwrite_with_gradient/decoder/layers/mlp/wi_1/Fp8DirectDotGeneralOp_0/output_grad_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0420 08:25:56.638193 136472689899328 maxtext_utils.py:1821]  params/_overwrite_with_gradient/decoder/layers/mlp/wo/Fp8DirectDotGeneralOp_0/input_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0420 08:25:56.638247 136472689899328 maxtext_utils.py:1821]  params/_overwrite_with_gradient/decoder/layers/mlp/wo/Fp8DirectDotGeneralOp_0/input_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0420 08:25:56.638297 136472689899328 maxtext_utils.py:1821]  params/_overwrite_with_gradient/decoder/layers/mlp/wo/Fp8DirectDotGeneralOp_0/kernel_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0420 08:25:56.638346 136472689899328 maxtext_utils.py:1821]  params/_overwrite_with_gradient/decoder/layers/mlp/wo/Fp8DirectDotGeneralOp_0/kernel_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0420 08:25:56.638394 136472689899328 maxtext_utils.py:1821]  params/_overwrite_with_gradient/decoder/layers/mlp/wo/Fp8DirectDotGeneralOp_0/output_grad_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0420 08:25:56.638445 136472689899328 maxtext_utils.py:1821]  params/_overwrite_with_gradient/decoder/layers/mlp/wo/Fp8DirectDotGeneralOp_0/output_grad_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0420 08:25:56.638485 136472689899328 maxtext_utils.py:1821]  params/_overwrite_with_gradient/decoder/layers/self_attention/key/Fp8DirectDotGeneralOp_0/input_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0420 08:25:56.638526 136472689899328 maxtext_utils.py:1821]  params/_overwrite_with_gradient/decoder/layers/self_attention/key/Fp8DirectDotGeneralOp_0/input_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0420 08:25:56.638556 136472689899328 maxtext_utils.py:1821]  params/_overwrite_with_gradient/decoder/layers/self_attention/key/Fp8DirectDotGeneralOp_0/kernel_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0420 08:25:56.638583 136472689899328 maxtext_utils.py:1821]  params/_overwrite_with_gradient/decoder/layers/self_attention/key/Fp8DirectDotGeneralOp_0/kernel_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0420 08:25:56.638611 136472689899328 maxtext_utils.py:1821]  params/_overwrite_with_gradient/decoder/layers/self_attention/key/Fp8DirectDotGeneralOp_0/output_grad_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0420 08:25:56.638639 136472689899328 maxtext_utils.py:1821]  params/_overwrite_with_gradient/decoder/layers/self_attention/key/Fp8DirectDotGeneralOp_0/output_grad_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0420 08:25:56.638667 136472689899328 maxtext_utils.py:1821]  params/_overwrite_with_gradient/decoder/layers/self_attention/out/Fp8DirectDotGeneralOp_0/input_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0420 08:25:56.638695 136472689899328 maxtext_utils.py:1821]  params/_overwrite_with_gradient/decoder/layers/self_attention/out/Fp8DirectDotGeneralOp_0/input_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0420 08:25:56.638751 136472689899328 maxtext_utils.py:1821]  params/_overwrite_with_gradient/decoder/layers/self_attention/out/Fp8DirectDotGeneralOp_0/kernel_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0420 08:25:56.638806 136472689899328 maxtext_utils.py:1821]  params/_overwrite_with_gradient/decoder/layers/self_attention/out/Fp8DirectDotGeneralOp_0/kernel_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0420 08:25:56.638859 136472689899328 maxtext_utils.py:1821]  params/_overwrite_with_gradient/decoder/layers/self_attention/out/Fp8DirectDotGeneralOp_0/output_grad_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0420 08:25:56.638910 136472689899328 maxtext_utils.py:1821]  params/_overwrite_with_gradient/decoder/layers/self_attention/out/Fp8DirectDotGeneralOp_0/output_grad_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0420 08:25:56.638965 136472689899328 maxtext_utils.py:1821]  params/_overwrite_with_gradient/decoder/layers/self_attention/query/Fp8DirectDotGeneralOp_0/input_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0420 08:25:56.639013 136472689899328 maxtext_utils.py:1821]  params/_overwrite_with_gradient/decoder/layers/self_attention/query/Fp8DirectDotGeneralOp_0/input_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0420 08:25:56.639057 136472689899328 maxtext_utils.py:1821]  params/_overwrite_with_gradient/decoder/layers/self_attention/query/Fp8DirectDotGeneralOp_0/kernel_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0420 08:25:56.639105 136472689899328 maxtext_utils.py:1821]  params/_overwrite_with_gradient/decoder/layers/self_attention/query/Fp8DirectDotGeneralOp_0/kernel_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0420 08:25:56.639149 136472689899328 maxtext_utils.py:1821]  params/_overwrite_with_gradient/decoder/layers/self_attention/query/Fp8DirectDotGeneralOp_0/output_grad_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0420 08:25:56.639193 136472689899328 maxtext_utils.py:1821]  params/_overwrite_with_gradient/decoder/layers/self_attention/query/Fp8DirectDotGeneralOp_0/output_grad_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0420 08:25:56.639237 136472689899328 maxtext_utils.py:1821]  params/_overwrite_with_gradient/decoder/layers/self_attention/value/Fp8DirectDotGeneralOp_0/input_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0420 08:25:56.639281 136472689899328 maxtext_utils.py:1821]  params/_overwrite_with_gradient/decoder/layers/self_attention/value/Fp8DirectDotGeneralOp_0/input_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0420 08:25:56.639323 136472689899328 maxtext_utils.py:1821]  params/_overwrite_with_gradient/decoder/layers/self_attention/value/Fp8DirectDotGeneralOp_0/kernel_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0420 08:25:56.639367 136472689899328 maxtext_utils.py:1821]  params/_overwrite_with_gradient/decoder/layers/self_attention/value/Fp8DirectDotGeneralOp_0/kernel_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0420 08:25:56.639410 136472689899328 maxtext_utils.py:1821]  params/_overwrite_with_gradient/decoder/layers/self_attention/value/Fp8DirectDotGeneralOp_0/output_grad_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0420 08:25:56.639453 136472689899328 maxtext_utils.py:1821]  params/_overwrite_with_gradient/decoder/layers/self_attention/value/Fp8DirectDotGeneralOp_0/output_grad_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0420 08:25:56.639554 136472689899328 maxtext_utils.py:1821]  params/params/decoder/decoder_norm/scale
    Shape:     float32[2048]
    Logical:   P('norm',)
    Physical:  (None,)
I0420 08:25:56.639646 136472689899328 maxtext_utils.py:1821]  params/params/decoder/layers/mlp/wi_0/kernel
    Shape:     float32[2048,16,7168]
    Logical:   P('embed', 'layers', 'mlp')
    Physical:  ('fsdp', None, None)
I0420 08:25:56.639724 136472689899328 maxtext_utils.py:1821]  params/params/decoder/layers/mlp/wi_1/kernel
    Shape:     float32[2048,16,7168]
    Logical:   P('embed', 'layers', 'mlp')
    Physical:  ('fsdp', None, None)
I0420 08:25:56.639811 136472689899328 maxtext_utils.py:1821]  params/params/decoder/layers/mlp/wo/kernel
    Shape:     float32[7168,16,2048]
    Logical:   P('mlp', 'layers', 'embed')
    Physical:  (None, None, 'fsdp')
I0420 08:25:56.639899 136472689899328 maxtext_utils.py:1821]  params/params/decoder/layers/post_self_attention_layer_norm/scale
    Shape:     float32[2048,16]
    Logical:   P('norm', 'layers')
    Physical:  (None, None)
I0420 08:25:56.639963 136472689899328 maxtext_utils.py:1821]  params/params/decoder/layers/pre_self_attention_layer_norm/scale
    Shape:     float32[2048,16]
    Logical:   P('norm', 'layers')
    Physical:  (None, None)
I0420 08:25:56.640044 136472689899328 maxtext_utils.py:1821]  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)
I0420 08:25:56.640127 136472689899328 maxtext_utils.py:1821]  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')
I0420 08:25:56.640190 136472689899328 maxtext_utils.py:1821]  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)
I0420 08:25:56.640246 136472689899328 maxtext_utils.py:1821]  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)
I0420 08:25:56.640319 136472689899328 maxtext_utils.py:1821]  params/params/decoder/logits_dense/kernel
    Shape:     float32[2048,32000]
    Logical:   P('embed_vocab', 'vocab')
    Physical:  ('fsdp', None)
I0420 08:25:56.640392 136472689899328 maxtext_utils.py:1821]  params/params/token_embedder/embedding
    Shape:     float32[32000,2048]
    Logical:   P('vocab', 'embed_vocab')
    Physical:  (None, 'fsdp')

I0420 08:25:58.283374 136472689899328 train.py:157] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0420 08:25:58.283475 136472689899328 train.py:157] train/xent Physical: float32[32,2048]............................................ ('fsdp', None).
I0420 08:25:58.298795 136472689899328 train.py:164] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0420 08:25:58.298856 136472689899328 train.py:164] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None).
I0420 08:26:11.834925 136472689899328 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.
I0420 08:26:11.835855 136472689899328 metric_logger.py:301] number parameters: 1.105 billion
I0420 08:26:26.718356 136472689899328 checkpointing.py:794] Waiting for step 0 to finish before checkpoint...
I0420 08:26:27.252981 136472689899328 checkpointing.py:798] Waited 0.5346012115478516 seconds for step 0 to finish before starting checkpointing.
I0420 08:26:27.255528 136472689899328 checkpoint_manager.py:2009] [process=1][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0420 08:26:27.257462 136472689899328 checkpoint_manager.py:1512] [process=1] Saving checkpoint at step 0
I0420 08:26:27.258846 136472689899328 event_tracking.py:70] [process=1] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260420_081533/linen_xpk_feat_nnx_trainstate_and_training_loop_20260420_081533_05_fp8/checkpoints/0.
I0420 08:26:28.021653 136472689899328 signaling_client.py:364] Using JaxDistributedSignalingClient
I0420 08:26:28.022763 136472689899328 jax_array_handlers.py:360] Scheduling D2H of 81 prioritized jax.Array.
I0420 08:26:28.022881 136472689899328 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
I0420 08:26:28.438570 136472689899328 base_pytree_checkpoint_handler.py:154] [process=1][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.417035s
I0420 08:26:28.438759 136472689899328 base_pytree_checkpoint_handler.py:130] [process=1] /jax/orbax/write/blocking_gbytes_per_sec: 3.629 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.42511701583862305 s) (per-host)
I0420 08:26:28.438817 136472689899328 base_pytree_checkpoint_handler.py:768] [process=1][thread=MainThread] Initiated Pytree async_save. Time taken: 0.425184s (batch_requests_ready=0.003354s, total_serialization_initiated=0.421756s, others=0.000075s)
I0420 08:26:28.438925 136472689899328 composite_checkpoint_handler.py:715] [process=1][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.429308s (all_items=0.000017s, per_item={'items': '0.00001717'}, temp_paths=0.429291)
I0420 08:26:28.439642 136472689899328 event_tracking.py:125] [process=1] [async] Finished blocking save in 1.18 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260420_081533/linen_xpk_feat_nnx_trainstate_and_training_loop_20260420_081533_05_fp8/checkpoints/0.
I0420 08:26:28.440013 136343117551360 async_checkpointer.py:76] [process=1][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-20 08:46:28.439975
I0420 08:26:28.442065 136472689899328 checkpoint_manager.py:1560] [process=1][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize
I0420 08:26:28.442312 136343082473216 async_checkpointer.py:280] [process=1][thread=save_finalize] Waiting for background save thread=async_save.
I0420 08:26:28.442431 136472689899328 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260420_081533/linen_xpk_feat_nnx_trainstate_and_training_loop_20260420_081533_05_fp8/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776673587.2555106, 'wait_for_prev_duration_secs': 6.079673767089844e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776673587.2575004, 'checkpointer_blocking_duration_secs': 1.1826763153076172, 'get_old_steps_start_time': 1776673588.4402013, 'get_old_steps_duration_secs': 3.0040740966796875e-05, 'checkpoint_manager_blocking_start_time': 1776673587.25359, 'checkpoint_manager_blocking_duration_secs': 1.1888031959533691}
I0420 08:26:28.442538 136472689899328 checkpointing.py:409] Started an asynchronous checkpoint save for step 0
I0420 08:26:28.442592 136472689899328 max_utils.py:750] 
Memstats: After params initialized:
I0420 08:26:28.442644 136472689899328 max_utils.py:756] 	Using (GB) 0.45 / 31.25 (1.440000%) on TPU_2(process=1,(2,0,0,0))
I0420 08:26:28.442678 136472689899328 max_utils.py:756] 	Using (GB) 0.45 / 31.25 (1.440000%) on TPU_3(process=1,(3,0,0,0))
I0420 08:26:28.442719 136472689899328 max_utils.py:756] 	Using (GB) 0.45 / 31.25 (1.440000%) on TPU_6(process=1,(2,1,0,0))
I0420 08:26:28.442748 136472689899328 max_utils.py:756] 	Using (GB) 0.45 / 31.25 (1.440000%) on TPU_7(process=1,(3,1,0,0))
I0420 08:26:28.752869 136472689899328 metric_logger.py:196] completed step: 0, seconds: 14.882, TFLOP/s/device: 0.913, Tokens/s/device: 137.612, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52776.805
I0420 08:26:28.954339 136472689899328 metric_logger.py:196] completed step: 1, seconds: 2.032, TFLOP/s/device: 6.685, Tokens/s/device: 1007.636, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52761.840
I0420 08:26:29.386265 136472689899328 metric_logger.py:196] completed step: 2, seconds: 0.044, TFLOP/s/device: 308.616, Tokens/s/device: 46517.967, total_weights: 65536, loss: 10.816, lm_loss: 10.816, perplexity: 49832.398
I0420 08:26:29.544511 136472689899328 metric_logger.py:196] completed step: 3, seconds: 0.432, TFLOP/s/device: 31.425, Tokens/s/device: 4736.706, total_weights: 65536, loss: 10.431, lm_loss: 10.431, perplexity: 33901.820
I0420 08:26:29.862064 136472689899328 metric_logger.py:196] completed step: 4, seconds: 0.164, TFLOP/s/device: 82.983, Tokens/s/device: 12508.169, total_weights: 65536, loss: 9.992, lm_loss: 9.992, perplexity: 21847.639
I0420 08:26:29.868675 136472689899328 metric_logger.py:196] completed step: 5, seconds: 0.158, TFLOP/s/device: 85.909, Tokens/s/device: 12949.076, total_weights: 65536, loss: 9.549, lm_loss: 9.549, perplexity: 14035.006
I0420 08:26:31.466036    2859 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com
I0420 08:26:33.970749 136343090865920 array_metadata_store.py:203] [process=1][thread=array_type_handler] Wrote 81 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260420_081533/linen_xpk_feat_nnx_trainstate_and_training_loop_20260420_081533_05_fp8/checkpoints/0/items/array_metadatas/process_1
I0420 08:26:48.080006 136343117551360 base_pytree_checkpoint_handler.py:130] [process=1] /jax/orbax/write/gbytes_per_sec: 78.727 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 20.066319227218628 s) (per-host)
I0420 08:26:48.080152 136343117551360 async_checkpointer.py:90] [process=1][thread=async_save] 3 Handler Commit operations completed. Time taken: 19.640021s.
I0420 08:26:54.187602 136472689899328 metric_logger.py:196] completed step: 6, seconds: 0.318, TFLOP/s/device: 42.685, Tokens/s/device: 6433.939, total_weights: 65536, loss: 9.111, lm_loss: 9.111, perplexity: 9058.688
I0420 08:26:54.346222 136472689899328 metric_logger.py:196] completed step: 7, seconds: 24.162, TFLOP/s/device: 0.562, Tokens/s/device: 84.762, total_weights: 65536, loss: 8.685, lm_loss: 8.685, perplexity: 5914.720
I0420 08:26:54.504799 136472689899328 metric_logger.py:196] completed step: 8, seconds: 0.164, TFLOP/s/device: 82.981, Tokens/s/device: 12507.863, total_weights: 65536, loss: 8.281, lm_loss: 8.281, perplexity: 3950.010
I0420 08:26:54.509690 136472689899328 checkpointing.py:794] Waiting for step 10 to finish before checkpoint...
I0420 08:26:54.821978 136472689899328 checkpointing.py:798] Waited 0.31225085258483887 seconds for step 10 to finish before starting checkpointing.
I0420 08:26:54.824647 136472689899328 checkpoint_manager.py:2020] [process=1][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0420 08:26:56.815745 136343117551360 async_checkpointer.py:160] [process=1][thread=async_save] Background save thread done. Time taken: 28.375577s.
I0420 08:26:56.816029 136343082473216 async_checkpointer.py:288] [process=1][thread=save_finalize] Done with waiting for background save thread=async_save.
I0420 08:26:56.816153 136343082473216 async_checkpointer.py:298] [process=1][thread=save_finalize] No errors found in background save thread=async_save.
I0420 08:26:56.816222 136343082473216 checkpoint_manager.py:2137] [process=1][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts...
I0420 08:26:56.817849 136343082473216 checkpoint_manager.py:2146] [process=1][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts.
I0420 08:26:56.818081 136472689899328 checkpoint_manager.py:2032] [process=1][thread=MainThread][step=0][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=0.
W0420 08:26:56.818221 136472689899328 checkpoint_manager.py:1452] Waiting for previous save to complete took 1.993575 seconds. If this number is high, consider checkpointing less frequently.
I0420 08:26:56.820107 136472689899328 checkpoint_manager.py:1512] [process=1] Saving checkpoint at step 10
I0420 08:26:56.822127 136472689899328 event_tracking.py:70] [process=1] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260420_081533/linen_xpk_feat_nnx_trainstate_and_training_loop_20260420_081533_05_fp8/checkpoints/10.
I0420 08:26:57.550034 136472689899328 jax_array_handlers.py:360] Scheduling D2H of 81 prioritized jax.Array.
I0420 08:26:57.550128 136472689899328 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
I0420 08:26:57.603179 136472689899328 base_pytree_checkpoint_handler.py:154] [process=1][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.054171s
I0420 08:26:57.603340 136472689899328 base_pytree_checkpoint_handler.py:130] [process=1] /jax/orbax/write/blocking_gbytes_per_sec: 25.634 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.06018257141113281 s) (per-host)
I0420 08:26:57.603390 136472689899328 base_pytree_checkpoint_handler.py:768] [process=1][thread=MainThread] Initiated Pytree async_save. Time taken: 0.060240s (batch_requests_ready=0.003000s, total_serialization_initiated=0.057175s, others=0.000065s)
I0420 08:26:57.603520 136472689899328 composite_checkpoint_handler.py:715] [process=1][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.064485s (all_items=0.000015s, per_item={'items': '0.00001526'}, temp_paths=0.064469)
I0420 08:26:57.604362 136472689899328 event_tracking.py:125] [process=1] [async] Finished blocking save in 0.78 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260420_081533/linen_xpk_feat_nnx_trainstate_and_training_loop_20260420_081533_05_fp8/checkpoints/10.
I0420 08:26:57.604656 136343082473216 async_checkpointer.py:76] [process=1][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-20 08:46:57.604622
I0420 08:26:57.606690 136472689899328 checkpoint_manager.py:1560] [process=1][thread=MainThread][step=10] Starting CheckpointManager Save Finalize thread=save_finalize
I0420 08:26:57.606989 136340440082176 async_checkpointer.py:280] [process=1][thread=save_finalize] Waiting for background save thread=async_save.
I0420 08:26:57.607154 136472689899328 standard_logger.py:34] {'step': 10, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260420_081533/linen_xpk_feat_nnx_trainstate_and_training_loop_20260420_081533_05_fp8/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776673614.8246167, 'wait_for_prev_duration_secs': 1.993574619293213, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776673616.820147, 'checkpointer_blocking_duration_secs': 0.7846899032592773, 'get_old_steps_start_time': 1776673617.6048596, 'get_old_steps_duration_secs': 2.8371810913085938e-05, 'checkpoint_manager_blocking_start_time': 1776673614.8228188, 'checkpoint_manager_blocking_duration_secs': 2.784301519393921}
I0420 08:26:57.607270 136472689899328 checkpointing.py:409] Started an asynchronous checkpoint save for step 10
I0420 08:26:57.607901 136472689899328 metric_logger.py:196] completed step: 9, seconds: 0.158, TFLOP/s/device: 85.974, Tokens/s/device: 12958.909, total_weights: 65536, loss: 7.908, lm_loss: 7.908, perplexity: 2720.116
I0420 08:26:57.614884 136472689899328 metric_logger.py:196] completed step: 10, seconds: 0.159, TFLOP/s/device: 85.715, Tokens/s/device: 12919.831, total_weights: 65536, loss: 7.568, lm_loss: 7.568, perplexity: 1936.198
I0420 08:26:57.772824 136472689899328 metric_logger.py:196] completed step: 11, seconds: 3.103, TFLOP/s/device: 4.379, Tokens/s/device: 660.015, total_weights: 65536, loss: 7.288, lm_loss: 7.288, perplexity: 1463.272
I0420 08:26:57.931108 136472689899328 metric_logger.py:196] completed step: 12, seconds: 0.006, TFLOP/s/device: 2199.632, Tokens/s/device: 331552.534, total_weights: 65536, loss: 7.032, lm_loss: 7.032, perplexity: 1132.061
I0420 08:26:58.089493 136472689899328 metric_logger.py:196] completed step: 13, seconds: 0.159, TFLOP/s/device: 85.576, Tokens/s/device: 12898.919, total_weights: 65536, loss: 6.815, lm_loss: 6.815, perplexity: 911.031
I0420 08:26:58.651065 136472689899328 metric_logger.py:196] completed step: 14, seconds: 0.158, TFLOP/s/device: 85.775, Tokens/s/device: 12928.885, total_weights: 65536, loss: 6.635, lm_loss: 6.635, perplexity: 761.080
I0420 08:26:58.809426 136472689899328 metric_logger.py:196] completed step: 15, seconds: 0.557, TFLOP/s/device: 24.403, Tokens/s/device: 3678.227, total_weights: 65536, loss: 6.492, lm_loss: 6.492, perplexity: 659.915
I0420 08:26:58.967696 136472689899328 metric_logger.py:196] completed step: 16, seconds: 0.163, TFLOP/s/device: 83.248, Tokens/s/device: 12548.097, total_weights: 65536, loss: 6.380, lm_loss: 6.380, perplexity: 589.948
I0420 08:26:59.126142 136472689899328 metric_logger.py:196] completed step: 17, seconds: 0.158, TFLOP/s/device: 85.834, Tokens/s/device: 12937.787, total_weights: 65536, loss: 6.293, lm_loss: 6.293, perplexity: 540.941
I0420 08:26:59.284526 136472689899328 metric_logger.py:196] completed step: 18, seconds: 0.158, TFLOP/s/device: 85.828, Tokens/s/device: 12936.888, total_weights: 65536, loss: 6.223, lm_loss: 6.223, perplexity: 504.108
I0420 08:26:59.442191 136472689899328 checkpointing.py:794] Waiting for step 19 to finish before checkpoint...
I0420 08:26:59.443187 136472689899328 checkpointing.py:798] Waited 0.0010175704956054688 seconds for step 19 to finish before starting checkpointing.
I0420 08:26:59.445206 136472689899328 checkpoint_manager.py:2020] [process=1][thread=MainThread][step=10][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0420 08:27:03.269574 136339357947648 array_metadata_store.py:203] [process=1][thread=array_type_handler] Wrote 81 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260420_081533/linen_xpk_feat_nnx_trainstate_and_training_loop_20260420_081533_05_fp8/checkpoints/10/items/array_metadatas/process_1
I0420 08:27:40.348488 136343082473216 base_pytree_checkpoint_handler.py:130] [process=1] /jax/orbax/write/gbytes_per_sec: 36.906 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 42.80528688430786 s) (per-host)
I0420 08:27:40.348616 136343082473216 async_checkpointer.py:90] [process=1][thread=async_save] 3 Handler Commit operations completed. Time taken: 42.743808s.
I0420 08:27:49.099895 136343082473216 async_checkpointer.py:160] [process=1][thread=async_save] Background save thread done. Time taken: 51.495071s.
I0420 08:27:49.100138 136340440082176 async_checkpointer.py:288] [process=1][thread=save_finalize] Done with waiting for background save thread=async_save.
I0420 08:27:49.100201 136340440082176 async_checkpointer.py:298] [process=1][thread=save_finalize] No errors found in background save thread=async_save.
I0420 08:27:49.100249 136340440082176 checkpoint_manager.py:2137] [process=1][thread=save_finalize][step=10] CheckpointManager Save Finalize is syncing with other hosts...
I0420 08:27:49.101782 136340440082176 checkpoint_manager.py:2146] [process=1][thread=save_finalize][step=10] CheckpointManager Save Finalize is done on all hosts.
I0420 08:27:49.101958 136472689899328 checkpoint_manager.py:2032] [process=1][thread=MainThread][step=10][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=10.
W0420 08:27:49.102093 136472689899328 checkpoint_manager.py:1452] Waiting for previous save to complete took 49.656891 seconds. If this number is high, consider checkpointing less frequently.
I0420 08:27:49.103592 136472689899328 checkpoint_manager.py:1512] [process=1] Saving checkpoint at step 19
I0420 08:27:49.105576 136472689899328 event_tracking.py:70] [process=1] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260420_081533/linen_xpk_feat_nnx_trainstate_and_training_loop_20260420_081533_05_fp8/checkpoints/19.
I0420 08:27:49.402894 136472689899328 jax_array_handlers.py:360] Scheduling D2H of 81 prioritized jax.Array.
I0420 08:27:49.402990 136472689899328 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
I0420 08:27:49.458295 136472689899328 base_pytree_checkpoint_handler.py:154] [process=1][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.056282s
I0420 08:27:49.458451 136472689899328 base_pytree_checkpoint_handler.py:130] [process=1] /jax/orbax/write/blocking_gbytes_per_sec: 24.713 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.06242561340332031 s) (per-host)
I0420 08:27:49.458503 136472689899328 base_pytree_checkpoint_handler.py:768] [process=1][thread=MainThread] Initiated Pytree async_save. Time taken: 0.062486s (batch_requests_ready=0.003011s, total_serialization_initiated=0.059411s, others=0.000065s)
I0420 08:27:49.458618 136472689899328 composite_checkpoint_handler.py:715] [process=1][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.066627s (all_items=0.000011s, per_item={'items': '0.00001097'}, temp_paths=0.066616)
I0420 08:27:49.459431 136472689899328 event_tracking.py:125] [process=1] [async] Finished blocking save in 0.36 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260420_081533/linen_xpk_feat_nnx_trainstate_and_training_loop_20260420_081533_05_fp8/checkpoints/19.
I0420 08:27:49.459797 136340440082176 async_checkpointer.py:76] [process=1][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-20 08:47:49.459759
I0420 08:27:49.461848 136472689899328 checkpoint_manager.py:1560] [process=1][thread=MainThread][step=19] Starting CheckpointManager Save Finalize thread=save_finalize
I0420 08:27:49.462096 136339357947648 async_checkpointer.py:280] [process=1][thread=save_finalize] Waiting for background save thread=async_save.
I0420 08:27:49.462227 136472689899328 standard_logger.py:34] {'step': 19, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260420_081533/linen_xpk_feat_nnx_trainstate_and_training_loop_20260420_081533_05_fp8/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776673619.445179, 'wait_for_prev_duration_secs': 49.656890869140625, 'time_between_consecutive_saves_sec': 2.6272835731506348, 'checkpointer_blocking_start_time': 1776673669.1036305, 'checkpointer_blocking_duration_secs': 0.3563191890716553, 'get_old_steps_start_time': 1776673669.459968, 'get_old_steps_duration_secs': 2.5033950805664062e-05, 'checkpoint_manager_blocking_start_time': 1776673619.4434588, 'checkpoint_manager_blocking_duration_secs': 50.01873564720154}
I0420 08:27:49.462388 136472689899328 checkpointing.py:409] Started an asynchronous checkpoint save for step 19
I0420 08:27:49.462432 136472689899328 checkpoint_manager.py:2020] [process=1][thread=MainThread][step=19][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0420 08:27:54.557334 136338837862144 array_metadata_store.py:203] [process=1][thread=array_type_handler] Wrote 81 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260420_081533/linen_xpk_feat_nnx_trainstate_and_training_loop_20260420_081533_05_fp8/checkpoints/19/items/array_metadatas/process_1
I0420 08:28:30.764234 136340440082176 base_pytree_checkpoint_handler.py:130] [process=1] /jax/orbax/write/gbytes_per_sec: 38.188 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 41.36816334724426 s) (per-host)
I0420 08:28:30.764353 136340440082176 async_checkpointer.py:90] [process=1][thread=async_save] 3 Handler Commit operations completed. Time taken: 41.304442s.
I0420 08:28:39.532284 136340440082176 async_checkpointer.py:160] [process=1][thread=async_save] Background save thread done. Time taken: 50.072356s.
I0420 08:28:39.532608 136339357947648 async_checkpointer.py:288] [process=1][thread=save_finalize] Done with waiting for background save thread=async_save.
I0420 08:28:39.532761 136339357947648 async_checkpointer.py:298] [process=1][thread=save_finalize] No errors found in background save thread=async_save.
I0420 08:28:39.532805 136339357947648 checkpoint_manager.py:2137] [process=1][thread=save_finalize][step=19] CheckpointManager Save Finalize is syncing with other hosts...
I0420 08:28:39.534259 136339357947648 checkpoint_manager.py:2146] [process=1][thread=save_finalize][step=19] CheckpointManager Save Finalize is done on all hosts.
I0420 08:28:39.534436 136472689899328 checkpoint_manager.py:2032] [process=1][thread=MainThread][step=19][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=19.
I0420 08:28:39.534586 136472689899328 checkpoint_manager.py:2009] [process=1][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0420 08:28:39.535652 136472689899328 metric_logger.py:196] completed step: 19, seconds: 0.158, TFLOP/s/device: 85.745, Tokens/s/device: 12924.479, 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: Mon Apr 20 08:28:51 UTC 2026
EXIT_CODE=0

Metrics: main (574ad3fb9) vs feat/nnx-trainstate-and-training-loop (574ad3fb9)

Metrics: `main (574ad3fb9)` vs `feat/nnx-trainstate-and-training-loop (574ad3fb9)`