Compare: 05_fp8 — MaxView

‹ 04_int8Case: 05_fp806_grad_accum ›

Metrics: `main (586e69205)` vs `feat/nnx-post-train-fixes (7f06c99ac)`

Metric	main `586e69205`	feat/nnx-post-train-fixes `7f06c99ac`	Diff (feat/nnx-post-train-fixes − main)
Parameters	1.105 billion	1.105 billion	—
Final loss	6.1670	6.1670	0
TFLOP/s	85.865	85.739	-0.126
Tok/s	12942.4	12923.6	-18.866
Avg s/step	1.825	1.683	-0.142
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-post-train-fixes
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-post-train-fixes
Total	1.5 GB	1.5 GB
Output	0.4 GB	0.4 GB
Temp	1.1 GB	1.1 GB
Argument	0.4 GB	0.4 GB

main

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

feat/nnx-post-train-fixes

Device	Used (GB)	Total (GB)	%
TPU_18(process=5,(2,4,0,0))	0.45	31.25	1.4%
TPU_19(process=5,(3,4,0,0))	0.45	31.25	1.4%
TPU_22(process=5,(2,5,0,0))	0.45	31.25	1.4%
TPU_23(process=5,(3,5,0,0))	0.45	31.25	1.4%

main TFLOP/s

feat/nnx-post-train-fixes TFLOP/s

Step	main Loss	feat/nnx-post-train-fixes Loss	main TFLOP/s	feat/nnx-post-train-fixes TFLOP/s	TFLOP/s Δ
0	10.874	10.874	0.898	0.902	+0.0040
1	10.874	10.874	10.731	8.162	-2.5690
2	10.816	10.816	300.255	588.697	+288.4420
3	10.431	10.431	29.404	32.752	+3.3480
4	9.992	9.992	82.747	82.861	+0.1140
5	9.549	9.549	85.854	85.526	-0.3280
6	9.111	9.111	42.737	42.742	+0.0050
7	8.685	8.685	0.525	0.574	+0.0490
8	8.281	8.281	83.006	82.661	-0.3450
9	7.908	7.908	85.911	86.319	+0.4080
10	7.568	7.568	85.718	85.657	-0.0610
11	7.288	7.288	3.110	3.809	+0.6990
12	7.032	7.032	2121.000	2139.032	+18.0320
13	6.815	6.815	23.984	24.363	+0.3790
14	6.635	6.635	83.216	83.123	-0.0930
15	6.492	6.492	85.870	85.938	+0.0680
16	6.380	6.380	85.898	85.832	-0.0660
17	6.293	6.293	85.833	85.846	+0.0130
18	6.223	6.223	86.035	85.822	-0.2130
19	6.167	6.167	85.865	85.739	-0.1260

main · 586e69205 · main_20260423_071538 · full log

XPK Start: Thu Apr 23 07:31:34 UTC 2026
PyTorch was not found. Models won't be available and only tokenizers, configuration and file/data utilities can be used.
`rope_parameters`'s factor field must be a float >= 1, got 40
`rope_parameters`'s beta_fast field must be a float, got 32
`rope_parameters`'s beta_slow field must be a float, got 1
DeepseekV32Config got `key=rope_scaling` in kwargs but hasn't set it as attribute. For RoPE standardization you need to set `self.rope_parameters` in model's config. 
2026-04-23 07:31:59.656801: 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)
I0423 07:31:59.870465 139373294970688 max_utils.py:273] Attempting to initialize the jax distributed system...
I0423 07:32:08.920689 139373294970688 distributed.py:149] Starting JAX distributed service on [::]:8482
I0423 07:32:08.923003 139373294970688 distributed.py:172] Connecting to JAX distributed service on mt-05-fp8-alg8t-slice-job-0-0.mt-05-fp8-alg8t:8482
I0423 07:32:10.788594 139373294970688 max_utils.py:284] Jax distributed system initialized!
I0423 07:32:16.918651 139373294970688 max_utils.py:800] System Information: Jax Version: 0.9.2
I0423 07:32:16.918759 139373294970688 max_utils.py:801] System Information: Jaxlib Version: 0.9.2
I0423 07:32:16.918799 139373294970688 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
I0423 07:32:16.918837 139373294970688 train_utils.py:361] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing.
I0423 07:32:17.615948 139373294970688 maxtext_utils.py:1604] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0423 07:32:17.616255 139373294970688 checkpointing.py:677] Setting up checkpoint logger...
I0423 07:32:17.616314 139373294970688 checkpointing.py:233] Creating checkpoint manager with ocdbt=True and zarr3=True
I0423 07:32:17.616358 139373294970688 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None
I0423 07:32:17.616718 139373294970688 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 0x7ec1bf7ee6c0>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB)
I0423 07:32:20.997365 139373294970688 checkpointing.py:265] Enabling policy for fixed interval checkpointing.
I0423 07:32:20.997617 139373294970688 checkpoint_manager.py:708] [process=6][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=('items',), item_handlers={'items': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7ead88209430>}, handler_registry=None
I0423 07:32:20.997881 139373294970688 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7ead88209430>` 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`.
I0423 07:32:20.997934 139373294970688 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7ead880a89b0>` 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`.
I0423 07:32:20.997978 139373294970688 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 0x7ead88209430>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7ead88209430>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7ead880a89b0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7ead880a89b0>}).
I0423 07:32:20.998344 139373294970688 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34
I0423 07:32:20.998443 139373294970688 async_checkpointer.py:192] [process=6][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7ead607a96c0> timeout: 1200 secs and primary_host=0 for async checkpoint writes
I0423 07:32:21.929824 139373294970688 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_main_20260423_071538/linen_xpk_main_20260423_071538_05_fp8/checkpoints
I0423 07:32:22.342839 139373294970688 checkpoint_manager.py:929] [process=6][thread=MainThread] CheckpointManager created,  primary_host=0, CheckpointManagerOptions=CheckpointManagerOptions(save_interval_steps=1, max_to_keep=None, keep_time_interval=None, keep_period=None, should_keep_fn=None, best_fn=None, best_mode='max', keep_checkpoints_without_metrics=True, step_prefix=None, step_format_fixed_length=None, step_name_format=None, create=True, cleanup_tmp_directories=False, save_on_steps=frozenset(), single_host_load_and_broadcast=False, todelete_subdir=None, todelete_full_path=None, enable_background_delete=False, read_only=False, enable_async_checkpointing=True, async_options=None, multiprocessing_options=MultiprocessingOptions(primary_host=0, active_processes=None, barrier_sync_key_prefix=None), should_save_fn=None, file_options=FileOptions(path_permission_mode=None), save_root_metadata=True, temporary_path_class=None, save_decision_policy=FixedIntervalPolicy(interval=10), preservation_policy=LatestN(n=None), prevent_write_metrics=False, enable_should_save_is_saving_in_progress_check=True, enable_per_process_directory_creation=False, lightweight_initialize=False), root_directory=gs://lance-maxtext/linen_ckpt_xpk_main_20260423_071538/linen_xpk_main_20260423_071538_05_fp8/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7ead880a9ac0>
I0423 07:32:22.343015 139373294970688 checkpointing.py:301] Checkpoint manager created!
I0423 07:32:23.426989 139373294970688 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed').
I0423 07:32:23.427126 139373294970688 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0423 07:32:23.815335 139373294970688 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch_attn', 'activation_length_attn', 'activation_embed_attn').
I0423 07:32:23.815428 139373294970688 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0423 07:32:23.831877 139373294970688 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch_attn', 'activation_length_attn', 'activation_embed_attn').
I0423 07:32:23.831938 139373294970688 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0423 07:32:23.884871 139373294970688 attentions.py:1154] attentions/query Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_length_attn', 'activation_kv_heads', 'activation_kv_head_dim').
I0423 07:32:23.884955 139373294970688 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0423 07:32:23.901546 139373294970688 attentions.py:1155] attentions/key Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_length_attn', 'activation_kv_heads', 'activation_kv_head_dim').
I0423 07:32:23.901607 139373294970688 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0423 07:32:23.918048 139373294970688 attentions.py:1156] attentions/value Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_length_attn', 'activation_kv_heads', 'activation_kv_head_dim').
I0423 07:32:23.918122 139373294970688 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0423 07:32:23.943580 139373294970688 attentions.py:1198] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch_attn', 'activation_length_attn', 'activation_heads', 'activation_kv').
I0423 07:32:23.943653 139373294970688 attentions.py:1198] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0423 07:32:23.994808 139373294970688 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp').
I0423 07:32:23.994889 139373294970688 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None).
I0423 07:32:24.426899 139373294970688 checkpointing.py:577] checkpoint manager exists so trying to load this run's existing checkpoint
I0423 07:32:24.427016 139373294970688 checkpointing.py:665] No existing checkpoints found, not restoring checkpoint.
fsdp: 32
I0423 07:32:26.345986 139373294970688 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/mlp/wi_0/Fp8DirectDotGeneralOp_0/input_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0423 07:32:26.346121 139373294970688 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/mlp/wi_0/Fp8DirectDotGeneralOp_0/input_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0423 07:32:26.346169 139373294970688 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/mlp/wi_0/Fp8DirectDotGeneralOp_0/kernel_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0423 07:32:26.346214 139373294970688 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/mlp/wi_0/Fp8DirectDotGeneralOp_0/kernel_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0423 07:32:26.346251 139373294970688 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/mlp/wi_0/Fp8DirectDotGeneralOp_0/output_grad_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0423 07:32:26.346287 139373294970688 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/mlp/wi_0/Fp8DirectDotGeneralOp_0/output_grad_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0423 07:32:26.346320 139373294970688 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/mlp/wi_1/Fp8DirectDotGeneralOp_0/input_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0423 07:32:26.346351 139373294970688 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/mlp/wi_1/Fp8DirectDotGeneralOp_0/input_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0423 07:32:26.346382 139373294970688 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/mlp/wi_1/Fp8DirectDotGeneralOp_0/kernel_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0423 07:32:26.346411 139373294970688 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/mlp/wi_1/Fp8DirectDotGeneralOp_0/kernel_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0423 07:32:26.346441 139373294970688 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/mlp/wi_1/Fp8DirectDotGeneralOp_0/output_grad_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0423 07:32:26.346477 139373294970688 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/mlp/wi_1/Fp8DirectDotGeneralOp_0/output_grad_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0423 07:32:26.346511 139373294970688 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/mlp/wo/Fp8DirectDotGeneralOp_0/input_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0423 07:32:26.346543 139373294970688 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/mlp/wo/Fp8DirectDotGeneralOp_0/input_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0423 07:32:26.346573 139373294970688 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/mlp/wo/Fp8DirectDotGeneralOp_0/kernel_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0423 07:32:26.346601 139373294970688 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/mlp/wo/Fp8DirectDotGeneralOp_0/kernel_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0423 07:32:26.346629 139373294970688 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/mlp/wo/Fp8DirectDotGeneralOp_0/output_grad_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0423 07:32:26.346658 139373294970688 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/mlp/wo/Fp8DirectDotGeneralOp_0/output_grad_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0423 07:32:26.346686 139373294970688 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/self_attention/key/Fp8DirectDotGeneralOp_0/input_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0423 07:32:26.346713 139373294970688 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/self_attention/key/Fp8DirectDotGeneralOp_0/input_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0423 07:32:26.346740 139373294970688 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/self_attention/key/Fp8DirectDotGeneralOp_0/kernel_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0423 07:32:26.346768 139373294970688 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/self_attention/key/Fp8DirectDotGeneralOp_0/kernel_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0423 07:32:26.346795 139373294970688 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/self_attention/key/Fp8DirectDotGeneralOp_0/output_grad_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0423 07:32:26.346821 139373294970688 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/self_attention/key/Fp8DirectDotGeneralOp_0/output_grad_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0423 07:32:26.346848 139373294970688 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/self_attention/out/Fp8DirectDotGeneralOp_0/input_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0423 07:32:26.346874 139373294970688 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/self_attention/out/Fp8DirectDotGeneralOp_0/input_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0423 07:32:26.346901 139373294970688 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/self_attention/out/Fp8DirectDotGeneralOp_0/kernel_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0423 07:32:26.346927 139373294970688 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/self_attention/out/Fp8DirectDotGeneralOp_0/kernel_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0423 07:32:26.346954 139373294970688 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/self_attention/out/Fp8DirectDotGeneralOp_0/output_grad_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0423 07:32:26.346981 139373294970688 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/self_attention/out/Fp8DirectDotGeneralOp_0/output_grad_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0423 07:32:26.347010 139373294970688 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/self_attention/query/Fp8DirectDotGeneralOp_0/input_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0423 07:32:26.347037 139373294970688 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/self_attention/query/Fp8DirectDotGeneralOp_0/input_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0423 07:32:26.347064 139373294970688 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/self_attention/query/Fp8DirectDotGeneralOp_0/kernel_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0423 07:32:26.347091 139373294970688 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/self_attention/query/Fp8DirectDotGeneralOp_0/kernel_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0423 07:32:26.347133 139373294970688 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/self_attention/query/Fp8DirectDotGeneralOp_0/output_grad_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0423 07:32:26.347162 139373294970688 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/self_attention/query/Fp8DirectDotGeneralOp_0/output_grad_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0423 07:32:26.347188 139373294970688 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/self_attention/value/Fp8DirectDotGeneralOp_0/input_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0423 07:32:26.347247 139373294970688 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/self_attention/value/Fp8DirectDotGeneralOp_0/input_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0423 07:32:26.347299 139373294970688 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/self_attention/value/Fp8DirectDotGeneralOp_0/kernel_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0423 07:32:26.347333 139373294970688 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/self_attention/value/Fp8DirectDotGeneralOp_0/kernel_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0423 07:32:26.347362 139373294970688 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/self_attention/value/Fp8DirectDotGeneralOp_0/output_grad_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0423 07:32:26.347390 139373294970688 maxtext_utils.py:1707]  params/_overwrite_with_gradient/decoder/layers/self_attention/value/Fp8DirectDotGeneralOp_0/output_grad_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0423 07:32:26.347440 139373294970688 maxtext_utils.py:1707]  params/params/decoder/decoder_norm/scale
    Shape:     float32[2048]
    Logical:   P('norm',)
    Physical:  (None,)
I0423 07:32:26.347500 139373294970688 maxtext_utils.py:1707]  params/params/decoder/layers/mlp/wi_0/kernel
    Shape:     float32[2048,16,7168]
    Logical:   P('embed', 'layers', 'mlp')
    Physical:  ('fsdp', None, None)
I0423 07:32:26.347538 139373294970688 maxtext_utils.py:1707]  params/params/decoder/layers/mlp/wi_1/kernel
    Shape:     float32[2048,16,7168]
    Logical:   P('embed', 'layers', 'mlp')
    Physical:  ('fsdp', None, None)
I0423 07:32:26.347585 139373294970688 maxtext_utils.py:1707]  params/params/decoder/layers/mlp/wo/kernel
    Shape:     float32[7168,16,2048]
    Logical:   P('mlp', 'layers', 'embed')
    Physical:  (None, None, 'fsdp')
I0423 07:32:26.347630 139373294970688 maxtext_utils.py:1707]  params/params/decoder/layers/post_self_attention_layer_norm/scale
    Shape:     float32[2048,16]
    Logical:   P('norm', 'layers')
    Physical:  (None, None)
I0423 07:32:26.347663 139373294970688 maxtext_utils.py:1707]  params/params/decoder/layers/pre_self_attention_layer_norm/scale
    Shape:     float32[2048,16]
    Logical:   P('norm', 'layers')
    Physical:  (None, None)
I0423 07:32:26.347710 139373294970688 maxtext_utils.py:1707]  params/params/decoder/layers/self_attention/key/kernel
    Shape:     float32[2048,16,16,128]
    Logical:   P('embed', 'layers', 'kv_heads', 'kv_head_dim')
    Physical:  ('fsdp', None, None, None)
I0423 07:32:26.347757 139373294970688 maxtext_utils.py:1707]  params/params/decoder/layers/self_attention/out/kernel
    Shape:     float32[16,16,128,2048]
    Logical:   P('heads', 'layers', 'kv', 'embed')
    Physical:  (None, None, None, 'fsdp')
I0423 07:32:26.347794 139373294970688 maxtext_utils.py:1707]  params/params/decoder/layers/self_attention/query/kernel
    Shape:     float32[2048,16,16,128]
    Logical:   P('embed', 'layers', 'q_heads', 'kv')
    Physical:  ('fsdp', None, None, None)
I0423 07:32:26.347827 139373294970688 maxtext_utils.py:1707]  params/params/decoder/layers/self_attention/value/kernel
    Shape:     float32[2048,16,16,128]
    Logical:   P('embed', 'layers', 'kv_heads', 'kv_head_dim')
    Physical:  ('fsdp', None, None, None)
I0423 07:32:26.347871 139373294970688 maxtext_utils.py:1707]  params/params/decoder/logits_dense/kernel
    Shape:     float32[2048,32000]
    Logical:   P('embed_vocab', 'vocab')
    Physical:  ('fsdp', None)
I0423 07:32:26.347914 139373294970688 maxtext_utils.py:1707]  params/params/token_embedder/embedding
    Shape:     float32[32000,2048]
    Logical:   P('vocab', 'embed_vocab')
    Physical:  (None, 'fsdp')

I0423 07:32:27.969732 139373294970688 train.py:155] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0423 07:32:27.969832 139373294970688 train.py:155] train/xent Physical: float32[32,2048]............................................ ('fsdp', None).
I0423 07:32:27.985569 139373294970688 train.py:162] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0423 07:32:27.985629 139373294970688 train.py:162] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None).
I0423 07:32:41.867264 139373294970688 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.
I0423 07:32:41.868116 139373294970688 metric_logger.py:301] number parameters: 1.105 billion
I0423 07:32:57.006255 139373294970688 checkpointing.py:772] Waiting for step 0 to finish before checkpoint...
I0423 07:32:57.179013 139373294970688 checkpointing.py:776] Waited 0.17274188995361328 seconds for step 0 to finish before starting checkpointing.
I0423 07:32:57.181473 139373294970688 checkpoint_manager.py:2009] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0423 07:32:57.183639 139373294970688 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 0
I0423 07:32:57.185044 139373294970688 event_tracking.py:70] [process=6] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260423_071538/linen_xpk_main_20260423_071538_05_fp8/checkpoints/0.
I0423 07:32:57.536808 139373294970688 signaling_client.py:364] Using JaxDistributedSignalingClient
I0423 07:32:57.539361 139373294970688 jax_array_handlers.py:360] Scheduling D2H of 81 prioritized jax.Array.
I0423 07:32:57.539422 139373294970688 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
I0423 07:32:57.956130 139373294970688 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.419444s
I0423 07:32:57.956305 139373294970688 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 3.607 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.4277513027191162 s) (per-host)
I0423 07:32:57.956357 139373294970688 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.427814s (batch_requests_ready=0.003576s, total_serialization_initiated=0.424168s, others=0.000070s)
I0423 07:32:57.956454 139373294970688 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.432122s (all_items=0.000018s, per_item={'items': '0.00001812'}, temp_paths=0.432104)
I0423 07:32:57.957205 139373294970688 event_tracking.py:125] [process=6] [async] Finished blocking save in 0.77 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260423_071538/linen_xpk_main_20260423_071538_05_fp8/checkpoints/0.
I0423 07:32:57.957540 139242940339968 async_checkpointer.py:76] [process=6][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-23 07:52:57.957501
I0423 07:32:57.959526 139373294970688 checkpoint_manager.py:1560] [process=6][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize
I0423 07:32:57.959757 139242428688128 async_checkpointer.py:280] [process=6][thread=save_finalize] Waiting for background save thread=async_save.
I0423 07:32:57.959854 139373294970688 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260423_071538/linen_xpk_main_20260423_071538_05_fp8/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776929577.1814551, 'wait_for_prev_duration_secs': 6.818771362304688e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776929577.1836772, 'checkpointer_blocking_duration_secs': 0.774022102355957, 'get_old_steps_start_time': 1776929577.9577246, 'get_old_steps_duration_secs': 3.0279159545898438e-05, 'checkpoint_manager_blocking_start_time': 1776929577.1796067, 'checkpoint_manager_blocking_duration_secs': 0.7802093029022217}
I0423 07:32:57.960004 139373294970688 checkpointing.py:408] Started an asynchronous checkpoint save for step 0
I0423 07:32:57.960055 139373294970688 max_utils.py:750] 
Memstats: After params initialized:
I0423 07:32:57.960118 139373294970688 max_utils.py:756] 	Using (GB) 0.45 / 31.25 (1.440000%) on TPU_24(process=6,(0,6,0,0))
I0423 07:32:57.960156 139373294970688 max_utils.py:756] 	Using (GB) 0.45 / 31.25 (1.440000%) on TPU_25(process=6,(1,6,0,0))
I0423 07:32:57.960185 139373294970688 max_utils.py:756] 	Using (GB) 0.45 / 31.25 (1.440000%) on TPU_28(process=6,(0,7,0,0))
I0423 07:32:57.960210 139373294970688 max_utils.py:756] 	Using (GB) 0.45 / 31.25 (1.440000%) on TPU_29(process=6,(1,7,0,0))
I0423 07:32:58.274454 139373294970688 metric_logger.py:196] completed step: 0, seconds: 15.138, TFLOP/s/device: 0.898, Tokens/s/device: 135.288, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52776.805
I0423 07:32:58.477155 139373294970688 metric_logger.py:196] completed step: 1, seconds: 1.266, TFLOP/s/device: 10.731, Tokens/s/device: 1617.485, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52761.840
I0423 07:32:58.939247 139373294970688 metric_logger.py:196] completed step: 2, seconds: 0.045, TFLOP/s/device: 300.255, Tokens/s/device: 45257.668, total_weights: 65536, loss: 10.816, lm_loss: 10.816, perplexity: 49832.398
I0423 07:32:59.097347 139373294970688 metric_logger.py:196] completed step: 3, seconds: 0.462, TFLOP/s/device: 29.404, Tokens/s/device: 4432.066, total_weights: 65536, loss: 10.431, lm_loss: 10.431, perplexity: 33901.820
I0423 07:32:59.415153 139373294970688 metric_logger.py:196] completed step: 4, seconds: 0.164, TFLOP/s/device: 82.747, Tokens/s/device: 12472.518, total_weights: 65536, loss: 9.992, lm_loss: 9.992, perplexity: 21847.639
I0423 07:32:59.421382 139373294970688 metric_logger.py:196] completed step: 5, seconds: 0.158, TFLOP/s/device: 85.854, Tokens/s/device: 12940.894, total_weights: 65536, loss: 9.549, lm_loss: 9.549, perplexity: 14035.006
I0423 07:33:02.193426    2825 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com
I0423 07:33:04.708017 139242437080832 array_metadata_store.py:203] [process=6][thread=array_type_handler] Wrote 81 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260423_071538/linen_xpk_main_20260423_071538_05_fp8/checkpoints/0/items/array_metadatas/process_6
I0423 07:33:18.789169 139242940339968 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 74.305 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 21.26058292388916 s) (per-host)
I0423 07:33:18.789277 139242940339968 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 20.831622s.
I0423 07:33:25.474045 139373294970688 metric_logger.py:196] completed step: 6, seconds: 0.318, TFLOP/s/device: 42.737, Tokens/s/device: 6441.811, total_weights: 65536, loss: 9.111, lm_loss: 9.111, perplexity: 9058.688
I0423 07:33:25.632261 139373294970688 metric_logger.py:196] completed step: 7, seconds: 25.895, TFLOP/s/device: 0.525, Tokens/s/device: 79.088, total_weights: 65536, loss: 8.685, lm_loss: 8.685, perplexity: 5914.720
I0423 07:33:25.790932 139373294970688 metric_logger.py:196] completed step: 8, seconds: 0.164, TFLOP/s/device: 83.006, Tokens/s/device: 12511.607, total_weights: 65536, loss: 8.281, lm_loss: 8.281, perplexity: 3950.010
I0423 07:33:25.795874 139373294970688 checkpointing.py:772] Waiting for step 10 to finish before checkpoint...
I0423 07:33:26.107756 139373294970688 checkpointing.py:776] Waited 0.3118460178375244 seconds for step 10 to finish before starting checkpointing.
I0423 07:33:26.110548 139373294970688 checkpoint_manager.py:2020] [process=6][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0423 07:33:29.500606 139242940339968 async_checkpointer.py:160] [process=6][thread=async_save] Background save thread done. Time taken: 31.542934s.
I0423 07:33:29.500828 139242428688128 async_checkpointer.py:288] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save.
I0423 07:33:29.500886 139242428688128 async_checkpointer.py:298] [process=6][thread=save_finalize] No errors found in background save thread=async_save.
I0423 07:33:29.500942 139242428688128 checkpoint_manager.py:2137] [process=6][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts...
I0423 07:33:29.502681 139242428688128 checkpoint_manager.py:2146] [process=6][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts.
I0423 07:33:29.502840 139373294970688 checkpoint_manager.py:2032] [process=6][thread=MainThread][step=0][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=0.
W0423 07:33:29.502973 139373294970688 checkpoint_manager.py:1452] Waiting for previous save to complete took 3.392425 seconds. If this number is high, consider checkpointing less frequently.
I0423 07:33:29.505077 139373294970688 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 10
I0423 07:33:29.507053 139373294970688 event_tracking.py:70] [process=6] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260423_071538/linen_xpk_main_20260423_071538_05_fp8/checkpoints/10.
I0423 07:33:30.097562 139373294970688 jax_array_handlers.py:360] Scheduling D2H of 81 prioritized jax.Array.
I0423 07:33:30.097655 139373294970688 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
I0423 07:33:30.155774 139373294970688 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.059222s
I0423 07:33:30.155932 139373294970688 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 23.515 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.06560540199279785 s) (per-host)
I0423 07:33:30.155980 139373294970688 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.065662s (batch_requests_ready=0.003061s, total_serialization_initiated=0.062536s, others=0.000065s)
I0423 07:33:30.156073 139373294970688 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.069979s (all_items=0.000020s, per_item={'items': '0.00001955'}, temp_paths=0.069960)
I0423 07:33:30.156833 139373294970688 event_tracking.py:125] [process=6] [async] Finished blocking save in 0.65 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260423_071538/linen_xpk_main_20260423_071538_05_fp8/checkpoints/10.
I0423 07:33:30.157213 139239702370048 async_checkpointer.py:76] [process=6][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-23 07:53:30.157176
I0423 07:33:30.159186 139373294970688 checkpoint_manager.py:1560] [process=6][thread=MainThread][step=10] Starting CheckpointManager Save Finalize thread=save_finalize
I0423 07:33:30.159416 139242428688128 async_checkpointer.py:280] [process=6][thread=save_finalize] Waiting for background save thread=async_save.
I0423 07:33:30.159518 139373294970688 standard_logger.py:34] {'step': 10, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260423_071538/linen_xpk_main_20260423_071538_05_fp8/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776929606.1105175, 'wait_for_prev_duration_secs': 3.3924248218536377, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776929609.505127, 'checkpointer_blocking_duration_secs': 0.6522328853607178, 'get_old_steps_start_time': 1776929610.1573834, 'get_old_steps_duration_secs': 2.9325485229492188e-05, 'checkpoint_manager_blocking_start_time': 1776929606.1086085, 'checkpoint_manager_blocking_duration_secs': 4.050875663757324}
I0423 07:33:30.159680 139373294970688 checkpointing.py:408] Started an asynchronous checkpoint save for step 10
I0423 07:33:30.160352 139373294970688 metric_logger.py:196] completed step: 9, seconds: 0.158, TFLOP/s/device: 85.911, Tokens/s/device: 12949.486, total_weights: 65536, loss: 7.908, lm_loss: 7.908, perplexity: 2720.116
I0423 07:33:30.167417 139373294970688 metric_logger.py:196] completed step: 10, seconds: 0.159, TFLOP/s/device: 85.718, Tokens/s/device: 12920.402, total_weights: 65536, loss: 7.568, lm_loss: 7.568, perplexity: 1936.198
I0423 07:33:30.325752 139373294970688 metric_logger.py:196] completed step: 11, seconds: 4.369, TFLOP/s/device: 3.110, Tokens/s/device: 468.729, total_weights: 65536, loss: 7.288, lm_loss: 7.288, perplexity: 1463.272
I0423 07:33:30.896339 139373294970688 metric_logger.py:196] completed step: 12, seconds: 0.006, TFLOP/s/device: 2121.000, Tokens/s/device: 319700.281, total_weights: 65536, loss: 7.032, lm_loss: 7.032, perplexity: 1132.061
I0423 07:33:31.054591 139373294970688 metric_logger.py:196] completed step: 13, seconds: 0.567, TFLOP/s/device: 23.984, Tokens/s/device: 3615.130, total_weights: 65536, loss: 6.815, lm_loss: 6.815, perplexity: 911.031
I0423 07:33:31.212941 139373294970688 metric_logger.py:196] completed step: 14, seconds: 0.163, TFLOP/s/device: 83.216, Tokens/s/device: 12543.178, total_weights: 65536, loss: 6.635, lm_loss: 6.635, perplexity: 761.080
I0423 07:33:31.371157 139373294970688 metric_logger.py:196] completed step: 15, seconds: 0.158, TFLOP/s/device: 85.870, Tokens/s/device: 12943.266, total_weights: 65536, loss: 6.492, lm_loss: 6.492, perplexity: 659.915
I0423 07:33:31.529217 139373294970688 metric_logger.py:196] completed step: 16, seconds: 0.158, TFLOP/s/device: 85.898, Tokens/s/device: 12947.439, total_weights: 65536, loss: 6.380, lm_loss: 6.380, perplexity: 589.948
I0423 07:33:31.687371 139373294970688 metric_logger.py:196] completed step: 17, seconds: 0.158, TFLOP/s/device: 85.833, Tokens/s/device: 12937.624, total_weights: 65536, loss: 6.293, lm_loss: 6.293, perplexity: 540.941
I0423 07:33:31.845711 139373294970688 metric_logger.py:196] completed step: 18, seconds: 0.158, TFLOP/s/device: 86.035, Tokens/s/device: 12968.099, total_weights: 65536, loss: 6.223, lm_loss: 6.223, perplexity: 504.108
I0423 07:33:32.003240 139373294970688 checkpointing.py:772] Waiting for step 19 to finish before checkpoint...
I0423 07:33:32.004149 139373294970688 checkpointing.py:776] Waited 0.0009305477142333984 seconds for step 19 to finish before starting checkpointing.
I0423 07:33:32.006573 139373294970688 checkpoint_manager.py:2020] [process=6][thread=MainThread][step=10][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0423 07:33:37.373457 139240262870784 array_metadata_store.py:203] [process=6][thread=array_type_handler] Wrote 81 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260423_071538/linen_xpk_main_20260423_071538_05_fp8/checkpoints/10/items/array_metadatas/process_6
I0423 07:34:12.478358 139239702370048 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 37.269 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 42.38797974586487 s) (per-host)
I0423 07:34:12.478490 139239702370048 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 42.321160s.
I0423 07:34:22.808341 139239702370048 async_checkpointer.py:160] [process=6][thread=async_save] Background save thread done. Time taken: 52.650996s.
I0423 07:34:22.808561 139242428688128 async_checkpointer.py:288] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save.
I0423 07:34:22.808625 139242428688128 async_checkpointer.py:298] [process=6][thread=save_finalize] No errors found in background save thread=async_save.
I0423 07:34:22.808672 139242428688128 checkpoint_manager.py:2137] [process=6][thread=save_finalize][step=10] CheckpointManager Save Finalize is syncing with other hosts...
I0423 07:34:22.811248 139242428688128 checkpoint_manager.py:2146] [process=6][thread=save_finalize][step=10] CheckpointManager Save Finalize is done on all hosts.
I0423 07:34:22.811438 139373294970688 checkpoint_manager.py:2032] [process=6][thread=MainThread][step=10][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=10.
W0423 07:34:22.811580 139373294970688 checkpoint_manager.py:1452] Waiting for previous save to complete took 50.805012 seconds. If this number is high, consider checkpointing less frequently.
I0423 07:34:22.813238 139373294970688 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 19
I0423 07:34:22.815365 139373294970688 event_tracking.py:70] [process=6] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260423_071538/linen_xpk_main_20260423_071538_05_fp8/checkpoints/19.
I0423 07:34:23.520359 139373294970688 jax_array_handlers.py:360] Scheduling D2H of 81 prioritized jax.Array.
I0423 07:34:23.520456 139373294970688 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
I0423 07:34:23.578991 139373294970688 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.059677s
I0423 07:34:23.579174 139373294970688 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 23.439 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.06581950187683105 s) (per-host)
I0423 07:34:23.579225 139373294970688 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.065882s (batch_requests_ready=0.002995s, total_serialization_initiated=0.062820s, others=0.000067s)
I0423 07:34:23.579337 139373294970688 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.070251s (all_items=0.000012s, per_item={'items': '0.00001168'}, temp_paths=0.070239)
I0423 07:34:23.580023 139373294970688 event_tracking.py:125] [process=6] [async] Finished blocking save in 0.77 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260423_071538/linen_xpk_main_20260423_071538_05_fp8/checkpoints/19.
I0423 07:34:23.580396 139242428688128 async_checkpointer.py:76] [process=6][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-23 07:54:23.580352
I0423 07:34:23.582364 139373294970688 checkpoint_manager.py:1560] [process=6][thread=MainThread][step=19] Starting CheckpointManager Save Finalize thread=save_finalize
I0423 07:34:23.582636 139240262870784 async_checkpointer.py:280] [process=6][thread=save_finalize] Waiting for background save thread=async_save.
I0423 07:34:23.582767 139373294970688 standard_logger.py:34] {'step': 19, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260423_071538/linen_xpk_main_20260423_071538_05_fp8/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776929612.0065439, 'wait_for_prev_duration_secs': 50.80501198768616, 'time_between_consecutive_saves_sec': 2.5038228034973145, 'checkpointer_blocking_start_time': 1776929662.8132777, 'checkpointer_blocking_duration_secs': 0.7672295570373535, 'get_old_steps_start_time': 1776929663.580526, 'get_old_steps_duration_secs': 2.4557113647460938e-05, 'checkpoint_manager_blocking_start_time': 1776929612.0044262, 'checkpoint_manager_blocking_duration_secs': 51.57831048965454}
I0423 07:34:23.582931 139373294970688 checkpointing.py:408] Started an asynchronous checkpoint save for step 19
I0423 07:34:23.582974 139373294970688 checkpoint_manager.py:2020] [process=6][thread=MainThread][step=19][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0423 07:34:30.448322 139242395117312 array_metadata_store.py:203] [process=6][thread=array_type_handler] Wrote 81 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260423_071538/linen_xpk_main_20260423_071538_05_fp8/checkpoints/19/items/array_metadatas/process_6
I0423 07:35:05.930865 139242428688128 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 37.243 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 42.41748046875 s) (per-host)
I0423 07:35:05.930973 139242428688128 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 42.349913s.
I0423 07:35:15.871556 139242428688128 async_checkpointer.py:160] [process=6][thread=async_save] Background save thread done. Time taken: 52.290481s.
I0423 07:35:15.871846 139240262870784 async_checkpointer.py:288] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save.
I0423 07:35:15.871965 139240262870784 async_checkpointer.py:298] [process=6][thread=save_finalize] No errors found in background save thread=async_save.
I0423 07:35:15.872006 139240262870784 checkpoint_manager.py:2137] [process=6][thread=save_finalize][step=19] CheckpointManager Save Finalize is syncing with other hosts...
I0423 07:35:15.874220 139240262870784 checkpoint_manager.py:2146] [process=6][thread=save_finalize][step=19] CheckpointManager Save Finalize is done on all hosts.
I0423 07:35:15.874398 139373294970688 checkpoint_manager.py:2032] [process=6][thread=MainThread][step=19][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=19.
I0423 07:35:15.874552 139373294970688 checkpoint_manager.py:2009] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0423 07:35:15.875467 139373294970688 metric_logger.py:196] completed step: 19, seconds: 0.158, TFLOP/s/device: 85.865, Tokens/s/device: 12942.448, 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: Thu Apr 23 07:35:24 UTC 2026
EXIT_CODE=0

feat/nnx-post-train-fixes · 7f06c99ac · feat_nnx_post_train_fixes_20260423_124541 · full log

XPK Start: Thu Apr 23 12:57:23 UTC 2026
PyTorch was not found. Models won't be available and only tokenizers, configuration and file/data utilities can be used.
`rope_parameters`'s factor field must be a float >= 1, got 40
`rope_parameters`'s beta_fast field must be a float, got 32
`rope_parameters`'s beta_slow field must be a float, got 1
DeepseekV32Config got `key=rope_scaling` in kwargs but hasn't set it as attribute. For RoPE standardization you need to set `self.rope_parameters` in model's config. 
2026-04-23 12:57:48.493307: 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)
I0423 12:57:48.707322 137766256609088 max_utils.py:273] Attempting to initialize the jax distributed system...
I0423 12:57:57.748845 137766256609088 distributed.py:149] Starting JAX distributed service on [::]:8482
I0423 12:57:57.751228 137766256609088 distributed.py:172] Connecting to JAX distributed service on mt-05-fp8-3kov9-slice-job-0-0.mt-05-fp8-3kov9:8482
I0423 12:57:58.952524 137766256609088 max_utils.py:284] Jax distributed system initialized!
I0423 12:58:05.234977 137766256609088 max_utils.py:800] System Information: Jax Version: 0.9.2
I0423 12:58:05.235082 137766256609088 max_utils.py:801] System Information: Jaxlib Version: 0.9.2
I0423 12:58:05.235122 137766256609088 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
I0423 12:58:05.235159 137766256609088 train_utils.py:391] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing.
I0423 12:58:05.945783 137766256609088 maxtext_utils.py:1771] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0423 12:58:05.946375 137766256609088 maxtext_utils.py:1771] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0423 12:58:05.946556 137766256609088 checkpointing.py:688] Setting up checkpoint logger...
I0423 12:58:05.946607 137766256609088 checkpointing.py:234] Creating checkpoint manager with ocdbt=True and zarr3=True
I0423 12:58:05.946659 137766256609088 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None
I0423 12:58:05.947002 137766256609088 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 0x7d4b96d181d0>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB)
I0423 12:58:08.786869 137766256609088 checkpointing.py:266] Enabling policy for fixed interval checkpointing.
I0423 12:58:08.787099 137766256609088 checkpoint_manager.py:708] [process=5][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=('items',), item_handlers={'items': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7d3800712b10>}, handler_registry=None
I0423 12:58:08.787353 137766256609088 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7d3800712b10>` 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`.
I0423 12:58:08.787405 137766256609088 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7d3818264ec0>` 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`.
I0423 12:58:08.787441 137766256609088 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 0x7d3800712b10>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7d3800712b10>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7d3818264ec0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7d3818264ec0>}).
I0423 12:58:08.787789 137766256609088 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34
I0423 12:58:08.787861 137766256609088 async_checkpointer.py:192] [process=5][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7d37e8459d00> timeout: 1200 secs and primary_host=0 for async checkpoint writes
I0423 12:58:10.368114 137766256609088 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260423_124541/linen_xpk_feat_nnx_post_train_fixes_20260423_124541_05_fp8/checkpoints
I0423 12:58:10.820600 137766256609088 checkpoint_manager.py:929] [process=5][thread=MainThread] CheckpointManager created,  primary_host=0, CheckpointManagerOptions=CheckpointManagerOptions(save_interval_steps=1, max_to_keep=None, keep_time_interval=None, keep_period=None, should_keep_fn=None, best_fn=None, best_mode='max', keep_checkpoints_without_metrics=True, step_prefix=None, step_format_fixed_length=None, step_name_format=None, create=True, cleanup_tmp_directories=False, save_on_steps=frozenset(), single_host_load_and_broadcast=False, todelete_subdir=None, todelete_full_path=None, enable_background_delete=False, read_only=False, enable_async_checkpointing=True, async_options=None, multiprocessing_options=MultiprocessingOptions(primary_host=0, active_processes=None, barrier_sync_key_prefix=None), should_save_fn=None, file_options=FileOptions(path_permission_mode=None), save_root_metadata=True, temporary_path_class=None, save_decision_policy=FixedIntervalPolicy(interval=10), preservation_policy=LatestN(n=None), prevent_write_metrics=False, enable_should_save_is_saving_in_progress_check=True, enable_per_process_directory_creation=False, lightweight_initialize=False), root_directory=gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260423_124541/linen_xpk_feat_nnx_post_train_fixes_20260423_124541_05_fp8/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7d4a70e382f0>
I0423 12:58:10.820804 137766256609088 checkpointing.py:302] Checkpoint manager created!
I0423 12:58:11.940890 137766256609088 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed').
I0423 12:58:11.941000 137766256609088 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0423 12:58:12.330400 137766256609088 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0423 12:58:12.330498 137766256609088 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0423 12:58:12.347068 137766256609088 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0423 12:58:12.347124 137766256609088 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0423 12:58:12.400186 137766256609088 attentions.py:1154] attentions/query Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0423 12:58:12.400277 137766256609088 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0423 12:58:12.417085 137766256609088 attentions.py:1155] attentions/key Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0423 12:58:12.417146 137766256609088 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0423 12:58:12.433761 137766256609088 attentions.py:1156] attentions/value Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim').
I0423 12:58:12.433818 137766256609088 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0423 12:58:12.459330 137766256609088 attentions.py:1198] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv').
I0423 12:58:12.459401 137766256609088 attentions.py:1198] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0423 12:58:12.511096 137766256609088 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp').
I0423 12:58:12.511172 137766256609088 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None).
I0423 12:58:12.961747 137766256609088 checkpointing.py:578] checkpoint manager exists so trying to load this run's existing checkpoint
I0423 12:58:12.961873 137766256609088 checkpointing.py:676] No existing checkpoints found, not restoring checkpoint.
fsdp: 32
I0423 12:58:14.890910 137766256609088 maxtext_utils.py:1880]  params/_overwrite_with_gradient/decoder/layers/mlp/wi_0/Fp8DirectDotGeneralOp_0/input_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0423 12:58:14.891032 137766256609088 maxtext_utils.py:1880]  params/_overwrite_with_gradient/decoder/layers/mlp/wi_0/Fp8DirectDotGeneralOp_0/input_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0423 12:58:14.891083 137766256609088 maxtext_utils.py:1880]  params/_overwrite_with_gradient/decoder/layers/mlp/wi_0/Fp8DirectDotGeneralOp_0/kernel_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0423 12:58:14.891124 137766256609088 maxtext_utils.py:1880]  params/_overwrite_with_gradient/decoder/layers/mlp/wi_0/Fp8DirectDotGeneralOp_0/kernel_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0423 12:58:14.891160 137766256609088 maxtext_utils.py:1880]  params/_overwrite_with_gradient/decoder/layers/mlp/wi_0/Fp8DirectDotGeneralOp_0/output_grad_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0423 12:58:14.891202 137766256609088 maxtext_utils.py:1880]  params/_overwrite_with_gradient/decoder/layers/mlp/wi_0/Fp8DirectDotGeneralOp_0/output_grad_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0423 12:58:14.891236 137766256609088 maxtext_utils.py:1880]  params/_overwrite_with_gradient/decoder/layers/mlp/wi_1/Fp8DirectDotGeneralOp_0/input_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0423 12:58:14.891269 137766256609088 maxtext_utils.py:1880]  params/_overwrite_with_gradient/decoder/layers/mlp/wi_1/Fp8DirectDotGeneralOp_0/input_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0423 12:58:14.891301 137766256609088 maxtext_utils.py:1880]  params/_overwrite_with_gradient/decoder/layers/mlp/wi_1/Fp8DirectDotGeneralOp_0/kernel_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0423 12:58:14.891332 137766256609088 maxtext_utils.py:1880]  params/_overwrite_with_gradient/decoder/layers/mlp/wi_1/Fp8DirectDotGeneralOp_0/kernel_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0423 12:58:14.891361 137766256609088 maxtext_utils.py:1880]  params/_overwrite_with_gradient/decoder/layers/mlp/wi_1/Fp8DirectDotGeneralOp_0/output_grad_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0423 12:58:14.891395 137766256609088 maxtext_utils.py:1880]  params/_overwrite_with_gradient/decoder/layers/mlp/wi_1/Fp8DirectDotGeneralOp_0/output_grad_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0423 12:58:14.891426 137766256609088 maxtext_utils.py:1880]  params/_overwrite_with_gradient/decoder/layers/mlp/wo/Fp8DirectDotGeneralOp_0/input_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0423 12:58:14.891457 137766256609088 maxtext_utils.py:1880]  params/_overwrite_with_gradient/decoder/layers/mlp/wo/Fp8DirectDotGeneralOp_0/input_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0423 12:58:14.891486 137766256609088 maxtext_utils.py:1880]  params/_overwrite_with_gradient/decoder/layers/mlp/wo/Fp8DirectDotGeneralOp_0/kernel_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0423 12:58:14.891515 137766256609088 maxtext_utils.py:1880]  params/_overwrite_with_gradient/decoder/layers/mlp/wo/Fp8DirectDotGeneralOp_0/kernel_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0423 12:58:14.891544 137766256609088 maxtext_utils.py:1880]  params/_overwrite_with_gradient/decoder/layers/mlp/wo/Fp8DirectDotGeneralOp_0/output_grad_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0423 12:58:14.891572 137766256609088 maxtext_utils.py:1880]  params/_overwrite_with_gradient/decoder/layers/mlp/wo/Fp8DirectDotGeneralOp_0/output_grad_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0423 12:58:14.891601 137766256609088 maxtext_utils.py:1880]  params/_overwrite_with_gradient/decoder/layers/self_attention/key/Fp8DirectDotGeneralOp_0/input_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0423 12:58:14.891628 137766256609088 maxtext_utils.py:1880]  params/_overwrite_with_gradient/decoder/layers/self_attention/key/Fp8DirectDotGeneralOp_0/input_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0423 12:58:14.891668 137766256609088 maxtext_utils.py:1880]  params/_overwrite_with_gradient/decoder/layers/self_attention/key/Fp8DirectDotGeneralOp_0/kernel_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0423 12:58:14.891699 137766256609088 maxtext_utils.py:1880]  params/_overwrite_with_gradient/decoder/layers/self_attention/key/Fp8DirectDotGeneralOp_0/kernel_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0423 12:58:14.891728 137766256609088 maxtext_utils.py:1880]  params/_overwrite_with_gradient/decoder/layers/self_attention/key/Fp8DirectDotGeneralOp_0/output_grad_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0423 12:58:14.891755 137766256609088 maxtext_utils.py:1880]  params/_overwrite_with_gradient/decoder/layers/self_attention/key/Fp8DirectDotGeneralOp_0/output_grad_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0423 12:58:14.891783 137766256609088 maxtext_utils.py:1880]  params/_overwrite_with_gradient/decoder/layers/self_attention/out/Fp8DirectDotGeneralOp_0/input_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0423 12:58:14.891810 137766256609088 maxtext_utils.py:1880]  params/_overwrite_with_gradient/decoder/layers/self_attention/out/Fp8DirectDotGeneralOp_0/input_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0423 12:58:14.891838 137766256609088 maxtext_utils.py:1880]  params/_overwrite_with_gradient/decoder/layers/self_attention/out/Fp8DirectDotGeneralOp_0/kernel_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0423 12:58:14.891865 137766256609088 maxtext_utils.py:1880]  params/_overwrite_with_gradient/decoder/layers/self_attention/out/Fp8DirectDotGeneralOp_0/kernel_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0423 12:58:14.891892 137766256609088 maxtext_utils.py:1880]  params/_overwrite_with_gradient/decoder/layers/self_attention/out/Fp8DirectDotGeneralOp_0/output_grad_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0423 12:58:14.891920 137766256609088 maxtext_utils.py:1880]  params/_overwrite_with_gradient/decoder/layers/self_attention/out/Fp8DirectDotGeneralOp_0/output_grad_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0423 12:58:14.891948 137766256609088 maxtext_utils.py:1880]  params/_overwrite_with_gradient/decoder/layers/self_attention/query/Fp8DirectDotGeneralOp_0/input_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0423 12:58:14.891976 137766256609088 maxtext_utils.py:1880]  params/_overwrite_with_gradient/decoder/layers/self_attention/query/Fp8DirectDotGeneralOp_0/input_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0423 12:58:14.892004 137766256609088 maxtext_utils.py:1880]  params/_overwrite_with_gradient/decoder/layers/self_attention/query/Fp8DirectDotGeneralOp_0/kernel_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0423 12:58:14.892031 137766256609088 maxtext_utils.py:1880]  params/_overwrite_with_gradient/decoder/layers/self_attention/query/Fp8DirectDotGeneralOp_0/kernel_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0423 12:58:14.892058 137766256609088 maxtext_utils.py:1880]  params/_overwrite_with_gradient/decoder/layers/self_attention/query/Fp8DirectDotGeneralOp_0/output_grad_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0423 12:58:14.892086 137766256609088 maxtext_utils.py:1880]  params/_overwrite_with_gradient/decoder/layers/self_attention/query/Fp8DirectDotGeneralOp_0/output_grad_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0423 12:58:14.892112 137766256609088 maxtext_utils.py:1880]  params/_overwrite_with_gradient/decoder/layers/self_attention/value/Fp8DirectDotGeneralOp_0/input_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0423 12:58:14.892139 137766256609088 maxtext_utils.py:1880]  params/_overwrite_with_gradient/decoder/layers/self_attention/value/Fp8DirectDotGeneralOp_0/input_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0423 12:58:14.892184 137766256609088 maxtext_utils.py:1880]  params/_overwrite_with_gradient/decoder/layers/self_attention/value/Fp8DirectDotGeneralOp_0/kernel_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0423 12:58:14.892214 137766256609088 maxtext_utils.py:1880]  params/_overwrite_with_gradient/decoder/layers/self_attention/value/Fp8DirectDotGeneralOp_0/kernel_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0423 12:58:14.892243 137766256609088 maxtext_utils.py:1880]  params/_overwrite_with_gradient/decoder/layers/self_attention/value/Fp8DirectDotGeneralOp_0/output_grad_amax_history
    Shape:     float32[16,1024]
    Logical:   P()
    Physical:  ()
I0423 12:58:14.892270 137766256609088 maxtext_utils.py:1880]  params/_overwrite_with_gradient/decoder/layers/self_attention/value/Fp8DirectDotGeneralOp_0/output_grad_scale
    Shape:     float32[16,1]
    Logical:   P()
    Physical:  ()
I0423 12:58:14.892321 137766256609088 maxtext_utils.py:1880]  params/params/decoder/decoder_norm/scale
    Shape:     float32[2048]
    Logical:   P('norm',)
    Physical:  (None,)
I0423 12:58:14.892374 137766256609088 maxtext_utils.py:1880]  params/params/decoder/layers/mlp/wi_0/kernel
    Shape:     float32[2048,16,7168]
    Logical:   P('embed', 'layers', 'mlp')
    Physical:  ('fsdp', None, None)
I0423 12:58:14.892413 137766256609088 maxtext_utils.py:1880]  params/params/decoder/layers/mlp/wi_1/kernel
    Shape:     float32[2048,16,7168]
    Logical:   P('embed', 'layers', 'mlp')
    Physical:  ('fsdp', None, None)
I0423 12:58:14.892461 137766256609088 maxtext_utils.py:1880]  params/params/decoder/layers/mlp/wo/kernel
    Shape:     float32[7168,16,2048]
    Logical:   P('mlp', 'layers', 'embed')
    Physical:  (None, None, 'fsdp')
I0423 12:58:14.892508 137766256609088 maxtext_utils.py:1880]  params/params/decoder/layers/post_self_attention_layer_norm/scale
    Shape:     float32[2048,16]
    Logical:   P('norm', 'layers')
    Physical:  (None, None)
I0423 12:58:14.892543 137766256609088 maxtext_utils.py:1880]  params/params/decoder/layers/pre_self_attention_layer_norm/scale
    Shape:     float32[2048,16]
    Logical:   P('norm', 'layers')
    Physical:  (None, None)
I0423 12:58:14.892590 137766256609088 maxtext_utils.py:1880]  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)
I0423 12:58:14.892638 137766256609088 maxtext_utils.py:1880]  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')
I0423 12:58:14.892711 137766256609088 maxtext_utils.py:1880]  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)
I0423 12:58:14.892750 137766256609088 maxtext_utils.py:1880]  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)
I0423 12:58:14.892798 137766256609088 maxtext_utils.py:1880]  params/params/decoder/logits_dense/kernel
    Shape:     float32[2048,32000]
    Logical:   P('embed_vocab', 'vocab')
    Physical:  ('fsdp', None)

I0423 12:58:14.892844 137766256609088 maxtext_utils.py:1880]  params/params/token_embedder/embedding
    Shape:     float32[32000,2048]
    Logical:   P('vocab', 'embed_vocab')
    Physical:  (None, 'fsdp')
I0423 12:58:16.547358 137766256609088 train.py:157] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0423 12:58:16.547453 137766256609088 train.py:157] train/xent Physical: float32[32,2048]............................................ ('fsdp', None).
I0423 12:58:16.562975 137766256609088 train.py:164] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0423 12:58:16.563042 137766256609088 train.py:164] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None).
I0423 12:58:30.651262 137766256609088 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.
I0423 12:58:30.652175 137766256609088 metric_logger.py:301] number parameters: 1.105 billion
I0423 12:58:45.719362 137766256609088 checkpointing.py:794] Waiting for step 0 to finish before checkpoint...
I0423 12:58:45.892343 137766256609088 checkpointing.py:798] Waited 0.1729569435119629 seconds for step 0 to finish before starting checkpointing.
I0423 12:58:45.894746 137766256609088 checkpoint_manager.py:2009] [process=5][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0423 12:58:45.896722 137766256609088 checkpoint_manager.py:1512] [process=5] Saving checkpoint at step 0
I0423 12:58:45.898169 137766256609088 event_tracking.py:70] [process=5] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260423_124541/linen_xpk_feat_nnx_post_train_fixes_20260423_124541_05_fp8/checkpoints/0.
I0423 12:58:46.636516 137766256609088 signaling_client.py:364] Using JaxDistributedSignalingClient
I0423 12:58:46.637436 137766256609088 jax_array_handlers.py:360] Scheduling D2H of 81 prioritized jax.Array.
I0423 12:58:46.637497 137766256609088 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
I0423 12:58:47.050123 137766256609088 base_pytree_checkpoint_handler.py:154] [process=5][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.413729s
I0423 12:58:47.050303 137766256609088 base_pytree_checkpoint_handler.py:130] [process=5] /jax/orbax/write/blocking_gbytes_per_sec: 3.653 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.4223034381866455 s) (per-host)
I0423 12:58:47.050358 137766256609088 base_pytree_checkpoint_handler.py:768] [process=5][thread=MainThread] Initiated Pytree async_save. Time taken: 0.422368s (batch_requests_ready=0.003514s, total_serialization_initiated=0.418782s, others=0.000072s)
I0423 12:58:47.050459 137766256609088 composite_checkpoint_handler.py:715] [process=5][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.426485s (all_items=0.000018s, per_item={'items': '0.00001764'}, temp_paths=0.426467)
I0423 12:58:47.051327 137766256609088 event_tracking.py:125] [process=5] [async] Finished blocking save in 1.15 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260423_124541/linen_xpk_feat_nnx_post_train_fixes_20260423_124541_05_fp8/checkpoints/0.
I0423 12:58:47.051671 137640853485312 async_checkpointer.py:76] [process=5][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-23 13:18:47.051619
I0423 12:58:47.055740 137766256609088 checkpoint_manager.py:1560] [process=5][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize
I0423 12:58:47.056048 137640338777856 async_checkpointer.py:280] [process=5][thread=save_finalize] Waiting for background save thread=async_save.
I0423 12:58:47.056210 137766256609088 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260423_124541/linen_xpk_feat_nnx_post_train_fixes_20260423_124541_05_fp8/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776949125.894728, 'wait_for_prev_duration_secs': 6.175041198730469e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776949125.8967605, 'checkpointer_blocking_duration_secs': 1.1550631523132324, 'get_old_steps_start_time': 1776949127.05185, 'get_old_steps_duration_secs': 2.956390380859375e-05, 'checkpoint_manager_blocking_start_time': 1776949125.8929722, 'checkpoint_manager_blocking_duration_secs': 1.163194179534912}
I0423 12:58:47.056323 137766256609088 checkpointing.py:409] Started an asynchronous checkpoint save for step 0
I0423 12:58:47.056372 137766256609088 max_utils.py:750] 
Memstats: After params initialized:
I0423 12:58:47.056422 137766256609088 max_utils.py:756] 	Using (GB) 0.45 / 31.25 (1.440000%) on TPU_18(process=5,(2,4,0,0))
I0423 12:58:47.056454 137766256609088 max_utils.py:756] 	Using (GB) 0.45 / 31.25 (1.440000%) on TPU_19(process=5,(3,4,0,0))
I0423 12:58:47.056481 137766256609088 max_utils.py:756] 	Using (GB) 0.45 / 31.25 (1.440000%) on TPU_22(process=5,(2,5,0,0))
I0423 12:58:47.056506 137766256609088 max_utils.py:756] 	Using (GB) 0.45 / 31.25 (1.440000%) on TPU_23(process=5,(3,5,0,0))
I0423 12:58:47.386138 137766256609088 metric_logger.py:196] completed step: 0, seconds: 15.067, TFLOP/s/device: 0.902, Tokens/s/device: 135.925, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52776.805
I0423 12:58:47.566607 137766256609088 metric_logger.py:196] completed step: 1, seconds: 1.665, TFLOP/s/device: 8.162, Tokens/s/device: 1230.224, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52761.840
I0423 12:58:47.981162 137766256609088 metric_logger.py:196] completed step: 2, seconds: 0.023, TFLOP/s/device: 588.697, Tokens/s/device: 88734.835, total_weights: 65536, loss: 10.816, lm_loss: 10.816, perplexity: 49832.398
I0423 12:58:48.140054 137766256609088 metric_logger.py:196] completed step: 3, seconds: 0.415, TFLOP/s/device: 32.752, Tokens/s/device: 4936.748, total_weights: 65536, loss: 10.431, lm_loss: 10.431, perplexity: 33901.820
I0423 12:58:48.457703 137766256609088 metric_logger.py:196] completed step: 4, seconds: 0.164, TFLOP/s/device: 82.861, Tokens/s/device: 12489.785, total_weights: 65536, loss: 9.992, lm_loss: 9.992, perplexity: 21847.639
I0423 12:58:48.463986 137766256609088 metric_logger.py:196] completed step: 5, seconds: 0.159, TFLOP/s/device: 85.526, Tokens/s/device: 12891.368, total_weights: 65536, loss: 9.549, lm_loss: 9.549, perplexity: 14035.006
I0423 12:58:50.081887    2775 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com
I0423 12:58:52.210256 137640347170560 array_metadata_store.py:203] [process=5][thread=array_type_handler] Wrote 81 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260423_124541/linen_xpk_feat_nnx_post_train_fixes_20260423_124541_05_fp8/checkpoints/0/items/array_metadatas/process_5
I0423 12:59:07.040245 137640853485312 base_pytree_checkpoint_handler.py:130] [process=5] /jax/orbax/write/gbytes_per_sec: 77.393 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 20.41221284866333 s) (per-host)
I0423 12:59:07.040359 137640853485312 async_checkpointer.py:90] [process=5][thread=async_save] 3 Handler Commit operations completed. Time taken: 19.988574s.
I0423 12:59:12.286720 137766256609088 metric_logger.py:196] completed step: 6, seconds: 0.318, TFLOP/s/device: 42.742, Tokens/s/device: 6442.602, total_weights: 65536, loss: 9.111, lm_loss: 9.111, perplexity: 9058.688
I0423 12:59:12.445367 137766256609088 metric_logger.py:196] completed step: 7, seconds: 23.665, TFLOP/s/device: 0.574, Tokens/s/device: 86.539, total_weights: 65536, loss: 8.685, lm_loss: 8.685, perplexity: 5914.720
I0423 12:59:12.603983 137766256609088 metric_logger.py:196] completed step: 8, seconds: 0.164, TFLOP/s/device: 82.661, Tokens/s/device: 12459.619, total_weights: 65536, loss: 8.281, lm_loss: 8.281, perplexity: 3950.010
I0423 12:59:12.608880 137766256609088 checkpointing.py:794] Waiting for step 10 to finish before checkpoint...
I0423 12:59:12.921725 137766256609088 checkpointing.py:798] Waited 0.31277966499328613 seconds for step 10 to finish before starting checkpointing.
I0423 12:59:12.924345 137766256609088 checkpoint_manager.py:2020] [process=5][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0423 12:59:15.750403 137640853485312 async_checkpointer.py:160] [process=5][thread=async_save] Background save thread done. Time taken: 28.698601s.
I0423 12:59:15.750725 137640338777856 async_checkpointer.py:288] [process=5][thread=save_finalize] Done with waiting for background save thread=async_save.
I0423 12:59:15.750846 137640338777856 async_checkpointer.py:298] [process=5][thread=save_finalize] No errors found in background save thread=async_save.
I0423 12:59:15.750898 137640338777856 checkpoint_manager.py:2137] [process=5][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts...
I0423 12:59:15.753515 137640338777856 checkpoint_manager.py:2146] [process=5][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts.
I0423 12:59:15.753670 137766256609088 checkpoint_manager.py:2032] [process=5][thread=MainThread][step=0][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=0.
W0423 12:59:15.753736 137766256609088 checkpoint_manager.py:1452] Waiting for previous save to complete took 2.829400 seconds. If this number is high, consider checkpointing less frequently.
I0423 12:59:15.755529 137766256609088 checkpoint_manager.py:1512] [process=5] Saving checkpoint at step 10
I0423 12:59:15.757486 137766256609088 event_tracking.py:70] [process=5] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260423_124541/linen_xpk_feat_nnx_post_train_fixes_20260423_124541_05_fp8/checkpoints/10.
I0423 12:59:16.112607 137766256609088 jax_array_handlers.py:360] Scheduling D2H of 81 prioritized jax.Array.
I0423 12:59:16.112714 137766256609088 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
I0423 12:59:16.166677 137766256609088 base_pytree_checkpoint_handler.py:154] [process=5][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.055115s
I0423 12:59:16.166843 137766256609088 base_pytree_checkpoint_handler.py:130] [process=5] /jax/orbax/write/blocking_gbytes_per_sec: 25.123 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.06140780448913574 s) (per-host)
I0423 12:59:16.166893 137766256609088 base_pytree_checkpoint_handler.py:768] [process=5][thread=MainThread] Initiated Pytree async_save. Time taken: 0.061467s (batch_requests_ready=0.003136s, total_serialization_initiated=0.058264s, others=0.000067s)
I0423 12:59:16.166994 137766256609088 composite_checkpoint_handler.py:715] [process=5][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.065532s (all_items=0.000016s, per_item={'items': '0.00001621'}, temp_paths=0.065516)
I0423 12:59:16.167687 137766256609088 event_tracking.py:125] [process=5] [async] Finished blocking save in 0.41 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260423_124541/linen_xpk_feat_nnx_post_train_fixes_20260423_124541_05_fp8/checkpoints/10.
I0423 12:59:16.168004 137636589074176 async_checkpointer.py:76] [process=5][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-23 13:19:16.167981
I0423 12:59:16.169901 137766256609088 checkpoint_manager.py:1560] [process=5][thread=MainThread][step=10] Starting CheckpointManager Save Finalize thread=save_finalize
I0423 12:59:16.170179 137640338777856 async_checkpointer.py:280] [process=5][thread=save_finalize] Waiting for background save thread=async_save.
I0423 12:59:16.170320 137766256609088 standard_logger.py:34] {'step': 10, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260423_124541/linen_xpk_feat_nnx_post_train_fixes_20260423_124541_05_fp8/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776949152.9243133, 'wait_for_prev_duration_secs': 2.829399585723877, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776949155.7555673, 'checkpointer_blocking_duration_secs': 0.4125368595123291, 'get_old_steps_start_time': 1776949156.1681278, 'get_old_steps_duration_secs': 2.8848648071289062e-05, 'checkpoint_manager_blocking_start_time': 1776949152.922503, 'checkpoint_manager_blocking_duration_secs': 3.247784376144409}
I0423 12:59:16.170482 137766256609088 checkpointing.py:409] Started an asynchronous checkpoint save for step 10
I0423 12:59:16.171059 137766256609088 metric_logger.py:196] completed step: 9, seconds: 0.157, TFLOP/s/device: 86.319, Tokens/s/device: 13010.940, total_weights: 65536, loss: 7.908, lm_loss: 7.908, perplexity: 2720.116
I0423 12:59:16.178238 137766256609088 metric_logger.py:196] completed step: 10, seconds: 0.159, TFLOP/s/device: 85.657, Tokens/s/device: 12911.116, total_weights: 65536, loss: 7.568, lm_loss: 7.568, perplexity: 1936.198
I0423 12:59:16.337056 137766256609088 metric_logger.py:196] completed step: 11, seconds: 3.567, TFLOP/s/device: 3.809, Tokens/s/device: 574.133, total_weights: 65536, loss: 7.288, lm_loss: 7.288, perplexity: 1463.272
I0423 12:59:16.898251 137766256609088 metric_logger.py:196] completed step: 12, seconds: 0.006, TFLOP/s/device: 2139.032, Tokens/s/device: 322418.136, total_weights: 65536, loss: 7.032, lm_loss: 7.032, perplexity: 1132.061
I0423 12:59:17.056515 137766256609088 metric_logger.py:196] completed step: 13, seconds: 0.558, TFLOP/s/device: 24.363, Tokens/s/device: 3672.265, total_weights: 65536, loss: 6.815, lm_loss: 6.815, perplexity: 911.031
I0423 12:59:17.214859 137766256609088 metric_logger.py:196] completed step: 14, seconds: 0.163, TFLOP/s/device: 83.123, Tokens/s/device: 12529.212, total_weights: 65536, loss: 6.635, lm_loss: 6.635, perplexity: 761.080
I0423 12:59:17.373148 137766256609088 metric_logger.py:196] completed step: 15, seconds: 0.158, TFLOP/s/device: 85.938, Tokens/s/device: 12953.581, total_weights: 65536, loss: 6.492, lm_loss: 6.492, perplexity: 659.915
I0423 12:59:17.531535 137766256609088 metric_logger.py:196] completed step: 16, seconds: 0.158, TFLOP/s/device: 85.832, Tokens/s/device: 12937.542, total_weights: 65536, loss: 6.380, lm_loss: 6.380, perplexity: 589.948
I0423 12:59:17.690035 137766256609088 metric_logger.py:196] completed step: 17, seconds: 0.158, TFLOP/s/device: 85.846, Tokens/s/device: 12939.586, total_weights: 65536, loss: 6.293, lm_loss: 6.293, perplexity: 540.941
I0423 12:59:17.848194 137766256609088 metric_logger.py:196] completed step: 18, seconds: 0.158, TFLOP/s/device: 85.822, Tokens/s/device: 12935.990, total_weights: 65536, loss: 6.223, lm_loss: 6.223, perplexity: 504.108
I0423 12:59:18.005964 137766256609088 checkpointing.py:794] Waiting for step 19 to finish before checkpoint...
I0423 12:59:18.006973 137766256609088 checkpointing.py:798] Waited 0.001028299331665039 seconds for step 19 to finish before starting checkpointing.
I0423 12:59:18.009876 137766256609088 checkpoint_manager.py:2020] [process=5][thread=MainThread][step=10][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0423 12:59:22.194953 137637100766976 array_metadata_store.py:203] [process=5][thread=array_type_handler] Wrote 81 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260423_124541/linen_xpk_feat_nnx_post_train_fixes_20260423_124541_05_fp8/checkpoints/10/items/array_metadatas/process_5
I0423 12:59:59.180174 137636589074176 base_pytree_checkpoint_handler.py:130] [process=5] /jax/orbax/write/gbytes_per_sec: 36.675 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 43.07469868659973 s) (per-host)
I0423 12:59:59.180307 137636589074176 async_checkpointer.py:90] [process=5][thread=async_save] 3 Handler Commit operations completed. Time taken: 43.012224s.
I0423 13:00:06.801047 137636589074176 async_checkpointer.py:160] [process=5][thread=async_save] Background save thread done. Time taken: 50.632949s.
I0423 13:00:06.801331 137640338777856 async_checkpointer.py:288] [process=5][thread=save_finalize] Done with waiting for background save thread=async_save.
I0423 13:00:06.801459 137640338777856 async_checkpointer.py:298] [process=5][thread=save_finalize] No errors found in background save thread=async_save.
I0423 13:00:06.801508 137640338777856 checkpoint_manager.py:2137] [process=5][thread=save_finalize][step=10] CheckpointManager Save Finalize is syncing with other hosts...
I0423 13:00:06.803175 137640338777856 checkpoint_manager.py:2146] [process=5][thread=save_finalize][step=10] CheckpointManager Save Finalize is done on all hosts.
I0423 13:00:06.803365 137766256609088 checkpoint_manager.py:2032] [process=5][thread=MainThread][step=10][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=10.
W0423 13:00:06.803491 137766256609088 checkpoint_manager.py:1452] Waiting for previous save to complete took 48.793621 seconds. If this number is high, consider checkpointing less frequently.
I0423 13:00:06.805197 137766256609088 checkpoint_manager.py:1512] [process=5] Saving checkpoint at step 19
I0423 13:00:06.807377 137766256609088 event_tracking.py:70] [process=5] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260423_124541/linen_xpk_feat_nnx_post_train_fixes_20260423_124541_05_fp8/checkpoints/19.
I0423 13:00:07.333661 137766256609088 jax_array_handlers.py:360] Scheduling D2H of 81 prioritized jax.Array.
I0423 13:00:07.333757 137766256609088 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
I0423 13:00:07.385032 137766256609088 base_pytree_checkpoint_handler.py:154] [process=5][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.052336s
I0423 13:00:07.385196 137766256609088 base_pytree_checkpoint_handler.py:130] [process=5] /jax/orbax/write/blocking_gbytes_per_sec: 26.324 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.05860543251037598 s) (per-host)
I0423 13:00:07.385243 137766256609088 base_pytree_checkpoint_handler.py:768] [process=5][thread=MainThread] Initiated Pytree async_save. Time taken: 0.058663s (batch_requests_ready=0.003140s, total_serialization_initiated=0.055461s, others=0.000062s)
I0423 13:00:07.385343 137766256609088 composite_checkpoint_handler.py:715] [process=5][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.062860s (all_items=0.000011s, per_item={'items': '0.00001097'}, temp_paths=0.062850)
I0423 13:00:07.386033 137766256609088 event_tracking.py:125] [process=5] [async] Finished blocking save in 0.58 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260423_124541/linen_xpk_feat_nnx_post_train_fixes_20260423_124541_05_fp8/checkpoints/19.
I0423 13:00:07.386294 137640338777856 async_checkpointer.py:76] [process=5][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-23 13:20:07.386270
I0423 13:00:07.388237 137766256609088 checkpoint_manager.py:1560] [process=5][thread=MainThread][step=19] Starting CheckpointManager Save Finalize thread=save_finalize
I0423 13:00:07.388457 137635490154240 async_checkpointer.py:280] [process=5][thread=save_finalize] Waiting for background save thread=async_save.
I0423 13:00:07.388552 137766256609088 standard_logger.py:34] {'step': 19, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260423_124541/linen_xpk_feat_nnx_post_train_fixes_20260423_124541_05_fp8/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776949158.0098476, 'wait_for_prev_duration_secs': 48.793620586395264, 'time_between_consecutive_saves_sec': 2.2562904357910156, 'checkpointer_blocking_start_time': 1776949206.805238, 'checkpointer_blocking_duration_secs': 0.5811562538146973, 'get_old_steps_start_time': 1776949207.3864148, 'get_old_steps_duration_secs': 2.384185791015625e-05, 'checkpoint_manager_blocking_start_time': 1776949158.007261, 'checkpoint_manager_blocking_duration_secs': 49.381258964538574}
I0423 13:00:07.388734 137766256609088 checkpointing.py:409] Started an asynchronous checkpoint save for step 19
I0423 13:00:07.388782 137766256609088 checkpoint_manager.py:2020] [process=5][thread=MainThread][step=19][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0423 13:00:12.847807 137637100766976 array_metadata_store.py:203] [process=5][thread=array_type_handler] Wrote 81 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260423_124541/linen_xpk_feat_nnx_post_train_fixes_20260423_124541_05_fp8/checkpoints/19/items/array_metadatas/process_5
I0423 13:00:49.792986 137640338777856 base_pytree_checkpoint_handler.py:130] [process=5] /jax/orbax/write/gbytes_per_sec: 37.200 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 42.46636176109314 s) (per-host)
I0423 13:00:49.793103 137640338777856 async_checkpointer.py:90] [process=5][thread=async_save] 3 Handler Commit operations completed. Time taken: 42.406730s.
I0423 13:00:57.953485 137640338777856 async_checkpointer.py:160] [process=5][thread=async_save] Background save thread done. Time taken: 50.567098s.
I0423 13:00:57.953733 137635490154240 async_checkpointer.py:288] [process=5][thread=save_finalize] Done with waiting for background save thread=async_save.
I0423 13:00:57.953796 137635490154240 async_checkpointer.py:298] [process=5][thread=save_finalize] No errors found in background save thread=async_save.
I0423 13:00:57.953839 137635490154240 checkpoint_manager.py:2137] [process=5][thread=save_finalize][step=19] CheckpointManager Save Finalize is syncing with other hosts...
I0423 13:00:57.955266 137635490154240 checkpoint_manager.py:2146] [process=5][thread=save_finalize][step=19] CheckpointManager Save Finalize is done on all hosts.
I0423 13:00:57.955424 137766256609088 checkpoint_manager.py:2032] [process=5][thread=MainThread][step=19][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=19.
I0423 13:00:57.955576 137766256609088 checkpoint_manager.py:2009] [process=5][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0423 13:00:57.956517 137766256609088 metric_logger.py:196] completed step: 19, seconds: 0.158, TFLOP/s/device: 85.739, Tokens/s/device: 12923.582, 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: Thu Apr 23 13:01:08 UTC 2026
EXIT_CODE=0

Metrics: main (586e69205) vs feat/nnx-post-train-fixes (7f06c99ac)

Metrics: `main (586e69205)` vs `feat/nnx-post-train-fixes (7f06c99ac)`