feat/nnx-trainstate-and-training-loop| Metric | Linen e27fc1e97 | NNX e27fc1e97 | Diff (NNX − Linen) |
|---|---|---|---|
| Parameters | 1.105 billion | 1.104 billion | — |
| Final loss | 6.1670 | 10.5490 | +4.382 |
| TFLOP/s | 85.939 | 86.537 | +0.598 |
| Tok/s | 12953.7 | 13043.8 | +90.175 |
| Avg s/step | 1.921 | 1.719 | -0.202 |
| 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: Tue Apr 21 08:46:28 UTC 2026 PyTorch was not found. Models won't be available and only tokenizers, configuration and file/data utilities can be used. 2026-04-21 08:46:52.612853: 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) I0421 08:46:53.150649 140642351560512 max_utils.py:273] Attempting to initialize the jax distributed system... I0421 08:47:02.192463 140642351560512 distributed.py:149] Starting JAX distributed service on [::]:8482 I0421 08:47:02.194844 140642351560512 distributed.py:172] Connecting to JAX distributed service on mt-05-fp8-536jq-slice-job-0-0.mt-05-fp8-536jq:8482 I0421 08:47:03.378498 140642351560512 max_utils.py:284] Jax distributed system initialized! I0421 08:47:09.461931 140642351560512 max_utils.py:800] System Information: Jax Version: 0.9.2 I0421 08:47:09.462034 140642351560512 max_utils.py:801] System Information: Jaxlib Version: 0.9.2 I0421 08:47:09.462076 140642351560512 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 I0421 08:47:09.462113 140642351560512 train_utils.py:378] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing. I0421 08:47:10.172908 140642351560512 maxtext_utils.py:1718] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0421 08:47:10.173495 140642351560512 maxtext_utils.py:1718] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0421 08:47:10.173679 140642351560512 checkpointing.py:688] Setting up checkpoint logger... I0421 08:47:10.173730 140642351560512 checkpointing.py:234] Creating checkpoint manager with ocdbt=True and zarr3=True I0421 08:47:10.173785 140642351560512 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None I0421 08:47:10.174113 140642351560512 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 0x7fe93d7182f0>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0421 08:47:13.079841 140642351560512 checkpointing.py:266] Enabling policy for fixed interval checkpointing. I0421 08:47:13.080082 140642351560512 checkpoint_manager.py:708] [process=7][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=('items',), item_handlers={'items': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7fd53c418710>}, handler_registry=None I0421 08:47:13.080316 140642351560512 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7fd53c418710>` 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`. I0421 08:47:13.080363 140642351560512 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7fd53c41a510>` 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`. I0421 08:47:13.080396 140642351560512 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 0x7fd53c418710>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7fd53c418710>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7fd53c41a510>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7fd53c41a510>}). I0421 08:47:13.080703 140642351560512 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34 I0421 08:47:13.080785 140642351560512 async_checkpointer.py:192] [process=7][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7fd43c767ec0> timeout: 1200 secs and primary_host=0 for async checkpoint writes I0421 08:47:14.643466 140642351560512 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260421_083238/linen_xpk_feat_nnx_trainstate_and_training_loop_20260421_083238_05_fp8/checkpoints I0421 08:47:14.665360 140642351560512 checkpoint_manager.py:929] [process=7][thread=MainThread] CheckpointManager created, primary_host=0, CheckpointManagerOptions=CheckpointManagerOptions(save_interval_steps=1, max_to_keep=None, keep_time_interval=None, keep_period=None, should_keep_fn=None, best_fn=None, best_mode='max', keep_checkpoints_without_metrics=True, step_prefix=None, step_format_fixed_length=None, step_name_format=None, create=True, cleanup_tmp_directories=False, save_on_steps=frozenset(), single_host_load_and_broadcast=False, todelete_subdir=None, todelete_full_path=None, enable_background_delete=False, read_only=False, enable_async_checkpointing=True, async_options=None, multiprocessing_options=MultiprocessingOptions(primary_host=0, active_processes=None, barrier_sync_key_prefix=None), should_save_fn=None, file_options=FileOptions(path_permission_mode=None), save_root_metadata=True, temporary_path_class=None, save_decision_policy=FixedIntervalPolicy(interval=10), preservation_policy=LatestN(n=None), prevent_write_metrics=False, enable_should_save_is_saving_in_progress_check=True, enable_per_process_directory_creation=False, lightweight_initialize=False), root_directory=gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260421_083238/linen_xpk_feat_nnx_trainstate_and_training_loop_20260421_083238_05_fp8/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7fd53c41bdd0> I0421 08:47:14.665490 140642351560512 checkpointing.py:302] Checkpoint manager created! I0421 08:47:15.751488 140642351560512 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0421 08:47:15.751591 140642351560512 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0421 08:47:16.141890 140642351560512 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0421 08:47:16.141981 140642351560512 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0421 08:47:16.158638 140642351560512 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0421 08:47:16.158697 140642351560512 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0421 08:47:16.211182 140642351560512 attentions.py:1154] attentions/query Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0421 08:47:16.211259 140642351560512 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0421 08:47:16.227998 140642351560512 attentions.py:1155] attentions/key Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0421 08:47:16.228055 140642351560512 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0421 08:47:16.244813 140642351560512 attentions.py:1156] attentions/value Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0421 08:47:16.244873 140642351560512 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0421 08:47:16.269720 140642351560512 attentions.py:1197] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv'). I0421 08:47:16.269800 140642351560512 attentions.py:1197] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0421 08:47:16.321338 140642351560512 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0421 08:47:16.321411 140642351560512 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None). I0421 08:47:16.758805 140642351560512 checkpointing.py:578] checkpoint manager exists so trying to load this run's existing checkpoint I0421 08:47:16.758911 140642351560512 checkpointing.py:676] No existing checkpoints found, not restoring checkpoint. fsdp: 32 I0421 08:47:18.681915 140642351560512 maxtext_utils.py:1821] params/_overwrite_with_gradient/decoder/layers/mlp/wi_0/Fp8DirectDotGeneralOp_0/input_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0421 08:47:18.682062 140642351560512 maxtext_utils.py:1821] params/_overwrite_with_gradient/decoder/layers/mlp/wi_0/Fp8DirectDotGeneralOp_0/input_scale Shape: float32[16,1] Logical: P() Physical: () I0421 08:47:18.682119 140642351560512 maxtext_utils.py:1821] params/_overwrite_with_gradient/decoder/layers/mlp/wi_0/Fp8DirectDotGeneralOp_0/kernel_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0421 08:47:18.682183 140642351560512 maxtext_utils.py:1821] params/_overwrite_with_gradient/decoder/layers/mlp/wi_0/Fp8DirectDotGeneralOp_0/kernel_scale Shape: float32[16,1] Logical: P() Physical: () I0421 08:47:18.682246 140642351560512 maxtext_utils.py:1821] params/_overwrite_with_gradient/decoder/layers/mlp/wi_0/Fp8DirectDotGeneralOp_0/output_grad_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0421 08:47:18.682306 140642351560512 maxtext_utils.py:1821] params/_overwrite_with_gradient/decoder/layers/mlp/wi_0/Fp8DirectDotGeneralOp_0/output_grad_scale Shape: float32[16,1] Logical: P() Physical: () I0421 08:47:18.682364 140642351560512 maxtext_utils.py:1821] params/_overwrite_with_gradient/decoder/layers/mlp/wi_1/Fp8DirectDotGeneralOp_0/input_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0421 08:47:18.682423 140642351560512 maxtext_utils.py:1821] params/_overwrite_with_gradient/decoder/layers/mlp/wi_1/Fp8DirectDotGeneralOp_0/input_scale Shape: float32[16,1] Logical: P() Physical: () I0421 08:47:18.682482 140642351560512 maxtext_utils.py:1821] params/_overwrite_with_gradient/decoder/layers/mlp/wi_1/Fp8DirectDotGeneralOp_0/kernel_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0421 08:47:18.682541 140642351560512 maxtext_utils.py:1821] params/_overwrite_with_gradient/decoder/layers/mlp/wi_1/Fp8DirectDotGeneralOp_0/kernel_scale Shape: float32[16,1] Logical: P() Physical: () I0421 08:47:18.682598 140642351560512 maxtext_utils.py:1821] params/_overwrite_with_gradient/decoder/layers/mlp/wi_1/Fp8DirectDotGeneralOp_0/output_grad_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0421 08:47:18.682656 140642351560512 maxtext_utils.py:1821] params/_overwrite_with_gradient/decoder/layers/mlp/wi_1/Fp8DirectDotGeneralOp_0/output_grad_scale Shape: float32[16,1] Logical: P() Physical: () I0421 08:47:18.682715 140642351560512 maxtext_utils.py:1821] params/_overwrite_with_gradient/decoder/layers/mlp/wo/Fp8DirectDotGeneralOp_0/input_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0421 08:47:18.682832 140642351560512 maxtext_utils.py:1821] params/_overwrite_with_gradient/decoder/layers/mlp/wo/Fp8DirectDotGeneralOp_0/input_scale Shape: float32[16,1] Logical: P() Physical: () I0421 08:47:18.682897 140642351560512 maxtext_utils.py:1821] params/_overwrite_with_gradient/decoder/layers/mlp/wo/Fp8DirectDotGeneralOp_0/kernel_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0421 08:47:18.682953 140642351560512 maxtext_utils.py:1821] params/_overwrite_with_gradient/decoder/layers/mlp/wo/Fp8DirectDotGeneralOp_0/kernel_scale Shape: float32[16,1] Logical: P() Physical: () I0421 08:47:18.683016 140642351560512 maxtext_utils.py:1821] params/_overwrite_with_gradient/decoder/layers/mlp/wo/Fp8DirectDotGeneralOp_0/output_grad_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0421 08:47:18.683074 140642351560512 maxtext_utils.py:1821] params/_overwrite_with_gradient/decoder/layers/mlp/wo/Fp8DirectDotGeneralOp_0/output_grad_scale Shape: float32[16,1] Logical: P() Physical: () I0421 08:47:18.683131 140642351560512 maxtext_utils.py:1821] params/_overwrite_with_gradient/decoder/layers/self_attention/key/Fp8DirectDotGeneralOp_0/input_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0421 08:47:18.683188 140642351560512 maxtext_utils.py:1821] params/_overwrite_with_gradient/decoder/layers/self_attention/key/Fp8DirectDotGeneralOp_0/input_scale Shape: float32[16,1] Logical: P() Physical: () I0421 08:47:18.683239 140642351560512 maxtext_utils.py:1821] params/_overwrite_with_gradient/decoder/layers/self_attention/key/Fp8DirectDotGeneralOp_0/kernel_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0421 08:47:18.683291 140642351560512 maxtext_utils.py:1821] params/_overwrite_with_gradient/decoder/layers/self_attention/key/Fp8DirectDotGeneralOp_0/kernel_scale Shape: float32[16,1] Logical: P() Physical: () I0421 08:47:18.683344 140642351560512 maxtext_utils.py:1821] params/_overwrite_with_gradient/decoder/layers/self_attention/key/Fp8DirectDotGeneralOp_0/output_grad_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0421 08:47:18.683442 140642351560512 maxtext_utils.py:1821] params/_overwrite_with_gradient/decoder/layers/self_attention/key/Fp8DirectDotGeneralOp_0/output_grad_scale Shape: float32[16,1] Logical: P() Physical: () I0421 08:47:18.683496 140642351560512 maxtext_utils.py:1821] params/_overwrite_with_gradient/decoder/layers/self_attention/out/Fp8DirectDotGeneralOp_0/input_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0421 08:47:18.683527 140642351560512 maxtext_utils.py:1821] params/_overwrite_with_gradient/decoder/layers/self_attention/out/Fp8DirectDotGeneralOp_0/input_scale Shape: float32[16,1] Logical: P() Physical: () I0421 08:47:18.683555 140642351560512 maxtext_utils.py:1821] params/_overwrite_with_gradient/decoder/layers/self_attention/out/Fp8DirectDotGeneralOp_0/kernel_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0421 08:47:18.683583 140642351560512 maxtext_utils.py:1821] params/_overwrite_with_gradient/decoder/layers/self_attention/out/Fp8DirectDotGeneralOp_0/kernel_scale Shape: float32[16,1] Logical: P() Physical: () I0421 08:47:18.683610 140642351560512 maxtext_utils.py:1821] params/_overwrite_with_gradient/decoder/layers/self_attention/out/Fp8DirectDotGeneralOp_0/output_grad_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0421 08:47:18.683637 140642351560512 maxtext_utils.py:1821] params/_overwrite_with_gradient/decoder/layers/self_attention/out/Fp8DirectDotGeneralOp_0/output_grad_scale Shape: float32[16,1] Logical: P() Physical: () I0421 08:47:18.683664 140642351560512 maxtext_utils.py:1821] params/_overwrite_with_gradient/decoder/layers/self_attention/query/Fp8DirectDotGeneralOp_0/input_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0421 08:47:18.683690 140642351560512 maxtext_utils.py:1821] params/_overwrite_with_gradient/decoder/layers/self_attention/query/Fp8DirectDotGeneralOp_0/input_scale Shape: float32[16,1] Logical: P() Physical: () I0421 08:47:18.683715 140642351560512 maxtext_utils.py:1821] params/_overwrite_with_gradient/decoder/layers/self_attention/query/Fp8DirectDotGeneralOp_0/kernel_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0421 08:47:18.683756 140642351560512 maxtext_utils.py:1821] params/_overwrite_with_gradient/decoder/layers/self_attention/query/Fp8DirectDotGeneralOp_0/kernel_scale Shape: float32[16,1] Logical: P() Physical: () I0421 08:47:18.683784 140642351560512 maxtext_utils.py:1821] params/_overwrite_with_gradient/decoder/layers/self_attention/query/Fp8DirectDotGeneralOp_0/output_grad_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0421 08:47:18.683811 140642351560512 maxtext_utils.py:1821] params/_overwrite_with_gradient/decoder/layers/self_attention/query/Fp8DirectDotGeneralOp_0/output_grad_scale Shape: float32[16,1] Logical: P() Physical: () I0421 08:47:18.683835 140642351560512 maxtext_utils.py:1821] params/_overwrite_with_gradient/decoder/layers/self_attention/value/Fp8DirectDotGeneralOp_0/input_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0421 08:47:18.683865 140642351560512 maxtext_utils.py:1821] params/_overwrite_with_gradient/decoder/layers/self_attention/value/Fp8DirectDotGeneralOp_0/input_scale Shape: float32[16,1] Logical: P() Physical: () I0421 08:47:18.683893 140642351560512 maxtext_utils.py:1821] params/_overwrite_with_gradient/decoder/layers/self_attention/value/Fp8DirectDotGeneralOp_0/kernel_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0421 08:47:18.683919 140642351560512 maxtext_utils.py:1821] params/_overwrite_with_gradient/decoder/layers/self_attention/value/Fp8DirectDotGeneralOp_0/kernel_scale Shape: float32[16,1] Logical: P() Physical: () I0421 08:47:18.683944 140642351560512 maxtext_utils.py:1821] params/_overwrite_with_gradient/decoder/layers/self_attention/value/Fp8DirectDotGeneralOp_0/output_grad_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0421 08:47:18.683971 140642351560512 maxtext_utils.py:1821] params/_overwrite_with_gradient/decoder/layers/self_attention/value/Fp8DirectDotGeneralOp_0/output_grad_scale Shape: float32[16,1] Logical: P() Physical: () I0421 08:47:18.684029 140642351560512 maxtext_utils.py:1821] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: P('norm',) Physical: (None,) I0421 08:47:18.684087 140642351560512 maxtext_utils.py:1821] params/params/decoder/layers/mlp/wi_0/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0421 08:47:18.684126 140642351560512 maxtext_utils.py:1821] params/params/decoder/layers/mlp/wi_1/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0421 08:47:18.684173 140642351560512 maxtext_utils.py:1821] params/params/decoder/layers/mlp/wo/kernel Shape: float32[7168,16,2048] Logical: P('mlp', 'layers', 'embed') Physical: (None, None, 'fsdp') I0421 08:47:18.684219 140642351560512 maxtext_utils.py:1821] params/params/decoder/layers/post_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0421 08:47:18.684252 140642351560512 maxtext_utils.py:1821] params/params/decoder/layers/pre_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0421 08:47:18.684298 140642351560512 maxtext_utils.py:1821] params/params/decoder/layers/self_attention/key/kernel Shape: float32[2048,16,16,128] Logical: P('embed', 'layers', 'kv_heads', 'kv_head_dim') Physical: ('fsdp', None, None, None) I0421 08:47:18.684344 140642351560512 maxtext_utils.py:1821] params/params/decoder/layers/self_attention/out/kernel Shape: float32[16,16,128,2048] Logical: P('heads', 'layers', 'kv', 'embed') Physical: (None, None, None, 'fsdp') I0421 08:47:18.684378 140642351560512 maxtext_utils.py:1821] params/params/decoder/layers/self_attention/query/kernel Shape: float32[2048,16,16,128] Logical: P('embed', 'layers', 'q_heads', 'kv') Physical: ('fsdp', None, None, None) I0421 08:47:18.684410 140642351560512 maxtext_utils.py:1821] params/params/decoder/layers/self_attention/value/kernel Shape: float32[2048,16,16,128] Logical: P('embed', 'layers', 'kv_heads', 'kv_head_dim') Physical: ('fsdp', None, None, None) I0421 08:47:18.684452 140642351560512 maxtext_utils.py:1821] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: P('embed_vocab', 'vocab') Physical: ('fsdp', None) I0421 08:47:18.684493 140642351560512 maxtext_utils.py:1821] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: P('vocab', 'embed_vocab') Physical: (None, 'fsdp') I0421 08:47:20.336683 140642351560512 train.py:157] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0421 08:47:20.336795 140642351560512 train.py:157] train/xent Physical: float32[32,2048]............................................ ('fsdp', None). I0421 08:47:20.352689 140642351560512 train.py:164] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0421 08:47:20.352762 140642351560512 train.py:164] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None). I0421 08:47:34.055128 140642351560512 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. I0421 08:47:34.056007 140642351560512 metric_logger.py:301] number parameters: 1.105 billion I0421 08:47:49.108026 140642351560512 checkpointing.py:794] Waiting for step 0 to finish before checkpoint... I0421 08:47:49.398776 140642351560512 checkpointing.py:798] Waited 0.290722131729126 seconds for step 0 to finish before starting checkpointing. I0421 08:47:49.401166 140642351560512 checkpoint_manager.py:2009] [process=7][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0421 08:47:49.403311 140642351560512 checkpoint_manager.py:1512] [process=7] Saving checkpoint at step 0 I0421 08:47:49.404903 140642351560512 event_tracking.py:70] [process=7] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260421_083238/linen_xpk_feat_nnx_trainstate_and_training_loop_20260421_083238_05_fp8/checkpoints/0. I0421 08:47:49.750866 140642351560512 signaling_client.py:364] Using JaxDistributedSignalingClient I0421 08:47:49.751859 140642351560512 jax_array_handlers.py:360] Scheduling D2H of 81 prioritized jax.Array. I0421 08:47:49.751981 140642351560512 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 I0421 08:47:50.167444 140642351560512 base_pytree_checkpoint_handler.py:154] [process=7][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.416727s I0421 08:47:50.167617 140642351560512 base_pytree_checkpoint_handler.py:130] [process=7] /jax/orbax/write/blocking_gbytes_per_sec: 3.628 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.4251730442047119 s) (per-host) I0421 08:47:50.167668 140642351560512 base_pytree_checkpoint_handler.py:768] [process=7][thread=MainThread] Initiated Pytree async_save. Time taken: 0.425234s (batch_requests_ready=0.003379s, total_serialization_initiated=0.421787s, others=0.000068s) I0421 08:47:50.167805 140642351560512 composite_checkpoint_handler.py:715] [process=7][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.429347s (all_items=0.000017s, per_item={'items': '0.00001740'}, temp_paths=0.429329) I0421 08:47:50.168604 140642351560512 event_tracking.py:125] [process=7] [async] Finished blocking save in 0.77 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260421_083238/linen_xpk_feat_nnx_trainstate_and_training_loop_20260421_083238_05_fp8/checkpoints/0. I0421 08:47:50.168970 140513643624192 async_checkpointer.py:76] [process=7][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-21 09:07:50.168933 I0421 08:47:50.179263 140642351560512 checkpoint_manager.py:1560] [process=7][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize I0421 08:47:50.179576 140512598157056 async_checkpointer.py:280] [process=7][thread=save_finalize] Waiting for background save thread=async_save. I0421 08:47:50.179744 140642351560512 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260421_083238/linen_xpk_feat_nnx_trainstate_and_training_loop_20260421_083238_05_fp8/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776761269.4011483, 'wait_for_prev_duration_secs': 6.222724914550781e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776761269.4033506, 'checkpointer_blocking_duration_secs': 0.7657670974731445, 'get_old_steps_start_time': 1776761270.1691408, 'get_old_steps_duration_secs': 3.0040740966796875e-05, 'checkpoint_manager_blocking_start_time': 1776761269.3992996, 'checkpoint_manager_blocking_duration_secs': 0.7803945541381836} I0421 08:47:50.179851 140642351560512 checkpointing.py:409] Started an asynchronous checkpoint save for step 0 I0421 08:47:50.179899 140642351560512 max_utils.py:750] Memstats: After params initialized: I0421 08:47:50.179948 140642351560512 max_utils.py:756] Using (GB) 0.45 / 31.25 (1.440000%) on TPU_26(process=7,(2,6,0,0)) I0421 08:47:50.179979 140642351560512 max_utils.py:756] Using (GB) 0.45 / 31.25 (1.440000%) on TPU_27(process=7,(3,6,0,0)) I0421 08:47:50.180005 140642351560512 max_utils.py:756] Using (GB) 0.45 / 31.25 (1.440000%) on TPU_30(process=7,(2,7,0,0)) I0421 08:47:50.180030 140642351560512 max_utils.py:756] Using (GB) 0.45 / 31.25 (1.440000%) on TPU_31(process=7,(3,7,0,0)) I0421 08:47:50.495319 140642351560512 metric_logger.py:196] completed step: 0, seconds: 15.052, TFLOP/s/device: 0.903, Tokens/s/device: 136.063, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52776.805 I0421 08:47:50.680987 140642351560512 metric_logger.py:196] completed step: 1, seconds: 1.385, TFLOP/s/device: 9.809, Tokens/s/device: 1478.454, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52761.840 I0421 08:47:51.118836 140642351560512 metric_logger.py:196] completed step: 2, seconds: 0.028, TFLOP/s/device: 481.711, Tokens/s/device: 72608.665, total_weights: 65536, loss: 10.816, lm_loss: 10.816, perplexity: 49832.398 I0421 08:47:51.277565 140642351560512 metric_logger.py:196] completed step: 3, seconds: 0.438, TFLOP/s/device: 31.007, Tokens/s/device: 4673.654, total_weights: 65536, loss: 10.431, lm_loss: 10.431, perplexity: 33901.820 I0421 08:47:51.595844 140642351560512 metric_logger.py:196] completed step: 4, seconds: 0.164, TFLOP/s/device: 82.867, Tokens/s/device: 12490.623, total_weights: 65536, loss: 9.992, lm_loss: 9.992, perplexity: 21847.639 I0421 08:47:51.602030 140642351560512 metric_logger.py:196] completed step: 5, seconds: 0.159, TFLOP/s/device: 85.608, Tokens/s/device: 12903.795, total_weights: 65536, loss: 9.549, lm_loss: 9.549, perplexity: 14035.006 I0421 08:47:58.017717 2851 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com I0421 08:48:00.108915 140512606549760 array_metadata_store.py:203] [process=7][thread=array_type_handler] Wrote 81 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260421_083238/linen_xpk_feat_nnx_trainstate_and_training_loop_20260421_083238_05_fp8/checkpoints/0/items/array_metadatas/process_7 I0421 08:48:15.351929 140513643624192 base_pytree_checkpoint_handler.py:130] [process=7] /jax/orbax/write/gbytes_per_sec: 61.687 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 25.609453439712524 s) (per-host) I0421 08:48:15.352044 140513643624192 async_checkpointer.py:90] [process=7][thread=async_save] 3 Handler Commit operations completed. Time taken: 25.182965s. I0421 08:48:15.919490 140642351560512 metric_logger.py:196] completed step: 6, seconds: 0.318, TFLOP/s/device: 42.661, Tokens/s/device: 6430.283, total_weights: 65536, loss: 9.111, lm_loss: 9.111, perplexity: 9058.688 I0421 08:48:16.077965 140642351560512 metric_logger.py:196] completed step: 7, seconds: 24.160, TFLOP/s/device: 0.562, Tokens/s/device: 84.767, total_weights: 65536, loss: 8.685, lm_loss: 8.685, perplexity: 5914.720 I0421 08:48:16.236391 140642351560512 metric_logger.py:196] completed step: 8, seconds: 0.163, TFLOP/s/device: 83.195, Tokens/s/device: 12540.106, total_weights: 65536, loss: 8.281, lm_loss: 8.281, perplexity: 3950.010 I0421 08:48:16.241319 140642351560512 checkpointing.py:794] Waiting for step 10 to finish before checkpoint... I0421 08:48:16.553010 140642351560512 checkpointing.py:798] Waited 0.31165337562561035 seconds for step 10 to finish before starting checkpointing. I0421 08:48:16.555814 140642351560512 checkpoint_manager.py:2020] [process=7][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0421 08:48:23.312061 140513643624192 async_checkpointer.py:160] [process=7][thread=async_save] Background save thread done. Time taken: 33.142967s. I0421 08:48:23.312341 140512598157056 async_checkpointer.py:288] [process=7][thread=save_finalize] Done with waiting for background save thread=async_save. I0421 08:48:23.312464 140512598157056 async_checkpointer.py:298] [process=7][thread=save_finalize] No errors found in background save thread=async_save. I0421 08:48:23.312514 140512598157056 checkpoint_manager.py:2137] [process=7][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts... I0421 08:48:23.313992 140512598157056 checkpoint_manager.py:2146] [process=7][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts. I0421 08:48:23.314167 140642351560512 checkpoint_manager.py:2032] [process=7][thread=MainThread][step=0][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=0. W0421 08:48:23.314296 140642351560512 checkpoint_manager.py:1452] Waiting for previous save to complete took 6.758482 seconds. If this number is high, consider checkpointing less frequently. I0421 08:48:23.316150 140642351560512 checkpoint_manager.py:1512] [process=7] Saving checkpoint at step 10 I0421 08:48:23.318500 140642351560512 event_tracking.py:70] [process=7] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260421_083238/linen_xpk_feat_nnx_trainstate_and_training_loop_20260421_083238_05_fp8/checkpoints/10. I0421 08:48:24.036188 140642351560512 jax_array_handlers.py:360] Scheduling D2H of 81 prioritized jax.Array. I0421 08:48:24.036285 140642351560512 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 I0421 08:48:24.086996 140642351560512 base_pytree_checkpoint_handler.py:154] [process=7][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.051820s I0421 08:48:24.087168 140642351560512 base_pytree_checkpoint_handler.py:130] [process=7] /jax/orbax/write/blocking_gbytes_per_sec: 26.608 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.05798029899597168 s) (per-host) I0421 08:48:24.087218 140642351560512 base_pytree_checkpoint_handler.py:768] [process=7][thread=MainThread] Initiated Pytree async_save. Time taken: 0.058041s (batch_requests_ready=0.003072s, total_serialization_initiated=0.054899s, others=0.000069s) I0421 08:48:24.087322 140642351560512 composite_checkpoint_handler.py:715] [process=7][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.062251s (all_items=0.000015s, per_item={'items': '0.00001478'}, temp_paths=0.062236) I0421 08:48:24.087982 140642351560512 event_tracking.py:125] [process=7] [async] Finished blocking save in 0.77 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260421_083238/linen_xpk_feat_nnx_trainstate_and_training_loop_20260421_083238_05_fp8/checkpoints/10. I0421 08:48:24.088369 140511494633216 async_checkpointer.py:76] [process=7][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-21 09:08:24.088333 I0421 08:48:24.090327 140642351560512 checkpoint_manager.py:1560] [process=7][thread=MainThread][step=10] Starting CheckpointManager Save Finalize thread=save_finalize I0421 08:48:24.090601 140512598157056 async_checkpointer.py:280] [process=7][thread=save_finalize] Waiting for background save thread=async_save. I0421 08:48:24.090769 140642351560512 standard_logger.py:34] {'step': 10, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260421_083238/linen_xpk_feat_nnx_trainstate_and_training_loop_20260421_083238_05_fp8/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776761296.5557845, 'wait_for_prev_duration_secs': 6.758482456207275, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776761303.3161867, 'checkpointer_blocking_duration_secs': 0.7722911834716797, 'get_old_steps_start_time': 1776761304.088498, 'get_old_steps_duration_secs': 2.956390380859375e-05, 'checkpoint_manager_blocking_start_time': 1776761296.5538652, 'checkpoint_manager_blocking_duration_secs': 7.536859750747681} I0421 08:48:24.090878 140642351560512 checkpointing.py:409] Started an asynchronous checkpoint save for step 10 I0421 08:48:24.091538 140642351560512 metric_logger.py:196] completed step: 9, seconds: 0.158, TFLOP/s/device: 85.807, Tokens/s/device: 12933.784, total_weights: 65536, loss: 7.908, lm_loss: 7.908, perplexity: 2720.116 I0421 08:48:24.098305 140642351560512 metric_logger.py:196] completed step: 10, seconds: 0.158, TFLOP/s/device: 85.794, Tokens/s/device: 12931.742, total_weights: 65536, loss: 7.568, lm_loss: 7.568, perplexity: 1936.198 I0421 08:48:24.256994 140642351560512 metric_logger.py:196] completed step: 11, seconds: 7.855, TFLOP/s/device: 1.730, Tokens/s/device: 260.727, total_weights: 65536, loss: 7.288, lm_loss: 7.288, perplexity: 1463.272 I0421 08:48:24.415906 140642351560512 metric_logger.py:196] completed step: 12, seconds: 0.006, TFLOP/s/device: 2299.396, Tokens/s/device: 346589.948, total_weights: 65536, loss: 7.032, lm_loss: 7.032, perplexity: 1132.061 I0421 08:48:24.973446 140642351560512 metric_logger.py:196] completed step: 13, seconds: 0.160, TFLOP/s/device: 85.021, Tokens/s/device: 12815.218, total_weights: 65536, loss: 6.815, lm_loss: 6.815, perplexity: 911.031 I0421 08:48:25.131870 140642351560512 metric_logger.py:196] completed step: 14, seconds: 0.553, TFLOP/s/device: 24.552, Tokens/s/device: 3700.692, total_weights: 65536, loss: 6.635, lm_loss: 6.635, perplexity: 761.080 I0421 08:48:25.290119 140642351560512 metric_logger.py:196] completed step: 15, seconds: 0.163, TFLOP/s/device: 83.407, Tokens/s/device: 12571.976, total_weights: 65536, loss: 6.492, lm_loss: 6.492, perplexity: 659.915 I0421 08:48:25.448150 140642351560512 metric_logger.py:196] completed step: 16, seconds: 0.158, TFLOP/s/device: 85.775, Tokens/s/device: 12928.885, total_weights: 65536, loss: 6.380, lm_loss: 6.380, perplexity: 589.948 I0421 08:48:25.606317 140642351560512 metric_logger.py:196] completed step: 17, seconds: 0.158, TFLOP/s/device: 85.826, Tokens/s/device: 12936.562, total_weights: 65536, loss: 6.293, lm_loss: 6.293, perplexity: 540.941 I0421 08:48:25.764596 140642351560512 metric_logger.py:196] completed step: 18, seconds: 0.158, TFLOP/s/device: 86.007, Tokens/s/device: 12963.912, total_weights: 65536, loss: 6.223, lm_loss: 6.223, perplexity: 504.108 I0421 08:48:25.922172 140642351560512 checkpointing.py:794] Waiting for step 19 to finish before checkpoint... I0421 08:48:25.923164 140642351560512 checkpointing.py:798] Waited 0.0010123252868652344 seconds for step 19 to finish before starting checkpointing. I0421 08:48:25.925209 140642351560512 checkpoint_manager.py:2020] [process=7][thread=MainThread][step=10][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0421 08:48:31.055653 140509905409792 array_metadata_store.py:203] [process=7][thread=array_type_handler] Wrote 81 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260421_083238/linen_xpk_feat_nnx_trainstate_and_training_loop_20260421_083238_05_fp8/checkpoints/10/items/array_metadatas/process_7 I0421 08:49:07.076150 140511494633216 base_pytree_checkpoint_handler.py:130] [process=7] /jax/orbax/write/gbytes_per_sec: 36.699 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 43.04692554473877 s) (per-host) I0421 08:49:07.076275 140511494633216 async_checkpointer.py:90] [process=7][thread=async_save] 3 Handler Commit operations completed. Time taken: 42.987248s. I0421 08:49:15.532143 140511494633216 async_checkpointer.py:160] [process=7][thread=async_save] Background save thread done. Time taken: 51.443100s. I0421 08:49:15.532424 140512598157056 async_checkpointer.py:288] [process=7][thread=save_finalize] Done with waiting for background save thread=async_save. I0421 08:49:15.532549 140512598157056 async_checkpointer.py:298] [process=7][thread=save_finalize] No errors found in background save thread=async_save. I0421 08:49:15.532599 140512598157056 checkpoint_manager.py:2137] [process=7][thread=save_finalize][step=10] CheckpointManager Save Finalize is syncing with other hosts... I0421 08:49:15.534027 140512598157056 checkpoint_manager.py:2146] [process=7][thread=save_finalize][step=10] CheckpointManager Save Finalize is done on all hosts. I0421 08:49:15.534203 140642351560512 checkpoint_manager.py:2032] [process=7][thread=MainThread][step=10][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=10. W0421 08:49:15.534324 140642351560512 checkpoint_manager.py:1452] Waiting for previous save to complete took 49.609122 seconds. If this number is high, consider checkpointing less frequently. I0421 08:49:15.535992 140642351560512 checkpoint_manager.py:1512] [process=7] Saving checkpoint at step 19 I0421 08:49:15.537937 140642351560512 event_tracking.py:70] [process=7] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260421_083238/linen_xpk_feat_nnx_trainstate_and_training_loop_20260421_083238_05_fp8/checkpoints/19. I0421 08:49:16.263781 140642351560512 jax_array_handlers.py:360] Scheduling D2H of 81 prioritized jax.Array. I0421 08:49:16.263878 140642351560512 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 I0421 08:49:16.316092 140642351560512 base_pytree_checkpoint_handler.py:154] [process=7][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.053190s I0421 08:49:16.316249 140642351560512 base_pytree_checkpoint_handler.py:130] [process=7] /jax/orbax/write/blocking_gbytes_per_sec: 25.960 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.05942797660827637 s) (per-host) I0421 08:49:16.316296 140642351560512 base_pytree_checkpoint_handler.py:768] [process=7][thread=MainThread] Initiated Pytree async_save. Time taken: 0.059484s (batch_requests_ready=0.003019s, total_serialization_initiated=0.056405s, others=0.000061s) I0421 08:49:16.316395 140642351560512 composite_checkpoint_handler.py:715] [process=7][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.063583s (all_items=0.000010s, per_item={'items': '0.00000978'}, temp_paths=0.063574) I0421 08:49:16.317056 140642351560512 event_tracking.py:125] [process=7] [async] Finished blocking save in 0.78 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260421_083238/linen_xpk_feat_nnx_trainstate_and_training_loop_20260421_083238_05_fp8/checkpoints/19. I0421 08:49:16.317385 140509905409792 async_checkpointer.py:76] [process=7][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-21 09:09:16.317349 I0421 08:49:16.319293 140642351560512 checkpoint_manager.py:1560] [process=7][thread=MainThread][step=19] Starting CheckpointManager Save Finalize thread=save_finalize I0421 08:49:16.319596 140512598157056 async_checkpointer.py:280] [process=7][thread=save_finalize] Waiting for background save thread=async_save. I0421 08:49:16.319759 140642351560512 standard_logger.py:34] {'step': 19, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260421_083238/linen_xpk_feat_nnx_trainstate_and_training_loop_20260421_083238_05_fp8/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776761305.9251795, 'wait_for_prev_duration_secs': 49.609121561050415, 'time_between_consecutive_saves_sec': 2.611144542694092, 'checkpointer_blocking_start_time': 1776761355.5360754, 'checkpointer_blocking_duration_secs': 0.7814509868621826, 'get_old_steps_start_time': 1776761356.3175416, 'get_old_steps_duration_secs': 2.6226043701171875e-05, 'checkpoint_manager_blocking_start_time': 1776761305.9234402, 'checkpoint_manager_blocking_duration_secs': 50.396281242370605} I0421 08:49:16.319865 140642351560512 checkpointing.py:409] Started an asynchronous checkpoint save for step 19 I0421 08:49:16.319907 140642351560512 checkpoint_manager.py:2020] [process=7][thread=MainThread][step=19][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0421 08:49:22.256639 140510979151616 array_metadata_store.py:203] [process=7][thread=array_type_handler] Wrote 81 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260421_083238/linen_xpk_feat_nnx_trainstate_and_training_loop_20260421_083238_05_fp8/checkpoints/19/items/array_metadatas/process_7 I0421 08:49:58.267204 140509905409792 base_pytree_checkpoint_handler.py:130] [process=7] /jax/orbax/write/gbytes_per_sec: 37.604 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 42.01034116744995 s) (per-host) I0421 08:49:58.267330 140509905409792 async_checkpointer.py:90] [process=7][thread=async_save] 3 Handler Commit operations completed. Time taken: 41.949833s. I0421 08:50:05.631032 140509905409792 async_checkpointer.py:160] [process=7][thread=async_save] Background save thread done. Time taken: 49.313521s. I0421 08:50:05.631361 140512598157056 async_checkpointer.py:288] [process=7][thread=save_finalize] Done with waiting for background save thread=async_save. I0421 08:50:05.631476 140512598157056 async_checkpointer.py:298] [process=7][thread=save_finalize] No errors found in background save thread=async_save. I0421 08:50:05.631518 140512598157056 checkpoint_manager.py:2137] [process=7][thread=save_finalize][step=19] CheckpointManager Save Finalize is syncing with other hosts... I0421 08:50:05.633282 140512598157056 checkpoint_manager.py:2146] [process=7][thread=save_finalize][step=19] CheckpointManager Save Finalize is done on all hosts. I0421 08:50:05.633456 140642351560512 checkpoint_manager.py:2032] [process=7][thread=MainThread][step=19][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=19. I0421 08:50:05.633606 140642351560512 checkpoint_manager.py:2009] [process=7][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0421 08:50:05.634551 140642351560512 metric_logger.py:196] completed step: 19, seconds: 0.158, TFLOP/s/device: 85.939, Tokens/s/device: 12953.663, total_weights: 65536, loss: 6.167, lm_loss: 6.167, perplexity: 476.947 Per train step: Total TFLOPs: 13.59 split as 93.93% learnable weight flops and 6.07% attention flops XPK End: Tue Apr 21 08:50:18 UTC 2026 EXIT_CODE=0
XPK Start: Tue Apr 21 10:18:40 UTC 2026 PyTorch was not found. Models won't be available and only tokenizers, configuration and file/data utilities can be used. 2026-04-21 10:19:04.728314: 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) I0421 10:19:05.268508 139561289512768 max_utils.py:273] Attempting to initialize the jax distributed system... I0421 10:19:14.310658 139561289512768 distributed.py:149] Starting JAX distributed service on [::]:8482 I0421 10:19:14.313053 139561289512768 distributed.py:172] Connecting to JAX distributed service on mt-05-fp8-v9w3p-slice-job-0-0.mt-05-fp8-v9w3p:8482 I0421 10:19:15.252977 139561289512768 max_utils.py:284] Jax distributed system initialized! I0421 10:19:20.395625 139561289512768 max_utils.py:800] System Information: Jax Version: 0.9.2 I0421 10:19:20.395725 139561289512768 max_utils.py:801] System Information: Jaxlib Version: 0.9.2 I0421 10:19:20.395788 139561289512768 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 I0421 10:19:20.395824 139561289512768 train_utils.py:378] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing. I0421 10:19:21.101146 139561289512768 maxtext_utils.py:1718] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0421 10:19:21.534214 139561289512768 checkpointing.py:688] Setting up checkpoint logger... I0421 10:19:21.534333 139561289512768 checkpointing.py:234] Creating checkpoint manager with ocdbt=True and zarr3=True I0421 10:19:21.534379 139561289512768 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None I0421 10:19:21.534600 139561289512768 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 0x7eed95d7a5d0>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0421 10:19:24.413335 139561289512768 checkpointing.py:266] Enabling policy for fixed interval checkpointing. I0421 10:19:24.413566 139561289512768 checkpoint_manager.py:708] [process=7][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=('items',), item_handlers={'items': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7ed9ec4deb70>}, handler_registry=None I0421 10:19:24.413830 139561289512768 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7ed9ec4deb70>` 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`. I0421 10:19:24.413880 139561289512768 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7ed9ec456480>` 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`. I0421 10:19:24.413916 139561289512768 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 0x7ed9ec4deb70>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7ed9ec4deb70>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7ed9ec456480>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7ed9ec456480>}). I0421 10:19:24.414259 139561289512768 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34 I0421 10:19:24.414330 139561289512768 async_checkpointer.py:192] [process=7][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7ed9ec3f3b00> timeout: 1200 secs and primary_host=0 for async checkpoint writes I0421 10:19:25.990352 139561289512768 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260421_083238/nnx_xpk_feat_nnx_trainstate_and_training_loop_20260421_083238_05_fp8/checkpoints I0421 10:19:26.426362 139561289512768 checkpoint_manager.py:929] [process=7][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_trainstate_and_training_loop_20260421_083238/nnx_xpk_feat_nnx_trainstate_and_training_loop_20260421_083238_05_fp8/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7ed9cc717c20> I0421 10:19:26.426532 139561289512768 checkpointing.py:302] Checkpoint manager created! I0421 10:19:27.716311 139561289512768 checkpointing.py:578] checkpoint manager exists so trying to load this run's existing checkpoint I0421 10:19:27.716428 139561289512768 checkpointing.py:676] No existing checkpoints found, not restoring checkpoint. fsdp: 32 I0421 10:19:30.265649 139561289512768 nnx_decoders.py:465] nnx_decoders/carry Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0421 10:19:30.265756 139561289512768 nnx_decoders.py:465] nnx_decoders/carry Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0421 10:19:30.271355 139561289512768 nnx_decoders.py:465] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0421 10:19:30.271412 139561289512768 nnx_decoders.py:465] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0421 10:19:30.287581 139561289512768 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0421 10:19:30.287639 139561289512768 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0421 10:19:30.303320 139561289512768 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0421 10:19:30.303378 139561289512768 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0421 10:19:30.364502 139561289512768 attentions.py:1154] attentions/query Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0421 10:19:30.364586 139561289512768 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0421 10:19:30.380256 139561289512768 attentions.py:1155] attentions/key Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0421 10:19:30.380315 139561289512768 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0421 10:19:30.395929 139561289512768 attentions.py:1156] attentions/value Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0421 10:19:30.395988 139561289512768 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0421 10:19:30.426196 139561289512768 attentions.py:1197] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv'). I0421 10:19:30.426269 139561289512768 attentions.py:1197] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0421 10:19:30.487162 139561289512768 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0421 10:19:30.487244 139561289512768 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None). I0421 10:19:44.494500 139561289512768 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. I0421 10:19:44.498476 139561289512768 metric_logger.py:301] number parameters: 1.104 billion I0421 10:19:57.993032 139561289512768 checkpointing.py:794] Waiting for step 0 to finish before checkpoint... I0421 10:19:58.154865 139561289512768 checkpointing.py:798] Waited 0.16181731224060059 seconds for step 0 to finish before starting checkpointing. I0421 10:19:58.157637 139561289512768 checkpoint_manager.py:2009] [process=7][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0421 10:19:58.159633 139561289512768 checkpoint_manager.py:1512] [process=7] Saving checkpoint at step 0 I0421 10:19:58.161190 139561289512768 event_tracking.py:70] [process=7] [async] Started save checkpoint @ gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260421_083238/nnx_xpk_feat_nnx_trainstate_and_training_loop_20260421_083238_05_fp8/checkpoints/0. I0421 10:19:58.529578 139561289512768 signaling_client.py:364] Using JaxDistributedSignalingClient I0421 10:19:58.530439 139561289512768 jax_array_handlers.py:360] Scheduling D2H of 153 prioritized jax.Array. I0421 10:19:58.530507 139561289512768 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 I0421 10:19:58.940761 139561289512768 base_pytree_checkpoint_handler.py:154] [process=7][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.412823s I0421 10:19:58.940942 139561289512768 base_pytree_checkpoint_handler.py:130] [process=7] /jax/orbax/write/blocking_gbytes_per_sec: 3.579 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.43099212646484375 s) (per-host) I0421 10:19:58.940995 139561289512768 base_pytree_checkpoint_handler.py:768] [process=7][thread=MainThread] Initiated Pytree async_save. Time taken: 0.431057s (batch_requests_ready=0.006145s, total_serialization_initiated=0.424839s, others=0.000073s) I0421 10:19:58.941119 139561289512768 composite_checkpoint_handler.py:715] [process=7][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.435204s (all_items=0.000017s, per_item={'items': '0.00001740'}, temp_paths=0.435186) I0421 10:19:58.941937 139561289512768 event_tracking.py:125] [process=7] [async] Finished blocking save in 0.78 seconds. Continuing save @ gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260421_083238/nnx_xpk_feat_nnx_trainstate_and_training_loop_20260421_083238_05_fp8/checkpoints/0. I0421 10:19:58.942293 139435027298048 async_checkpointer.py:76] [process=7][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-21 10:39:58.942254 I0421 10:19:58.952356 139561289512768 checkpoint_manager.py:1560] [process=7][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize I0421 10:19:58.952659 139434989164288 async_checkpointer.py:280] [process=7][thread=save_finalize] Waiting for background save thread=async_save. I0421 10:19:58.952807 139561289512768 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260421_083238/nnx_xpk_feat_nnx_trainstate_and_training_loop_20260421_083238_05_fp8/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776766798.1576178, 'wait_for_prev_duration_secs': 6.127357482910156e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776766798.1596718, 'checkpointer_blocking_duration_secs': 0.7827794551849365, 'get_old_steps_start_time': 1776766798.942475, 'get_old_steps_duration_secs': 2.86102294921875e-05, 'checkpoint_manager_blocking_start_time': 1776766798.1553402, 'checkpoint_manager_blocking_duration_secs': 0.7974300384521484} I0421 10:19:58.952977 139561289512768 checkpointing.py:409] Started an asynchronous checkpoint save for step 0 I0421 10:19:58.953054 139561289512768 max_utils.py:750] Memstats: After params initialized: I0421 10:19:58.953102 139561289512768 max_utils.py:756] Using (GB) 0.45 / 31.25 (1.440000%) on TPU_26(process=7,(2,6,0,0)) I0421 10:19:58.953135 139561289512768 max_utils.py:756] Using (GB) 0.45 / 31.25 (1.440000%) on TPU_27(process=7,(3,6,0,0)) I0421 10:19:58.953162 139561289512768 max_utils.py:756] Using (GB) 0.45 / 31.25 (1.440000%) on TPU_30(process=7,(2,7,0,0)) I0421 10:19:58.953186 139561289512768 max_utils.py:756] Using (GB) 0.45 / 31.25 (1.440000%) on TPU_31(process=7,(3,7,0,0)) I0421 10:19:59.273945 139561289512768 metric_logger.py:196] completed step: 0, seconds: 13.492, TFLOP/s/device: 1.007, Tokens/s/device: 151.790, total_weights: 65536, loss: 10.872, lm_loss: 10.872, perplexity: 52680.742 I0421 10:19:59.454940 139561289512768 metric_logger.py:196] completed step: 1, seconds: 1.278, TFLOP/s/device: 10.631, Tokens/s/device: 1602.478, total_weights: 65536, loss: 10.872, lm_loss: 10.872, perplexity: 52680.742 I0421 10:19:59.894028 139561289512768 metric_logger.py:196] completed step: 2, seconds: 0.029, TFLOP/s/device: 470.045, Tokens/s/device: 70850.342, total_weights: 65536, loss: 10.856, lm_loss: 10.856, perplexity: 51864.762 I0421 10:20:00.050866 139561289512768 metric_logger.py:196] completed step: 3, seconds: 0.415, TFLOP/s/device: 32.748, Tokens/s/device: 4936.177, total_weights: 65536, loss: 10.824, lm_loss: 10.824, perplexity: 50225.512 I0421 10:20:00.365849 139561289512768 metric_logger.py:196] completed step: 4, seconds: 0.187, TFLOP/s/device: 72.650, Tokens/s/device: 10950.525, total_weights: 65536, loss: 10.793, lm_loss: 10.793, perplexity: 48689.156 I0421 10:20:00.375267 139561289512768 metric_logger.py:196] completed step: 5, seconds: 0.157, TFLOP/s/device: 86.774, Tokens/s/device: 13079.576, total_weights: 65536, loss: 10.763, lm_loss: 10.763, perplexity: 47229.012 I0421 10:20:02.612548 2756 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com I0421 10:20:05.058184 139435008407296 array_metadata_store.py:203] [process=7][thread=array_type_handler] Wrote 153 array_metadata.ArrayMetadata to gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260421_083238/nnx_xpk_feat_nnx_trainstate_and_training_loop_20260421_083238_05_fp8/checkpoints/0/items/array_metadatas/process_7 I0421 10:20:19.554264 139435027298048 base_pytree_checkpoint_handler.py:130] [process=7] /jax/orbax/write/gbytes_per_sec: 75.068 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 21.044276475906372 s) (per-host) I0421 10:20:19.554389 139435027298048 async_checkpointer.py:90] [process=7][thread=async_save] 3 Handler Commit operations completed. Time taken: 20.611982s. I0421 10:20:24.241041 139561289512768 metric_logger.py:196] completed step: 6, seconds: 0.315, TFLOP/s/device: 43.113, Tokens/s/device: 6498.493, total_weights: 65536, loss: 10.733, lm_loss: 10.733, perplexity: 45864.258 I0421 10:20:24.398012 139561289512768 metric_logger.py:196] completed step: 7, seconds: 23.712, TFLOP/s/device: 0.573, Tokens/s/device: 86.368, total_weights: 65536, loss: 10.706, lm_loss: 10.706, perplexity: 44620.242 I0421 10:20:24.555152 139561289512768 metric_logger.py:196] completed step: 8, seconds: 0.163, TFLOP/s/device: 83.338, Tokens/s/device: 12561.643, total_weights: 65536, loss: 10.680, lm_loss: 10.680, perplexity: 43477.055 I0421 10:20:24.563246 139561289512768 checkpointing.py:794] Waiting for step 10 to finish before checkpoint... I0421 10:20:24.869291 139561289512768 checkpointing.py:798] Waited 0.3060169219970703 seconds for step 10 to finish before starting checkpointing. I0421 10:20:24.871976 139561289512768 checkpoint_manager.py:2020] [process=7][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0421 10:20:28.149790 139435027298048 async_checkpointer.py:160] [process=7][thread=async_save] Background save thread done. Time taken: 29.207366s. I0421 10:20:28.150078 139434989164288 async_checkpointer.py:288] [process=7][thread=save_finalize] Done with waiting for background save thread=async_save. I0421 10:20:28.150194 139434989164288 async_checkpointer.py:298] [process=7][thread=save_finalize] No errors found in background save thread=async_save. I0421 10:20:28.150248 139434989164288 checkpoint_manager.py:2137] [process=7][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts... I0421 10:20:28.152776 139434989164288 checkpoint_manager.py:2146] [process=7][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts. I0421 10:20:28.152957 139561289512768 checkpoint_manager.py:2032] [process=7][thread=MainThread][step=0][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=0. W0421 10:20:28.153093 139561289512768 checkpoint_manager.py:1452] Waiting for previous save to complete took 3.281118 seconds. If this number is high, consider checkpointing less frequently. I0421 10:20:28.154784 139561289512768 checkpoint_manager.py:1512] [process=7] Saving checkpoint at step 10 I0421 10:20:28.156811 139561289512768 event_tracking.py:70] [process=7] [async] Started save checkpoint @ gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260421_083238/nnx_xpk_feat_nnx_trainstate_and_training_loop_20260421_083238_05_fp8/checkpoints/10. I0421 10:20:28.470136 139561289512768 jax_array_handlers.py:360] Scheduling D2H of 153 prioritized jax.Array. I0421 10:20:28.470269 139561289512768 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 I0421 10:20:28.506722 139561289512768 base_pytree_checkpoint_handler.py:154] [process=7][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.039185s I0421 10:20:28.506910 139561289512768 base_pytree_checkpoint_handler.py:130] [process=7] /jax/orbax/write/blocking_gbytes_per_sec: 28.392 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.0543370246887207 s) (per-host) I0421 10:20:28.506963 139561289512768 base_pytree_checkpoint_handler.py:768] [process=7][thread=MainThread] Initiated Pytree async_save. Time taken: 0.054400s (batch_requests_ready=0.005777s, total_serialization_initiated=0.048552s, others=0.000071s) I0421 10:20:28.507096 139561289512768 composite_checkpoint_handler.py:715] [process=7][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.058549s (all_items=0.000015s, per_item={'items': '0.00001550'}, temp_paths=0.058534) I0421 10:20:28.507846 139561289512768 event_tracking.py:125] [process=7] [async] Finished blocking save in 0.35 seconds. Continuing save @ gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260421_083238/nnx_xpk_feat_nnx_trainstate_and_training_loop_20260421_083238_05_fp8/checkpoints/10. I0421 10:20:28.508124 139433420515072 async_checkpointer.py:76] [process=7][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-21 10:40:28.508092 I0421 10:20:28.964525 139561289512768 checkpoint_manager.py:1560] [process=7][thread=MainThread][step=10] Starting CheckpointManager Save Finalize thread=save_finalize I0421 10:20:28.964995 139434989164288 async_checkpointer.py:280] [process=7][thread=save_finalize] Waiting for background save thread=async_save. I0421 10:20:28.965188 139561289512768 standard_logger.py:34] {'step': 10, 'event_type': 'save', 'directory': 'gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260421_083238/nnx_xpk_feat_nnx_trainstate_and_training_loop_20260421_083238_05_fp8/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776766824.8719454, 'wait_for_prev_duration_secs': 3.2811179161071777, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776766828.1548276, 'checkpointer_blocking_duration_secs': 0.3534221649169922, 'get_old_steps_start_time': 1776766828.5082781, 'get_old_steps_duration_secs': 3.24249267578125e-05, 'checkpoint_manager_blocking_start_time': 1776766824.869757, 'checkpoint_manager_blocking_duration_secs': 4.095392942428589} I0421 10:20:28.965315 139561289512768 checkpointing.py:409] Started an asynchronous checkpoint save for step 10 I0421 10:20:28.966121 139561289512768 metric_logger.py:196] completed step: 9, seconds: 0.157, TFLOP/s/device: 86.642, Tokens/s/device: 13059.642, total_weights: 65536, loss: 10.656, lm_loss: 10.656, perplexity: 42467.320 I0421 10:20:28.977348 139561289512768 metric_logger.py:196] completed step: 10, seconds: 0.157, TFLOP/s/device: 86.603, Tokens/s/device: 13053.732, total_weights: 65536, loss: 10.636, lm_loss: 10.636, perplexity: 41601.941 I0421 10:20:29.130327 139561289512768 metric_logger.py:196] completed step: 11, seconds: 4.411, TFLOP/s/device: 3.080, Tokens/s/device: 464.243, total_weights: 65536, loss: 10.618, lm_loss: 10.618, perplexity: 40849.570 I0421 10:20:29.287353 139561289512768 metric_logger.py:196] completed step: 12, seconds: 0.009, TFLOP/s/device: 1433.998, Tokens/s/device: 216147.757, total_weights: 65536, loss: 10.602, lm_loss: 10.602, perplexity: 40203.926 I0421 10:20:29.853219 139561289512768 metric_logger.py:196] completed step: 13, seconds: 0.154, TFLOP/s/device: 87.946, Tokens/s/device: 13256.178, total_weights: 65536, loss: 10.588, lm_loss: 10.588, perplexity: 39652.145 I0421 10:20:30.010108 139561289512768 metric_logger.py:196] completed step: 14, seconds: 0.560, TFLOP/s/device: 24.260, Tokens/s/device: 3656.666, total_weights: 65536, loss: 10.577, lm_loss: 10.577, perplexity: 39212.879 I0421 10:20:30.329165 139561289512768 metric_logger.py:196] completed step: 15, seconds: 0.163, TFLOP/s/device: 83.405, Tokens/s/device: 12571.744, total_weights: 65536, loss: 10.568, lm_loss: 10.568, perplexity: 38879.223 I0421 10:20:30.485962 139561289512768 metric_logger.py:196] completed step: 16, seconds: 0.313, TFLOP/s/device: 43.346, Tokens/s/device: 6533.612, total_weights: 65536, loss: 10.562, lm_loss: 10.562, perplexity: 38619.105 I0421 10:20:30.642902 139561289512768 metric_logger.py:196] completed step: 17, seconds: 0.162, TFLOP/s/device: 83.681, Tokens/s/device: 12613.245, total_weights: 65536, loss: 10.556, lm_loss: 10.556, perplexity: 38395.816 I0421 10:20:30.799766 139561289512768 metric_logger.py:196] completed step: 18, seconds: 0.157, TFLOP/s/device: 86.692, Tokens/s/device: 13067.225, total_weights: 65536, loss: 10.552, lm_loss: 10.552, perplexity: 38236.652 I0421 10:20:30.958541 139561289512768 checkpointing.py:794] Waiting for step 19 to finish before checkpoint... I0421 10:20:30.959813 139561289512768 checkpointing.py:798] Waited 0.0012836456298828125 seconds for step 19 to finish before starting checkpointing. I0421 10:20:30.962477 139561289512768 checkpoint_manager.py:2020] [process=7][thread=MainThread][step=10][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0421 10:20:34.334133 139432314709760 array_metadata_store.py:203] [process=7][thread=array_type_handler] Wrote 153 array_metadata.ArrayMetadata to gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260421_083238/nnx_xpk_feat_nnx_trainstate_and_training_loop_20260421_083238_05_fp8/checkpoints/10/items/array_metadatas/process_7 I0421 10:20:50.178169 139433420515072 base_pytree_checkpoint_handler.py:130] [process=7] /jax/orbax/write/gbytes_per_sec: 72.714 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 21.725571632385254 s) (per-host) I0421 10:20:50.178276 139433420515072 async_checkpointer.py:90] [process=7][thread=async_save] 3 Handler Commit operations completed. Time taken: 21.670053s. I0421 10:20:57.164722 139433420515072 async_checkpointer.py:160] [process=7][thread=async_save] Background save thread done. Time taken: 28.656483s. I0421 10:20:57.164957 139434989164288 async_checkpointer.py:288] [process=7][thread=save_finalize] Done with waiting for background save thread=async_save. I0421 10:20:57.165031 139434989164288 async_checkpointer.py:298] [process=7][thread=save_finalize] No errors found in background save thread=async_save. I0421 10:20:57.165079 139434989164288 checkpoint_manager.py:2137] [process=7][thread=save_finalize][step=10] CheckpointManager Save Finalize is syncing with other hosts... I0421 10:20:57.167641 139434989164288 checkpoint_manager.py:2146] [process=7][thread=save_finalize][step=10] CheckpointManager Save Finalize is done on all hosts. I0421 10:20:57.167844 139561289512768 checkpoint_manager.py:2032] [process=7][thread=MainThread][step=10][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=10. W0421 10:20:57.167978 139561289512768 checkpoint_manager.py:1452] Waiting for previous save to complete took 26.205505 seconds. If this number is high, consider checkpointing less frequently. I0421 10:20:57.169505 139561289512768 checkpoint_manager.py:1512] [process=7] Saving checkpoint at step 19 I0421 10:20:57.171652 139561289512768 event_tracking.py:70] [process=7] [async] Started save checkpoint @ gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260421_083238/nnx_xpk_feat_nnx_trainstate_and_training_loop_20260421_083238_05_fp8/checkpoints/19. I0421 10:20:57.464795 139561289512768 jax_array_handlers.py:360] Scheduling D2H of 153 prioritized jax.Array. I0421 10:20:57.464891 139561289512768 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 I0421 10:20:57.501201 139561289512768 base_pytree_checkpoint_handler.py:154] [process=7][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.038867s I0421 10:20:57.501336 139561289512768 base_pytree_checkpoint_handler.py:130] [process=7] /jax/orbax/write/blocking_gbytes_per_sec: 28.804 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.05356001853942871 s) (per-host) I0421 10:20:57.501382 139561289512768 base_pytree_checkpoint_handler.py:768] [process=7][thread=MainThread] Initiated Pytree async_save. Time taken: 0.053617s (batch_requests_ready=0.005670s, total_serialization_initiated=0.047886s, others=0.000062s) I0421 10:20:57.501478 139561289512768 composite_checkpoint_handler.py:715] [process=7][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.057656s (all_items=0.000011s, per_item={'items': '0.00001073'}, temp_paths=0.057645) I0421 10:20:57.502232 139561289512768 event_tracking.py:125] [process=7] [async] Finished blocking save in 0.33 seconds. Continuing save @ gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260421_083238/nnx_xpk_feat_nnx_trainstate_and_training_loop_20260421_083238_05_fp8/checkpoints/19. I0421 10:20:57.502511 139431247853312 async_checkpointer.py:76] [process=7][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-21 10:40:57.502480 I0421 10:20:57.510272 139561289512768 checkpoint_manager.py:1560] [process=7][thread=MainThread][step=19] Starting CheckpointManager Save Finalize thread=save_finalize I0421 10:20:57.510581 139434989164288 async_checkpointer.py:280] [process=7][thread=save_finalize] Waiting for background save thread=async_save. I0421 10:20:57.510708 139561289512768 standard_logger.py:34] {'step': 19, 'event_type': 'save', 'directory': 'gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260421_083238/nnx_xpk_feat_nnx_trainstate_and_training_loop_20260421_083238_05_fp8/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776766830.9624476, 'wait_for_prev_duration_secs': 26.20550537109375, 'time_between_consecutive_saves_sec': 2.8096187114715576, 'checkpointer_blocking_start_time': 1776766857.1695437, 'checkpointer_blocking_duration_secs': 0.33310914039611816, 'get_old_steps_start_time': 1776766857.5026705, 'get_old_steps_duration_secs': 2.574920654296875e-05, 'checkpoint_manager_blocking_start_time': 1776766830.960131, 'checkpoint_manager_blocking_duration_secs': 26.550544261932373} I0421 10:20:57.510841 139561289512768 checkpointing.py:409] Started an asynchronous checkpoint save for step 19 I0421 10:20:57.510884 139561289512768 checkpoint_manager.py:2020] [process=7][thread=MainThread][step=19][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0421 10:21:02.623328 139431273031424 array_metadata_store.py:203] [process=7][thread=array_type_handler] Wrote 153 array_metadata.ArrayMetadata to gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260421_083238/nnx_xpk_feat_nnx_trainstate_and_training_loop_20260421_083238_05_fp8/checkpoints/19/items/array_metadatas/process_7 I0421 10:21:18.305551 139431247853312 base_pytree_checkpoint_handler.py:130] [process=7] /jax/orbax/write/gbytes_per_sec: 75.740 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 20.85773754119873 s) (per-host) I0421 10:21:18.305676 139431247853312 async_checkpointer.py:90] [process=7][thread=async_save] 3 Handler Commit operations completed. Time taken: 20.803053s. I0421 10:21:25.910361 139431247853312 async_checkpointer.py:160] [process=7][thread=async_save] Background save thread done. Time taken: 28.407720s. I0421 10:21:25.910578 139434989164288 async_checkpointer.py:288] [process=7][thread=save_finalize] Done with waiting for background save thread=async_save. I0421 10:21:25.910638 139434989164288 async_checkpointer.py:298] [process=7][thread=save_finalize] No errors found in background save thread=async_save. I0421 10:21:25.910689 139434989164288 checkpoint_manager.py:2137] [process=7][thread=save_finalize][step=19] CheckpointManager Save Finalize is syncing with other hosts... I0421 10:21:25.913148 139434989164288 checkpoint_manager.py:2146] [process=7][thread=save_finalize][step=19] CheckpointManager Save Finalize is done on all hosts. I0421 10:21:25.913331 139561289512768 checkpoint_manager.py:2032] [process=7][thread=MainThread][step=19][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=19. I0421 10:21:25.913509 139561289512768 checkpoint_manager.py:2009] [process=7][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0421 10:21:25.914666 139561289512768 metric_logger.py:196] completed step: 19, seconds: 0.157, TFLOP/s/device: 86.537, Tokens/s/device: 13043.838, total_weights: 65536, loss: 10.549, lm_loss: 10.549, perplexity: 38125.074 Per train step: Total TFLOPs: 13.59 split as 93.93% learnable weight flops and 6.07% attention flops XPK End: Tue Apr 21 10:21:36 UTC 2026 EXIT_CODE=0