2026-04-17 13:46:23.119139: E external/local_xla/xla/stream_executor/cuda/cuda_platform.cc:51] failed call to cuInit: INTERNAL: CUDA error: Failed call to cuInit: UNKNOWN ERROR (303) I0417 13:46:23.236891 140132447517824 max_utils.py:238] Skipping jax distributed system due to skip_jax_distributed_system=True flag. I0417 13:47:16.328656 140132447517824 max_utils.py:800] System Information: Jax Version: 0.9.2 I0417 13:47:16.328764 140132447517824 max_utils.py:801] System Information: Jaxlib Version: 0.9.2 I0417 13:47:16.328799 140132447517824 max_utils.py:802] System Information: Jax Backend: PJRT C API TFRT TPU v6 lite Built on Apr 6 2026 20:48:10 (1775533690) cl/895581894 I0417 13:47:16.328827 140132447517824 train_utils.py:335] WARNING: 'dataset_path' might be pointing your local file system I0417 13:47:16.328851 140132447517824 train_utils.py:348] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing. I0417 13:47:16.328954 140132447517824 train.py:671] [DECOUPLED NO-OP] skipping cloud diagnostics wrapper. W0417 13:47:16.421745 489912 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0417 13:47:16.806341 140132447517824 maxtext_utils.py:1548] Num_devices: 8, shape (1, 1, 1, 8, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0417 13:47:16.806676 140132447517824 checkpointing.py:677] Setting up checkpoint logger... I0417 13:47:16.806726 140132447517824 checkpointing.py:233] Creating checkpoint manager with ocdbt=True and zarr3=True I0417 13:47:16.806780 140132447517824 pytree_checkpoint_handler.py:577] save_device_host_concurrent_bytes=None I0417 13:47:16.807332 140132447517824 base_pytree_checkpoint_handler.py:411] 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 0x7f7289df8c20>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0417 13:47:19.222179 140132447517824 checkpointing.py:265] Enabling policy for fixed interval checkpointing. I0417 13:47:19.222364 140132447517824 checkpoint_manager.py:702] [process=0][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=('items',), item_handlers={'items': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7f6d8bb1c1a0>}, handler_registry=None I0417 13:47:19.222646 140132447517824 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7f6d8bb1c1a0>` for item "items" and save args `<class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeSaveArgs'>` and restore args `<class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>` to `_handler_registry`. I0417 13:47:19.222687 140132447517824 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7f6c0db32510>` for item "metrics" and save args `<class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>` and restore args `<class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>` to `_handler_registry`. I0417 13:47:19.222715 140132447517824 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 0x7f6d8bb1c1a0>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7f6d8bb1c1a0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7f6c0db32510>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7f6c0db32510>}). I0417 13:47:19.223019 140132447517824 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.28 I0417 13:47:19.223088 140132447517824 async_checkpointer.py:177] [process=0][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>.<lambda> at 0x7f6c0db2e200> timeout: 600 secs and primary_host=0 for async checkpoint writes I0417 13:47:19.349833 140132447517824 checkpoint_manager.py:1788] Found 0 checkpoint steps in gs://wanglance-maxtext/linen_ckpt_main_20260417_125240/linen_main_20260417_125240_15_ocdbt_true/checkpoints I0417 13:47:19.350089 140132447517824 checkpoint_manager.py:921] [process=0][thread=MainThread] CheckpointManager created, primary_host=0, CheckpointManagerOptions=CheckpointManagerOptions(save_interval_steps=1, max_to_keep=None, keep_time_interval=None, keep_period=None, should_keep_fn=None, best_fn=None, best_mode='max', keep_checkpoints_without_metrics=True, step_prefix=None, step_format_fixed_length=None, step_name_format=None, create=True, cleanup_tmp_directories=False, save_on_steps=frozenset(), single_host_load_and_broadcast=False, todelete_subdir=None, todelete_full_path=None, enable_hns=False, 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=5), preservation_policy=LatestN(n=None), prevent_write_metrics=False, enable_should_save_is_saving_in_progress_check=True, enable_per_process_directory_creation=False), root_directory=gs://wanglance-maxtext/linen_ckpt_main_20260417_125240/linen_main_20260417_125240_15_ocdbt_true/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7f6c0e398b00> I0417 13:47:19.350172 140132447517824 checkpointing.py:301] Checkpoint manager created! W0417 13:47:19.366923 489912 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0417 13:47:19.908121 140132447517824 nnx_wrappers.py:437] Unknown Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0417 13:47:19.908232 140132447517824 nnx_wrappers.py:437] Unknown Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0417 13:47:19.991580 140132447517824 attentions.py:1088] attentions/inputs_q Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0417 13:47:19.991665 140132447517824 attentions.py:1088] attentions/inputs_q Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0417 13:47:20.005816 140132447517824 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0417 13:47:20.005864 140132447517824 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0417 13:47:20.026336 140132447517824 attentions.py:1154] attentions/query Logical: bfloat16[8,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0417 13:47:20.026389 140132447517824 attentions.py:1154] attentions/query Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0417 13:47:20.040576 140132447517824 attentions.py:1155] attentions/key Logical: bfloat16[8,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0417 13:47:20.040624 140132447517824 attentions.py:1155] attentions/key Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0417 13:47:20.054777 140132447517824 attentions.py:1156] attentions/value Logical: bfloat16[8,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0417 13:47:20.054826 140132447517824 attentions.py:1156] attentions/value Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0417 13:47:20.076337 140132447517824 attentions.py:1197] attentions/out Logical: bfloat16[8,2048,16,128]..................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv'). I0417 13:47:20.076396 140132447517824 attentions.py:1197] attentions/out Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0417 13:47:20.094641 140132447517824 linears.py:525] linears/x Logical: bfloat16[8,2048,7168]....................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0417 13:47:20.094692 140132447517824 linears.py:525] linears/x Physical: bfloat16[8,2048,7168]....................................... ('fsdp', None, None). I0417 13:47:20.288980 140132447517824 checkpointing.py:577] checkpoint manager exists so trying to load this run's existing checkpoint I0417 13:47:20.289084 140132447517824 checkpointing.py:665] No existing checkpoints found, not restoring checkpoint. W0417 13:47:20.339146 489912 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. [DECOUPLED NO-OP] gcs_storage: using stubs. [DECOUPLED NO-OP] mldiagnostics: using stub. [DECOUPLED NO-OP] mldiagnostics: using stub. [DECOUPLED NO-OP] mldiagnostics: using stub. [DECOUPLED NO-OP] workload_monitor: using stub. [DECOUPLED NO-OP] vertex_tensorboard: using stub. fsdp: 8 I0417 13:47:20.719910 140132447517824 maxtext_utils.py:1651] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: P('norm',) Physical: (None,) I0417 13:47:20.720018 140132447517824 maxtext_utils.py:1651] params/params/decoder/layers/mlp/wi_0/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0417 13:47:20.720074 140132447517824 maxtext_utils.py:1651] params/params/decoder/layers/mlp/wi_1/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0417 13:47:20.720123 140132447517824 maxtext_utils.py:1651] params/params/decoder/layers/mlp/wo/kernel Shape: float32[7168,16,2048] Logical: P('mlp', 'layers', 'embed') Physical: (None, None, 'fsdp') I0417 13:47:20.720164 140132447517824 maxtext_utils.py:1651] params/params/decoder/layers/post_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0417 13:47:20.720195 140132447517824 maxtext_utils.py:1651] params/params/decoder/layers/pre_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0417 13:47:20.720236 140132447517824 maxtext_utils.py:1651] params/params/decoder/layers/self_attention/key/kernel Shape: float32[2048,16,16,128] Logical: P('embed', 'layers', 'kv_heads', 'kv_head_dim') Physical: ('fsdp', None, None, None) I0417 13:47:20.720275 140132447517824 maxtext_utils.py:1651] params/params/decoder/layers/self_attention/out/kernel Shape: float32[16,16,128,2048] Logical: P('heads', 'layers', 'kv', 'embed') Physical: (None, None, None, 'fsdp') I0417 13:47:20.720303 140132447517824 maxtext_utils.py:1651] params/params/decoder/layers/self_attention/query/kernel Shape: float32[2048,16,16,128] Logical: P('embed', 'layers', 'q_heads', 'kv') Physical: ('fsdp', None, None, None) I0417 13:47:20.720329 140132447517824 maxtext_utils.py:1651] params/params/decoder/layers/self_attention/value/kernel Shape: float32[2048,16,16,128] Logical: P('embed', 'layers', 'kv_heads', 'kv_head_dim') Physical: ('fsdp', None, None, None) I0417 13:47:20.720364 140132447517824 maxtext_utils.py:1651] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: P('embed_vocab', 'vocab') Physical: ('fsdp', None) I0417 13:47:20.720398 140132447517824 maxtext_utils.py:1651] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: P('vocab', 'embed_vocab') Physical: (None, 'fsdp') I0417 13:47:21.140743 140132447517824 train.py:155] train/xent Logical: float32[8,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length'). I0417 13:47:21.140826 140132447517824 train.py:155] train/xent Physical: float32[8,2048]............................................. ('fsdp', None). I0417 13:47:21.154381 140132447517824 train.py:162] train/z_loss Logical: float32[8,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length'). I0417 13:47:21.154432 140132447517824 train.py:162] train/z_loss Physical: float32[8,2048]............................................. ('fsdp', None). W0417 13:47:21.404251 489912 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0417 13:47:21.526279 140132447517824 max_utils.py:791] Total memory size: 3.6 GB, Output size: 1.5 GB, Temp size: 2.0 GB, Argument size: 1.5 GB, Host temp size: 0.0 GB. I0417 13:47:21.526738 140132447517824 max_utils.py:194] tensorboardX not available; using no-op SummaryWriter. I0417 13:47:21.526888 140132447517824 metric_logger.py:301] number parameters: 1.104 billion W0417 13:47:22.367264 489912 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0417 13:47:22.461705 140132447517824 checkpointing.py:772] Waiting for step 0 to finish before checkpoint... I0417 13:47:22.578649 140132447517824 checkpointing.py:776] Waited 0.11696696281433105 seconds for step 0 to finish before starting checkpointing. I0417 13:47:22.579426 140132447517824 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0417 13:47:22.579626 140132447517824 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 0 I0417 13:47:22.580290 140132447517824 async_checkpointer.py:452] [process=0] Started async saving checkpoint to gs://wanglance-maxtext/linen_ckpt_main_20260417_125240/linen_main_20260417_125240_15_ocdbt_true/checkpoints/0. I0417 13:47:22.694813 140132447517824 signaling_client.py:373] Using ThreadSafeKeyValueSignalingClient I0417 13:47:22.778304 140024997742144 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260417_125240/linen_main_20260417_125240_15_ocdbt_true/checkpoints/0 I0417 13:47:22.787181 140132447517824 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array. I0417 13:47:22.787270 140132447517824 replica_slices.py:410] 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 W0417 13:47:23.000778 489912 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0417 13:47:23.010924 489912 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0417 13:47:23.028943 489912 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0417 13:47:23.043229 489912 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0417 13:47:23.051292 489912 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0417 13:47:23.056896 489912 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0417 13:47:23.062832 489912 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0417 13:47:23.066903 489912 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0417 13:47:23.454271 140132447517824 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.667542s I0417 13:47:23.454867 140132447517824 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 16.268 GiB/s (total gbytes: 12.3 GiB) (time elapsed: 758 milliseconds) (per-host) I0417 13:47:23.455008 140132447517824 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 0.758864s (batch_requests_ready=0.087176s, total_serialization_initiated=0.671177s, others=0.000510s) I0417 13:47:23.455140 140132447517824 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.759828s (all_items=0.000056s, per_item={'items': '0.00005603'}, temp_paths=0.759772) I0417 13:47:23.455956 140024930633280 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started. I0417 13:47:23.456062 140132447517824 async_checkpointer.py:561] Finished blocking save. Time taken: 0.876377s. Continuing background save to gs://wanglance-maxtext/linen_ckpt_main_20260417_125240/linen_main_20260417_125240_15_ocdbt_true/checkpoints/0. I0417 13:47:23.456278 140132447517824 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize I0417 13:47:23.456498 140132447517824 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://wanglance-maxtext/linen_ckpt_main_20260417_125240/linen_main_20260417_125240_15_ocdbt_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776433642.5794072, 'wait_for_prev_duration_secs': 4.8160552978515625e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776433642.5796475, 'checkpointer_blocking_duration_secs': 0.8765316009521484, 'get_old_steps_start_time': 1776433643.4562035, 'get_old_steps_duration_secs': 3.147125244140625e-05, 'checkpoint_manager_blocking_start_time': 1776433642.5792184, 'checkpoint_manager_blocking_duration_secs': 0.8772516250610352} I0417 13:47:23.456577 140132447517824 checkpointing.py:408] Started an asynchronous checkpoint save for step 0 I0417 13:47:23.456636 140132447517824 max_utils.py:750] Memstats: After params initialized: I0417 13:47:23.456731 140132447517824 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_0(process=0,(0,0,0,0)) I0417 13:47:23.456761 140132447517824 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_1(process=0,(1,0,0,0)) I0417 13:47:23.456783 140132447517824 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_2(process=0,(0,1,0,0)) I0417 13:47:23.456803 140132447517824 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_3(process=0,(1,1,0,0)) I0417 13:47:23.456822 140132447517824 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_4(process=0,(0,2,0,0)) I0417 13:47:23.456840 140132447517824 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_5(process=0,(1,2,0,0)) I0417 13:47:23.456859 140132447517824 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_6(process=0,(0,3,0,0)) I0417 13:47:23.456877 140132447517824 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_7(process=0,(1,3,0,0)) W0417 13:47:23.462832 489912 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0417 13:47:23.468327 489912 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0417 13:47:23.473062 489912 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0417 13:47:23.494446 140024987256384 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260417_125240/linen_main_20260417_125240_15_ocdbt_true/checkpoints/0/items I0417 13:47:23.640184 140024951604800 checkpoint.py:188] Wrote Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776433643397986168, 'commit_timestamp_nsecs': None, 'custom_metadata': {}}, json={"item_handlers": null, "metrics": {}, "performance_metrics": {}, "init_timestamp_nsecs": 1776433643397986168, "commit_timestamp_nsecs": null, "custom_metadata": {}} to gs://wanglance-maxtext/linen_ckpt_main_20260417_125240/linen_main_20260417_125240_15_ocdbt_true/checkpoints/0/_CHECKPOINT_METADATA I0417 13:47:23.640430 140025008227904 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0417 13:47:23.663785 140132447517824 metric_logger.py:196] completed step: 0, seconds: 0.934, TFLOP/s/device: 14.543, Tokens/s/device: 2192.102, total_weights: 16384, loss: 10.874, lm_loss: 10.874, perplexity: 52770.941 I0417 13:47:23.664757 140132447517824 metric_logger.py:281] To see full metrics 'tensorboard --logdir=gs://wanglance-maxtext/linen_ckpt_main_20260417_125240/linen_main_20260417_125240_15_ocdbt_true/tensorboard/' I0417 13:47:23.779923 140132447517824 metric_logger.py:196] completed step: 1, seconds: 1.200, TFLOP/s/device: 11.320, Tokens/s/device: 1706.264, total_weights: 16384, loss: 10.874, lm_loss: 10.874, perplexity: 52770.941 I0417 13:47:23.898192 140132447517824 metric_logger.py:196] completed step: 2, seconds: 0.006, TFLOP/s/device: 2377.866, Tokens/s/device: 358417.921, total_weights: 16384, loss: 10.276, lm_loss: 10.276, perplexity: 29019.973 I0417 13:47:23.970491 493442 google_auth_provider.cc:149] Using credentials at ~/.config/gcloud/application_default_credentials.json I0417 13:47:23.970539 493442 google_auth_provider.cc:156] Using OAuth2 AuthProvider I0417 13:47:24.016861 140132447517824 metric_logger.py:196] completed step: 3, seconds: 0.117, TFLOP/s/device: 116.353, Tokens/s/device: 17538.000, total_weights: 16384, loss: 9.282, lm_loss: 9.282, perplexity: 10739.224 I0417 13:47:24.021425 140132447517824 checkpointing.py:772] Waiting for step 5 to finish before checkpoint... I0417 13:47:24.253123 140132447517824 checkpointing.py:776] Waited 0.23166561126708984 seconds for step 5 to finish before starting checkpointing. I0417 13:47:24.253501 140132447517824 checkpoint_manager.py:1994] [process=0][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0417 13:47:24.728029 140024976770624 array_metadata_store.py:203] [process=0][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://wanglance-maxtext/linen_ckpt_main_20260417_125240/linen_main_20260417_125240_15_ocdbt_true/checkpoints/0/items/array_metadatas/process_0 I0417 13:47:51.031937 140024941119040 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 27.066934s (commit=26.623792s, array_metadata_write=0.443142s) I0417 13:47:51.032979 140024930633280 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 445.950 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 28 seconds) (per-host) I0417 13:47:51.033096 140024930633280 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 27.576989s. I0417 13:47:51.273786 140024930633280 checkpoint.py:228] Read Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776433643397986168, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} from gs://wanglance-maxtext/linen_ckpt_main_20260417_125240/linen_main_20260417_125240_15_ocdbt_true/checkpoints/0/_CHECKPOINT_METADATA I0417 13:47:51.452622 140024930633280 array_metadata_store.py:367] [process=0][thread=async_save] Skipped cross-host ArrayMetadata validation because only one process is found: process_index=0. I0417 13:47:51.665123 140024951604800 checkpoint.py:247] Updated Metadata={'item_handlers': {'items': 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler'}, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776433643397986168, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} to gs://wanglance-maxtext/linen_ckpt_main_20260417_125240/linen_main_20260417_125240_15_ocdbt_true/checkpoints/0/_CHECKPOINT_METADATA I0417 13:47:51.903241 140024930633280 ocdbt_utils.py:56] Param validation support for Zarr3 will be added later (b/362328389). I0417 13:47:51.903860 140024930633280 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.588409s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://wanglance-maxtext/linen_ckpt_main_20260417_125240/linen_main_20260417_125240_15_ocdbt_true/checkpoints/0/items I0417 13:47:51.904516 140024930633280 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260417_125240/linen_main_20260417_125240_15_ocdbt_true/checkpoints/0/items I0417 13:47:52.147543 140024930633280 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260417_125240/linen_main_20260417_125240_15_ocdbt_true/checkpoints/0 I0417 13:47:52.842314 140024930633280 atomicity.py:794] [process=0][thread=async_save] Finished saving checkpoint (finalized tmp dir) to `gs://wanglance-maxtext/linen_ckpt_main_20260417_125240/linen_main_20260417_125240_15_ocdbt_true/checkpoints/0`. I0417 13:47:52.842922 140024930633280 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 30.263256s. directory=gs://wanglance-maxtext/linen_ckpt_main_20260417_125240/linen_main_20260417_125240_15_ocdbt_true/checkpoints/0 I0417 13:47:52.842999 140024930633280 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 29.386894s. I0417 13:47:52.843170 140025008227904 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0417 13:47:52.843274 140025008227904 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0417 13:47:52.843344 140025008227904 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts... I0417 13:47:52.843394 140025008227904 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts. I0417 13:47:52.843581 140132447517824 checkpoint_manager.py:2006] [process=0][thread=MainThread][step=0][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=0. W0417 13:47:52.844156 140132447517824 checkpoint_manager.py:1441] Waiting for previous save to complete took 28.590665 seconds. If this number is high, consider checkpointing less frequently. I0417 13:47:52.844909 140132447517824 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 5 I0417 13:47:52.845192 140132447517824 async_checkpointer.py:452] [process=0] Started async saving checkpoint to gs://wanglance-maxtext/linen_ckpt_main_20260417_125240/linen_main_20260417_125240_15_ocdbt_true/checkpoints/5. I0417 13:47:53.024111 140132447517824 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array. I0417 13:47:53.024200 140132447517824 replica_slices.py:410] Transferring arrays to host memory with options: use_replica_parallel=True, min_slice_bytes_for_replica_parallel=None, max_replicas_for_replica_parallel=None, enable_pinned_host_transfer=False I0417 13:47:53.032015 140025008227904 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260417_125240/linen_main_20260417_125240_15_ocdbt_true/checkpoints/5 I0417 13:47:53.701440 140024987256384 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260417_125240/linen_main_20260417_125240_15_ocdbt_true/checkpoints/5/items I0417 13:48:01.291011 140132447517824 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 8.267339s I0417 13:48:01.295211 140132447517824 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 1.476 GiB/s (total gbytes: 12.3 GiB) (time elapsed: 8 seconds) (per-host) I0417 13:48:01.295276 140132447517824 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 8.361767s (batch_requests_ready=0.087359s, total_serialization_initiated=8.270339s, others=0.004069s) I0417 13:48:01.295349 140132447517824 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 8.362262s (all_items=0.000018s, per_item={'items': '0.00001788'}, temp_paths=8.362244) I0417 13:48:01.296858 140024941119040 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started. I0417 13:48:01.296954 140132447517824 async_checkpointer.py:561] Finished blocking save. Time taken: 8.452003s. Continuing background save to gs://wanglance-maxtext/linen_ckpt_main_20260417_125240/linen_main_20260417_125240_15_ocdbt_true/checkpoints/5. I0417 13:48:01.297162 140132447517824 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=5] Starting CheckpointManager Save Finalize thread=save_finalize I0417 13:48:01.297321 140024930633280 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0417 13:48:01.297423 140132447517824 standard_logger.py:34] {'step': 5, 'event_type': 'save', 'directory': 'gs://wanglance-maxtext/linen_ckpt_main_20260417_125240/linen_main_20260417_125240_15_ocdbt_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776433644.2534657, 'wait_for_prev_duration_secs': 28.590664625167847, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776433672.844932, 'checkpointer_blocking_duration_secs': 8.452143907546997, 'get_old_steps_start_time': 1776433681.2970963, 'get_old_steps_duration_secs': 2.8371810913085938e-05, 'checkpoint_manager_blocking_start_time': 1776433644.2534003, 'checkpoint_manager_blocking_duration_secs': 37.04399394989014} I0417 13:48:01.297542 140132447517824 checkpointing.py:408] Started an asynchronous checkpoint save for step 5 I0417 13:48:01.298164 140132447517824 metric_logger.py:196] completed step: 4, seconds: 0.118, TFLOP/s/device: 114.917, Tokens/s/device: 17321.583, total_weights: 16384, loss: 8.327, lm_loss: 8.327, perplexity: 4134.812 I0417 13:48:01.304911 140132447517824 metric_logger.py:196] completed step: 5, seconds: 0.119, TFLOP/s/device: 114.523, Tokens/s/device: 17262.161, total_weights: 16384, loss: 7.367, lm_loss: 7.367, perplexity: 1583.522 I0417 13:48:01.421516 140132447517824 metric_logger.py:196] completed step: 6, seconds: 37.282, TFLOP/s/device: 0.364, Tokens/s/device: 54.933, total_weights: 16384, loss: 6.459, lm_loss: 6.459, perplexity: 638.363 I0417 13:48:01.540037 140132447517824 metric_logger.py:196] completed step: 7, seconds: 0.005, TFLOP/s/device: 2487.119, Tokens/s/device: 374885.594, total_weights: 16384, loss: 5.622, lm_loss: 5.622, perplexity: 276.577 I0417 13:48:01.658178 140132447517824 metric_logger.py:196] completed step: 8, seconds: 0.118, TFLOP/s/device: 115.556, Tokens/s/device: 17417.928, total_weights: 16384, loss: 4.864, lm_loss: 4.864, perplexity: 129.523 I0417 13:48:01.776174 140132447517824 checkpointing.py:772] Waiting for step 9 to finish before checkpoint... I0417 13:48:01.776995 140132447517824 checkpointing.py:776] Waited 0.0008437633514404297 seconds for step 9 to finish before starting checkpointing. I0417 13:48:01.777256 140132447517824 checkpoint_manager.py:1994] [process=0][thread=MainThread][step=5][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0417 13:48:02.056262 140024976770624 array_metadata_store.py:203] [process=0][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://wanglance-maxtext/linen_ckpt_main_20260417_125240/linen_main_20260417_125240_15_ocdbt_true/checkpoints/5/items/array_metadatas/process_0 I0417 13:48:33.773284 140024987256384 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 32.477372s (commit=32.035709s, array_metadata_write=0.441663s) I0417 13:48:33.774399 140024941119040 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 309.416 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 40 seconds) (per-host) I0417 13:48:33.774528 140024941119040 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 32.477528s. I0417 13:48:34.185772 140024941119040 array_metadata_store.py:367] [process=0][thread=async_save] Skipped cross-host ArrayMetadata validation because only one process is found: process_index=0. I0417 13:48:34.618318 140024941119040 ocdbt_utils.py:56] Param validation support for Zarr3 will be added later (b/362328389). I0417 13:48:34.618492 140024941119040 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.571395s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://wanglance-maxtext/linen_ckpt_main_20260417_125240/linen_main_20260417_125240_15_ocdbt_true/checkpoints/5/items I0417 13:48:34.619058 140024941119040 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260417_125240/linen_main_20260417_125240_15_ocdbt_true/checkpoints/5/items I0417 13:48:34.861292 140024941119040 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260417_125240/linen_main_20260417_125240_15_ocdbt_true/checkpoints/5 I0417 13:48:35.569326 140024941119040 atomicity.py:794] [process=0][thread=async_save] Finished saving checkpoint (finalized tmp dir) to `gs://wanglance-maxtext/linen_ckpt_main_20260417_125240/linen_main_20260417_125240_15_ocdbt_true/checkpoints/5`. I0417 13:48:35.569960 140024941119040 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 42.725014s. directory=gs://wanglance-maxtext/linen_ckpt_main_20260417_125240/linen_main_20260417_125240_15_ocdbt_true/checkpoints/5 I0417 13:48:35.570031 140024941119040 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 34.273034s. I0417 13:48:35.570224 140024930633280 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0417 13:48:35.570330 140024930633280 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0417 13:48:35.570394 140024930633280 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=5] CheckpointManager Save Finalize is syncing with other hosts... I0417 13:48:35.570434 140024930633280 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=5] CheckpointManager Save Finalize is done on all hosts. I0417 13:48:35.570584 140132447517824 checkpoint_manager.py:2006] [process=0][thread=MainThread][step=5][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=5. W0417 13:48:35.570698 140132447517824 checkpoint_manager.py:1441] Waiting for previous save to complete took 33.793450 seconds. If this number is high, consider checkpointing less frequently. I0417 13:48:35.570922 140132447517824 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 9 I0417 13:48:35.571208 140132447517824 async_checkpointer.py:452] [process=0] Started async saving checkpoint to gs://wanglance-maxtext/linen_ckpt_main_20260417_125240/linen_main_20260417_125240_15_ocdbt_true/checkpoints/9. I0417 13:48:35.746307 140132447517824 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array. I0417 13:48:35.746397 140132447517824 replica_slices.py:410] Transferring arrays to host memory with options: use_replica_parallel=True, min_slice_bytes_for_replica_parallel=None, max_replicas_for_replica_parallel=None, enable_pinned_host_transfer=False I0417 13:48:35.754538 140024930633280 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260417_125240/linen_main_20260417_125240_15_ocdbt_true/checkpoints/9 I0417 13:48:36.428078 140024880301632 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260417_125240/linen_main_20260417_125240_15_ocdbt_true/checkpoints/9/items I0417 13:48:44.954917 140132447517824 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 9.208985s I0417 13:48:44.959411 140132447517824 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 1.327 GiB/s (total gbytes: 12.3 GiB) (time elapsed: 9 seconds) (per-host) I0417 13:48:44.959482 140132447517824 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 9.297569s (batch_requests_ready=0.080640s, total_serialization_initiated=9.212567s, others=0.004362s) I0417 13:48:44.959588 140132447517824 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 9.298144s (all_items=0.000019s, per_item={'items': '0.00001884'}, temp_paths=9.298125) I0417 13:48:44.960994 140024976770624 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started. I0417 13:48:44.961100 140132447517824 async_checkpointer.py:561] Finished blocking save. Time taken: 9.390134s. Continuing background save to gs://wanglance-maxtext/linen_ckpt_main_20260417_125240/linen_main_20260417_125240_15_ocdbt_true/checkpoints/9. I0417 13:48:44.961287 140132447517824 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize I0417 13:48:44.961457 140024859330112 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0417 13:48:44.961541 140132447517824 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://wanglance-maxtext/linen_ckpt_main_20260417_125240/linen_main_20260417_125240_15_ocdbt_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776433681.7772317, 'wait_for_prev_duration_secs': 33.79344964027405, 'time_between_consecutive_saves_sec': 8.933823108673096, 'checkpointer_blocking_start_time': 1776433715.5709445, 'checkpointer_blocking_duration_secs': 9.390267372131348, 'get_old_steps_start_time': 1776433724.9612253, 'get_old_steps_duration_secs': 2.5033950805664062e-05, 'checkpoint_manager_blocking_start_time': 1776433681.7771983, 'checkpoint_manager_blocking_duration_secs': 43.18431830406189} I0417 13:48:44.961673 140132447517824 checkpointing.py:408] Started an asynchronous checkpoint save for step 9 I0417 13:48:44.961704 140132447517824 checkpoint_manager.py:1994] [process=0][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0417 13:48:45.704395 140025008227904 array_metadata_store.py:203] [process=0][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://wanglance-maxtext/linen_ckpt_main_20260417_125240/linen_main_20260417_125240_15_ocdbt_true/checkpoints/9/items/array_metadatas/process_0 I0417 13:49:28.366680 140024987256384 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 43.406708s (commit=42.981242s, array_metadata_write=0.425466s) I0417 13:49:28.367668 140024976770624 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 239.761 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 52 seconds) (per-host) I0417 13:49:28.367773 140024976770624 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 43.406624s. I0417 13:49:28.828294 140024976770624 array_metadata_store.py:367] [process=0][thread=async_save] Skipped cross-host ArrayMetadata validation because only one process is found: process_index=0. I0417 13:49:29.325393 140024976770624 ocdbt_utils.py:56] Param validation support for Zarr3 will be added later (b/362328389). I0417 13:49:29.325566 140024976770624 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.639780s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://wanglance-maxtext/linen_ckpt_main_20260417_125240/linen_main_20260417_125240_15_ocdbt_true/checkpoints/9/items I0417 13:49:29.326203 140024976770624 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260417_125240/linen_main_20260417_125240_15_ocdbt_true/checkpoints/9/items I0417 13:49:29.555774 140024976770624 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260417_125240/linen_main_20260417_125240_15_ocdbt_true/checkpoints/9 I0417 13:49:30.256563 140024976770624 atomicity.py:794] [process=0][thread=async_save] Finished saving checkpoint (finalized tmp dir) to `gs://wanglance-maxtext/linen_ckpt_main_20260417_125240/linen_main_20260417_125240_15_ocdbt_true/checkpoints/9`. I0417 13:49:30.257189 140024976770624 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 54.686229s. directory=gs://wanglance-maxtext/linen_ckpt_main_20260417_125240/linen_main_20260417_125240_15_ocdbt_true/checkpoints/9 I0417 13:49:30.257258 140024976770624 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 45.296114s. I0417 13:49:30.257368 140024859330112 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0417 13:49:30.257423 140024859330112 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0417 13:49:30.257487 140024859330112 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts... I0417 13:49:30.257530 140024859330112 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts. I0417 13:49:30.257812 140132447517824 checkpoint_manager.py:2006] [process=0][thread=MainThread][step=9][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=9. I0417 13:49:30.258375 140132447517824 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0417 13:49:30.259144 140132447517824 metric_logger.py:196] completed step: 9, seconds: 0.119, TFLOP/s/device: 114.573, Tokens/s/device: 17269.730, total_weights: 16384, loss: 4.189, lm_loss: 4.189, perplexity: 65.979 Per train step: Total TFLOPs: 13.59 split as 93.93% learnable weight flops and 6.07% attention flops