main (0ba93e21a) vs feat/nnx-post-train-fixes (d8cde296b)| Metric | main 0ba93e21a | feat/nnx-post-train-fixes d8cde296b | Diff (feat/nnx-post-train-fixes − main) |
|---|---|---|---|
| Parameters | 1.104 billion | 1.104 billion | — |
| Final loss | 5.9150 | 5.9150 | 0 |
| TFLOP/s | 114.582 | 114.923 | +0.341 |
| Tok/s | 17271.0 | 17322.5 | +51.421 |
| Avg s/step | 2.449 | 2.458 | +0.009 |
| Memory % | 5.09 | 5.09 | 0 |
| JAX | 0.9.2 | 0.9.2 | — |
Diff = branch value − main value. Green = branch improved. Red = branch regressed.
2026-04-16 13:11:50.968908: 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) I0416 13:11:51.084475 126548021759104 max_utils.py:238] Skipping jax distributed system due to skip_jax_distributed_system=True flag. I0416 13:12:47.494298 126548021759104 max_utils.py:800] System Information: Jax Version: 0.9.2 I0416 13:12:47.494412 126548021759104 max_utils.py:801] System Information: Jaxlib Version: 0.9.2 I0416 13:12:47.494446 126548021759104 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 I0416 13:12:47.494473 126548021759104 train_utils.py:334] WARNING: 'dataset_path' might be pointing your local file system I0416 13:12:47.494498 126548021759104 train_utils.py:347] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing. I0416 13:12:47.494580 126548021759104 train.py:683] [DECOUPLED NO-OP] skipping cloud diagnostics wrapper. W0416 13:12:47.587602 3856933 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0416 13:12:47.992786 126548021759104 maxtext_utils.py:1517] Num_devices: 8, shape (1, 1, 1, 8, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0416 13:12:47.993119 126548021759104 checkpointing.py:677] Setting up checkpoint logger... I0416 13:12:47.993165 126548021759104 checkpointing.py:233] Creating checkpoint manager with ocdbt=True and zarr3=True I0416 13:12:47.993205 126548021759104 pytree_checkpoint_handler.py:577] save_device_host_concurrent_bytes=None I0416 13:12:47.993834 126548021759104 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 0x7317ab9db740>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0416 13:12:50.409807 126548021759104 checkpointing.py:265] Enabling policy for fixed interval checkpointing. I0416 13:12:50.409972 126548021759104 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 0x7311305cbaa0>}, handler_registry=None I0416 13:12:50.410234 126548021759104 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7311305cbaa0>` 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`. I0416 13:12:50.410278 126548021759104 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x73112ff67b60>` 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`. I0416 13:12:50.410307 126548021759104 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 0x7311305cbaa0>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7311305cbaa0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x73112ff67b60>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x73112ff67b60>}). I0416 13:12:50.410598 126548021759104 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.28 I0416 13:12:50.410649 126548021759104 async_checkpointer.py:177] [process=0][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>.<lambda> at 0x73112f131080> timeout: 600 secs and primary_host=0 for async checkpoint writes I0416 13:12:50.542996 126548021759104 checkpoint_manager.py:1788] Found 0 checkpoint steps in gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_11_optimizer_offload_false/checkpoints I0416 13:12:50.543248 126548021759104 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=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), root_directory=gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_11_optimizer_offload_false/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x73112f13dfd0> I0416 13:12:50.543324 126548021759104 checkpointing.py:301] Checkpoint manager created! W0416 13:12:50.559501 3856933 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0416 13:12:50.738502 126548021759104 nnx_wrappers.py:437] Unknown Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0416 13:12:50.738717 126548021759104 nnx_wrappers.py:437] Unknown Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0416 13:12:50.829035 126548021759104 attentions.py:1088] attentions/inputs_q Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0416 13:12:50.829126 126548021759104 attentions.py:1088] attentions/inputs_q Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0416 13:12:50.843394 126548021759104 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0416 13:12:50.843441 126548021759104 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0416 13:12:50.864184 126548021759104 attentions.py:1154] attentions/query Logical: bfloat16[8,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0416 13:12:50.864236 126548021759104 attentions.py:1154] attentions/query Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0416 13:12:50.878452 126548021759104 attentions.py:1155] attentions/key Logical: bfloat16[8,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0416 13:12:50.878502 126548021759104 attentions.py:1155] attentions/key Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0416 13:12:50.892747 126548021759104 attentions.py:1156] attentions/value Logical: bfloat16[8,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0416 13:12:50.892796 126548021759104 attentions.py:1156] attentions/value Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0416 13:12:50.914760 126548021759104 attentions.py:1197] attentions/out Logical: bfloat16[8,2048,16,128]..................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv'). I0416 13:12:50.914815 126548021759104 attentions.py:1197] attentions/out Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0416 13:12:50.933000 126548021759104 linears.py:525] linears/x Logical: bfloat16[8,2048,7168]....................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0416 13:12:50.933061 126548021759104 linears.py:525] linears/x Physical: bfloat16[8,2048,7168]....................................... ('fsdp', None, None). I0416 13:12:51.114853 126548021759104 checkpointing.py:577] checkpoint manager exists so trying to load this run's existing checkpoint I0416 13:12:51.114946 126548021759104 checkpointing.py:665] No existing checkpoints found, not restoring checkpoint. W0416 13:12:51.165861 3856933 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 I0416 13:12:51.543303 126548021759104 maxtext_utils.py:1620] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: P('norm',) Physical: (None,) I0416 13:12:51.543411 126548021759104 maxtext_utils.py:1620] params/params/decoder/layers/mlp/wi_0/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0416 13:12:51.543451 126548021759104 maxtext_utils.py:1620] params/params/decoder/layers/mlp/wi_1/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0416 13:12:51.543493 126548021759104 maxtext_utils.py:1620] params/params/decoder/layers/mlp/wo/kernel Shape: float32[7168,16,2048] Logical: P('mlp', 'layers', 'embed') Physical: (None, None, 'fsdp') I0416 13:12:51.543532 126548021759104 maxtext_utils.py:1620] params/params/decoder/layers/post_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0416 13:12:51.543559 126548021759104 maxtext_utils.py:1620] params/params/decoder/layers/pre_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0416 13:12:51.543606 126548021759104 maxtext_utils.py:1620] 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) I0416 13:12:51.543646 126548021759104 maxtext_utils.py:1620] 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') I0416 13:12:51.543674 126548021759104 maxtext_utils.py:1620] 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) I0416 13:12:51.543700 126548021759104 maxtext_utils.py:1620] 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) I0416 13:12:51.543735 126548021759104 maxtext_utils.py:1620] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: P('embed_vocab', 'vocab') Physical: ('fsdp', None) I0416 13:12:51.543768 126548021759104 maxtext_utils.py:1620] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: P('vocab', 'embed_vocab') Physical: (None, 'fsdp') I0416 13:12:51.962566 126548021759104 train.py:155] train/xent Logical: float32[8,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length'). I0416 13:12:51.962648 126548021759104 train.py:155] train/xent Physical: float32[8,2048]............................................. ('fsdp', None). I0416 13:12:51.976177 126548021759104 train.py:162] train/z_loss Logical: float32[8,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length'). I0416 13:12:51.976229 126548021759104 train.py:162] train/z_loss Physical: float32[8,2048]............................................. ('fsdp', None). W0416 13:12:52.578064 3856933 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0416 13:12:52.703635 126548021759104 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. I0416 13:12:52.704116 126548021759104 max_utils.py:194] tensorboardX not available; using no-op SummaryWriter. I0416 13:12:52.704252 126548021759104 metric_logger.py:289] number parameters: 1.104 billion W0416 13:12:53.537066 3856933 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0416 13:12:53.633864 126548021759104 checkpointing.py:772] Waiting for step 0 to finish before checkpoint... I0416 13:12:53.750961 126548021759104 checkpointing.py:776] Waited 0.11708474159240723 seconds for step 0 to finish before starting checkpointing. I0416 13:12:53.751420 126548021759104 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0416 13:12:53.751574 126548021759104 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 0 I0416 13:12:53.751944 126548021759104 async_checkpointer.py:452] [process=0] Started async saving checkpoint to gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_11_optimizer_offload_false/checkpoints/0. I0416 13:12:53.844645 126548021759104 signaling_client.py:373] Using ThreadSafeKeyValueSignalingClient I0416 13:12:53.941276 126548021759104 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array. I0416 13:12:53.941420 126440561444416 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_11_optimizer_offload_false/checkpoints/0 I0416 13:12:53.941825 126548021759104 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 W0416 13:12:54.158121 3856933 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 13:12:54.169186 3856933 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 13:12:54.175522 3856933 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 13:12:54.185696 3856933 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 13:12:54.199885 3856933 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 13:12:54.213572 3856933 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 13:12:54.219728 3856933 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 13:12:54.224926 3856933 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0416 13:12:54.651411 126440550958656 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_11_optimizer_offload_false/checkpoints/0/items I0416 13:12:54.791453 126440519501376 checkpoint.py:188] Wrote Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776345174544131959, 'commit_timestamp_nsecs': None, 'custom_metadata': {}}, json={"item_handlers": null, "metrics": {}, "performance_metrics": {}, "init_timestamp_nsecs": 1776345174544131959, "commit_timestamp_nsecs": null, "custom_metadata": {}} to gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_11_optimizer_offload_false/checkpoints/0/_CHECKPOINT_METADATA I0416 13:12:55.044742 126548021759104 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 1.104360s I0416 13:12:55.045169 126548021759104 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 10.291 GiB/s (total gbytes: 12.3 GiB) (time elapsed: a second) (per-host) I0416 13:12:55.045220 126548021759104 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 1.199318s (batch_requests_ready=0.090783s, total_serialization_initiated=1.108207s, others=0.000329s) I0416 13:12:55.045278 126548021759104 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 1.200151s (all_items=0.000020s, per_item={'items': '0.00001955'}, temp_paths=1.200132) I0416 13:12:55.045988 126440498529856 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started. I0416 13:12:55.046092 126548021759104 async_checkpointer.py:561] Finished blocking save. Time taken: 1.294471s. Continuing background save to gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_11_optimizer_offload_false/checkpoints/0. I0416 13:12:55.046275 126548021759104 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize I0416 13:12:55.046447 126440571930176 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0416 13:12:55.046550 126548021759104 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_11_optimizer_offload_false/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776345173.751408, 'wait_for_prev_duration_secs': 4.0531158447265625e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776345173.7516022, 'checkpointer_blocking_duration_secs': 1.2946014404296875, 'get_old_steps_start_time': 1776345175.0462215, 'get_old_steps_duration_secs': 2.0503997802734375e-05, 'checkpoint_manager_blocking_start_time': 1776345173.7513258, 'checkpoint_manager_blocking_duration_secs': 1.2951982021331787} I0416 13:12:55.046658 126548021759104 checkpointing.py:408] Started an asynchronous checkpoint save for step 0 I0416 13:12:55.046696 126548021759104 max_utils.py:750] Memstats: After params initialized: I0416 13:12:55.046770 126548021759104 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_0(process=0,(0,0,0,0)) I0416 13:12:55.046796 126548021759104 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_1(process=0,(1,0,0,0)) I0416 13:12:55.046815 126548021759104 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_2(process=0,(0,1,0,0)) I0416 13:12:55.046832 126548021759104 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_3(process=0,(1,1,0,0)) I0416 13:12:55.046855 126548021759104 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_4(process=0,(0,2,0,0)) I0416 13:12:55.046873 126548021759104 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_5(process=0,(1,2,0,0)) I0416 13:12:55.046890 126548021759104 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_6(process=0,(0,3,0,0)) I0416 13:12:55.046905 126548021759104 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_7(process=0,(1,3,0,0)) W0416 13:12:55.052506 3856933 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 13:12:55.058356 3856933 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 13:12:55.063276 3856933 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0416 13:12:55.165989 3859099 google_auth_provider.cc:149] Using credentials at ~/.config/gcloud/application_default_credentials.json I0416 13:12:55.166064 3859099 google_auth_provider.cc:156] Using OAuth2 AuthProvider I0416 13:12:55.408700 126548021759104 metric_logger.py:185] completed step: 0, seconds: 0.929, TFLOP/s/device: 14.623, Tokens/s/device: 2204.179, total_weights: 16384, loss: 10.874 I0416 13:12:55.409728 126548021759104 metric_logger.py:269] To see full metrics 'tensorboard --logdir=gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_11_optimizer_offload_false/tensorboard/' I0416 13:12:55.524522 126548021759104 metric_logger.py:185] completed step: 1, seconds: 1.773, TFLOP/s/device: 7.664, Tokens/s/device: 1155.244, total_weights: 16384, loss: 10.874 I0416 13:12:55.643238 126548021759104 metric_logger.py:185] completed step: 2, seconds: 0.007, TFLOP/s/device: 1975.448, Tokens/s/device: 297760.977, total_weights: 16384, loss: 9.775 I0416 13:12:55.762150 126548021759104 metric_logger.py:185] completed step: 3, seconds: 0.117, TFLOP/s/device: 116.600, Tokens/s/device: 17575.175, total_weights: 16384, loss: 8.838 I0416 13:12:55.996293 126440540472896 array_metadata_store.py:203] [process=0][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_11_optimizer_offload_false/checkpoints/0/items/array_metadatas/process_0 I0416 13:13:15.421543 126548021759104 metric_logger.py:185] completed step: 4, seconds: 0.119, TFLOP/s/device: 114.595, Tokens/s/device: 17273.080, total_weights: 16384, loss: 7.965 I0416 13:13:15.428908 126548021759104 metric_logger.py:185] completed step: 5, seconds: 0.119, TFLOP/s/device: 114.062, Tokens/s/device: 17192.602, total_weights: 16384, loss: 7.233 I0416 13:13:15.545884 126548021759104 metric_logger.py:185] completed step: 6, seconds: 19.660, TFLOP/s/device: 0.691, Tokens/s/device: 104.170, total_weights: 16384, loss: 6.676 I0416 13:13:15.664312 126548021759104 metric_logger.py:185] completed step: 7, seconds: 0.005, TFLOP/s/device: 2903.853, Tokens/s/device: 437700.363, total_weights: 16384, loss: 6.293 I0416 13:13:15.782961 126548021759104 metric_logger.py:185] completed step: 8, seconds: 0.118, TFLOP/s/device: 115.312, Tokens/s/device: 17381.120, total_weights: 16384, loss: 6.055 I0416 13:13:15.901223 126548021759104 checkpointing.py:772] Waiting for step 9 to finish before checkpoint... I0416 13:13:15.902075 126548021759104 checkpointing.py:776] Waited 0.0008802413940429688 seconds for step 9 to finish before starting checkpointing. I0416 13:13:15.902331 126548021759104 checkpoint_manager.py:1994] [process=0][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0416 13:13:24.921638 126440509015616 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 29.758655s (commit=29.275246s, array_metadata_write=0.483409s) I0416 13:13:24.922673 126440498529856 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 406.632 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 31 seconds) (per-host) I0416 13:13:24.922777 126440498529856 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 29.876638s. I0416 13:13:25.169290 126440498529856 checkpoint.py:228] Read Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776345174544131959, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} from gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_11_optimizer_offload_false/checkpoints/0/_CHECKPOINT_METADATA I0416 13:13:25.368488 126440498529856 array_metadata_store.py:367] [process=0][thread=async_save] Skipped cross-host ArrayMetadata validation because only one process is found: process_index=0. I0416 13:13:25.559220 126440519501376 checkpoint.py:247] Updated Metadata={'item_handlers': {'items': 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler'}, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776345174544131959, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} to gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_11_optimizer_offload_false/checkpoints/0/_CHECKPOINT_METADATA I0416 13:13:25.818826 126440498529856 ocdbt_utils.py:56] Param validation support for Zarr3 will be added later (b/362328389). I0416 13:13:25.819464 126440498529856 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.601813s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_11_optimizer_offload_false/checkpoints/0/items I0416 13:13:25.820118 126440498529856 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_11_optimizer_offload_false/checkpoints/0/items I0416 13:13:26.076198 126440498529856 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_11_optimizer_offload_false/checkpoints/0 I0416 13:13:26.765753 126440498529856 atomicity.py:794] [process=0][thread=async_save] Finished saving checkpoint (finalized tmp dir) to `gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_11_optimizer_offload_false/checkpoints/0`. I0416 13:13:26.766386 126440498529856 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 33.014771s. directory=gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_11_optimizer_offload_false/checkpoints/0 I0416 13:13:26.766461 126440498529856 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 31.720323s. I0416 13:13:26.766631 126440571930176 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0416 13:13:26.766747 126440571930176 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0416 13:13:26.766805 126440571930176 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts... I0416 13:13:26.766854 126440571930176 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts. I0416 13:13:26.766947 126548021759104 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. W0416 13:13:26.767018 126548021759104 checkpoint_manager.py:1441] Waiting for previous save to complete took 10.864694 seconds. If this number is high, consider checkpointing less frequently. I0416 13:13:26.767726 126548021759104 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 9 I0416 13:13:26.768013 126548021759104 async_checkpointer.py:452] [process=0] Started async saving checkpoint to gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_11_optimizer_offload_false/checkpoints/9. I0416 13:13:26.946711 126548021759104 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array. I0416 13:13:26.946802 126548021759104 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 I0416 13:13:26.950309 126440571930176 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_11_optimizer_offload_false/checkpoints/9 I0416 13:13:27.684839 126440561444416 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_11_optimizer_offload_false/checkpoints/9/items I0416 13:13:37.131621 126548021759104 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 10.185308s I0416 13:13:37.135937 126548021759104 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 1.201 GiB/s (total gbytes: 12.3 GiB) (time elapsed: 10 seconds) (per-host) I0416 13:13:37.135996 126548021759104 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 10.276965s (batch_requests_ready=0.083918s, total_serialization_initiated=10.188843s, others=0.004205s) I0416 13:13:37.136086 126548021759104 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 10.277744s (all_items=0.000018s, per_item={'items': '0.00001764'}, temp_paths=10.277726) I0416 13:13:37.137654 126440571930176 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started. I0416 13:13:37.137758 126548021759104 async_checkpointer.py:561] Finished blocking save. Time taken: 10.369981s. Continuing background save to gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_11_optimizer_offload_false/checkpoints/9. I0416 13:13:37.137994 126548021759104 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize I0416 13:13:37.138154 126440498529856 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0416 13:13:37.138244 126548021759104 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_11_optimizer_offload_false/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776345195.9023, 'wait_for_prev_duration_secs': 10.864693880081177, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776345206.7677498, 'checkpointer_blocking_duration_secs': 10.370151281356812, 'get_old_steps_start_time': 1776345217.1379209, 'get_old_steps_duration_secs': 3.170967102050781e-05, 'checkpoint_manager_blocking_start_time': 1776345195.9022593, 'checkpoint_manager_blocking_duration_secs': 21.235954999923706} I0416 13:13:37.138331 126548021759104 checkpointing.py:408] Started an asynchronous checkpoint save for step 9 I0416 13:13:37.138365 126548021759104 checkpoint_manager.py:1994] [process=0][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0416 13:13:37.899403 126440540472896 array_metadata_store.py:203] [process=0][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_11_optimizer_offload_false/checkpoints/9/items/array_metadatas/process_0 I0416 13:14:11.488225 126440561444416 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 34.351403s (commit=33.904584s, array_metadata_write=0.446819s) I0416 13:14:11.489335 126440571930176 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 283.144 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 44 seconds) (per-host) I0416 13:14:11.489448 126440571930176 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 34.351633s. I0416 13:14:11.933846 126440571930176 array_metadata_store.py:367] [process=0][thread=async_save] Skipped cross-host ArrayMetadata validation because only one process is found: process_index=0. I0416 13:14:12.341196 126440571930176 ocdbt_utils.py:56] Param validation support for Zarr3 will be added later (b/362328389). I0416 13:14:12.341892 126440571930176 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.556726s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_11_optimizer_offload_false/checkpoints/9/items I0416 13:14:12.342571 126440571930176 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_11_optimizer_offload_false/checkpoints/9/items I0416 13:14:12.580826 126440571930176 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_11_optimizer_offload_false/checkpoints/9 I0416 13:14:13.296232 126440571930176 atomicity.py:794] [process=0][thread=async_save] Finished saving checkpoint (finalized tmp dir) to `gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_11_optimizer_offload_false/checkpoints/9`. I0416 13:14:13.297440 126440571930176 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 46.529675s. directory=gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_11_optimizer_offload_false/checkpoints/9 I0416 13:14:13.297522 126440571930176 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 36.159710s. I0416 13:14:13.297660 126440498529856 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0416 13:14:13.297710 126440498529856 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0416 13:14:13.297765 126440498529856 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts... I0416 13:14:13.297804 126440498529856 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts. I0416 13:14:13.298063 126548021759104 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. I0416 13:14:13.298200 126548021759104 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0416 13:14:13.298826 126548021759104 metric_logger.py:185] completed step: 9, seconds: 0.119, TFLOP/s/device: 114.582, Tokens/s/device: 17271.041, total_weights: 16384, loss: 5.915 Per train step: Total TFLOPs: 13.59 split as 93.93% learnable weight flops and 6.07% attention flops
2026-04-16 18:51:38.867253: 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) I0416 18:51:38.983134 137270208547968 max_utils.py:238] Skipping jax distributed system due to skip_jax_distributed_system=True flag. I0416 18:52:10.312076 137270208547968 max_utils.py:800] System Information: Jax Version: 0.9.2 I0416 18:52:10.312214 137270208547968 max_utils.py:801] System Information: Jaxlib Version: 0.9.2 I0416 18:52:10.312251 137270208547968 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 I0416 18:52:10.312278 137270208547968 train_utils.py:364] WARNING: 'dataset_path' might be pointing your local file system I0416 18:52:10.312301 137270208547968 train_utils.py:377] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing. I0416 18:52:10.312440 137270208547968 train.py:811] [DECOUPLED NO-OP] skipping cloud diagnostics wrapper. W0416 18:52:10.407884 4006445 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0416 18:52:10.816525 137270208547968 maxtext_utils.py:1687] Num_devices: 8, shape (1, 1, 1, 8, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0416 18:52:10.816705 137270208547968 maxtext_utils.py:1687] Num_devices: 8, shape (1, 1, 1, 8, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0416 18:52:10.816917 137270208547968 checkpointing.py:688] Setting up checkpoint logger... I0416 18:52:10.816961 137270208547968 checkpointing.py:234] Creating checkpoint manager with ocdbt=True and zarr3=True I0416 18:52:10.817013 137270208547968 pytree_checkpoint_handler.py:577] save_device_host_concurrent_bytes=None I0416 18:52:10.817673 137270208547968 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 0x7cd81f1f8bf0>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0416 18:52:13.197435 137270208547968 checkpointing.py:266] Enabling policy for fixed interval checkpointing. I0416 18:52:13.197705 137270208547968 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 0x7cd37e3f2990>}, handler_registry=None I0416 18:52:13.198027 137270208547968 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7cd37e3f2990>` 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`. I0416 18:52:13.198086 137270208547968 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7cd1c7d3d8e0>` 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`. I0416 18:52:13.198116 137270208547968 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 0x7cd37e3f2990>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7cd37e3f2990>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7cd1c7d3d8e0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7cd1c7d3d8e0>}). I0416 18:52:13.198441 137270208547968 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.28 I0416 18:52:13.198493 137270208547968 async_checkpointer.py:177] [process=0][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>.<lambda> at 0x7cd1c7d35ee0> timeout: 600 secs and primary_host=0 for async checkpoint writes I0416 18:52:13.320204 137270208547968 checkpoint_manager.py:1788] Found 0 checkpoint steps in gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_11_optimizer_offload_false/checkpoints I0416 18:52:13.320456 137270208547968 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=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), root_directory=gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_11_optimizer_offload_false/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7cd1c87a0b00> I0416 18:52:13.320538 137270208547968 checkpointing.py:302] Checkpoint manager created! W0416 18:52:13.337861 4006445 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0416 18:52:13.543485 137270208547968 nnx_wrappers.py:437] Unknown Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0416 18:52:13.543607 137270208547968 nnx_wrappers.py:437] Unknown Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0416 18:52:13.635514 137270208547968 attentions.py:1088] attentions/inputs_q Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0416 18:52:13.635602 137270208547968 attentions.py:1088] attentions/inputs_q Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0416 18:52:13.649970 137270208547968 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0416 18:52:13.650023 137270208547968 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0416 18:52:13.671502 137270208547968 attentions.py:1154] attentions/query Logical: bfloat16[8,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0416 18:52:13.671564 137270208547968 attentions.py:1154] attentions/query Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0416 18:52:13.685878 137270208547968 attentions.py:1155] attentions/key Logical: bfloat16[8,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0416 18:52:13.685930 137270208547968 attentions.py:1155] attentions/key Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0416 18:52:13.700276 137270208547968 attentions.py:1156] attentions/value Logical: bfloat16[8,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0416 18:52:13.700327 137270208547968 attentions.py:1156] attentions/value Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0416 18:52:13.722232 137270208547968 attentions.py:1197] attentions/out Logical: bfloat16[8,2048,16,128]..................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv'). I0416 18:52:13.722292 137270208547968 attentions.py:1197] attentions/out Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0416 18:52:13.740359 137270208547968 linears.py:525] linears/x Logical: bfloat16[8,2048,7168]....................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0416 18:52:13.740417 137270208547968 linears.py:525] linears/x Physical: bfloat16[8,2048,7168]....................................... ('fsdp', None, None). I0416 18:52:13.929517 137270208547968 checkpointing.py:578] checkpoint manager exists so trying to load this run's existing checkpoint I0416 18:52:13.929619 137270208547968 checkpointing.py:676] No existing checkpoints found, not restoring checkpoint. W0416 18:52:13.981130 4006445 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 I0416 18:52:14.359088 137270208547968 maxtext_utils.py:1796] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: P('norm',) Physical: (None,) I0416 18:52:14.359201 137270208547968 maxtext_utils.py:1796] params/params/decoder/layers/mlp/wi_0/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0416 18:52:14.359241 137270208547968 maxtext_utils.py:1796] params/params/decoder/layers/mlp/wi_1/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0416 18:52:14.359284 137270208547968 maxtext_utils.py:1796] params/params/decoder/layers/mlp/wo/kernel Shape: float32[7168,16,2048] Logical: P('mlp', 'layers', 'embed') Physical: (None, None, 'fsdp') I0416 18:52:14.359323 137270208547968 maxtext_utils.py:1796] params/params/decoder/layers/post_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0416 18:52:14.359351 137270208547968 maxtext_utils.py:1796] params/params/decoder/layers/pre_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0416 18:52:14.359390 137270208547968 maxtext_utils.py:1796] params/params/decoder/layers/self_attention/key/kernel Shape: float32[2048,16,16,128] Logical: P('embed', 'layers', 'kv_heads', 'kv_head_dim') Physical: ('fsdp', None, None, None) I0416 18:52:14.359427 137270208547968 maxtext_utils.py:1796] params/params/decoder/layers/self_attention/out/kernel Shape: float32[16,16,128,2048] Logical: P('heads', 'layers', 'kv', 'embed') Physical: (None, None, None, 'fsdp') I0416 18:52:14.359454 137270208547968 maxtext_utils.py:1796] params/params/decoder/layers/self_attention/query/kernel Shape: float32[2048,16,16,128] Logical: P('embed', 'layers', 'q_heads', 'kv') Physical: ('fsdp', None, None, None) I0416 18:52:14.359481 137270208547968 maxtext_utils.py:1796] params/params/decoder/layers/self_attention/value/kernel Shape: float32[2048,16,16,128] Logical: P('embed', 'layers', 'kv_heads', 'kv_head_dim') Physical: ('fsdp', None, None, None) I0416 18:52:14.359515 137270208547968 maxtext_utils.py:1796] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: P('embed_vocab', 'vocab') Physical: ('fsdp', None) I0416 18:52:14.359548 137270208547968 maxtext_utils.py:1796] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: P('vocab', 'embed_vocab') Physical: (None, 'fsdp') I0416 18:52:14.783499 137270208547968 train.py:157] train/xent Logical: float32[8,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length'). I0416 18:52:14.783592 137270208547968 train.py:157] train/xent Physical: float32[8,2048]............................................. ('fsdp', None). I0416 18:52:14.797493 137270208547968 train.py:164] train/z_loss Logical: float32[8,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length'). I0416 18:52:14.797545 137270208547968 train.py:164] train/z_loss Physical: float32[8,2048]............................................. ('fsdp', None). W0416 18:52:15.458054 4006445 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0416 18:52:15.585552 137270208547968 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. I0416 18:52:15.586143 137270208547968 max_utils.py:194] tensorboardX not available; using no-op SummaryWriter. I0416 18:52:15.586292 137270208547968 metric_logger.py:289] number parameters: 1.104 billion W0416 18:52:16.467305 4006445 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0416 18:52:16.565726 137270208547968 checkpointing.py:794] Waiting for step 0 to finish before checkpoint... I0416 18:52:16.683308 137270208547968 checkpointing.py:798] Waited 0.11756658554077148 seconds for step 0 to finish before starting checkpointing. I0416 18:52:16.683798 137270208547968 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0416 18:52:16.683954 137270208547968 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 0 I0416 18:52:16.684381 137270208547968 async_checkpointer.py:452] [process=0] Started async saving checkpoint to gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_11_optimizer_offload_false/checkpoints/0. I0416 18:52:16.773706 137270208547968 signaling_client.py:373] Using ThreadSafeKeyValueSignalingClient I0416 18:52:16.862428 137270208547968 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array. I0416 18:52:16.862603 137162718709312 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_11_optimizer_offload_false/checkpoints/0 I0416 18:52:16.863076 137270208547968 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 W0416 18:52:17.093401 4006445 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 18:52:17.104722 4006445 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 18:52:17.113946 4006445 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 18:52:17.127487 4006445 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 18:52:17.136462 4006445 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 18:52:17.145069 4006445 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 18:52:17.151069 4006445 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 18:52:17.155492 4006445 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0416 18:52:17.536034 137162708223552 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_11_optimizer_offload_false/checkpoints/0/items I0416 18:52:17.696257 137162672571968 checkpoint.py:188] Wrote Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776365537448572042, 'commit_timestamp_nsecs': None, 'custom_metadata': {}}, json={"item_handlers": null, "metrics": {}, "performance_metrics": {}, "init_timestamp_nsecs": 1776365537448572042, "commit_timestamp_nsecs": null, "custom_metadata": {}} to gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_11_optimizer_offload_false/checkpoints/0/_CHECKPOINT_METADATA I0416 18:52:17.841545 137270208547968 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.980012s I0416 18:52:17.841941 137270208547968 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 11.569 GiB/s (total gbytes: 12.3 GiB) (time elapsed: a second) (per-host) I0416 18:52:17.842006 137270208547968 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 1.066778s (batch_requests_ready=0.082491s, total_serialization_initiated=0.984011s, others=0.000276s) I0416 18:52:17.842087 137270208547968 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 1.067784s (all_items=0.000022s, per_item={'items': '0.00002241'}, temp_paths=1.067761) I0416 18:52:17.842937 137162651600448 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started. I0416 18:52:17.843085 137270208547968 async_checkpointer.py:561] Finished blocking save. Time taken: 1.159088s. Continuing background save to gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_11_optimizer_offload_false/checkpoints/0. I0416 18:52:17.843334 137270208547968 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize I0416 18:52:17.843542 137162641114688 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0416 18:52:17.843647 137270208547968 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_11_optimizer_offload_false/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776365536.6837847, 'wait_for_prev_duration_secs': 4.1484832763671875e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776365536.683973, 'checkpointer_blocking_duration_secs': 1.1592586040496826, 'get_old_steps_start_time': 1776365537.843259, 'get_old_steps_duration_secs': 3.0040740966796875e-05, 'checkpoint_manager_blocking_start_time': 1776365536.683689, 'checkpoint_manager_blocking_duration_secs': 1.1599199771881104} I0416 18:52:17.843755 137270208547968 checkpointing.py:409] Started an asynchronous checkpoint save for step 0 I0416 18:52:17.843803 137270208547968 max_utils.py:750] Memstats: After params initialized: I0416 18:52:17.843897 137270208547968 max_utils.py:756] Using (GB) 1.6 / 31.25 (5.120000%) on TPU_0(process=0,(0,0,0,0)) I0416 18:52:17.843924 137270208547968 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_1(process=0,(1,0,0,0)) I0416 18:52:17.843950 137270208547968 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_2(process=0,(0,1,0,0)) I0416 18:52:17.843969 137270208547968 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_3(process=0,(1,1,0,0)) I0416 18:52:17.843986 137270208547968 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_4(process=0,(0,2,0,0)) I0416 18:52:17.844002 137270208547968 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_5(process=0,(1,2,0,0)) I0416 18:52:17.844018 137270208547968 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_6(process=0,(0,3,0,0)) I0416 18:52:17.844035 137270208547968 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_7(process=0,(1,3,0,0)) W0416 18:52:17.851242 4006445 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 18:52:17.857611 4006445 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 18:52:17.863037 4006445 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0416 18:52:18.018941 4008563 google_auth_provider.cc:149] Using credentials at ~/.config/gcloud/application_default_credentials.json I0416 18:52:18.019023 4008563 google_auth_provider.cc:156] Using OAuth2 AuthProvider I0416 18:52:18.205875 137270208547968 metric_logger.py:185] completed step: 0, seconds: 0.979, TFLOP/s/device: 13.880, Tokens/s/device: 2092.087, total_weights: 16384, loss: 10.874 I0416 18:52:18.206856 137270208547968 metric_logger.py:269] To see full metrics 'tensorboard --logdir=gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_11_optimizer_offload_false/tensorboard/' I0416 18:52:18.321884 137270208547968 metric_logger.py:185] completed step: 1, seconds: 1.638, TFLOP/s/device: 8.294, Tokens/s/device: 1250.164, total_weights: 16384, loss: 10.874 I0416 18:52:18.445756 137270208547968 metric_logger.py:185] completed step: 2, seconds: 0.007, TFLOP/s/device: 2014.998, Tokens/s/device: 303722.379, total_weights: 16384, loss: 9.775 I0416 18:52:18.564477 137270208547968 metric_logger.py:185] completed step: 3, seconds: 0.117, TFLOP/s/device: 116.277, Tokens/s/device: 17526.594, total_weights: 16384, loss: 8.838 I0416 18:52:18.786719 137162718709312 array_metadata_store.py:203] [process=0][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_11_optimizer_offload_false/checkpoints/0/items/array_metadatas/process_0 I0416 18:52:38.440500 137270208547968 metric_logger.py:185] completed step: 4, seconds: 0.124, TFLOP/s/device: 109.734, Tokens/s/device: 16540.272, total_weights: 16384, loss: 7.965 I0416 18:52:38.448179 137270208547968 metric_logger.py:185] completed step: 5, seconds: 0.119, TFLOP/s/device: 114.285, Tokens/s/device: 17226.297, total_weights: 16384, loss: 7.233 I0416 18:52:38.565082 137270208547968 metric_logger.py:185] completed step: 6, seconds: 19.877, TFLOP/s/device: 0.684, Tokens/s/device: 103.034, total_weights: 16384, loss: 6.676 I0416 18:52:38.683264 137270208547968 metric_logger.py:185] completed step: 7, seconds: 0.005, TFLOP/s/device: 2848.455, Tokens/s/device: 429350.105, total_weights: 16384, loss: 6.293 I0416 18:52:38.801806 137270208547968 metric_logger.py:185] completed step: 8, seconds: 0.118, TFLOP/s/device: 115.477, Tokens/s/device: 17405.937, total_weights: 16384, loss: 6.055 I0416 18:52:38.919653 137270208547968 checkpointing.py:794] Waiting for step 9 to finish before checkpoint... I0416 18:52:38.920580 137270208547968 checkpointing.py:798] Waited 0.0009565353393554688 seconds for step 9 to finish before starting checkpointing. I0416 18:52:38.920889 137270208547968 checkpoint_manager.py:1994] [process=0][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0416 18:52:46.975779 137162697737792 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 28.953763s (commit=28.515764s, array_metadata_write=0.437999s) I0416 18:52:46.976998 137162651600448 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 418.413 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 30 seconds) (per-host) I0416 18:52:46.977073 137162651600448 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 29.133930s. I0416 18:52:47.201910 137162651600448 checkpoint.py:228] Read Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776365537448572042, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} from gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_11_optimizer_offload_false/checkpoints/0/_CHECKPOINT_METADATA I0416 18:52:47.381816 137162651600448 array_metadata_store.py:367] [process=0][thread=async_save] Skipped cross-host ArrayMetadata validation because only one process is found: process_index=0. I0416 18:52:47.602321 137162672571968 checkpoint.py:247] Updated Metadata={'item_handlers': {'items': 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler'}, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776365537448572042, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} to gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_11_optimizer_offload_false/checkpoints/0/_CHECKPOINT_METADATA I0416 18:52:47.825002 137162651600448 ocdbt_utils.py:56] Param validation support for Zarr3 will be added later (b/362328389). I0416 18:52:47.825726 137162651600448 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.581194s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_11_optimizer_offload_false/checkpoints/0/items I0416 18:52:47.826531 137162651600448 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_11_optimizer_offload_false/checkpoints/0/items I0416 18:52:48.088013 137162651600448 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_11_optimizer_offload_false/checkpoints/0 I0416 18:52:48.785381 137162651600448 atomicity.py:794] [process=0][thread=async_save] Finished saving checkpoint (finalized tmp dir) to `gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_11_optimizer_offload_false/checkpoints/0`. I0416 18:52:48.786194 137162651600448 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 32.102209s. directory=gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_11_optimizer_offload_false/checkpoints/0 I0416 18:52:48.786269 137162651600448 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 30.943128s. I0416 18:52:48.786390 137162641114688 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0416 18:52:48.786440 137162641114688 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0416 18:52:48.786494 137162641114688 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts... I0416 18:52:48.786538 137162641114688 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts. I0416 18:52:48.786719 137270208547968 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. W0416 18:52:48.786874 137270208547968 checkpoint_manager.py:1441] Waiting for previous save to complete took 9.866000 seconds. If this number is high, consider checkpointing less frequently. I0416 18:52:48.787718 137270208547968 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 9 I0416 18:52:48.788068 137270208547968 async_checkpointer.py:452] [process=0] Started async saving checkpoint to gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_11_optimizer_offload_false/checkpoints/9. I0416 18:52:48.958712 137270208547968 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array. I0416 18:52:48.958815 137270208547968 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 I0416 18:52:48.960529 137162641114688 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_11_optimizer_offload_false/checkpoints/9 I0416 18:52:49.672495 137162580297280 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_11_optimizer_offload_false/checkpoints/9/items I0416 18:52:57.603763 137270208547968 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 8.645533s I0416 18:52:57.608625 137270208547968 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 1.412 GiB/s (total gbytes: 12.3 GiB) (time elapsed: 8 seconds) (per-host) I0416 18:52:57.608696 137270208547968 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 8.740270s (batch_requests_ready=0.086720s, total_serialization_initiated=8.648840s, others=0.004709s) I0416 18:52:57.608858 137270208547968 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 8.740925s (all_items=0.000020s, per_item={'items': '0.00001979'}, temp_paths=8.740905) I0416 18:52:57.610463 137162641114688 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started. I0416 18:52:57.610552 137270208547968 async_checkpointer.py:561] Finished blocking save. Time taken: 8.822792s. Continuing background save to gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_11_optimizer_offload_false/checkpoints/9. I0416 18:52:57.610774 137270208547968 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize I0416 18:52:57.610990 137162651600448 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0416 18:52:57.611138 137270208547968 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_11_optimizer_offload_false/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776365558.9208455, 'wait_for_prev_duration_secs': 9.865999937057495, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776365568.7877412, 'checkpointer_blocking_duration_secs': 8.822929382324219, 'get_old_steps_start_time': 1776365577.6106915, 'get_old_steps_duration_secs': 3.886222839355469e-05, 'checkpoint_manager_blocking_start_time': 1776365558.9207952, 'checkpoint_manager_blocking_duration_secs': 18.69031310081482} I0416 18:52:57.611295 137270208547968 checkpointing.py:409] Started an asynchronous checkpoint save for step 9 I0416 18:52:57.611330 137270208547968 checkpoint_manager.py:1994] [process=0][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0416 18:52:58.381464 137162718709312 array_metadata_store.py:203] [process=0][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_11_optimizer_offload_false/checkpoints/9/items/array_metadatas/process_0 I0416 18:53:28.770101 137162580297280 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 31.160520s (commit=30.716269s, array_metadata_write=0.444252s) I0416 18:53:28.771482 137162641114688 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 316.688 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 39 seconds) (per-host) I0416 18:53:28.771549 137162641114688 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 31.160951s. I0416 18:53:29.184652 137162641114688 array_metadata_store.py:367] [process=0][thread=async_save] Skipped cross-host ArrayMetadata validation because only one process is found: process_index=0. I0416 18:53:29.612694 137162641114688 ocdbt_utils.py:56] Param validation support for Zarr3 will be added later (b/362328389). I0416 18:53:29.612908 137162641114688 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.566378s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_11_optimizer_offload_false/checkpoints/9/items I0416 18:53:29.613738 137162641114688 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_11_optimizer_offload_false/checkpoints/9/items I0416 18:53:29.852977 137162641114688 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_11_optimizer_offload_false/checkpoints/9 I0416 18:53:30.535752 137162641114688 atomicity.py:794] [process=0][thread=async_save] Finished saving checkpoint (finalized tmp dir) to `gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_11_optimizer_offload_false/checkpoints/9`. I0416 18:53:30.536538 137162641114688 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 41.748784s. directory=gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_11_optimizer_offload_false/checkpoints/9 I0416 18:53:30.536617 137162641114688 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 32.926020s. I0416 18:53:30.536872 137162651600448 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0416 18:53:30.537003 137162651600448 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0416 18:53:30.537102 137162651600448 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts... I0416 18:53:30.537150 137162651600448 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts. I0416 18:53:30.537694 137270208547968 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. I0416 18:53:30.538524 137270208547968 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0416 18:53:30.539749 137270208547968 metric_logger.py:185] completed step: 9, seconds: 0.118, TFLOP/s/device: 114.923, Tokens/s/device: 17322.462, total_weights: 16384, loss: 5.915 Per train step: Total TFLOPs: 13.59 split as 93.93% learnable weight flops and 6.07% attention flops