feat/nnx-trainstate-and-training-loop| Metric | Linen 77202a8e9 | NNX 77202a8e9 | Diff (NNX − Linen) |
|---|---|---|---|
| Parameters | 1.105 billion | 1.104 billion | — |
| Final loss | 6.1670 | 10.5490 | +4.382 |
| TFLOP/s | 85.866 | 86.725 | +0.859 |
| Tok/s | 12942.7 | 13072.1 | +129.369 |
| Avg s/step | 1.613 | 1.579 | -0.034 |
| 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: Sat Apr 25 09:30:37 UTC 2026 PyTorch was not found. Models won't be available and only tokenizers, configuration and file/data utilities can be used. `rope_parameters`'s factor field must be a float >= 1, got 40 `rope_parameters`'s beta_fast field must be a float, got 32 `rope_parameters`'s beta_slow field must be a float, got 1 DeepseekV32Config got `key=rope_scaling` in kwargs but hasn't set it as attribute. For RoPE standardization you need to set `self.rope_parameters` in model's config. 2026-04-25 09:31:02.506291: 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) I0425 09:31:02.718650 133066620835648 max_utils.py:273] Attempting to initialize the jax distributed system... I0425 09:31:11.760326 133066620835648 distributed.py:149] Starting JAX distributed service on [::]:8482 I0425 09:31:11.762790 133066620835648 distributed.py:172] Connecting to JAX distributed service on mt-05-fp8-4f2w6-slice-job-0-0.mt-05-fp8-4f2w6:8482 I0425 09:31:13.134604 133066620835648 max_utils.py:284] Jax distributed system initialized! I0425 09:31:19.156157 133066620835648 max_utils.py:800] System Information: Jax Version: 0.9.2 I0425 09:31:19.156262 133066620835648 max_utils.py:801] System Information: Jaxlib Version: 0.9.2 I0425 09:31:19.156304 133066620835648 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 I0425 09:31:19.156340 133066620835648 train_utils.py:391] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing. I0425 09:31:19.847989 133066620835648 maxtext_utils.py:1732] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0425 09:31:19.848593 133066620835648 maxtext_utils.py:1732] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0425 09:31:19.848784 133066620835648 checkpointing.py:688] Setting up checkpoint logger... I0425 09:31:19.848834 133066620835648 checkpointing.py:234] Creating checkpoint manager with ocdbt=True and zarr3=True I0425 09:31:19.848877 133066620835648 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None I0425 09:31:19.849221 133066620835648 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 0x79054504f740>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0425 09:31:22.679074 133066620835648 checkpointing.py:266] Enabling policy for fixed interval checkpointing. I0425 09:31:22.679308 133066620835648 checkpoint_manager.py:708] [process=3][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=('items',), item_handlers={'items': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x78f0443ae7e0>}, handler_registry=None I0425 09:31:22.679556 133066620835648 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x78f0443ae7e0>` 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`. I0425 09:31:22.679608 133066620835648 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x78f0443b23c0>` 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`. I0425 09:31:22.679644 133066620835648 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 0x78f0443ae7e0>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x78f0443ae7e0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x78f0443b23c0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x78f0443b23c0>}). I0425 09:31:22.679977 133066620835648 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34 I0425 09:31:22.680048 133066620835648 async_checkpointer.py:192] [process=3][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x78f0246ddbc0> timeout: 1200 secs and primary_host=0 for async checkpoint writes I0425 09:31:23.734682 133066620835648 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260425_092114/linen_xpk_feat_nnx_trainstate_and_training_loop_20260425_092114_05_fp8/checkpoints I0425 09:31:23.899883 133066620835648 checkpoint_manager.py:929] [process=3][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_20260425_092114/linen_xpk_feat_nnx_trainstate_and_training_loop_20260425_092114_05_fp8/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x78f0443af230> I0425 09:31:23.900054 133066620835648 checkpointing.py:302] Checkpoint manager created! I0425 09:31:24.993728 133066620835648 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0425 09:31:24.993839 133066620835648 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0425 09:31:25.380084 133066620835648 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0425 09:31:25.380179 133066620835648 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0425 09:31:25.396550 133066620835648 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0425 09:31:25.396610 133066620835648 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0425 09:31:25.449362 133066620835648 attentions.py:1154] attentions/query Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0425 09:31:25.449463 133066620835648 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0425 09:31:25.465892 133066620835648 attentions.py:1155] attentions/key Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0425 09:31:25.465958 133066620835648 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0425 09:31:25.482246 133066620835648 attentions.py:1156] attentions/value Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0425 09:31:25.482308 133066620835648 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0425 09:31:25.507693 133066620835648 attentions.py:1198] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv'). I0425 09:31:25.507767 133066620835648 attentions.py:1198] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0425 09:31:25.558403 133066620835648 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0425 09:31:25.558492 133066620835648 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None). I0425 09:31:26.009889 133066620835648 checkpointing.py:578] checkpoint manager exists so trying to load this run's existing checkpoint I0425 09:31:26.010009 133066620835648 checkpointing.py:676] No existing checkpoints found, not restoring checkpoint. fsdp: 32 I0425 09:31:27.941812 133066620835648 maxtext_utils.py:1835] params/_overwrite_with_gradient/decoder/layers/mlp/wi_0/Fp8DirectDotGeneralOp_0/input_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0425 09:31:27.941940 133066620835648 maxtext_utils.py:1835] params/_overwrite_with_gradient/decoder/layers/mlp/wi_0/Fp8DirectDotGeneralOp_0/input_scale Shape: float32[16,1] Logical: P() Physical: () I0425 09:31:27.941990 133066620835648 maxtext_utils.py:1835] params/_overwrite_with_gradient/decoder/layers/mlp/wi_0/Fp8DirectDotGeneralOp_0/kernel_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0425 09:31:27.942029 133066620835648 maxtext_utils.py:1835] params/_overwrite_with_gradient/decoder/layers/mlp/wi_0/Fp8DirectDotGeneralOp_0/kernel_scale Shape: float32[16,1] Logical: P() Physical: () I0425 09:31:27.942066 133066620835648 maxtext_utils.py:1835] params/_overwrite_with_gradient/decoder/layers/mlp/wi_0/Fp8DirectDotGeneralOp_0/output_grad_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0425 09:31:27.942098 133066620835648 maxtext_utils.py:1835] params/_overwrite_with_gradient/decoder/layers/mlp/wi_0/Fp8DirectDotGeneralOp_0/output_grad_scale Shape: float32[16,1] Logical: P() Physical: () I0425 09:31:27.942131 133066620835648 maxtext_utils.py:1835] params/_overwrite_with_gradient/decoder/layers/mlp/wi_1/Fp8DirectDotGeneralOp_0/input_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0425 09:31:27.942162 133066620835648 maxtext_utils.py:1835] params/_overwrite_with_gradient/decoder/layers/mlp/wi_1/Fp8DirectDotGeneralOp_0/input_scale Shape: float32[16,1] Logical: P() Physical: () I0425 09:31:27.942191 133066620835648 maxtext_utils.py:1835] params/_overwrite_with_gradient/decoder/layers/mlp/wi_1/Fp8DirectDotGeneralOp_0/kernel_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0425 09:31:27.942221 133066620835648 maxtext_utils.py:1835] params/_overwrite_with_gradient/decoder/layers/mlp/wi_1/Fp8DirectDotGeneralOp_0/kernel_scale Shape: float32[16,1] Logical: P() Physical: () I0425 09:31:27.942249 133066620835648 maxtext_utils.py:1835] params/_overwrite_with_gradient/decoder/layers/mlp/wi_1/Fp8DirectDotGeneralOp_0/output_grad_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0425 09:31:27.942280 133066620835648 maxtext_utils.py:1835] params/_overwrite_with_gradient/decoder/layers/mlp/wi_1/Fp8DirectDotGeneralOp_0/output_grad_scale Shape: float32[16,1] Logical: P() Physical: () I0425 09:31:27.942309 133066620835648 maxtext_utils.py:1835] params/_overwrite_with_gradient/decoder/layers/mlp/wo/Fp8DirectDotGeneralOp_0/input_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0425 09:31:27.942338 133066620835648 maxtext_utils.py:1835] params/_overwrite_with_gradient/decoder/layers/mlp/wo/Fp8DirectDotGeneralOp_0/input_scale Shape: float32[16,1] Logical: P() Physical: () I0425 09:31:27.942366 133066620835648 maxtext_utils.py:1835] params/_overwrite_with_gradient/decoder/layers/mlp/wo/Fp8DirectDotGeneralOp_0/kernel_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0425 09:31:27.942393 133066620835648 maxtext_utils.py:1835] params/_overwrite_with_gradient/decoder/layers/mlp/wo/Fp8DirectDotGeneralOp_0/kernel_scale Shape: float32[16,1] Logical: P() Physical: () I0425 09:31:27.942442 133066620835648 maxtext_utils.py:1835] params/_overwrite_with_gradient/decoder/layers/mlp/wo/Fp8DirectDotGeneralOp_0/output_grad_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0425 09:31:27.942479 133066620835648 maxtext_utils.py:1835] params/_overwrite_with_gradient/decoder/layers/mlp/wo/Fp8DirectDotGeneralOp_0/output_grad_scale Shape: float32[16,1] Logical: P() Physical: () I0425 09:31:27.942510 133066620835648 maxtext_utils.py:1835] params/_overwrite_with_gradient/decoder/layers/self_attention/key/Fp8DirectDotGeneralOp_0/input_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0425 09:31:27.942538 133066620835648 maxtext_utils.py:1835] params/_overwrite_with_gradient/decoder/layers/self_attention/key/Fp8DirectDotGeneralOp_0/input_scale Shape: float32[16,1] Logical: P() Physical: () I0425 09:31:27.942566 133066620835648 maxtext_utils.py:1835] params/_overwrite_with_gradient/decoder/layers/self_attention/key/Fp8DirectDotGeneralOp_0/kernel_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0425 09:31:27.942593 133066620835648 maxtext_utils.py:1835] params/_overwrite_with_gradient/decoder/layers/self_attention/key/Fp8DirectDotGeneralOp_0/kernel_scale Shape: float32[16,1] Logical: P() Physical: () I0425 09:31:27.942622 133066620835648 maxtext_utils.py:1835] params/_overwrite_with_gradient/decoder/layers/self_attention/key/Fp8DirectDotGeneralOp_0/output_grad_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0425 09:31:27.942651 133066620835648 maxtext_utils.py:1835] params/_overwrite_with_gradient/decoder/layers/self_attention/key/Fp8DirectDotGeneralOp_0/output_grad_scale Shape: float32[16,1] Logical: P() Physical: () I0425 09:31:27.942678 133066620835648 maxtext_utils.py:1835] params/_overwrite_with_gradient/decoder/layers/self_attention/out/Fp8DirectDotGeneralOp_0/input_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0425 09:31:27.942706 133066620835648 maxtext_utils.py:1835] params/_overwrite_with_gradient/decoder/layers/self_attention/out/Fp8DirectDotGeneralOp_0/input_scale Shape: float32[16,1] Logical: P() Physical: () I0425 09:31:27.942734 133066620835648 maxtext_utils.py:1835] params/_overwrite_with_gradient/decoder/layers/self_attention/out/Fp8DirectDotGeneralOp_0/kernel_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0425 09:31:27.942760 133066620835648 maxtext_utils.py:1835] params/_overwrite_with_gradient/decoder/layers/self_attention/out/Fp8DirectDotGeneralOp_0/kernel_scale Shape: float32[16,1] Logical: P() Physical: () I0425 09:31:27.942787 133066620835648 maxtext_utils.py:1835] params/_overwrite_with_gradient/decoder/layers/self_attention/out/Fp8DirectDotGeneralOp_0/output_grad_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0425 09:31:27.942813 133066620835648 maxtext_utils.py:1835] params/_overwrite_with_gradient/decoder/layers/self_attention/out/Fp8DirectDotGeneralOp_0/output_grad_scale Shape: float32[16,1] Logical: P() Physical: () I0425 09:31:27.942840 133066620835648 maxtext_utils.py:1835] params/_overwrite_with_gradient/decoder/layers/self_attention/query/Fp8DirectDotGeneralOp_0/input_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0425 09:31:27.942867 133066620835648 maxtext_utils.py:1835] params/_overwrite_with_gradient/decoder/layers/self_attention/query/Fp8DirectDotGeneralOp_0/input_scale Shape: float32[16,1] Logical: P() Physical: () I0425 09:31:27.942899 133066620835648 maxtext_utils.py:1835] params/_overwrite_with_gradient/decoder/layers/self_attention/query/Fp8DirectDotGeneralOp_0/kernel_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0425 09:31:27.942929 133066620835648 maxtext_utils.py:1835] params/_overwrite_with_gradient/decoder/layers/self_attention/query/Fp8DirectDotGeneralOp_0/kernel_scale Shape: float32[16,1] Logical: P() Physical: () I0425 09:31:27.942959 133066620835648 maxtext_utils.py:1835] params/_overwrite_with_gradient/decoder/layers/self_attention/query/Fp8DirectDotGeneralOp_0/output_grad_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0425 09:31:27.942986 133066620835648 maxtext_utils.py:1835] params/_overwrite_with_gradient/decoder/layers/self_attention/query/Fp8DirectDotGeneralOp_0/output_grad_scale Shape: float32[16,1] Logical: P() Physical: () I0425 09:31:27.943013 133066620835648 maxtext_utils.py:1835] params/_overwrite_with_gradient/decoder/layers/self_attention/value/Fp8DirectDotGeneralOp_0/input_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0425 09:31:27.943039 133066620835648 maxtext_utils.py:1835] params/_overwrite_with_gradient/decoder/layers/self_attention/value/Fp8DirectDotGeneralOp_0/input_scale Shape: float32[16,1] Logical: P() Physical: () I0425 09:31:27.943066 133066620835648 maxtext_utils.py:1835] params/_overwrite_with_gradient/decoder/layers/self_attention/value/Fp8DirectDotGeneralOp_0/kernel_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0425 09:31:27.943094 133066620835648 maxtext_utils.py:1835] params/_overwrite_with_gradient/decoder/layers/self_attention/value/Fp8DirectDotGeneralOp_0/kernel_scale Shape: float32[16,1] Logical: P() Physical: () I0425 09:31:27.943122 133066620835648 maxtext_utils.py:1835] params/_overwrite_with_gradient/decoder/layers/self_attention/value/Fp8DirectDotGeneralOp_0/output_grad_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0425 09:31:27.943150 133066620835648 maxtext_utils.py:1835] params/_overwrite_with_gradient/decoder/layers/self_attention/value/Fp8DirectDotGeneralOp_0/output_grad_scale Shape: float32[16,1] Logical: P() Physical: () I0425 09:31:27.943200 133066620835648 maxtext_utils.py:1835] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: P('norm',) Physical: (None,) I0425 09:31:27.943254 133066620835648 maxtext_utils.py:1835] params/params/decoder/layers/mlp/wi_0/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0425 09:31:27.943292 133066620835648 maxtext_utils.py:1835] params/params/decoder/layers/mlp/wi_1/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0425 09:31:27.943340 133066620835648 maxtext_utils.py:1835] params/params/decoder/layers/mlp/wo/kernel Shape: float32[7168,16,2048] Logical: P('mlp', 'layers', 'embed') Physical: (None, None, 'fsdp') I0425 09:31:27.943386 133066620835648 maxtext_utils.py:1835] params/params/decoder/layers/post_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0425 09:31:27.943499 133066620835648 maxtext_utils.py:1835] params/params/decoder/layers/pre_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0425 09:31:27.943559 133066620835648 maxtext_utils.py:1835] 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) I0425 09:31:27.943610 133066620835648 maxtext_utils.py:1835] 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') I0425 09:31:27.943647 133066620835648 maxtext_utils.py:1835] 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) I0425 09:31:27.943680 133066620835648 maxtext_utils.py:1835] 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) I0425 09:31:27.943723 133066620835648 maxtext_utils.py:1835] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: P('embed_vocab', 'vocab') Physical: ('fsdp', None) I0425 09:31:27.943765 133066620835648 maxtext_utils.py:1835] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: P('vocab', 'embed_vocab') Physical: (None, 'fsdp') I0425 09:31:29.578534 133066620835648 train.py:157] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0425 09:31:29.578629 133066620835648 train.py:157] train/xent Physical: float32[32,2048]............................................ ('fsdp', None). I0425 09:31:29.594175 133066620835648 train.py:164] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0425 09:31:29.594237 133066620835648 train.py:164] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None). I0425 09:31:43.395953 133066620835648 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. I0425 09:31:43.396803 133066620835648 metric_logger.py:301] number parameters: 1.105 billion I0425 09:31:58.772197 133066620835648 checkpointing.py:794] Waiting for step 0 to finish before checkpoint... I0425 09:31:58.939587 133066620835648 checkpointing.py:798] Waited 0.1673750877380371 seconds for step 0 to finish before starting checkpointing. I0425 09:31:58.942053 133066620835648 checkpoint_manager.py:2009] [process=3][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0425 09:31:58.943915 133066620835648 checkpoint_manager.py:1512] [process=3] Saving checkpoint at step 0 I0425 09:31:58.945516 133066620835648 event_tracking.py:70] [process=3] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260425_092114/linen_xpk_feat_nnx_trainstate_and_training_loop_20260425_092114_05_fp8/checkpoints/0. I0425 09:31:59.264237 133066620835648 signaling_client.py:364] Using JaxDistributedSignalingClient I0425 09:31:59.265321 133066620835648 jax_array_handlers.py:360] Scheduling D2H of 81 prioritized jax.Array. I0425 09:31:59.265381 133066620835648 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 I0425 09:31:59.677167 133066620835648 base_pytree_checkpoint_handler.py:154] [process=3][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.413051s I0425 09:31:59.677344 133066620835648 base_pytree_checkpoint_handler.py:130] [process=3] /jax/orbax/write/blocking_gbytes_per_sec: 3.662 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.42124295234680176 s) (per-host) I0425 09:31:59.677396 133066620835648 base_pytree_checkpoint_handler.py:768] [process=3][thread=MainThread] Initiated Pytree async_save. Time taken: 0.421305s (batch_requests_ready=0.003387s, total_serialization_initiated=0.417848s, others=0.000069s) I0425 09:31:59.677512 133066620835648 composite_checkpoint_handler.py:715] [process=3][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.425486s (all_items=0.000018s, per_item={'items': '0.00001788'}, temp_paths=0.425468) I0425 09:31:59.678333 133066620835648 event_tracking.py:125] [process=3] [async] Finished blocking save in 0.73 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260425_092114/linen_xpk_feat_nnx_trainstate_and_training_loop_20260425_092114_05_fp8/checkpoints/0. I0425 09:31:59.678719 132936988866304 async_checkpointer.py:76] [process=3][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-25 09:51:59.678681 I0425 09:31:59.695358 133066620835648 checkpoint_manager.py:1560] [process=3][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize I0425 09:31:59.695655 132936456558336 async_checkpointer.py:280] [process=3][thread=save_finalize] Waiting for background save thread=async_save. I0425 09:31:59.695815 133066620835648 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260425_092114/linen_xpk_feat_nnx_trainstate_and_training_loop_20260425_092114_05_fp8/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1777109518.9420347, 'wait_for_prev_duration_secs': 6.341934204101562e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1777109518.9439535, 'checkpointer_blocking_duration_secs': 0.7349326610565186, 'get_old_steps_start_time': 1777109519.678913, 'get_old_steps_duration_secs': 3.814697265625e-05, 'checkpoint_manager_blocking_start_time': 1777109518.9401839, 'checkpoint_manager_blocking_duration_secs': 0.7555923461914062} I0425 09:31:59.695931 133066620835648 checkpointing.py:409] Started an asynchronous checkpoint save for step 0 I0425 09:31:59.695984 133066620835648 max_utils.py:750] Memstats: After params initialized: I0425 09:31:59.696041 133066620835648 max_utils.py:756] Using (GB) 0.45 / 31.25 (1.440000%) on TPU_10(process=3,(2,2,0,0)) I0425 09:31:59.696072 133066620835648 max_utils.py:756] Using (GB) 0.45 / 31.25 (1.440000%) on TPU_11(process=3,(3,2,0,0)) I0425 09:31:59.696101 133066620835648 max_utils.py:756] Using (GB) 0.45 / 31.25 (1.440000%) on TPU_14(process=3,(2,3,0,0)) I0425 09:31:59.696125 133066620835648 max_utils.py:756] Using (GB) 0.45 / 31.25 (1.440000%) on TPU_15(process=3,(3,3,0,0)) I0425 09:32:00.009050 133066620835648 metric_logger.py:196] completed step: 0, seconds: 15.375, TFLOP/s/device: 0.884, Tokens/s/device: 133.201, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52776.805 I0425 09:32:00.206792 133066620835648 metric_logger.py:196] completed step: 1, seconds: 1.235, TFLOP/s/device: 11.003, Tokens/s/device: 1658.540, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52761.840 I0425 09:32:00.611503 133066620835648 metric_logger.py:196] completed step: 2, seconds: 0.040, TFLOP/s/device: 338.738, Tokens/s/device: 51058.313, total_weights: 65536, loss: 10.816, lm_loss: 10.816, perplexity: 49832.398 I0425 09:32:00.769769 133066620835648 metric_logger.py:196] completed step: 3, seconds: 0.405, TFLOP/s/device: 33.547, Tokens/s/device: 5056.503, total_weights: 65536, loss: 10.431, lm_loss: 10.431, perplexity: 33901.820 I0425 09:32:01.087048 133066620835648 metric_logger.py:196] completed step: 4, seconds: 0.164, TFLOP/s/device: 82.841, Tokens/s/device: 12486.739, total_weights: 65536, loss: 9.992, lm_loss: 9.992, perplexity: 21847.639 I0425 09:32:01.093432 133066620835648 metric_logger.py:196] completed step: 5, seconds: 0.158, TFLOP/s/device: 85.903, Tokens/s/device: 12948.176, total_weights: 65536, loss: 9.549, lm_loss: 9.549, perplexity: 14035.006 I0425 09:32:02.555681 2859 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com I0425 09:32:05.020566 132936464951040 array_metadata_store.py:203] [process=3][thread=array_type_handler] Wrote 81 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260425_092114/linen_xpk_feat_nnx_trainstate_and_training_loop_20260425_092114_05_fp8/checkpoints/0/items/array_metadatas/process_3 I0425 09:32:18.408836 132936988866304 base_pytree_checkpoint_handler.py:130] [process=3] /jax/orbax/write/gbytes_per_sec: 82.482 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 19.152687072753906 s) (per-host) I0425 09:32:18.408959 132936988866304 async_checkpointer.py:90] [process=3][thread=async_save] 3 Handler Commit operations completed. Time taken: 18.730119s. I0425 09:32:25.890308 133066620835648 metric_logger.py:196] completed step: 6, seconds: 0.318, TFLOP/s/device: 42.753, Tokens/s/device: 6444.203, total_weights: 65536, loss: 9.111, lm_loss: 9.111, perplexity: 9058.688 I0425 09:32:26.048835 133066620835648 metric_logger.py:196] completed step: 7, seconds: 24.640, TFLOP/s/device: 0.551, Tokens/s/device: 83.118, total_weights: 65536, loss: 8.685, lm_loss: 8.685, perplexity: 5914.720 I0425 09:32:26.207315 133066620835648 metric_logger.py:196] completed step: 8, seconds: 0.163, TFLOP/s/device: 83.121, Tokens/s/device: 12528.829, total_weights: 65536, loss: 8.281, lm_loss: 8.281, perplexity: 3950.010 I0425 09:32:26.212299 133066620835648 checkpointing.py:794] Waiting for step 10 to finish before checkpoint... I0425 09:32:26.523927 133066620835648 checkpointing.py:798] Waited 0.31159234046936035 seconds for step 10 to finish before starting checkpointing. I0425 09:32:26.526629 133066620835648 checkpoint_manager.py:2020] [process=3][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0425 09:32:27.540337 132936988866304 async_checkpointer.py:160] [process=3][thread=async_save] Background save thread done. Time taken: 27.861480s. I0425 09:32:27.540628 132936456558336 async_checkpointer.py:288] [process=3][thread=save_finalize] Done with waiting for background save thread=async_save. I0425 09:32:27.540754 132936456558336 async_checkpointer.py:298] [process=3][thread=save_finalize] No errors found in background save thread=async_save. I0425 09:32:27.540807 132936456558336 checkpoint_manager.py:2137] [process=3][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts... I0425 09:32:27.543214 132936456558336 checkpoint_manager.py:2146] [process=3][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts. I0425 09:32:27.543373 133066620835648 checkpoint_manager.py:2032] [process=3][thread=MainThread][step=0][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=0. W0425 09:32:27.543513 133066620835648 checkpoint_manager.py:1452] Waiting for previous save to complete took 1.016884 seconds. If this number is high, consider checkpointing less frequently. I0425 09:32:27.545324 133066620835648 checkpoint_manager.py:1512] [process=3] Saving checkpoint at step 10 I0425 09:32:27.547243 133066620835648 event_tracking.py:70] [process=3] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260425_092114/linen_xpk_feat_nnx_trainstate_and_training_loop_20260425_092114_05_fp8/checkpoints/10. I0425 09:32:27.830691 133066620835648 jax_array_handlers.py:360] Scheduling D2H of 81 prioritized jax.Array. I0425 09:32:27.830785 133066620835648 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 I0425 09:32:27.885389 133066620835648 base_pytree_checkpoint_handler.py:154] [process=3][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.055641s I0425 09:32:27.885572 133066620835648 base_pytree_checkpoint_handler.py:130] [process=3] /jax/orbax/write/blocking_gbytes_per_sec: 24.928 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.06188845634460449 s) (per-host) I0425 09:32:27.885622 133066620835648 base_pytree_checkpoint_handler.py:768] [process=3][thread=MainThread] Initiated Pytree async_save. Time taken: 0.061948s (batch_requests_ready=0.003058s, total_serialization_initiated=0.058823s, others=0.000068s) I0425 09:32:27.885728 133066620835648 composite_checkpoint_handler.py:715] [process=3][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.066145s (all_items=0.000017s, per_item={'items': '0.00001693'}, temp_paths=0.066128) I0425 09:32:27.886482 133066620835648 event_tracking.py:125] [process=3] [async] Finished blocking save in 0.34 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260425_092114/linen_xpk_feat_nnx_trainstate_and_training_loop_20260425_092114_05_fp8/checkpoints/10. I0425 09:32:27.886819 132936456558336 async_checkpointer.py:76] [process=3][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-25 09:52:27.886782 I0425 09:32:27.888783 133066620835648 checkpoint_manager.py:1560] [process=3][thread=MainThread][step=10] Starting CheckpointManager Save Finalize thread=save_finalize I0425 09:32:27.889007 132935378253568 async_checkpointer.py:280] [process=3][thread=save_finalize] Waiting for background save thread=async_save. I0425 09:32:27.889141 133066620835648 standard_logger.py:34] {'step': 10, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260425_092114/linen_xpk_feat_nnx_trainstate_and_training_loop_20260425_092114_05_fp8/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1777109546.526598, 'wait_for_prev_duration_secs': 1.0168840885162354, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1777109547.545362, 'checkpointer_blocking_duration_secs': 0.34156084060668945, 'get_old_steps_start_time': 1777109547.8869472, 'get_old_steps_duration_secs': 3.218650817871094e-05, 'checkpoint_manager_blocking_start_time': 1777109546.52477, 'checkpoint_manager_blocking_duration_secs': 1.3643379211425781} I0425 09:32:27.889312 133066620835648 checkpointing.py:409] Started an asynchronous checkpoint save for step 10 I0425 09:32:27.889940 133066620835648 metric_logger.py:196] completed step: 9, seconds: 0.158, TFLOP/s/device: 85.785, Tokens/s/device: 12930.436, total_weights: 65536, loss: 7.908, lm_loss: 7.908, perplexity: 2720.116 I0425 09:32:27.897070 133066620835648 metric_logger.py:196] completed step: 10, seconds: 0.158, TFLOP/s/device: 85.754, Tokens/s/device: 12925.784, total_weights: 65536, loss: 7.568, lm_loss: 7.568, perplexity: 1936.198 I0425 09:32:28.055376 133066620835648 metric_logger.py:196] completed step: 11, seconds: 1.683, TFLOP/s/device: 8.075, Tokens/s/device: 1217.225, total_weights: 65536, loss: 7.288, lm_loss: 7.288, perplexity: 1463.272 I0425 09:32:28.622099 133066620835648 metric_logger.py:196] completed step: 12, seconds: 0.006, TFLOP/s/device: 2172.898, Tokens/s/device: 327522.789, total_weights: 65536, loss: 7.032, lm_loss: 7.032, perplexity: 1132.061 I0425 09:32:28.780502 133066620835648 metric_logger.py:196] completed step: 13, seconds: 0.563, TFLOP/s/device: 24.144, Tokens/s/device: 3639.239, total_weights: 65536, loss: 6.815, lm_loss: 6.815, perplexity: 911.031 I0425 09:32:28.938756 133066620835648 metric_logger.py:196] completed step: 14, seconds: 0.163, TFLOP/s/device: 83.229, Tokens/s/device: 12545.253, total_weights: 65536, loss: 6.635, lm_loss: 6.635, perplexity: 761.080 I0425 09:32:29.097120 133066620835648 metric_logger.py:196] completed step: 15, seconds: 0.158, TFLOP/s/device: 85.784, Tokens/s/device: 12930.354, total_weights: 65536, loss: 6.492, lm_loss: 6.492, perplexity: 659.915 I0425 09:32:29.255266 133066620835648 metric_logger.py:196] completed step: 16, seconds: 0.158, TFLOP/s/device: 85.903, Tokens/s/device: 12948.176, total_weights: 65536, loss: 6.380, lm_loss: 6.380, perplexity: 589.948 I0425 09:32:29.413417 133066620835648 metric_logger.py:196] completed step: 17, seconds: 0.158, TFLOP/s/device: 85.765, Tokens/s/device: 12927.416, total_weights: 65536, loss: 6.293, lm_loss: 6.293, perplexity: 540.941 I0425 09:32:29.571657 133066620835648 metric_logger.py:196] completed step: 18, seconds: 0.158, TFLOP/s/device: 85.968, Tokens/s/device: 12958.089, total_weights: 65536, loss: 6.223, lm_loss: 6.223, perplexity: 504.108 I0425 09:32:29.729284 133066620835648 checkpointing.py:794] Waiting for step 19 to finish before checkpoint... I0425 09:32:29.730294 133066620835648 checkpointing.py:798] Waited 0.0010280609130859375 seconds for step 19 to finish before starting checkpointing. I0425 09:32:29.732369 133066620835648 checkpoint_manager.py:2020] [process=3][thread=MainThread][step=10][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0425 09:32:33.728752 132933750855424 array_metadata_store.py:203] [process=3][thread=array_type_handler] Wrote 81 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260425_092114/linen_xpk_feat_nnx_trainstate_and_training_loop_20260425_092114_05_fp8/checkpoints/10/items/array_metadatas/process_3 I0425 09:33:09.825095 132936456558336 base_pytree_checkpoint_handler.py:130] [process=3] /jax/orbax/write/gbytes_per_sec: 37.612 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 42.001359939575195 s) (per-host) I0425 09:33:09.825223 132936456558336 async_checkpointer.py:90] [process=3][thread=async_save] 3 Handler Commit operations completed. Time taken: 41.937728s. I0425 09:33:17.589012 132936456558336 async_checkpointer.py:160] [process=3][thread=async_save] Background save thread done. Time taken: 49.701502s. I0425 09:33:17.589313 132935378253568 async_checkpointer.py:288] [process=3][thread=save_finalize] Done with waiting for background save thread=async_save. I0425 09:33:17.589469 132935378253568 async_checkpointer.py:298] [process=3][thread=save_finalize] No errors found in background save thread=async_save. I0425 09:33:17.589521 132935378253568 checkpoint_manager.py:2137] [process=3][thread=save_finalize][step=10] CheckpointManager Save Finalize is syncing with other hosts... I0425 09:33:17.591034 132935378253568 checkpoint_manager.py:2146] [process=3][thread=save_finalize][step=10] CheckpointManager Save Finalize is done on all hosts. I0425 09:33:17.591226 133066620835648 checkpoint_manager.py:2032] [process=3][thread=MainThread][step=10][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=10. W0425 09:33:17.591361 133066620835648 checkpoint_manager.py:1452] Waiting for previous save to complete took 47.858996 seconds. If this number is high, consider checkpointing less frequently. I0425 09:33:17.592992 133066620835648 checkpoint_manager.py:1512] [process=3] Saving checkpoint at step 19 I0425 09:33:17.594955 133066620835648 event_tracking.py:70] [process=3] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260425_092114/linen_xpk_feat_nnx_trainstate_and_training_loop_20260425_092114_05_fp8/checkpoints/19. I0425 09:33:17.879847 133066620835648 jax_array_handlers.py:360] Scheduling D2H of 81 prioritized jax.Array. I0425 09:33:17.879939 133066620835648 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 I0425 09:33:17.937376 133066620835648 base_pytree_checkpoint_handler.py:154] [process=3][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.058467s I0425 09:33:17.937555 133066620835648 base_pytree_checkpoint_handler.py:130] [process=3] /jax/orbax/write/blocking_gbytes_per_sec: 23.879 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.0646059513092041 s) (per-host) I0425 09:33:17.937607 133066620835648 base_pytree_checkpoint_handler.py:768] [process=3][thread=MainThread] Initiated Pytree async_save. Time taken: 0.064669s (batch_requests_ready=0.002995s, total_serialization_initiated=0.061606s, others=0.000068s) I0425 09:33:17.937717 133066620835648 composite_checkpoint_handler.py:715] [process=3][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.068924s (all_items=0.000013s, per_item={'items': '0.00001264'}, temp_paths=0.068912) I0425 09:33:17.938458 133066620835648 event_tracking.py:125] [process=3] [async] Finished blocking save in 0.35 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260425_092114/linen_xpk_feat_nnx_trainstate_and_training_loop_20260425_092114_05_fp8/checkpoints/19. I0425 09:33:17.938804 132935378253568 async_checkpointer.py:76] [process=3][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-25 09:53:17.938768 I0425 09:33:17.940747 133066620835648 checkpoint_manager.py:1560] [process=3][thread=MainThread][step=19] Starting CheckpointManager Save Finalize thread=save_finalize I0425 09:33:17.940958 132932157028096 async_checkpointer.py:280] [process=3][thread=save_finalize] Waiting for background save thread=async_save. I0425 09:33:17.941087 133066620835648 standard_logger.py:34] {'step': 19, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260425_092114/linen_xpk_feat_nnx_trainstate_and_training_loop_20260425_092114_05_fp8/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1777109549.7323406, 'wait_for_prev_duration_secs': 47.85899567604065, 'time_between_consecutive_saves_sec': 2.189087390899658, 'checkpointer_blocking_start_time': 1777109597.5930321, 'checkpointer_blocking_duration_secs': 0.3459153175354004, 'get_old_steps_start_time': 1777109597.9389665, 'get_old_steps_duration_secs': 2.6226043701171875e-05, 'checkpoint_manager_blocking_start_time': 1777109549.7305815, 'checkpoint_manager_blocking_duration_secs': 48.2104754447937} I0425 09:33:17.941247 133066620835648 checkpointing.py:409] Started an asynchronous checkpoint save for step 19 I0425 09:33:17.941290 133066620835648 checkpoint_manager.py:2020] [process=3][thread=MainThread][step=19][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0425 09:33:23.316375 132933750855424 array_metadata_store.py:203] [process=3][thread=array_type_handler] Wrote 81 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260425_092114/linen_xpk_feat_nnx_trainstate_and_training_loop_20260425_092114_05_fp8/checkpoints/19/items/array_metadatas/process_3 I0425 09:33:59.349762 132935378253568 base_pytree_checkpoint_handler.py:130] [process=3] /jax/orbax/write/gbytes_per_sec: 38.088 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 41.47676992416382 s) (per-host) I0425 09:33:59.349887 132935378253568 async_checkpointer.py:90] [process=3][thread=async_save] 3 Handler Commit operations completed. Time taken: 41.410970s. I0425 09:34:08.024710 132935378253568 async_checkpointer.py:160] [process=3][thread=async_save] Background save thread done. Time taken: 50.085777s. I0425 09:34:08.024971 132932157028096 async_checkpointer.py:288] [process=3][thread=save_finalize] Done with waiting for background save thread=async_save. I0425 09:34:08.025099 132932157028096 async_checkpointer.py:298] [process=3][thread=save_finalize] No errors found in background save thread=async_save. I0425 09:34:08.025141 132932157028096 checkpoint_manager.py:2137] [process=3][thread=save_finalize][step=19] CheckpointManager Save Finalize is syncing with other hosts... I0425 09:34:08.026715 132932157028096 checkpoint_manager.py:2146] [process=3][thread=save_finalize][step=19] CheckpointManager Save Finalize is done on all hosts. I0425 09:34:08.026898 133066620835648 checkpoint_manager.py:2032] [process=3][thread=MainThread][step=19][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=19. I0425 09:34:08.027046 133066620835648 checkpoint_manager.py:2009] [process=3][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0425 09:34:08.028032 133066620835648 metric_logger.py:196] completed step: 19, seconds: 0.158, TFLOP/s/device: 85.866, Tokens/s/device: 12942.693, 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: Sat Apr 25 09:34:17 UTC 2026 EXIT_CODE=0
XPK Start: Sat Apr 25 10:57:59 UTC 2026 PyTorch was not found. Models won't be available and only tokenizers, configuration and file/data utilities can be used. `rope_parameters`'s factor field must be a float >= 1, got 40 `rope_parameters`'s beta_fast field must be a float, got 32 `rope_parameters`'s beta_slow field must be a float, got 1 DeepseekV32Config got `key=rope_scaling` in kwargs but hasn't set it as attribute. For RoPE standardization you need to set `self.rope_parameters` in model's config. 2026-04-25 10:58:24.129242: 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) I0425 10:58:24.343617 139025714255680 max_utils.py:273] Attempting to initialize the jax distributed system... I0425 10:58:33.384110 139025714255680 distributed.py:149] Starting JAX distributed service on [::]:8482 I0425 10:58:33.386693 139025714255680 distributed.py:172] Connecting to JAX distributed service on mt-05-fp8-twaho-slice-job-0-0.mt-05-fp8-twaho:8482 I0425 10:58:34.307609 139025714255680 max_utils.py:284] Jax distributed system initialized! I0425 10:58:40.406351 139025714255680 max_utils.py:800] System Information: Jax Version: 0.9.2 I0425 10:58:40.406461 139025714255680 max_utils.py:801] System Information: Jaxlib Version: 0.9.2 I0425 10:58:40.406500 139025714255680 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 I0425 10:58:40.406537 139025714255680 train_utils.py:391] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing. I0425 10:58:41.098713 139025714255680 maxtext_utils.py:1771] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0425 10:58:41.529741 139025714255680 checkpointing.py:688] Setting up checkpoint logger... I0425 10:58:41.529858 139025714255680 checkpointing.py:234] Creating checkpoint manager with ocdbt=True and zarr3=True I0425 10:58:41.529904 139025714255680 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None I0425 10:58:41.530137 139025714255680 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 0x7e70d917e5d0>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0425 10:58:44.415996 139025714255680 checkpointing.py:266] Enabling policy for fixed interval checkpointing. I0425 10:58:44.416240 139025714255680 checkpoint_manager.py:708] [process=3][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=('items',), item_handlers={'items': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7e5c9410e270>}, handler_registry=None I0425 10:58:44.416482 139025714255680 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7e5c9410e270>` 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`. I0425 10:58:44.416530 139025714255680 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7e5c7c4d6a20>` 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`. I0425 10:58:44.416567 139025714255680 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 0x7e5c9410e270>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7e5c9410e270>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7e5c7c4d6a20>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7e5c7c4d6a20>}). I0425 10:58:44.416908 139025714255680 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34 I0425 10:58:44.416991 139025714255680 async_checkpointer.py:192] [process=3][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7e5c940aae80> timeout: 1200 secs and primary_host=0 for async checkpoint writes I0425 10:58:45.530763 139025714255680 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260425_092114/nnx_xpk_feat_nnx_trainstate_and_training_loop_20260425_092114_05_fp8/checkpoints I0425 10:58:45.588759 139025714255680 checkpoint_manager.py:929] [process=3][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_20260425_092114/nnx_xpk_feat_nnx_trainstate_and_training_loop_20260425_092114_05_fp8/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7e5c7c487530> I0425 10:58:45.588920 139025714255680 checkpointing.py:302] Checkpoint manager created! I0425 10:58:46.863490 139025714255680 checkpointing.py:578] checkpoint manager exists so trying to load this run's existing checkpoint I0425 10:58:46.863613 139025714255680 checkpointing.py:676] No existing checkpoints found, not restoring checkpoint. fsdp: 32 I0425 10:58:49.393156 139025714255680 nnx_decoders.py:465] nnx_decoders/carry Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0425 10:58:49.393256 139025714255680 nnx_decoders.py:465] nnx_decoders/carry Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0425 10:58:49.398799 139025714255680 nnx_decoders.py:465] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0425 10:58:49.398861 139025714255680 nnx_decoders.py:465] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0425 10:58:49.415367 139025714255680 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch_attn', 'activation_length_attn', 'activation_embed_attn'). I0425 10:58:49.415433 139025714255680 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0425 10:58:49.431406 139025714255680 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch_attn', 'activation_length_attn', 'activation_embed_attn'). I0425 10:58:49.431466 139025714255680 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0425 10:58:49.490781 139025714255680 attentions.py:1154] attentions/query Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_length_attn', 'activation_kv_heads', 'activation_kv_head_dim'). I0425 10:58:49.490866 139025714255680 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0425 10:58:49.506788 139025714255680 attentions.py:1155] attentions/key Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_length_attn', 'activation_kv_heads', 'activation_kv_head_dim'). I0425 10:58:49.506852 139025714255680 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0425 10:58:49.522691 139025714255680 attentions.py:1156] attentions/value Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_length_attn', 'activation_kv_heads', 'activation_kv_head_dim'). I0425 10:58:49.522751 139025714255680 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0425 10:58:49.554411 139025714255680 attentions.py:1198] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch_attn', 'activation_length_attn', 'activation_heads', 'activation_kv'). I0425 10:58:49.554485 139025714255680 attentions.py:1198] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0425 10:58:49.614007 139025714255680 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0425 10:58:49.614092 139025714255680 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None). I0425 10:59:03.637450 139025714255680 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. I0425 10:59:03.641333 139025714255680 metric_logger.py:301] number parameters: 1.104 billion I0425 10:59:17.281382 139025714255680 checkpointing.py:794] Waiting for step 0 to finish before checkpoint... I0425 10:59:17.443669 139025714255680 checkpointing.py:798] Waited 0.1622781753540039 seconds for step 0 to finish before starting checkpointing. I0425 10:59:17.446316 139025714255680 checkpoint_manager.py:2009] [process=3][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0425 10:59:17.448475 139025714255680 checkpoint_manager.py:1512] [process=3] Saving checkpoint at step 0 I0425 10:59:17.449947 139025714255680 event_tracking.py:70] [process=3] [async] Started save checkpoint @ gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260425_092114/nnx_xpk_feat_nnx_trainstate_and_training_loop_20260425_092114_05_fp8/checkpoints/0. I0425 10:59:17.962777 139025714255680 signaling_client.py:364] Using JaxDistributedSignalingClient I0425 10:59:17.963790 139025714255680 jax_array_handlers.py:360] Scheduling D2H of 153 prioritized jax.Array. I0425 10:59:17.963855 139025714255680 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 I0425 10:59:18.373424 139025714255680 base_pytree_checkpoint_handler.py:154] [process=3][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.412303s I0425 10:59:18.373598 139025714255680 base_pytree_checkpoint_handler.py:130] [process=3] /jax/orbax/write/blocking_gbytes_per_sec: 3.596 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.42902398109436035 s) (per-host) I0425 10:59:18.373649 139025714255680 base_pytree_checkpoint_handler.py:768] [process=3][thread=MainThread] Initiated Pytree async_save. Time taken: 0.429085s (batch_requests_ready=0.005499s, total_serialization_initiated=0.423517s, others=0.000068s) I0425 10:59:18.373770 139025714255680 composite_checkpoint_handler.py:715] [process=3][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.433033s (all_items=0.000020s, per_item={'items': '0.00002003'}, temp_paths=0.433013) I0425 10:59:18.374533 139025714255680 event_tracking.py:125] [process=3] [async] Finished blocking save in 0.93 seconds. Continuing save @ gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260425_092114/nnx_xpk_feat_nnx_trainstate_and_training_loop_20260425_092114_05_fp8/checkpoints/0. I0425 10:59:18.374844 138896459597568 async_checkpointer.py:76] [process=3][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-25 11:19:18.374806 I0425 10:59:18.383857 139025714255680 checkpoint_manager.py:1560] [process=3][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize I0425 10:59:18.384165 138895921178368 async_checkpointer.py:280] [process=3][thread=save_finalize] Waiting for background save thread=async_save. I0425 10:59:18.384320 139025714255680 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260425_092114/nnx_xpk_feat_nnx_trainstate_and_training_loop_20260425_092114_05_fp8/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1777114757.4462974, 'wait_for_prev_duration_secs': 6.341934204101562e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1777114757.4485137, 'checkpointer_blocking_duration_secs': 0.9264364242553711, 'get_old_steps_start_time': 1777114758.3749764, 'get_old_steps_duration_secs': 2.86102294921875e-05, 'checkpoint_manager_blocking_start_time': 1777114757.4441497, 'checkpoint_manager_blocking_duration_secs': 0.9401328563690186} I0425 10:59:18.384433 139025714255680 checkpointing.py:409] Started an asynchronous checkpoint save for step 0 I0425 10:59:18.384505 139025714255680 max_utils.py:750] Memstats: After params initialized: I0425 10:59:18.384551 139025714255680 max_utils.py:756] Using (GB) 0.45 / 31.25 (1.440000%) on TPU_10(process=3,(2,2,0,0)) I0425 10:59:18.384583 139025714255680 max_utils.py:756] Using (GB) 0.45 / 31.25 (1.440000%) on TPU_11(process=3,(3,2,0,0)) I0425 10:59:18.384609 139025714255680 max_utils.py:756] Using (GB) 0.45 / 31.25 (1.440000%) on TPU_14(process=3,(2,3,0,0)) I0425 10:59:18.384633 139025714255680 max_utils.py:756] Using (GB) 0.45 / 31.25 (1.440000%) on TPU_15(process=3,(3,3,0,0)) I0425 10:59:18.703114 139025714255680 metric_logger.py:196] completed step: 0, seconds: 13.638, TFLOP/s/device: 0.996, Tokens/s/device: 150.171, total_weights: 65536, loss: 10.872, lm_loss: 10.872, perplexity: 52680.742 I0425 10:59:18.886156 139025714255680 metric_logger.py:196] completed step: 1, seconds: 1.419, TFLOP/s/device: 9.575, Tokens/s/device: 1443.241, total_weights: 65536, loss: 10.872, lm_loss: 10.872, perplexity: 52680.742 I0425 10:59:19.331307 139025714255680 metric_logger.py:196] completed step: 2, seconds: 0.030, TFLOP/s/device: 449.607, Tokens/s/device: 67769.689, total_weights: 65536, loss: 10.856, lm_loss: 10.856, perplexity: 51864.762 I0425 10:59:19.488119 139025714255680 metric_logger.py:196] completed step: 3, seconds: 0.422, TFLOP/s/device: 32.183, Tokens/s/device: 4851.000, total_weights: 65536, loss: 10.824, lm_loss: 10.824, perplexity: 50225.512 I0425 10:59:19.803033 139025714255680 metric_logger.py:196] completed step: 4, seconds: 0.186, TFLOP/s/device: 72.871, Tokens/s/device: 10983.942, total_weights: 65536, loss: 10.793, lm_loss: 10.793, perplexity: 48689.156 I0425 10:59:19.813426 139025714255680 metric_logger.py:196] completed step: 5, seconds: 0.157, TFLOP/s/device: 86.788, Tokens/s/device: 13081.581, total_weights: 65536, loss: 10.763, lm_loss: 10.763, perplexity: 47229.012 I0425 10:59:21.404722 2798 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com I0425 10:59:23.597281 138895929571072 array_metadata_store.py:203] [process=3][thread=array_type_handler] Wrote 153 array_metadata.ArrayMetadata to gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260425_092114/nnx_xpk_feat_nnx_trainstate_and_training_loop_20260425_092114_05_fp8/checkpoints/0/items/array_metadatas/process_3 I0425 10:59:38.305193 138896459597568 base_pytree_checkpoint_handler.py:130] [process=3] /jax/orbax/write/gbytes_per_sec: 77.589 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 20.360581874847412 s) (per-host) I0425 10:59:38.305321 138896459597568 async_checkpointer.py:90] [process=3][thread=async_save] 3 Handler Commit operations completed. Time taken: 19.929837s. I0425 10:59:43.760964 139025714255680 metric_logger.py:196] completed step: 6, seconds: 0.316, TFLOP/s/device: 42.980, Tokens/s/device: 6478.368, total_weights: 65536, loss: 10.733, lm_loss: 10.733, perplexity: 45864.258 I0425 10:59:43.917723 139025714255680 metric_logger.py:196] completed step: 7, seconds: 23.794, TFLOP/s/device: 0.571, Tokens/s/device: 86.072, total_weights: 65536, loss: 10.706, lm_loss: 10.706, perplexity: 44620.242 I0425 10:59:44.229822 139025714255680 metric_logger.py:196] completed step: 8, seconds: 0.163, TFLOP/s/device: 83.440, Tokens/s/device: 12576.917, total_weights: 65536, loss: 10.680, lm_loss: 10.680, perplexity: 43477.055 I0425 10:59:44.238140 139025714255680 checkpointing.py:794] Waiting for step 10 to finish before checkpoint... I0425 10:59:44.543896 139025714255680 checkpointing.py:798] Waited 0.3057265281677246 seconds for step 10 to finish before starting checkpointing. I0425 10:59:44.546290 139025714255680 checkpoint_manager.py:2020] [process=3][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0425 10:59:45.408214 138896459597568 async_checkpointer.py:160] [process=3][thread=async_save] Background save thread done. Time taken: 27.032713s. I0425 10:59:45.408504 138895921178368 async_checkpointer.py:288] [process=3][thread=save_finalize] Done with waiting for background save thread=async_save. I0425 10:59:45.408630 138895921178368 async_checkpointer.py:298] [process=3][thread=save_finalize] No errors found in background save thread=async_save. I0425 10:59:45.408689 138895921178368 checkpoint_manager.py:2137] [process=3][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts... I0425 10:59:45.411072 138895921178368 checkpoint_manager.py:2146] [process=3][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts. I0425 10:59:45.411256 139025714255680 checkpoint_manager.py:2032] [process=3][thread=MainThread][step=0][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=0. I0425 10:59:45.412780 139025714255680 checkpoint_manager.py:1512] [process=3] Saving checkpoint at step 10 I0425 10:59:45.414874 139025714255680 event_tracking.py:70] [process=3] [async] Started save checkpoint @ gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260425_092114/nnx_xpk_feat_nnx_trainstate_and_training_loop_20260425_092114_05_fp8/checkpoints/10. I0425 10:59:45.719407 139025714255680 jax_array_handlers.py:360] Scheduling D2H of 153 prioritized jax.Array. I0425 10:59:45.719515 139025714255680 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 I0425 10:59:45.767952 139025714255680 base_pytree_checkpoint_handler.py:154] [process=3][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.051015s I0425 10:59:45.768125 139025714255680 base_pytree_checkpoint_handler.py:130] [process=3] /jax/orbax/write/blocking_gbytes_per_sec: 23.518 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.0655984878540039 s) (per-host) I0425 10:59:45.768178 139025714255680 base_pytree_checkpoint_handler.py:768] [process=3][thread=MainThread] Initiated Pytree async_save. Time taken: 0.065661s (batch_requests_ready=0.005470s, total_serialization_initiated=0.060121s, others=0.000070s) I0425 10:59:45.768300 139025714255680 composite_checkpoint_handler.py:715] [process=3][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.069686s (all_items=0.000022s, per_item={'items': '0.00002193'}, temp_paths=0.069664) I0425 10:59:45.769038 139025714255680 event_tracking.py:125] [process=3] [async] Finished blocking save in 0.36 seconds. Continuing save @ gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260425_092114/nnx_xpk_feat_nnx_trainstate_and_training_loop_20260425_092114_05_fp8/checkpoints/10. I0425 10:59:45.769305 138895921178368 async_checkpointer.py:76] [process=3][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-25 11:19:45.769282 I0425 10:59:45.771265 139025714255680 checkpoint_manager.py:1560] [process=3][thread=MainThread][step=10] Starting CheckpointManager Save Finalize thread=save_finalize I0425 10:59:45.771477 138894326617856 async_checkpointer.py:280] [process=3][thread=save_finalize] Waiting for background save thread=async_save. I0425 10:59:45.771592 139025714255680 standard_logger.py:34] {'step': 10, 'event_type': 'save', 'directory': 'gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260425_092114/nnx_xpk_feat_nnx_trainstate_and_training_loop_20260425_092114_05_fp8/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1777114784.5462594, 'wait_for_prev_duration_secs': 0.8650994300842285, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1777114785.4128215, 'checkpointer_blocking_duration_secs': 0.3565857410430908, 'get_old_steps_start_time': 1777114785.76943, 'get_old_steps_duration_secs': 2.9087066650390625e-05, 'checkpoint_manager_blocking_start_time': 1777114784.5443513, 'checkpoint_manager_blocking_duration_secs': 1.2272088527679443} I0425 10:59:45.771704 139025714255680 checkpointing.py:409] Started an asynchronous checkpoint save for step 10 I0425 10:59:45.772345 139025714255680 metric_logger.py:196] completed step: 9, seconds: 0.306, TFLOP/s/device: 44.332, Tokens/s/device: 6682.263, total_weights: 65536, loss: 10.656, lm_loss: 10.656, perplexity: 42467.320 I0425 10:59:45.783314 139025714255680 metric_logger.py:196] completed step: 10, seconds: 0.162, TFLOP/s/device: 83.668, Tokens/s/device: 12611.381, total_weights: 65536, loss: 10.636, lm_loss: 10.636, perplexity: 41601.941 I0425 10:59:45.936900 139025714255680 metric_logger.py:196] completed step: 11, seconds: 1.542, TFLOP/s/device: 8.809, Tokens/s/device: 1327.791, total_weights: 65536, loss: 10.618, lm_loss: 10.618, perplexity: 40849.570 I0425 10:59:46.498335 139025714255680 metric_logger.py:196] completed step: 12, seconds: 0.010, TFLOP/s/device: 1370.776, Tokens/s/device: 206618.241, total_weights: 65536, loss: 10.602, lm_loss: 10.602, perplexity: 40203.926 I0425 10:59:46.655293 139025714255680 metric_logger.py:196] completed step: 13, seconds: 0.553, TFLOP/s/device: 24.565, Tokens/s/device: 3702.766, total_weights: 65536, loss: 10.588, lm_loss: 10.588, perplexity: 39652.145 I0425 10:59:46.812177 139025714255680 metric_logger.py:196] completed step: 14, seconds: 0.163, TFLOP/s/device: 83.196, Tokens/s/device: 12540.260, total_weights: 65536, loss: 10.577, lm_loss: 10.577, perplexity: 39212.879 I0425 10:59:46.969111 139025714255680 metric_logger.py:196] completed step: 15, seconds: 0.157, TFLOP/s/device: 86.730, Tokens/s/device: 13072.897, total_weights: 65536, loss: 10.568, lm_loss: 10.568, perplexity: 38879.223 I0425 10:59:47.125875 139025714255680 metric_logger.py:196] completed step: 16, seconds: 0.157, TFLOP/s/device: 86.569, Tokens/s/device: 13048.658, total_weights: 65536, loss: 10.562, lm_loss: 10.562, perplexity: 38619.105 I0425 10:59:47.282633 139025714255680 metric_logger.py:196] completed step: 17, seconds: 0.157, TFLOP/s/device: 86.587, Tokens/s/device: 13051.403, total_weights: 65536, loss: 10.556, lm_loss: 10.556, perplexity: 38395.816 I0425 10:59:47.439586 139025714255680 metric_logger.py:196] completed step: 18, seconds: 0.157, TFLOP/s/device: 86.666, Tokens/s/device: 13063.307, total_weights: 65536, loss: 10.552, lm_loss: 10.552, perplexity: 38236.652 I0425 10:59:47.598012 139025714255680 checkpointing.py:794] Waiting for step 19 to finish before checkpoint... I0425 10:59:47.599250 139025714255680 checkpointing.py:798] Waited 0.001249551773071289 seconds for step 19 to finish before starting checkpointing. I0425 10:59:47.601509 139025714255680 checkpoint_manager.py:2020] [process=3][thread=MainThread][step=10][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0425 10:59:52.263111 138892707612416 array_metadata_store.py:203] [process=3][thread=array_type_handler] Wrote 153 array_metadata.ArrayMetadata to gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260425_092114/nnx_xpk_feat_nnx_trainstate_and_training_loop_20260425_092114_05_fp8/checkpoints/10/items/array_metadatas/process_3 I0425 11:00:07.755494 138895921178368 base_pytree_checkpoint_handler.py:130] [process=3] /jax/orbax/write/gbytes_per_sec: 71.635 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 22.052945613861084 s) (per-host) I0425 11:00:07.755602 138895921178368 async_checkpointer.py:90] [process=3][thread=async_save] 3 Handler Commit operations completed. Time taken: 21.986218s. I0425 11:00:14.067419 138895921178368 async_checkpointer.py:160] [process=3][thread=async_save] Background save thread done. Time taken: 28.298018s. I0425 11:00:14.067652 138894326617856 async_checkpointer.py:288] [process=3][thread=save_finalize] Done with waiting for background save thread=async_save. I0425 11:00:14.067722 138894326617856 async_checkpointer.py:298] [process=3][thread=save_finalize] No errors found in background save thread=async_save. I0425 11:00:14.067767 138894326617856 checkpoint_manager.py:2137] [process=3][thread=save_finalize][step=10] CheckpointManager Save Finalize is syncing with other hosts... I0425 11:00:14.069326 138894326617856 checkpoint_manager.py:2146] [process=3][thread=save_finalize][step=10] CheckpointManager Save Finalize is done on all hosts. I0425 11:00:14.069509 139025714255680 checkpoint_manager.py:2032] [process=3][thread=MainThread][step=10][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=10. W0425 11:00:14.069656 139025714255680 checkpoint_manager.py:1452] Waiting for previous save to complete took 26.468141 seconds. If this number is high, consider checkpointing less frequently. I0425 11:00:14.071535 139025714255680 checkpoint_manager.py:1512] [process=3] Saving checkpoint at step 19 I0425 11:00:14.073537 139025714255680 event_tracking.py:70] [process=3] [async] Started save checkpoint @ gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260425_092114/nnx_xpk_feat_nnx_trainstate_and_training_loop_20260425_092114_05_fp8/checkpoints/19. I0425 11:00:14.814232 139025714255680 jax_array_handlers.py:360] Scheduling D2H of 153 prioritized jax.Array. I0425 11:00:14.814344 139025714255680 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 I0425 11:00:14.867633 139025714255680 base_pytree_checkpoint_handler.py:154] [process=3][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.056281s I0425 11:00:14.867799 139025714255680 base_pytree_checkpoint_handler.py:130] [process=3] /jax/orbax/write/blocking_gbytes_per_sec: 2.965 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.5203089714050293 s) (per-host) I0425 11:00:14.867850 139025714255680 base_pytree_checkpoint_handler.py:768] [process=3][thread=MainThread] Initiated Pytree async_save. Time taken: 0.520370s (batch_requests_ready=0.455079s, total_serialization_initiated=0.065225s, others=0.000066s) I0425 11:00:14.867964 139025714255680 composite_checkpoint_handler.py:715] [process=3][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.524512s (all_items=0.000011s, per_item={'items': '0.00001144'}, temp_paths=0.524501) I0425 11:00:14.868717 139025714255680 event_tracking.py:125] [process=3] [async] Finished blocking save in 0.80 seconds. Continuing save @ gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260425_092114/nnx_xpk_feat_nnx_trainstate_and_training_loop_20260425_092114_05_fp8/checkpoints/19. I0425 11:00:14.869029 138894326617856 async_checkpointer.py:76] [process=3][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-25 11:20:14.868996 I0425 11:00:14.870997 139025714255680 checkpoint_manager.py:1560] [process=3][thread=MainThread][step=19] Starting CheckpointManager Save Finalize thread=save_finalize I0425 11:00:14.871276 138893252876032 async_checkpointer.py:280] [process=3][thread=save_finalize] Waiting for background save thread=async_save. I0425 11:00:14.871434 139025714255680 standard_logger.py:34] {'step': 19, 'event_type': 'save', 'directory': 'gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260425_092114/nnx_xpk_feat_nnx_trainstate_and_training_loop_20260425_092114_05_fp8/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1777114787.60148, 'wait_for_prev_duration_secs': 26.468140602111816, 'time_between_consecutive_saves_sec': 2.190366744995117, 'checkpointer_blocking_start_time': 1777114814.071574, 'checkpointer_blocking_duration_secs': 0.7975938320159912, 'get_old_steps_start_time': 1777114814.8691857, 'get_old_steps_duration_secs': 2.574920654296875e-05, 'checkpoint_manager_blocking_start_time': 1777114787.5995429, 'checkpoint_manager_blocking_duration_secs': 27.271860361099243} I0425 11:00:14.871554 139025714255680 checkpointing.py:409] Started an asynchronous checkpoint save for step 19 I0425 11:00:14.871596 139025714255680 checkpoint_manager.py:2020] [process=3][thread=MainThread][step=19][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0425 11:00:19.824408 138894343403264 array_metadata_store.py:203] [process=3][thread=array_type_handler] Wrote 153 array_metadata.ArrayMetadata to gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260425_092114/nnx_xpk_feat_nnx_trainstate_and_training_loop_20260425_092114_05_fp8/checkpoints/19/items/array_metadatas/process_3 I0425 11:00:36.179052 138894326617856 base_pytree_checkpoint_handler.py:130] [process=3] /jax/orbax/write/gbytes_per_sec: 72.361 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 21.83151149749756 s) (per-host) I0425 11:00:36.179171 138894326617856 async_checkpointer.py:90] [process=3][thread=async_save] 3 Handler Commit operations completed. Time taken: 21.310031s. I0425 11:00:42.170537 138894326617856 async_checkpointer.py:160] [process=3][thread=async_save] Background save thread done. Time taken: 27.301381s. I0425 11:00:42.170897 138893252876032 async_checkpointer.py:288] [process=3][thread=save_finalize] Done with waiting for background save thread=async_save. I0425 11:00:42.171034 138893252876032 async_checkpointer.py:298] [process=3][thread=save_finalize] No errors found in background save thread=async_save. I0425 11:00:42.171076 138893252876032 checkpoint_manager.py:2137] [process=3][thread=save_finalize][step=19] CheckpointManager Save Finalize is syncing with other hosts... I0425 11:00:42.173492 138893252876032 checkpoint_manager.py:2146] [process=3][thread=save_finalize][step=19] CheckpointManager Save Finalize is done on all hosts. I0425 11:00:42.173698 139025714255680 checkpoint_manager.py:2032] [process=3][thread=MainThread][step=19][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=19. I0425 11:00:42.173879 139025714255680 checkpoint_manager.py:2009] [process=3][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0425 11:00:42.174879 139025714255680 metric_logger.py:196] completed step: 19, seconds: 0.157, TFLOP/s/device: 86.725, Tokens/s/device: 13072.062, 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: Sat Apr 25 11:00:54 UTC 2026 EXIT_CODE=0