feat/nnx-post-train-fixes| Metric | Linen 2fbc8c2f1 | NNX 2fbc8c2f1 | Diff (NNX − Linen) |
|---|---|---|---|
| Parameters | 1.105 billion | 1.104 billion | — |
| Final loss | 6.1670 | 10.5490 | +4.382 |
| TFLOP/s | 86.020 | 88.558 | +2.538 |
| Tok/s | 12965.9 | 13348.4 | +382.486 |
| Avg s/step | 1.721 | 1.694 | -0.027 |
| Memory % | 1.44 | 1.44 | 0 |
| JAX | 0.9.2 | 0.9.2 | — |
Diff = NNX value − Linen value. Green = NNX improved. Red = NNX regressed.
XPK Start: Fri Apr 17 02:54:27 UTC 2026 PyTorch was not found. Models won't be available and only tokenizers, configuration and file/data utilities can be used. 2026-04-17 02:54:51.482971: 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) I0417 02:54:52.015590 138719853598528 max_utils.py:273] Attempting to initialize the jax distributed system... I0417 02:55:01.054616 138719853598528 distributed.py:149] Starting JAX distributed service on [::]:8482 I0417 02:55:01.056950 138719853598528 distributed.py:172] Connecting to JAX distributed service on mt-05-fp8-a4sw4-slice-job-0-0.mt-05-fp8-a4sw4:8482 I0417 02:55:02.118947 138719853598528 max_utils.py:284] Jax distributed system initialized! I0417 02:55:08.342916 138719853598528 max_utils.py:800] System Information: Jax Version: 0.9.2 I0417 02:55:08.343017 138719853598528 max_utils.py:801] System Information: Jaxlib Version: 0.9.2 I0417 02:55:08.343056 138719853598528 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 I0417 02:55:08.343089 138719853598528 train_utils.py:377] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing. I0417 02:55:09.004114 138719853598528 maxtext_utils.py:1687] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0417 02:55:09.004753 138719853598528 maxtext_utils.py:1687] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0417 02:55:09.004953 138719853598528 checkpointing.py:688] Setting up checkpoint logger... I0417 02:55:09.005007 138719853598528 checkpointing.py:234] Creating checkpoint manager with ocdbt=True and zarr3=True I0417 02:55:09.005050 138719853598528 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None I0417 02:55:09.005385 138719853598528 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 0x7e2983d51bb0>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0417 02:55:11.887714 138719853598528 checkpointing.py:266] Enabling policy for fixed interval checkpointing. I0417 02:55:11.887958 138719853598528 checkpoint_manager.py:708] [process=0][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=('items',), item_handlers={'items': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7e15502e1070>}, handler_registry=None I0417 02:55:11.888245 138719853598528 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7e15502e1070>` 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`. I0417 02:55:11.888302 138719853598528 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7e15502e4b00>` 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`. I0417 02:55:11.888340 138719853598528 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 0x7e15502e1070>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7e15502e1070>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7e15502e4b00>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7e15502e4b00>}). I0417 02:55:11.888660 138719853598528 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34 I0417 02:55:11.888731 138719853598528 async_checkpointer.py:192] [process=0][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7e1550250680> timeout: 1200 secs and primary_host=0 for async checkpoint writes I0417 02:55:13.468558 138719853598528 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_05_fp8/checkpoints I0417 02:55:13.929044 138719853598528 checkpoint_manager.py:929] [process=0][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_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_05_fp8/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7e15502d65a0> I0417 02:55:13.929222 138719853598528 checkpointing.py:302] Checkpoint manager created! I0417 02:55:14.993643 138719853598528 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0417 02:55:14.993742 138719853598528 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0417 02:55:15.378467 138719853598528 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0417 02:55:15.378559 138719853598528 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0417 02:55:15.394737 138719853598528 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0417 02:55:15.394798 138719853598528 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0417 02:55:15.447156 138719853598528 attentions.py:1154] attentions/query Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0417 02:55:15.447238 138719853598528 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0417 02:55:15.463500 138719853598528 attentions.py:1155] attentions/key Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0417 02:55:15.463566 138719853598528 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0417 02:55:15.479990 138719853598528 attentions.py:1156] attentions/value Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0417 02:55:15.480055 138719853598528 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0417 02:55:15.504809 138719853598528 attentions.py:1197] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv'). I0417 02:55:15.504895 138719853598528 attentions.py:1197] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0417 02:55:15.556782 138719853598528 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0417 02:55:15.556878 138719853598528 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None). I0417 02:55:16.000612 138719853598528 checkpointing.py:578] checkpoint manager exists so trying to load this run's existing checkpoint I0417 02:55:16.000720 138719853598528 checkpointing.py:676] No existing checkpoints found, not restoring checkpoint. fsdp: 32 I0417 02:55:17.903960 138719853598528 maxtext_utils.py:1796] params/_overwrite_with_gradient/decoder/layers/mlp/wi_0/Fp8DirectDotGeneralOp_0/input_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0417 02:55:17.904082 138719853598528 maxtext_utils.py:1796] params/_overwrite_with_gradient/decoder/layers/mlp/wi_0/Fp8DirectDotGeneralOp_0/input_scale Shape: float32[16,1] Logical: P() Physical: () I0417 02:55:17.904136 138719853598528 maxtext_utils.py:1796] params/_overwrite_with_gradient/decoder/layers/mlp/wi_0/Fp8DirectDotGeneralOp_0/kernel_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0417 02:55:17.904176 138719853598528 maxtext_utils.py:1796] params/_overwrite_with_gradient/decoder/layers/mlp/wi_0/Fp8DirectDotGeneralOp_0/kernel_scale Shape: float32[16,1] Logical: P() Physical: () I0417 02:55:17.904212 138719853598528 maxtext_utils.py:1796] params/_overwrite_with_gradient/decoder/layers/mlp/wi_0/Fp8DirectDotGeneralOp_0/output_grad_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0417 02:55:17.904246 138719853598528 maxtext_utils.py:1796] params/_overwrite_with_gradient/decoder/layers/mlp/wi_0/Fp8DirectDotGeneralOp_0/output_grad_scale Shape: float32[16,1] Logical: P() Physical: () I0417 02:55:17.904277 138719853598528 maxtext_utils.py:1796] params/_overwrite_with_gradient/decoder/layers/mlp/wi_1/Fp8DirectDotGeneralOp_0/input_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0417 02:55:17.904308 138719853598528 maxtext_utils.py:1796] params/_overwrite_with_gradient/decoder/layers/mlp/wi_1/Fp8DirectDotGeneralOp_0/input_scale Shape: float32[16,1] Logical: P() Physical: () I0417 02:55:17.904337 138719853598528 maxtext_utils.py:1796] params/_overwrite_with_gradient/decoder/layers/mlp/wi_1/Fp8DirectDotGeneralOp_0/kernel_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0417 02:55:17.904366 138719853598528 maxtext_utils.py:1796] params/_overwrite_with_gradient/decoder/layers/mlp/wi_1/Fp8DirectDotGeneralOp_0/kernel_scale Shape: float32[16,1] Logical: P() Physical: () I0417 02:55:17.904395 138719853598528 maxtext_utils.py:1796] params/_overwrite_with_gradient/decoder/layers/mlp/wi_1/Fp8DirectDotGeneralOp_0/output_grad_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0417 02:55:17.904423 138719853598528 maxtext_utils.py:1796] params/_overwrite_with_gradient/decoder/layers/mlp/wi_1/Fp8DirectDotGeneralOp_0/output_grad_scale Shape: float32[16,1] Logical: P() Physical: () I0417 02:55:17.904452 138719853598528 maxtext_utils.py:1796] params/_overwrite_with_gradient/decoder/layers/mlp/wo/Fp8DirectDotGeneralOp_0/input_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0417 02:55:17.904481 138719853598528 maxtext_utils.py:1796] params/_overwrite_with_gradient/decoder/layers/mlp/wo/Fp8DirectDotGeneralOp_0/input_scale Shape: float32[16,1] Logical: P() Physical: () I0417 02:55:17.904510 138719853598528 maxtext_utils.py:1796] params/_overwrite_with_gradient/decoder/layers/mlp/wo/Fp8DirectDotGeneralOp_0/kernel_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0417 02:55:17.904539 138719853598528 maxtext_utils.py:1796] params/_overwrite_with_gradient/decoder/layers/mlp/wo/Fp8DirectDotGeneralOp_0/kernel_scale Shape: float32[16,1] Logical: P() Physical: () I0417 02:55:17.904569 138719853598528 maxtext_utils.py:1796] params/_overwrite_with_gradient/decoder/layers/mlp/wo/Fp8DirectDotGeneralOp_0/output_grad_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0417 02:55:17.904619 138719853598528 maxtext_utils.py:1796] params/_overwrite_with_gradient/decoder/layers/mlp/wo/Fp8DirectDotGeneralOp_0/output_grad_scale Shape: float32[16,1] Logical: P() Physical: () I0417 02:55:17.904649 138719853598528 maxtext_utils.py:1796] params/_overwrite_with_gradient/decoder/layers/self_attention/key/Fp8DirectDotGeneralOp_0/input_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0417 02:55:17.904678 138719853598528 maxtext_utils.py:1796] params/_overwrite_with_gradient/decoder/layers/self_attention/key/Fp8DirectDotGeneralOp_0/input_scale Shape: float32[16,1] Logical: P() Physical: () I0417 02:55:17.904705 138719853598528 maxtext_utils.py:1796] params/_overwrite_with_gradient/decoder/layers/self_attention/key/Fp8DirectDotGeneralOp_0/kernel_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0417 02:55:17.904732 138719853598528 maxtext_utils.py:1796] params/_overwrite_with_gradient/decoder/layers/self_attention/key/Fp8DirectDotGeneralOp_0/kernel_scale Shape: float32[16,1] Logical: P() Physical: () I0417 02:55:17.904759 138719853598528 maxtext_utils.py:1796] params/_overwrite_with_gradient/decoder/layers/self_attention/key/Fp8DirectDotGeneralOp_0/output_grad_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0417 02:55:17.904786 138719853598528 maxtext_utils.py:1796] params/_overwrite_with_gradient/decoder/layers/self_attention/key/Fp8DirectDotGeneralOp_0/output_grad_scale Shape: float32[16,1] Logical: P() Physical: () I0417 02:55:17.904813 138719853598528 maxtext_utils.py:1796] params/_overwrite_with_gradient/decoder/layers/self_attention/out/Fp8DirectDotGeneralOp_0/input_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0417 02:55:17.904854 138719853598528 maxtext_utils.py:1796] params/_overwrite_with_gradient/decoder/layers/self_attention/out/Fp8DirectDotGeneralOp_0/input_scale Shape: float32[16,1] Logical: P() Physical: () I0417 02:55:17.904886 138719853598528 maxtext_utils.py:1796] params/_overwrite_with_gradient/decoder/layers/self_attention/out/Fp8DirectDotGeneralOp_0/kernel_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0417 02:55:17.904913 138719853598528 maxtext_utils.py:1796] params/_overwrite_with_gradient/decoder/layers/self_attention/out/Fp8DirectDotGeneralOp_0/kernel_scale Shape: float32[16,1] Logical: P() Physical: () I0417 02:55:17.904940 138719853598528 maxtext_utils.py:1796] params/_overwrite_with_gradient/decoder/layers/self_attention/out/Fp8DirectDotGeneralOp_0/output_grad_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0417 02:55:17.904968 138719853598528 maxtext_utils.py:1796] params/_overwrite_with_gradient/decoder/layers/self_attention/out/Fp8DirectDotGeneralOp_0/output_grad_scale Shape: float32[16,1] Logical: P() Physical: () I0417 02:55:17.904995 138719853598528 maxtext_utils.py:1796] params/_overwrite_with_gradient/decoder/layers/self_attention/query/Fp8DirectDotGeneralOp_0/input_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0417 02:55:17.905021 138719853598528 maxtext_utils.py:1796] params/_overwrite_with_gradient/decoder/layers/self_attention/query/Fp8DirectDotGeneralOp_0/input_scale Shape: float32[16,1] Logical: P() Physical: () I0417 02:55:17.905048 138719853598528 maxtext_utils.py:1796] params/_overwrite_with_gradient/decoder/layers/self_attention/query/Fp8DirectDotGeneralOp_0/kernel_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0417 02:55:17.905074 138719853598528 maxtext_utils.py:1796] params/_overwrite_with_gradient/decoder/layers/self_attention/query/Fp8DirectDotGeneralOp_0/kernel_scale Shape: float32[16,1] Logical: P() Physical: () I0417 02:55:17.905105 138719853598528 maxtext_utils.py:1796] params/_overwrite_with_gradient/decoder/layers/self_attention/query/Fp8DirectDotGeneralOp_0/output_grad_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0417 02:55:17.905133 138719853598528 maxtext_utils.py:1796] params/_overwrite_with_gradient/decoder/layers/self_attention/query/Fp8DirectDotGeneralOp_0/output_grad_scale Shape: float32[16,1] Logical: P() Physical: () I0417 02:55:17.905160 138719853598528 maxtext_utils.py:1796] params/_overwrite_with_gradient/decoder/layers/self_attention/value/Fp8DirectDotGeneralOp_0/input_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0417 02:55:17.905186 138719853598528 maxtext_utils.py:1796] params/_overwrite_with_gradient/decoder/layers/self_attention/value/Fp8DirectDotGeneralOp_0/input_scale Shape: float32[16,1] Logical: P() Physical: () I0417 02:55:17.905213 138719853598528 maxtext_utils.py:1796] params/_overwrite_with_gradient/decoder/layers/self_attention/value/Fp8DirectDotGeneralOp_0/kernel_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0417 02:55:17.905239 138719853598528 maxtext_utils.py:1796] params/_overwrite_with_gradient/decoder/layers/self_attention/value/Fp8DirectDotGeneralOp_0/kernel_scale Shape: float32[16,1] Logical: P() Physical: () I0417 02:55:17.905265 138719853598528 maxtext_utils.py:1796] params/_overwrite_with_gradient/decoder/layers/self_attention/value/Fp8DirectDotGeneralOp_0/output_grad_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0417 02:55:17.905292 138719853598528 maxtext_utils.py:1796] params/_overwrite_with_gradient/decoder/layers/self_attention/value/Fp8DirectDotGeneralOp_0/output_grad_scale Shape: float32[16,1] Logical: P() Physical: () I0417 02:55:17.905340 138719853598528 maxtext_utils.py:1796] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: P('norm',) Physical: (None,) I0417 02:55:17.905393 138719853598528 maxtext_utils.py:1796] params/params/decoder/layers/mlp/wi_0/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0417 02:55:17.905429 138719853598528 maxtext_utils.py:1796] params/params/decoder/layers/mlp/wi_1/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0417 02:55:17.905474 138719853598528 maxtext_utils.py:1796] params/params/decoder/layers/mlp/wo/kernel Shape: float32[7168,16,2048] Logical: P('mlp', 'layers', 'embed') Physical: (None, None, 'fsdp') I0417 02:55:17.905519 138719853598528 maxtext_utils.py:1796] params/params/decoder/layers/post_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0417 02:55:17.905551 138719853598528 maxtext_utils.py:1796] params/params/decoder/layers/pre_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0417 02:55:17.905597 138719853598528 maxtext_utils.py:1796] 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) I0417 02:55:17.905642 138719853598528 maxtext_utils.py:1796] 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') I0417 02:55:17.905678 138719853598528 maxtext_utils.py:1796] 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) I0417 02:55:17.905712 138719853598528 maxtext_utils.py:1796] 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) I0417 02:55:17.905754 138719853598528 maxtext_utils.py:1796] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: P('embed_vocab', 'vocab') Physical: ('fsdp', None) I0417 02:55:17.905796 138719853598528 maxtext_utils.py:1796] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: P('vocab', 'embed_vocab') Physical: (None, 'fsdp') I0417 02:55:19.549542 138719853598528 train.py:157] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0417 02:55:19.549637 138719853598528 train.py:157] train/xent Physical: float32[32,2048]............................................ ('fsdp', None). I0417 02:55:19.565416 138719853598528 train.py:164] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0417 02:55:19.565481 138719853598528 train.py:164] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None). I0417 02:55:33.183582 138719853598528 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. I0417 02:55:33.191106 138719853598528 metric_logger.py:289] number parameters: 1.105 billion I0417 02:55:48.704413 138719853598528 checkpointing.py:794] Waiting for step 0 to finish before checkpoint... I0417 02:55:48.873579 138719853598528 checkpointing.py:798] Waited 0.1691427230834961 seconds for step 0 to finish before starting checkpointing. I0417 02:55:48.875939 138719853598528 checkpoint_manager.py:2009] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0417 02:55:48.877923 138719853598528 checkpoint_manager.py:1512] [process=0] Saving checkpoint at step 0 I0417 02:55:48.879290 138719853598528 event_tracking.py:70] [process=0] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_05_fp8/checkpoints/0. I0417 02:55:49.621336 138719853598528 signaling_client.py:364] Using JaxDistributedSignalingClient I0417 02:55:49.639158 138719853598528 jax_array_handlers.py:360] Scheduling D2H of 81 prioritized jax.Array. I0417 02:55:49.639226 138719853598528 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 I0417 02:55:49.876007 138589400725248 atomicity.py:140] Creating tmp directory gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_05_fp8/checkpoints/0 I0417 02:55:50.037578 138719853598528 base_pytree_checkpoint_handler.py:154] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.399212s I0417 02:55:50.038458 138719853598528 base_pytree_checkpoint_handler.py:130] [process=0] /jax/orbax/write/blocking_gbytes_per_sec: 3.770 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.4092414379119873 s) (per-host) I0417 02:55:50.038551 138719853598528 base_pytree_checkpoint_handler.py:768] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 0.409357s (batch_requests_ready=0.003439s, total_serialization_initiated=0.405102s, others=0.000815s) I0417 02:55:50.038665 138719853598528 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.413305s (all_items=0.000017s, per_item={'items': '0.00001717'}, temp_paths=0.413288) I0417 02:55:50.039507 138719853598528 event_tracking.py:125] [process=0] [async] Finished blocking save in 1.16 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_05_fp8/checkpoints/0. I0417 02:55:50.039852 138589881108224 async_checkpointer.py:76] [process=0][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-17 03:15:50.039799 I0417 02:55:50.066569 138719853598528 checkpoint_manager.py:1560] [process=0][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize I0417 02:55:50.066895 138588284434176 async_checkpointer.py:280] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0417 02:55:50.067057 138719853598528 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_05_fp8/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776394548.8759212, 'wait_for_prev_duration_secs': 5.9604644775390625e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776394548.877961, 'checkpointer_blocking_duration_secs': 1.1620397567749023, 'get_old_steps_start_time': 1776394550.0400248, 'get_old_steps_duration_secs': 3.0040740966796875e-05, 'checkpoint_manager_blocking_start_time': 1776394548.8741922, 'checkpoint_manager_blocking_duration_secs': 1.1928248405456543} I0417 02:55:50.067169 138719853598528 checkpointing.py:409] Started an asynchronous checkpoint save for step 0 I0417 02:55:50.067221 138719853598528 max_utils.py:750] Memstats: After params initialized: I0417 02:55:50.067276 138719853598528 max_utils.py:756] Using (GB) 0.45 / 31.25 (1.440000%) on TPU_0(process=0,(0,0,0,0)) I0417 02:55:50.067308 138719853598528 max_utils.py:756] Using (GB) 0.45 / 31.25 (1.440000%) on TPU_1(process=0,(1,0,0,0)) I0417 02:55:50.067337 138719853598528 max_utils.py:756] Using (GB) 0.45 / 31.25 (1.440000%) on TPU_4(process=0,(0,1,0,0)) I0417 02:55:50.067362 138719853598528 max_utils.py:756] Using (GB) 0.45 / 31.25 (1.440000%) on TPU_5(process=0,(1,1,0,0)) I0417 02:55:50.505244 138719853598528 metric_logger.py:185] completed step: 0, seconds: 15.081, TFLOP/s/device: 0.901, Tokens/s/device: 135.796, total_weights: 65536, loss: 10.874 I0417 02:55:50.508135 138719853598528 metric_logger.py:269] To see full metrics 'tensorboard --logdir=gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_05_fp8/tensorboard/' I0417 02:55:50.948040 138719853598528 metric_logger.py:185] completed step: 1, seconds: 1.799, TFLOP/s/device: 7.554, Tokens/s/device: 1138.689, total_weights: 65536, loss: 10.874 I0417 02:55:51.104524 138719853598528 metric_logger.py:185] completed step: 2, seconds: 0.444, TFLOP/s/device: 30.604, Tokens/s/device: 4612.966, total_weights: 65536, loss: 10.816 I0417 02:55:51.262877 138719853598528 metric_logger.py:185] completed step: 3, seconds: 0.009, TFLOP/s/device: 1474.458, Tokens/s/device: 222246.337, total_weights: 65536, loss: 10.431 I0417 02:55:52.501232 138588326983424 atomicity.py:140] Creating tmp directory gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_05_fp8/checkpoints/0/items I0417 02:55:52.632384 138587774834432 checkpoint.py:188] Wrote Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776394552219310455, 'commit_timestamp_nsecs': None, 'custom_metadata': {}}, json={"item_handlers": null, "metrics": {}, "performance_metrics": {}, "init_timestamp_nsecs": 1776394552219310455, "commit_timestamp_nsecs": null, "custom_metadata": {}} to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_05_fp8/checkpoints/0/_CHECKPOINT_METADATA I0417 02:55:54.064561 2953 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com I0417 02:55:56.267084 138588301219584 array_metadata_store.py:203] [process=0][thread=array_type_handler] Wrote 81 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_05_fp8/checkpoints/0/items/array_metadatas/process_0 I0417 02:56:13.267854 138588292826880 base_pytree_checkpoint_handler.py:1282] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 19.214660s (commit=18.124806s, array_metadata_write=1.089854s) I0417 02:56:13.269060 138589881108224 base_pytree_checkpoint_handler.py:130] [process=0] /jax/orbax/write/gbytes_per_sec: 66.826 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 23.639822006225586 s) (per-host) I0417 02:56:13.269188 138589881108224 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 23.229221s. I0417 02:56:13.936311 138589881108224 checkpoint.py:228] Read Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776394552219310455, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} from gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_05_fp8/checkpoints/0/_CHECKPOINT_METADATA I0417 02:56:15.521162 138719853598528 metric_logger.py:185] completed step: 4, seconds: 0.158, TFLOP/s/device: 86.173, Tokens/s/device: 12988.990, total_weights: 65536, loss: 9.992 I0417 02:56:15.530771 138719853598528 metric_logger.py:185] completed step: 5, seconds: 0.158, TFLOP/s/device: 85.817, Tokens/s/device: 12935.336, total_weights: 65536, loss: 9.549 I0417 02:56:15.687257 138719853598528 metric_logger.py:185] completed step: 6, seconds: 24.257, TFLOP/s/device: 0.560, Tokens/s/device: 84.428, total_weights: 65536, loss: 9.111 I0417 02:56:15.735851 138587774834432 checkpoint.py:247] Updated Metadata={'item_handlers': {'items': 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler'}, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776394552219310455, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_05_fp8/checkpoints/0/_CHECKPOINT_METADATA I0417 02:56:15.845657 138719853598528 metric_logger.py:185] completed step: 7, seconds: 0.008, TFLOP/s/device: 1800.574, Tokens/s/device: 271402.067, total_weights: 65536, loss: 8.685 I0417 02:56:16.004004 138719853598528 metric_logger.py:185] completed step: 8, seconds: 0.159, TFLOP/s/device: 85.194, Tokens/s/device: 12841.414, total_weights: 65536, loss: 8.281 I0417 02:56:16.011399 138719853598528 checkpointing.py:794] Waiting for step 10 to finish before checkpoint... I0417 02:56:16.024610 138589881108224 array_metadata_store.py:411] [process=0][thread=async_save] Validated ArrayMetadata from all 8 hosts. Time taken: 0.000472s. I0417 02:56:16.136794 138589881108224 ocdbt_utils.py:49] Param validation support for Zarr3 will be added later (b/362328389). I0417 02:56:16.321863 138719853598528 checkpointing.py:798] Waited 0.31040525436401367 seconds for step 10 to finish before starting checkpointing. I0417 02:56:16.324477 138719853598528 checkpoint_manager.py:2020] [process=0][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0417 02:56:17.225530 138589881108224 base_pytree_checkpoint_handler.py:1406] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 3.168289s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_05_fp8/checkpoints/0/items I0417 02:56:17.227360 138589881108224 atomicity.py:666] Finalizing gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_05_fp8/checkpoints/0/items I0417 02:56:17.767925 138589881108224 atomicity.py:666] Finalizing gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_05_fp8/checkpoints/0 I0417 02:56:19.139910 138589881108224 atomicity.py:847] [process=0][thread=async_save] Finished saving checkpoint (finalized tmp dir) to `gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_05_fp8/checkpoints/0`. I0417 02:56:19.140782 138589881108224 event_tracking.py:138] [process=0] [async] Finished save (blocking + background) in 30.26 seconds @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_05_fp8/checkpoints/0 I0417 02:56:19.142046 138589881108224 async_checkpointer.py:160] [process=0][thread=async_save] Background save thread done. Time taken: 29.102077s. I0417 02:56:19.142188 138588284434176 async_checkpointer.py:288] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0417 02:56:19.142248 138588284434176 async_checkpointer.py:298] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0417 02:56:19.142316 138588284434176 checkpoint_manager.py:2137] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts... I0417 02:56:19.145501 138588284434176 checkpoint_manager.py:2146] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts. I0417 02:56:19.145630 138719853598528 checkpoint_manager.py:2032] [process=0][thread=MainThread][step=0][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=0. W0417 02:56:19.145719 138719853598528 checkpoint_manager.py:1452] Waiting for previous save to complete took 2.821245 seconds. If this number is high, consider checkpointing less frequently. I0417 02:56:19.147554 138719853598528 checkpoint_manager.py:1512] [process=0] Saving checkpoint at step 10 I0417 02:56:19.149486 138719853598528 event_tracking.py:70] [process=0] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_05_fp8/checkpoints/10. I0417 02:56:19.883178 138719853598528 jax_array_handlers.py:360] Scheduling D2H of 81 prioritized jax.Array. I0417 02:56:19.883277 138719853598528 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 I0417 02:56:19.922750 138719853598528 base_pytree_checkpoint_handler.py:154] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.040372s I0417 02:56:19.923624 138719853598528 base_pytree_checkpoint_handler.py:130] [process=0] /jax/orbax/write/blocking_gbytes_per_sec: 32.201 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.04790902137756348 s) (per-host) I0417 02:56:19.923737 138719853598528 base_pytree_checkpoint_handler.py:768] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 0.048051s (batch_requests_ready=0.003130s, total_serialization_initiated=0.044112s, others=0.000809s) I0417 02:56:19.923866 138719853598528 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.052340s (all_items=0.000016s, per_item={'items': '0.00001621'}, temp_paths=0.052324) I0417 02:56:19.924591 138719853598528 event_tracking.py:125] [process=0] [async] Finished blocking save in 0.78 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_05_fp8/checkpoints/10. I0417 02:56:19.924952 138587228796672 async_checkpointer.py:76] [process=0][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-17 03:16:19.924917 I0417 02:56:19.926992 138719853598528 checkpoint_manager.py:1560] [process=0][thread=MainThread][step=10] Starting CheckpointManager Save Finalize thread=save_finalize I0417 02:56:19.927225 138588284434176 async_checkpointer.py:280] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0417 02:56:19.927350 138719853598528 standard_logger.py:34] {'step': 10, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_05_fp8/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776394576.3244445, 'wait_for_prev_duration_secs': 2.821244955062866, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776394579.1475925, 'checkpointer_blocking_duration_secs': 0.7774951457977295, 'get_old_steps_start_time': 1776394579.9251122, 'get_old_steps_duration_secs': 3.0279159545898438e-05, 'checkpoint_manager_blocking_start_time': 1776394576.3227174, 'checkpoint_manager_blocking_duration_secs': 3.604597806930542} I0417 02:56:19.927462 138719853598528 checkpointing.py:409] Started an asynchronous checkpoint save for step 10 I0417 02:56:19.927883 138719853598528 metric_logger.py:185] completed step: 9, seconds: 0.156, TFLOP/s/device: 86.975, Tokens/s/device: 13109.885, total_weights: 65536, loss: 7.908 I0417 02:56:19.937306 138719853598528 metric_logger.py:185] completed step: 10, seconds: 0.158, TFLOP/s/device: 85.836, Tokens/s/device: 12938.114, total_weights: 65536, loss: 7.568 I0417 02:56:20.092538 138719853598528 metric_logger.py:185] completed step: 11, seconds: 3.924, TFLOP/s/device: 3.463, Tokens/s/device: 521.938, total_weights: 65536, loss: 7.288 I0417 02:56:20.124592 138589881108224 atomicity.py:140] Creating tmp directory gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_05_fp8/checkpoints/10 I0417 02:56:20.250805 138719853598528 metric_logger.py:185] completed step: 12, seconds: 0.008, TFLOP/s/device: 1645.927, Tokens/s/device: 248092.065, total_weights: 65536, loss: 7.032 I0417 02:56:20.408896 138719853598528 metric_logger.py:185] completed step: 13, seconds: 0.156, TFLOP/s/device: 87.068, Tokens/s/device: 13123.831, total_weights: 65536, loss: 6.815 I0417 02:56:20.567799 138719853598528 metric_logger.py:185] completed step: 14, seconds: 0.158, TFLOP/s/device: 85.856, Tokens/s/device: 12941.139, total_weights: 65536, loss: 6.635 I0417 02:56:20.726115 138719853598528 metric_logger.py:185] completed step: 15, seconds: 0.158, TFLOP/s/device: 85.861, Tokens/s/device: 12941.957, total_weights: 65536, loss: 6.492 I0417 02:56:21.239207 138719853598528 metric_logger.py:185] completed step: 16, seconds: 0.159, TFLOP/s/device: 85.616, Tokens/s/device: 12905.015, total_weights: 65536, loss: 6.380 I0417 02:56:21.396171 138719853598528 metric_logger.py:185] completed step: 17, seconds: 0.664, TFLOP/s/device: 20.471, Tokens/s/device: 3085.578, total_weights: 65536, loss: 6.293 I0417 02:56:21.554342 138719853598528 metric_logger.py:185] completed step: 18, seconds: 0.006, TFLOP/s/device: 2105.879, Tokens/s/device: 317420.955, total_weights: 65536, loss: 6.223 I0417 02:56:21.712430 138719853598528 checkpointing.py:794] Waiting for step 19 to finish before checkpoint... I0417 02:56:21.713465 138719853598528 checkpointing.py:798] Waited 0.0010557174682617188 seconds for step 19 to finish before starting checkpointing. I0417 02:56:21.715526 138719853598528 checkpoint_manager.py:2020] [process=0][thread=MainThread][step=10][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0417 02:56:22.798981 138586707977984 atomicity.py:140] Creating tmp directory gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_05_fp8/checkpoints/10/items I0417 02:56:26.416069 138587783378688 array_metadata_store.py:203] [process=0][thread=array_type_handler] Wrote 81 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_05_fp8/checkpoints/10/items/array_metadatas/process_0 I0417 02:57:05.580528 138587749656320 base_pytree_checkpoint_handler.py:1282] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 41.648458s (commit=40.423589s, array_metadata_write=1.224869s) I0417 02:57:05.581789 138587228796672 base_pytree_checkpoint_handler.py:130] [process=0] /jax/orbax/write/gbytes_per_sec: 34.564 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 45.70605444908142 s) (per-host) I0417 02:57:05.581933 138587228796672 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 45.656868s. I0417 02:57:07.643125 138587228796672 ocdbt_utils.py:49] Param validation support for Zarr3 will be added later (b/362328389). I0417 02:57:07.757303 138587228796672 array_metadata_store.py:411] [process=0][thread=async_save] Validated ArrayMetadata from all 8 hosts. Time taken: 0.000469s. I0417 02:57:08.692631 138587228796672 base_pytree_checkpoint_handler.py:1406] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 1.919400s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_05_fp8/checkpoints/10/items I0417 02:57:08.694400 138587228796672 atomicity.py:666] Finalizing gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_05_fp8/checkpoints/10/items I0417 02:57:09.213774 138587228796672 atomicity.py:666] Finalizing gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_05_fp8/checkpoints/10 I0417 02:57:10.659397 138587228796672 atomicity.py:847] [process=0][thread=async_save] Finished saving checkpoint (finalized tmp dir) to `gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_05_fp8/checkpoints/10`. I0417 02:57:10.660285 138587228796672 event_tracking.py:138] [process=0] [async] Finished save (blocking + background) in 51.51 seconds @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_05_fp8/checkpoints/10 I0417 02:57:10.661875 138587228796672 async_checkpointer.py:160] [process=0][thread=async_save] Background save thread done. Time taken: 50.736809s. I0417 02:57:10.662095 138588284434176 async_checkpointer.py:288] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0417 02:57:10.662218 138588284434176 async_checkpointer.py:298] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0417 02:57:10.662282 138588284434176 checkpoint_manager.py:2137] [process=0][thread=save_finalize][step=10] CheckpointManager Save Finalize is syncing with other hosts... I0417 02:57:10.663903 138588284434176 checkpoint_manager.py:2146] [process=0][thread=save_finalize][step=10] CheckpointManager Save Finalize is done on all hosts. I0417 02:57:10.664036 138719853598528 checkpoint_manager.py:2032] [process=0][thread=MainThread][step=10][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=10. W0417 02:57:10.664139 138719853598528 checkpoint_manager.py:1452] Waiting for previous save to complete took 48.948615 seconds. If this number is high, consider checkpointing less frequently. I0417 02:57:10.665793 138719853598528 checkpoint_manager.py:1512] [process=0] Saving checkpoint at step 19 I0417 02:57:10.667947 138719853598528 event_tracking.py:70] [process=0] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_05_fp8/checkpoints/19. I0417 02:57:11.424042 138719853598528 jax_array_handlers.py:360] Scheduling D2H of 81 prioritized jax.Array. I0417 02:57:11.424140 138719853598528 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 I0417 02:57:11.462011 138719853598528 base_pytree_checkpoint_handler.py:154] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.038784s I0417 02:57:11.462809 138719853598528 base_pytree_checkpoint_handler.py:130] [process=0] /jax/orbax/write/blocking_gbytes_per_sec: 33.454 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.04611492156982422 s) (per-host) I0417 02:57:11.462889 138719853598528 base_pytree_checkpoint_handler.py:768] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 0.046212s (batch_requests_ready=0.003017s, total_serialization_initiated=0.042439s, others=0.000756s) I0417 02:57:11.462971 138719853598528 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.050189s (all_items=0.000012s, per_item={'items': '0.00001192'}, temp_paths=0.050177) I0417 02:57:11.463673 138719853598528 event_tracking.py:125] [process=0] [async] Finished blocking save in 0.80 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_05_fp8/checkpoints/19. I0417 02:57:11.463976 138588284434176 async_checkpointer.py:76] [process=0][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-17 03:17:11.463942 I0417 02:57:11.466160 138719853598528 checkpoint_manager.py:1560] [process=0][thread=MainThread][step=19] Starting CheckpointManager Save Finalize thread=save_finalize I0417 02:57:11.466408 138589400725248 async_checkpointer.py:280] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0417 02:57:11.466534 138719853598528 standard_logger.py:34] {'step': 19, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_05_fp8/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776394581.715497, 'wait_for_prev_duration_secs': 48.948615312576294, 'time_between_consecutive_saves_sec': 2.5699522495269775, 'checkpointer_blocking_start_time': 1776394630.6658309, 'checkpointer_blocking_duration_secs': 0.7982800006866455, 'get_old_steps_start_time': 1776394631.4641287, 'get_old_steps_duration_secs': 2.47955322265625e-05, 'checkpoint_manager_blocking_start_time': 1776394581.713751, 'checkpoint_manager_blocking_duration_secs': 49.75274848937988} I0417 02:57:11.466643 138719853598528 checkpointing.py:409] Started an asynchronous checkpoint save for step 19 I0417 02:57:11.466685 138719853598528 checkpoint_manager.py:2020] [process=0][thread=MainThread][step=19][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0417 02:57:11.655683 138587228796672 atomicity.py:140] Creating tmp directory gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_05_fp8/checkpoints/19 I0417 02:57:13.801571 138587783378688 atomicity.py:140] Creating tmp directory gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_05_fp8/checkpoints/19/items I0417 02:57:17.118712 138592059873024 array_metadata_store.py:203] [process=0][thread=array_type_handler] Wrote 81 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_05_fp8/checkpoints/19/items/array_metadatas/process_0 I0417 02:57:57.067484 138589881108224 base_pytree_checkpoint_handler.py:1282] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 41.709712s (commit=40.536648s, array_metadata_write=1.173064s) I0417 02:57:57.068691 138588284434176 base_pytree_checkpoint_handler.py:130] [process=0] /jax/orbax/write/gbytes_per_sec: 34.604 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 45.6519672870636 s) (per-host) I0417 02:57:57.068817 138588284434176 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 45.604730s. I0417 02:57:58.754400 138588284434176 ocdbt_utils.py:49] Param validation support for Zarr3 will be added later (b/362328389). I0417 02:57:58.950366 138588284434176 array_metadata_store.py:411] [process=0][thread=async_save] Validated ArrayMetadata from all 8 hosts. Time taken: 0.000466s. I0417 02:57:59.841068 138588284434176 base_pytree_checkpoint_handler.py:1406] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 1.914574s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_05_fp8/checkpoints/19/items I0417 02:57:59.842607 138588284434176 atomicity.py:666] Finalizing gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_05_fp8/checkpoints/19/items I0417 02:58:00.370525 138588284434176 atomicity.py:666] Finalizing gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_05_fp8/checkpoints/19 I0417 02:58:02.309966 138588284434176 atomicity.py:847] [process=0][thread=async_save] Finished saving checkpoint (finalized tmp dir) to `gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_05_fp8/checkpoints/19`. I0417 02:58:02.310765 138588284434176 event_tracking.py:138] [process=0] [async] Finished save (blocking + background) in 51.64 seconds @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_05_fp8/checkpoints/19 I0417 02:58:02.312226 138588284434176 async_checkpointer.py:160] [process=0][thread=async_save] Background save thread done. Time taken: 50.848137s. I0417 02:58:02.312351 138589400725248 async_checkpointer.py:288] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0417 02:58:02.312406 138589400725248 async_checkpointer.py:298] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0417 02:58:02.312460 138589400725248 checkpoint_manager.py:2137] [process=0][thread=save_finalize][step=19] CheckpointManager Save Finalize is syncing with other hosts... I0417 02:58:02.314545 138589400725248 checkpoint_manager.py:2146] [process=0][thread=save_finalize][step=19] CheckpointManager Save Finalize is done on all hosts. I0417 02:58:02.314713 138719853598528 checkpoint_manager.py:2032] [process=0][thread=MainThread][step=19][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=19. I0417 02:58:02.314879 138719853598528 checkpoint_manager.py:2009] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0417 02:58:02.315570 138719853598528 metric_logger.py:185] completed step: 19, seconds: 0.158, TFLOP/s/device: 86.020, Tokens/s/device: 12965.882, total_weights: 65536, loss: 6.167 Per train step: Total TFLOPs: 13.59 split as 93.93% learnable weight flops and 6.07% attention flops XPK End: Fri Apr 17 02:58:13 UTC 2026 EXIT_CODE=0
XPK Start: Fri Apr 17 04:55:13 UTC 2026 PyTorch was not found. Models won't be available and only tokenizers, configuration and file/data utilities can be used. 2026-04-17 04:55:37.129573: 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) I0417 04:55:37.305566 137044570789696 max_utils.py:273] Attempting to initialize the jax distributed system... I0417 04:55:46.344915 137044570789696 distributed.py:149] Starting JAX distributed service on [::]:8482 I0417 04:55:46.347131 137044570789696 distributed.py:172] Connecting to JAX distributed service on mt-05-fp8-7n1qo-slice-job-0-0.mt-05-fp8-7n1qo:8482 I0417 04:55:47.842010 137044570789696 max_utils.py:284] Jax distributed system initialized! I0417 04:55:53.930458 137044570789696 max_utils.py:800] System Information: Jax Version: 0.9.2 I0417 04:55:53.930557 137044570789696 max_utils.py:801] System Information: Jaxlib Version: 0.9.2 I0417 04:55:53.930597 137044570789696 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 I0417 04:55:53.930631 137044570789696 train_utils.py:377] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing. I0417 04:55:54.596069 137044570789696 maxtext_utils.py:1687] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0417 04:55:55.026422 137044570789696 checkpointing.py:688] Setting up checkpoint logger... I0417 04:55:55.026541 137044570789696 checkpointing.py:234] Creating checkpoint manager with ocdbt=True and zarr3=True I0417 04:55:55.026586 137044570789696 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None I0417 04:55:55.026803 137044570789696 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 0x7ca3753504d0>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0417 04:55:58.144815 137044570789696 checkpointing.py:266] Enabling policy for fixed interval checkpointing. I0417 04:55:58.145187 137044570789696 checkpoint_manager.py:708] [process=0][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=('items',), item_handlers={'items': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7c8e3c4a97f0>}, handler_registry=None I0417 04:55:58.145430 137044570789696 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7c8e3c4a97f0>` 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`. I0417 04:55:58.145481 137044570789696 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7c8e3c78dfa0>` 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`. I0417 04:55:58.145514 137044570789696 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 0x7c8e3c4a97f0>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7c8e3c4a97f0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7c8e3c78dfa0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7c8e3c78dfa0>}). I0417 04:55:58.145855 137044570789696 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34 I0417 04:55:58.145931 137044570789696 async_checkpointer.py:192] [process=0][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7c8e3c7cdb20> timeout: 1200 secs and primary_host=0 for async checkpoint writes I0417 04:56:00.378856 137044570789696 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/nnx_xpk_feat_nnx_post_train_fixes_20260417_023710_05_fp8/checkpoints I0417 04:56:00.398098 137044570789696 checkpoint_manager.py:929] [process=0][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/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/nnx_xpk_feat_nnx_post_train_fixes_20260417_023710_05_fp8/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7c8e3c4abb60> I0417 04:56:00.398239 137044570789696 checkpointing.py:302] Checkpoint manager created! I0417 04:56:01.645884 137044570789696 checkpointing.py:578] checkpoint manager exists so trying to load this run's existing checkpoint I0417 04:56:01.645996 137044570789696 checkpointing.py:676] No existing checkpoints found, not restoring checkpoint. fsdp: 32 I0417 04:56:03.739109 137044570789696 maxtext_utils.py:1805] decoder/decoder_norm/scale/value Shape: float32[2048] Physical: (None,) I0417 04:56:03.739223 137044570789696 maxtext_utils.py:1805] decoder/layers/mlp/wi_0/kernel/value Shape: float32[2048,16,7168] Physical: ('fsdp', None, None) I0417 04:56:03.739283 137044570789696 maxtext_utils.py:1805] decoder/layers/mlp/wi_1/kernel/value Shape: float32[2048,16,7168] Physical: ('fsdp', None, None) I0417 04:56:03.739327 137044570789696 maxtext_utils.py:1805] decoder/layers/mlp/wo/kernel/value Shape: float32[7168,16,2048] Physical: (None, None, 'fsdp') I0417 04:56:03.739368 137044570789696 maxtext_utils.py:1805] decoder/layers/post_self_attention_layer_norm/scale/value Shape: float32[2048,16] Physical: (None, None) I0417 04:56:03.739423 137044570789696 maxtext_utils.py:1805] decoder/layers/pre_self_attention_layer_norm/scale/value Shape: float32[2048,16] Physical: (None, None) I0417 04:56:03.739474 137044570789696 maxtext_utils.py:1805] decoder/layers/self_attention/key/kernel/value Shape: float32[2048,16,16,128] Physical: ('fsdp', None, None, None) I0417 04:56:03.739511 137044570789696 maxtext_utils.py:1805] decoder/layers/self_attention/out/kernel/value Shape: float32[16,16,128,2048] Physical: (None, None, None, 'fsdp') I0417 04:56:03.739547 137044570789696 maxtext_utils.py:1805] decoder/layers/self_attention/query/kernel/value Shape: float32[2048,16,16,128] Physical: ('fsdp', None, None, None) I0417 04:56:03.739582 137044570789696 maxtext_utils.py:1805] decoder/layers/self_attention/value/kernel/value Shape: float32[2048,16,16,128] Physical: ('fsdp', None, None, None) I0417 04:56:03.739614 137044570789696 maxtext_utils.py:1805] decoder/logits_dense/kernel/value Shape: float32[2048,32000] Physical: ('fsdp', None) I0417 04:56:03.739647 137044570789696 maxtext_utils.py:1805] token_embedder/embedding/value Shape: float32[32000,2048] Physical: (None, 'fsdp') I0417 04:56:04.156651 137044570789696 nnx_decoders.py:465] nnx_decoders/carry Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0417 04:56:04.156740 137044570789696 nnx_decoders.py:465] nnx_decoders/carry Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0417 04:56:04.162181 137044570789696 nnx_decoders.py:465] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0417 04:56:04.162235 137044570789696 nnx_decoders.py:465] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0417 04:56:04.178802 137044570789696 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0417 04:56:04.178878 137044570789696 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0417 04:56:04.194417 137044570789696 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0417 04:56:04.194473 137044570789696 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0417 04:56:04.254537 137044570789696 attentions.py:1154] attentions/query Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0417 04:56:04.254627 137044570789696 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0417 04:56:04.270359 137044570789696 attentions.py:1155] attentions/key Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0417 04:56:04.270421 137044570789696 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0417 04:56:04.286036 137044570789696 attentions.py:1156] attentions/value Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0417 04:56:04.286096 137044570789696 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0417 04:56:04.315657 137044570789696 attentions.py:1197] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv'). I0417 04:56:04.315728 137044570789696 attentions.py:1197] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0417 04:56:04.376015 137044570789696 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0417 04:56:04.376094 137044570789696 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None). I0417 04:56:18.302630 137044570789696 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. I0417 04:56:18.313398 137044570789696 metric_logger.py:289] number parameters: 1.104 billion I0417 04:56:32.338643 137044570789696 checkpointing.py:794] Waiting for step 0 to finish before checkpoint... I0417 04:56:32.499281 137044570789696 checkpointing.py:798] Waited 0.16062259674072266 seconds for step 0 to finish before starting checkpointing. I0417 04:56:32.501621 137044570789696 checkpoint_manager.py:2009] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0417 04:56:32.503527 137044570789696 checkpoint_manager.py:1512] [process=0] Saving checkpoint at step 0 I0417 04:56:32.504939 137044570789696 event_tracking.py:70] [process=0] [async] Started save checkpoint @ gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/nnx_xpk_feat_nnx_post_train_fixes_20260417_023710_05_fp8/checkpoints/0. I0417 04:56:33.232515 137044570789696 signaling_client.py:364] Using JaxDistributedSignalingClient I0417 04:56:33.262051 137044570789696 jax_array_handlers.py:360] Scheduling D2H of 153 prioritized jax.Array. I0417 04:56:33.262130 137044570789696 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 I0417 04:56:33.504940 136914010375936 atomicity.py:140] Creating tmp directory gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/nnx_xpk_feat_nnx_post_train_fixes_20260417_023710_05_fp8/checkpoints/0 I0417 04:56:33.675528 137044570789696 base_pytree_checkpoint_handler.py:154] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.415893s I0417 04:56:33.676282 137044570789696 base_pytree_checkpoint_handler.py:130] [process=0] /jax/orbax/write/blocking_gbytes_per_sec: 3.537 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.4361696243286133 s) (per-host) I0417 04:56:33.676349 137044570789696 base_pytree_checkpoint_handler.py:768] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 0.436256s (batch_requests_ready=0.006160s, total_serialization_initiated=0.429424s, others=0.000672s) I0417 04:56:33.676437 137044570789696 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.440076s (all_items=0.000018s, per_item={'items': '0.00001812'}, temp_paths=0.440058) I0417 04:56:33.677157 137044570789696 event_tracking.py:125] [process=0] [async] Finished blocking save in 1.17 seconds. Continuing save @ gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/nnx_xpk_feat_nnx_post_train_fixes_20260417_023710_05_fp8/checkpoints/0. I0417 04:56:33.677510 136914018768640 async_checkpointer.py:76] [process=0][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-17 05:16:33.677473 I0417 04:56:33.688996 137044570789696 checkpoint_manager.py:1560] [process=0][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize I0417 04:56:33.689311 136913482671872 async_checkpointer.py:280] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0417 04:56:33.689475 137044570789696 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/nnx_xpk_feat_nnx_post_train_fixes_20260417_023710_05_fp8/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776401792.5016024, 'wait_for_prev_duration_secs': 6.175041198730469e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776401792.5035658, 'checkpointer_blocking_duration_secs': 1.1740882396697998, 'get_old_steps_start_time': 1776401793.6776772, 'get_old_steps_duration_secs': 2.6941299438476562e-05, 'checkpoint_manager_blocking_start_time': 1776401792.4997468, 'checkpoint_manager_blocking_duration_secs': 1.1896858215332031} I0417 04:56:33.689597 137044570789696 checkpointing.py:409] Started an asynchronous checkpoint save for step 0 I0417 04:56:33.689677 137044570789696 max_utils.py:750] Memstats: After params initialized: I0417 04:56:33.689729 137044570789696 max_utils.py:756] Using (GB) 0.45 / 31.25 (1.440000%) on TPU_0(process=0,(0,0,0,0)) I0417 04:56:33.689764 137044570789696 max_utils.py:756] Using (GB) 0.45 / 31.25 (1.440000%) on TPU_1(process=0,(1,0,0,0)) I0417 04:56:33.689791 137044570789696 max_utils.py:756] Using (GB) 0.45 / 31.25 (1.440000%) on TPU_4(process=0,(0,1,0,0)) I0417 04:56:33.689815 137044570789696 max_utils.py:756] Using (GB) 0.45 / 31.25 (1.440000%) on TPU_5(process=0,(1,1,0,0)) I0417 04:56:34.029515 137044570789696 metric_logger.py:185] completed step: 0, seconds: 13.611, TFLOP/s/device: 0.998, Tokens/s/device: 150.462, total_weights: 65536, loss: 10.872 I0417 04:56:34.032421 137044570789696 metric_logger.py:269] To see full metrics 'tensorboard --logdir=gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/nnx_xpk_feat_nnx_post_train_fixes_20260417_023710_05_fp8/tensorboard/' I0417 04:56:34.473246 137044570789696 metric_logger.py:185] completed step: 1, seconds: 1.685, TFLOP/s/device: 8.065, Tokens/s/device: 1215.631, total_weights: 65536, loss: 10.872 I0417 04:56:34.619543 137044570789696 metric_logger.py:185] completed step: 2, seconds: 0.445, TFLOP/s/device: 30.514, Tokens/s/device: 4599.415, total_weights: 65536, loss: 10.856 I0417 04:56:34.776196 137044570789696 metric_logger.py:185] completed step: 3, seconds: 0.018, TFLOP/s/device: 766.552, Tokens/s/device: 115543.018, total_weights: 65536, loss: 10.824 I0417 04:56:36.292743 136913993590528 atomicity.py:140] Creating tmp directory gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/nnx_xpk_feat_nnx_post_train_fixes_20260417_023710_05_fp8/checkpoints/0/items I0417 04:56:36.874659 136912935827200 checkpoint.py:188] Wrote Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776401796042295206, 'commit_timestamp_nsecs': None, 'custom_metadata': {}}, json={"item_handlers": null, "metrics": {}, "performance_metrics": {}, "init_timestamp_nsecs": 1776401796042295206, "commit_timestamp_nsecs": null, "custom_metadata": {}} to gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/nnx_xpk_feat_nnx_post_train_fixes_20260417_023710_05_fp8/checkpoints/0/_CHECKPOINT_METADATA I0417 04:56:37.448727 2869 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com I0417 04:56:39.500908 136913970145024 array_metadata_store.py:203] [process=0][thread=array_type_handler] Wrote 153 array_metadata.ArrayMetadata to gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/nnx_xpk_feat_nnx_post_train_fixes_20260417_023710_05_fp8/checkpoints/0/items/array_metadatas/process_0 I0417 04:56:57.040228 136913491064576 base_pytree_checkpoint_handler.py:1282] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 19.602865s (commit=17.978097s, array_metadata_write=1.624768s) I0417 04:56:57.041494 136914018768640 base_pytree_checkpoint_handler.py:130] [process=0] /jax/orbax/write/gbytes_per_sec: 66.373 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 23.801358699798584 s) (per-host) I0417 04:56:57.041604 136914018768640 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 23.363985s. I0417 04:56:58.073459 136914018768640 checkpoint.py:228] Read Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776401796042295206, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} from gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/nnx_xpk_feat_nnx_post_train_fixes_20260417_023710_05_fp8/checkpoints/0/_CHECKPOINT_METADATA I0417 04:56:58.729018 137044570789696 metric_logger.py:185] completed step: 4, seconds: 0.146, TFLOP/s/device: 93.070, Tokens/s/device: 14028.550, total_weights: 65536, loss: 10.793 I0417 04:56:58.741307 137044570789696 metric_logger.py:185] completed step: 5, seconds: 0.157, TFLOP/s/device: 86.603, Tokens/s/device: 13053.815, total_weights: 65536, loss: 10.763 I0417 04:56:58.893153 137044570789696 metric_logger.py:185] completed step: 6, seconds: 23.950, TFLOP/s/device: 0.567, Tokens/s/device: 85.510, total_weights: 65536, loss: 10.733 I0417 04:56:59.050795 137044570789696 metric_logger.py:185] completed step: 7, seconds: 0.010, TFLOP/s/device: 1379.824, Tokens/s/device: 207982.127, total_weights: 65536, loss: 10.706 I0417 04:56:59.207789 137044570789696 metric_logger.py:185] completed step: 8, seconds: 0.157, TFLOP/s/device: 86.601, Tokens/s/device: 13053.482, total_weights: 65536, loss: 10.680 I0417 04:56:59.218770 137044570789696 checkpointing.py:794] Waiting for step 10 to finish before checkpoint... I0417 04:56:59.435542 136914018768640 ocdbt_utils.py:49] Param validation support for Zarr3 will be added later (b/362328389). I0417 04:56:59.459677 136914018768640 array_metadata_store.py:411] [process=0][thread=async_save] Validated ArrayMetadata from all 8 hosts. Time taken: 0.000921s. I0417 04:56:59.512952 136912935827200 checkpoint.py:247] Updated Metadata={'item_handlers': {'items': 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler'}, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776401796042295206, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} to gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/nnx_xpk_feat_nnx_post_train_fixes_20260417_023710_05_fp8/checkpoints/0/_CHECKPOINT_METADATA I0417 04:56:59.522158 137044570789696 checkpointing.py:798] Waited 0.3033571243286133 seconds for step 10 to finish before starting checkpointing. I0417 04:56:59.524533 137044570789696 checkpoint_manager.py:2020] [process=0][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0417 04:57:00.408266 136914018768640 base_pytree_checkpoint_handler.py:1406] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 2.210847s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/nnx_xpk_feat_nnx_post_train_fixes_20260417_023710_05_fp8/checkpoints/0/items I0417 04:57:00.410144 136914018768640 atomicity.py:666] Finalizing gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/nnx_xpk_feat_nnx_post_train_fixes_20260417_023710_05_fp8/checkpoints/0/items I0417 04:57:00.917009 136914018768640 atomicity.py:666] Finalizing gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/nnx_xpk_feat_nnx_post_train_fixes_20260417_023710_05_fp8/checkpoints/0 I0417 04:57:02.283664 136914018768640 atomicity.py:847] [process=0][thread=async_save] Finished saving checkpoint (finalized tmp dir) to `gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/nnx_xpk_feat_nnx_post_train_fixes_20260417_023710_05_fp8/checkpoints/0`. I0417 04:57:02.284504 136914018768640 event_tracking.py:138] [process=0] [async] Finished save (blocking + background) in 29.78 seconds @ gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/nnx_xpk_feat_nnx_post_train_fixes_20260417_023710_05_fp8/checkpoints/0 I0417 04:57:02.285800 136914018768640 async_checkpointer.py:160] [process=0][thread=async_save] Background save thread done. Time taken: 28.608178s. I0417 04:57:02.286016 136913482671872 async_checkpointer.py:288] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0417 04:57:02.286153 136913482671872 async_checkpointer.py:298] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0417 04:57:02.286244 136913482671872 checkpoint_manager.py:2137] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts... I0417 04:57:02.289187 136913482671872 checkpoint_manager.py:2146] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts. I0417 04:57:02.289309 137044570789696 checkpoint_manager.py:2032] [process=0][thread=MainThread][step=0][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=0. W0417 04:57:02.289376 137044570789696 checkpoint_manager.py:1452] Waiting for previous save to complete took 2.764850 seconds. If this number is high, consider checkpointing less frequently. I0417 04:57:02.291370 137044570789696 checkpoint_manager.py:1512] [process=0] Saving checkpoint at step 10 I0417 04:57:02.293425 137044570789696 event_tracking.py:70] [process=0] [async] Started save checkpoint @ gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/nnx_xpk_feat_nnx_post_train_fixes_20260417_023710_05_fp8/checkpoints/10. I0417 04:57:03.032800 137044570789696 jax_array_handlers.py:360] Scheduling D2H of 153 prioritized jax.Array. I0417 04:57:03.032915 137044570789696 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 I0417 04:57:03.072148 137044570789696 base_pytree_checkpoint_handler.py:154] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.041654s I0417 04:57:03.072883 137044570789696 base_pytree_checkpoint_handler.py:130] [process=0] /jax/orbax/write/blocking_gbytes_per_sec: 25.862 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.059652090072631836 s) (per-host) I0417 04:57:03.072952 137044570789696 base_pytree_checkpoint_handler.py:768] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 0.059737s (batch_requests_ready=0.005896s, total_serialization_initiated=0.053163s, others=0.000678s) I0417 04:57:03.073033 137044570789696 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.063812s (all_items=0.000016s, per_item={'items': '0.00001621'}, temp_paths=0.063796) I0417 04:57:03.073746 137044570789696 event_tracking.py:125] [process=0] [async] Finished blocking save in 0.78 seconds. Continuing save @ gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/nnx_xpk_feat_nnx_post_train_fixes_20260417_023710_05_fp8/checkpoints/10. I0417 04:57:03.074049 136913482671872 async_checkpointer.py:76] [process=0][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-17 05:17:03.074018 I0417 04:57:03.076053 137044570789696 checkpoint_manager.py:1560] [process=0][thread=MainThread][step=10] Starting CheckpointManager Save Finalize thread=save_finalize I0417 04:57:03.076293 136911872059136 async_checkpointer.py:280] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0417 04:57:03.076400 137044570789696 standard_logger.py:34] {'step': 10, 'event_type': 'save', 'directory': 'gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/nnx_xpk_feat_nnx_post_train_fixes_20260417_023710_05_fp8/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776401819.5245, 'wait_for_prev_duration_secs': 2.76485013961792, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776401822.2914248, 'checkpointer_blocking_duration_secs': 0.7827706336975098, 'get_old_steps_start_time': 1776401823.0742226, 'get_old_steps_duration_secs': 3.504753112792969e-05, 'checkpoint_manager_blocking_start_time': 1776401819.5226107, 'checkpoint_manager_blocking_duration_secs': 3.5537564754486084} I0417 04:57:03.076581 137044570789696 checkpointing.py:409] Started an asynchronous checkpoint save for step 10 I0417 04:57:03.077013 137044570789696 metric_logger.py:185] completed step: 9, seconds: 0.154, TFLOP/s/device: 88.102, Tokens/s/device: 13279.730, total_weights: 65536, loss: 10.656 I0417 04:57:03.091024 137044570789696 metric_logger.py:185] completed step: 10, seconds: 0.157, TFLOP/s/device: 86.331, Tokens/s/device: 13012.759, total_weights: 65536, loss: 10.636 I0417 04:57:03.240563 137044570789696 metric_logger.py:185] completed step: 11, seconds: 3.869, TFLOP/s/device: 3.512, Tokens/s/device: 529.295, total_weights: 65536, loss: 10.618 I0417 04:57:03.266059 136914018768640 atomicity.py:140] Creating tmp directory gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/nnx_xpk_feat_nnx_post_train_fixes_20260417_023710_05_fp8/checkpoints/10 I0417 04:57:03.397380 137044570789696 metric_logger.py:185] completed step: 12, seconds: 0.012, TFLOP/s/device: 1104.914, Tokens/s/device: 166544.686, total_weights: 65536, loss: 10.602 I0417 04:57:03.554283 137044570789696 metric_logger.py:185] completed step: 13, seconds: 0.151, TFLOP/s/device: 89.817, Tokens/s/device: 13538.258, total_weights: 65536, loss: 10.588 I0417 04:57:03.710894 137044570789696 metric_logger.py:185] completed step: 14, seconds: 0.157, TFLOP/s/device: 86.755, Tokens/s/device: 13076.653, total_weights: 65536, loss: 10.577 I0417 04:57:03.867770 137044570789696 metric_logger.py:185] completed step: 15, seconds: 0.157, TFLOP/s/device: 86.724, Tokens/s/device: 13071.979, total_weights: 65536, loss: 10.568 I0417 04:57:04.371639 137044570789696 metric_logger.py:185] completed step: 16, seconds: 0.157, TFLOP/s/device: 86.770, Tokens/s/device: 13078.991, total_weights: 65536, loss: 10.562 I0417 04:57:04.523741 137044570789696 metric_logger.py:185] completed step: 17, seconds: 0.649, TFLOP/s/device: 20.920, Tokens/s/device: 3153.287, total_weights: 65536, loss: 10.556 I0417 04:57:04.680993 137044570789696 metric_logger.py:185] completed step: 18, seconds: 0.010, TFLOP/s/device: 1378.984, Tokens/s/device: 207855.475, total_weights: 65536, loss: 10.552 I0417 04:57:04.839975 137044570789696 checkpointing.py:794] Waiting for step 19 to finish before checkpoint... I0417 04:57:04.841246 137044570789696 checkpointing.py:798] Waited 0.0012807846069335938 seconds for step 19 to finish before starting checkpointing. I0417 04:57:04.843410 137044570789696 checkpoint_manager.py:2020] [process=0][thread=MainThread][step=10][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0417 04:57:05.089672 136911880451840 atomicity.py:140] Creating tmp directory gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/nnx_xpk_feat_nnx_post_train_fixes_20260417_023710_05_fp8/checkpoints/10/items I0417 04:57:08.670809 136912400537344 array_metadata_store.py:203] [process=0][thread=array_type_handler] Wrote 153 array_metadata.ArrayMetadata to gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/nnx_xpk_feat_nnx_post_train_fixes_20260417_023710_05_fp8/checkpoints/10/items/array_metadatas/process_0 I0417 04:57:27.041225 136912383751936 base_pytree_checkpoint_handler.py:1282] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 20.815294s (commit=19.155414s, array_metadata_write=1.659880s) I0417 04:57:27.042509 136913482671872 base_pytree_checkpoint_handler.py:130] [process=0] /jax/orbax/write/gbytes_per_sec: 65.744 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 24.029250144958496 s) (per-host) I0417 04:57:27.042619 136913482671872 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 23.968453s. I0417 04:57:29.527686 136913482671872 ocdbt_utils.py:49] Param validation support for Zarr3 will be added later (b/362328389). I0417 04:57:29.754334 136913482671872 array_metadata_store.py:411] [process=0][thread=async_save] Validated ArrayMetadata from all 8 hosts. Time taken: 0.000881s. I0417 04:57:30.459597 136913482671872 base_pytree_checkpoint_handler.py:1406] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 2.154770s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/nnx_xpk_feat_nnx_post_train_fixes_20260417_023710_05_fp8/checkpoints/10/items I0417 04:57:30.462266 136913482671872 atomicity.py:666] Finalizing gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/nnx_xpk_feat_nnx_post_train_fixes_20260417_023710_05_fp8/checkpoints/10/items I0417 04:57:30.980590 136913482671872 atomicity.py:666] Finalizing gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/nnx_xpk_feat_nnx_post_train_fixes_20260417_023710_05_fp8/checkpoints/10 I0417 04:57:32.346663 136913482671872 atomicity.py:847] [process=0][thread=async_save] Finished saving checkpoint (finalized tmp dir) to `gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/nnx_xpk_feat_nnx_post_train_fixes_20260417_023710_05_fp8/checkpoints/10`. I0417 04:57:32.347491 136913482671872 event_tracking.py:138] [process=0] [async] Finished save (blocking + background) in 30.06 seconds @ gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/nnx_xpk_feat_nnx_post_train_fixes_20260417_023710_05_fp8/checkpoints/10 I0417 04:57:32.349018 136913482671872 async_checkpointer.py:160] [process=0][thread=async_save] Background save thread done. Time taken: 29.274850s. I0417 04:57:32.349222 136911872059136 async_checkpointer.py:288] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0417 04:57:32.349346 136911872059136 async_checkpointer.py:298] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0417 04:57:32.349407 136911872059136 checkpoint_manager.py:2137] [process=0][thread=save_finalize][step=10] CheckpointManager Save Finalize is syncing with other hosts... I0417 04:57:32.350952 136911872059136 checkpoint_manager.py:2146] [process=0][thread=save_finalize][step=10] CheckpointManager Save Finalize is done on all hosts. I0417 04:57:32.351125 137044570789696 checkpoint_manager.py:2032] [process=0][thread=MainThread][step=10][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=10. W0417 04:57:32.351251 137044570789696 checkpoint_manager.py:1452] Waiting for previous save to complete took 27.507845 seconds. If this number is high, consider checkpointing less frequently. I0417 04:57:32.352744 137044570789696 checkpoint_manager.py:1512] [process=0] Saving checkpoint at step 19 I0417 04:57:32.354715 137044570789696 event_tracking.py:70] [process=0] [async] Started save checkpoint @ gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/nnx_xpk_feat_nnx_post_train_fixes_20260417_023710_05_fp8/checkpoints/19. I0417 04:57:33.507041 137044570789696 jax_array_handlers.py:360] Scheduling D2H of 153 prioritized jax.Array. I0417 04:57:33.507204 137044570789696 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 I0417 04:57:33.547025 137044570789696 base_pytree_checkpoint_handler.py:154] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.042357s I0417 04:57:33.547769 137044570789696 base_pytree_checkpoint_handler.py:130] [process=0] /jax/orbax/write/blocking_gbytes_per_sec: 25.625 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.060204505920410156 s) (per-host) I0417 04:57:33.547853 137044570789696 base_pytree_checkpoint_handler.py:768] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 0.060298s (batch_requests_ready=0.005712s, total_serialization_initiated=0.053915s, others=0.000671s) I0417 04:57:33.547944 137044570789696 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.064413s (all_items=0.000012s, per_item={'items': '0.00001168'}, temp_paths=0.064402) I0417 04:57:33.548667 137044570789696 event_tracking.py:125] [process=0] [async] Finished blocking save in 1.20 seconds. Continuing save @ gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/nnx_xpk_feat_nnx_post_train_fixes_20260417_023710_05_fp8/checkpoints/19. I0417 04:57:33.549007 136912400537344 async_checkpointer.py:76] [process=0][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-17 05:17:33.548963 I0417 04:57:33.551000 137044570789696 checkpoint_manager.py:1560] [process=0][thread=MainThread][step=19] Starting CheckpointManager Save Finalize thread=save_finalize I0417 04:57:33.551288 136911872059136 async_checkpointer.py:280] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0417 04:57:33.551425 137044570789696 standard_logger.py:34] {'step': 19, 'event_type': 'save', 'directory': 'gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/nnx_xpk_feat_nnx_post_train_fixes_20260417_023710_05_fp8/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776401824.8433814, 'wait_for_prev_duration_secs': 27.507845401763916, 'time_between_consecutive_saves_sec': 2.5541512966156006, 'checkpointer_blocking_start_time': 1776401852.352783, 'checkpointer_blocking_duration_secs': 1.196378231048584, 'get_old_steps_start_time': 1776401853.549181, 'get_old_steps_duration_secs': 2.47955322265625e-05, 'checkpoint_manager_blocking_start_time': 1776401824.8416135, 'checkpoint_manager_blocking_duration_secs': 28.709779262542725} I0417 04:57:33.551603 137044570789696 checkpointing.py:409] Started an asynchronous checkpoint save for step 19 I0417 04:57:33.551649 137044570789696 checkpoint_manager.py:2020] [process=0][thread=MainThread][step=19][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0417 04:57:33.752943 136920420030208 atomicity.py:140] Creating tmp directory gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/nnx_xpk_feat_nnx_post_train_fixes_20260417_023710_05_fp8/checkpoints/19 I0417 04:57:35.821428 136914018768640 atomicity.py:140] Creating tmp directory gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/nnx_xpk_feat_nnx_post_train_fixes_20260417_023710_05_fp8/checkpoints/19/items I0417 04:57:38.752990 136913985197824 array_metadata_store.py:203] [process=0][thread=array_type_handler] Wrote 153 array_metadata.ArrayMetadata to gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/nnx_xpk_feat_nnx_post_train_fixes_20260417_023710_05_fp8/checkpoints/19/items/array_metadatas/process_0 I0417 04:57:56.899021 136913482684160 base_pytree_checkpoint_handler.py:1282] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 19.922117s (commit=18.220132s, array_metadata_write=1.701985s) I0417 04:57:56.900375 136912400537344 base_pytree_checkpoint_handler.py:130] [process=0] /jax/orbax/write/gbytes_per_sec: 67.475 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 23.41278052330017 s) (per-host) I0417 04:57:56.900502 136912400537344 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 23.351380s. I0417 04:57:59.046091 136912400537344 ocdbt_utils.py:49] Param validation support for Zarr3 will be added later (b/362328389). I0417 04:57:59.103136 136912400537344 array_metadata_store.py:411] [process=0][thread=async_save] Validated ArrayMetadata from all 8 hosts. Time taken: 0.000939s. I0417 04:57:59.694106 136912400537344 base_pytree_checkpoint_handler.py:1406] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 2.005469s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/nnx_xpk_feat_nnx_post_train_fixes_20260417_023710_05_fp8/checkpoints/19/items I0417 04:57:59.695743 136912400537344 atomicity.py:666] Finalizing gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/nnx_xpk_feat_nnx_post_train_fixes_20260417_023710_05_fp8/checkpoints/19/items I0417 04:58:00.633875 136912400537344 atomicity.py:666] Finalizing gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/nnx_xpk_feat_nnx_post_train_fixes_20260417_023710_05_fp8/checkpoints/19 I0417 04:58:02.017382 136912400537344 atomicity.py:847] [process=0][thread=async_save] Finished saving checkpoint (finalized tmp dir) to `gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/nnx_xpk_feat_nnx_post_train_fixes_20260417_023710_05_fp8/checkpoints/19`. I0417 04:58:02.018069 136912400537344 event_tracking.py:138] [process=0] [async] Finished save (blocking + background) in 29.67 seconds @ gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/nnx_xpk_feat_nnx_post_train_fixes_20260417_023710_05_fp8/checkpoints/19 I0417 04:58:02.019352 136912400537344 async_checkpointer.py:160] [process=0][thread=async_save] Background save thread done. Time taken: 28.470227s. I0417 04:58:02.019478 136911872059136 async_checkpointer.py:288] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0417 04:58:02.019538 136911872059136 async_checkpointer.py:298] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0417 04:58:02.019607 136911872059136 checkpoint_manager.py:2137] [process=0][thread=save_finalize][step=19] CheckpointManager Save Finalize is syncing with other hosts... I0417 04:58:02.021731 136911872059136 checkpoint_manager.py:2146] [process=0][thread=save_finalize][step=19] CheckpointManager Save Finalize is done on all hosts. I0417 04:58:02.021907 137044570789696 checkpoint_manager.py:2032] [process=0][thread=MainThread][step=19][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=19. I0417 04:58:02.022091 137044570789696 checkpoint_manager.py:2009] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0417 04:58:02.022739 137044570789696 metric_logger.py:185] completed step: 19, seconds: 0.153, TFLOP/s/device: 88.558, Tokens/s/device: 13348.368, total_weights: 65536, loss: 10.549 Per train step: Total TFLOPs: 13.59 split as 93.93% learnable weight flops and 6.07% attention flops XPK End: Fri Apr 17 04:58:14 UTC 2026 EXIT_CODE=0