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 | 4.1890 | 4.1890 | 0 |
| TFLOP/s | 114.586 | 115.021 | +0.435 |
| Tok/s | 17271.6 | 17337.3 | +65.65 |
| Avg s/step | 4.805 | 7.112 | +2.307 |
| 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 13:49:01.226747: 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 13:49:01.346541 125324629183616 max_utils.py:238] Skipping jax distributed system due to skip_jax_distributed_system=True flag. I0416 13:49:38.389983 125324629183616 max_utils.py:800] System Information: Jax Version: 0.9.2 I0416 13:49:38.390120 125324629183616 max_utils.py:801] System Information: Jaxlib Version: 0.9.2 I0416 13:49:38.390158 125324629183616 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 13:49:38.390189 125324629183616 train_utils.py:334] WARNING: 'dataset_path' might be pointing your local file system I0416 13:49:38.390214 125324629183616 train_utils.py:347] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing. I0416 13:49:38.390316 125324629183616 train.py:683] [DECOUPLED NO-OP] skipping cloud diagnostics wrapper. W0416 13:49:38.487126 3890411 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0416 13:49:38.906245 125324629183616 maxtext_utils.py:1517] Num_devices: 8, shape (1, 1, 1, 8, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0416 13:49:38.906632 125324629183616 checkpointing.py:677] Setting up checkpoint logger... I0416 13:49:38.906684 125324629183616 checkpointing.py:233] Creating checkpoint manager with ocdbt=True and zarr3=False I0416 13:49:38.906750 125324629183616 pytree_checkpoint_handler.py:577] save_device_host_concurrent_bytes=None I0416 13:49:38.907444 125324629183616 base_pytree_checkpoint_handler.py:411] Created BasePyTreeCheckpointHandler: use_ocdbt=True, use_zarr3=False, pytree_metadata_options=PyTreeMetadataOptions(support_rich_types=False), array_metadata_store=<orbax.checkpoint._src.metadata.array_metadata_store.Store object at 0x71fad3100860>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0416 13:49:41.312289 125324629183616 checkpointing.py:265] Enabling policy for fixed interval checkpointing. I0416 13:49:41.312515 125324629183616 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 0x71f45733c440>}, handler_registry=None I0416 13:49:41.312860 125324629183616 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x71f45733c440>` 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 13:49:41.312904 125324629183616 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x71f458126db0>` 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 13:49:41.312934 125324629183616 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 0x71f45733c440>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x71f45733c440>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x71f458126db0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x71f458126db0>}). I0416 13:49:41.313289 125324629183616 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.28 I0416 13:49:41.313348 125324629183616 async_checkpointer.py:177] [process=0][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>.<lambda> at 0x71f457331300> timeout: 600 secs and primary_host=0 for async checkpoint writes I0416 13:49:41.441837 125324629183616 checkpoint_manager.py:1788] Found 0 checkpoint steps in gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_16_zarr3_false/checkpoints I0416 13:49:41.442191 125324629183616 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_20260416_123638/linen_main_20260416_123638_16_zarr3_false/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x71f457749910> I0416 13:49:41.442284 125324629183616 checkpointing.py:301] Checkpoint manager created! W0416 13:49:41.459949 3890411 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0416 13:49:41.647155 125324629183616 nnx_wrappers.py:437] Unknown Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0416 13:49:41.647285 125324629183616 nnx_wrappers.py:437] Unknown Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0416 13:49:41.741729 125324629183616 attentions.py:1088] attentions/inputs_q Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0416 13:49:41.741816 125324629183616 attentions.py:1088] attentions/inputs_q Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0416 13:49:41.756273 125324629183616 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0416 13:49:41.756325 125324629183616 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0416 13:49:41.777164 125324629183616 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 13:49:41.777218 125324629183616 attentions.py:1154] attentions/query Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0416 13:49:41.791686 125324629183616 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 13:49:41.791739 125324629183616 attentions.py:1155] attentions/key Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0416 13:49:41.806197 125324629183616 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 13:49:41.806251 125324629183616 attentions.py:1156] attentions/value Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0416 13:49:41.828094 125324629183616 attentions.py:1197] attentions/out Logical: bfloat16[8,2048,16,128]..................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv'). I0416 13:49:41.828152 125324629183616 attentions.py:1197] attentions/out Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0416 13:49:41.846790 125324629183616 linears.py:525] linears/x Logical: bfloat16[8,2048,7168]....................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0416 13:49:41.846843 125324629183616 linears.py:525] linears/x Physical: bfloat16[8,2048,7168]....................................... ('fsdp', None, None). I0416 13:49:42.030090 125324629183616 checkpointing.py:577] checkpoint manager exists so trying to load this run's existing checkpoint I0416 13:49:42.030193 125324629183616 checkpointing.py:665] No existing checkpoints found, not restoring checkpoint. W0416 13:49:42.081108 3890411 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 13:49:42.460265 125324629183616 maxtext_utils.py:1620] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: P('norm',) Physical: (None,) I0416 13:49:42.460376 125324629183616 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 13:49:42.460417 125324629183616 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 13:49:42.460462 125324629183616 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 13:49:42.460500 125324629183616 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 13:49:42.460531 125324629183616 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 13:49:42.460570 125324629183616 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 13:49:42.460609 125324629183616 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 13:49:42.460637 125324629183616 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 13:49:42.460663 125324629183616 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 13:49:42.460697 125324629183616 maxtext_utils.py:1620] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: P('embed_vocab', 'vocab') Physical: ('fsdp', None) I0416 13:49:42.460736 125324629183616 maxtext_utils.py:1620] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: P('vocab', 'embed_vocab') Physical: (None, 'fsdp') I0416 13:49:42.883996 125324629183616 train.py:155] train/xent Logical: float32[8,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length'). I0416 13:49:42.884099 125324629183616 train.py:155] train/xent Physical: float32[8,2048]............................................. ('fsdp', None). I0416 13:49:42.897787 125324629183616 train.py:162] train/z_loss Logical: float32[8,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length'). I0416 13:49:42.897839 125324629183616 train.py:162] train/z_loss Physical: float32[8,2048]............................................. ('fsdp', None). W0416 13:49:43.565124 3890411 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0416 13:49:43.686992 125324629183616 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 13:49:43.687552 125324629183616 max_utils.py:194] tensorboardX not available; using no-op SummaryWriter. I0416 13:49:43.687701 125324629183616 metric_logger.py:289] number parameters: 1.104 billion W0416 13:49:44.532707 3890411 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0416 13:49:44.626548 125324629183616 checkpointing.py:772] Waiting for step 0 to finish before checkpoint... I0416 13:49:44.744206 125324629183616 checkpointing.py:776] Waited 0.11763811111450195 seconds for step 0 to finish before starting checkpointing. I0416 13:49:44.744673 125324629183616 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0416 13:49:44.744824 125324629183616 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 0 I0416 13:49:44.745250 125324629183616 async_checkpointer.py:452] [process=0] Started async saving checkpoint to gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_16_zarr3_false/checkpoints/0. I0416 13:49:44.836066 125324629183616 signaling_client.py:373] Using ThreadSafeKeyValueSignalingClient I0416 13:49:44.927783 125324629183616 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array. I0416 13:49:44.927980 125324629183616 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 13:49:44.933928 125217160562240 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_16_zarr3_false/checkpoints/0 W0416 13:49:45.157365 3890411 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 13:49:45.171263 3890411 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 13:49:45.183456 3890411 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 13:49:45.200745 3890411 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 13:49:45.213575 3890411 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 13:49:45.217999 3890411 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 13:49:45.222471 3890411 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 13:49:45.226844 3890411 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0416 13:49:45.417313 125324629183616 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.490361s I0416 13:49:45.417704 125324629183616 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 21.271 GiB/s (total gbytes: 12.3 GiB) (time elapsed: 580 milliseconds) (per-host) I0416 13:49:45.417755 125324629183616 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 0.580270s (batch_requests_ready=0.085753s, total_serialization_initiated=0.494246s, others=0.000271s) I0416 13:49:45.417816 125324629183616 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.581196s (all_items=0.000021s, per_item={'items': '0.00002050'}, temp_paths=0.581176) I0416 13:49:45.418587 125217097647680 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started. I0416 13:49:45.418687 125324629183616 async_checkpointer.py:561] Finished blocking save. Time taken: 0.673823s. Continuing background save to gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_16_zarr3_false/checkpoints/0. I0416 13:49:45.418855 125324629183616 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize I0416 13:49:45.419039 125217171048000 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0416 13:49:45.419166 125324629183616 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_16_zarr3_false/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776347384.7446613, 'wait_for_prev_duration_secs': 3.933906555175781e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776347384.7448459, 'checkpointer_blocking_duration_secs': 0.6739368438720703, 'get_old_steps_start_time': 1776347385.4188004, 'get_old_steps_duration_secs': 2.193450927734375e-05, 'checkpoint_manager_blocking_start_time': 1776347384.7445748, 'checkpoint_manager_blocking_duration_secs': 0.6745619773864746} I0416 13:49:45.419279 125324629183616 checkpointing.py:408] Started an asynchronous checkpoint save for step 0 I0416 13:49:45.419318 125324629183616 max_utils.py:750] Memstats: After params initialized: I0416 13:49:45.419393 125324629183616 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_0(process=0,(0,0,0,0)) I0416 13:49:45.419425 125324629183616 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_1(process=0,(1,0,0,0)) I0416 13:49:45.419444 125324629183616 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_2(process=0,(0,1,0,0)) I0416 13:49:45.419461 125324629183616 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_3(process=0,(1,1,0,0)) I0416 13:49:45.419478 125324629183616 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_4(process=0,(0,2,0,0)) I0416 13:49:45.419494 125324629183616 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_5(process=0,(1,2,0,0)) I0416 13:49:45.419511 125324629183616 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_6(process=0,(0,3,0,0)) I0416 13:49:45.419526 125324629183616 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_7(process=0,(1,3,0,0)) W0416 13:49:45.426099 3890411 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 13:49:45.432306 3890411 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 13:49:45.437179 3890411 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0416 13:49:45.597386 125217150076480 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_16_zarr3_false/checkpoints/0/items I0416 13:49:45.635696 125324629183616 metric_logger.py:185] completed step: 0, seconds: 0.938, TFLOP/s/device: 14.480, Tokens/s/device: 2182.624, total_weights: 16384, loss: 10.874 I0416 13:49:45.636607 125324629183616 metric_logger.py:269] To see full metrics 'tensorboard --logdir=gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_16_zarr3_false/tensorboard/' I0416 13:49:45.751746 125324629183616 metric_logger.py:185] completed step: 1, seconds: 1.007, TFLOP/s/device: 13.487, Tokens/s/device: 2032.916, total_weights: 16384, loss: 10.874 I0416 13:49:45.757115 125217078773312 checkpoint.py:188] Wrote Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776347385509550486, 'commit_timestamp_nsecs': None, 'custom_metadata': {}}, json={"item_handlers": null, "metrics": {}, "performance_metrics": {}, "init_timestamp_nsecs": 1776347385509550486, "commit_timestamp_nsecs": null, "custom_metadata": {}} to gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_16_zarr3_false/checkpoints/0/_CHECKPOINT_METADATA I0416 13:49:45.870272 125324629183616 metric_logger.py:185] completed step: 2, seconds: 0.006, TFLOP/s/device: 2430.613, Tokens/s/device: 366368.515, total_weights: 16384, loss: 10.276 I0416 13:49:45.989201 125324629183616 metric_logger.py:185] completed step: 3, seconds: 0.117, TFLOP/s/device: 116.110, Tokens/s/device: 17501.431, total_weights: 16384, loss: 9.282 I0416 13:49:45.994136 125324629183616 checkpointing.py:772] Waiting for step 5 to finish before checkpoint... I0416 13:49:46.065958 3892594 google_auth_provider.cc:149] Using credentials at ~/.config/gcloud/application_default_credentials.json I0416 13:49:46.066026 3892594 google_auth_provider.cc:156] Using OAuth2 AuthProvider I0416 13:49:46.226068 125324629183616 checkpointing.py:776] Waited 0.23187637329101562 seconds for step 5 to finish before starting checkpointing. I0416 13:49:46.226495 125324629183616 checkpoint_manager.py:1994] [process=0][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0416 13:49:46.852181 125217118619200 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_16_zarr3_false/checkpoints/0/items/array_metadatas/process_0 I0416 13:50:14.937244 125217108133440 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 28.868652s (commit=28.440824s, array_metadata_write=0.427828s) I0416 13:50:14.938453 125217097647680 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 419.814 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 30 seconds) (per-host) I0416 13:50:14.938511 125217097647680 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 29.519780s. I0416 13:50:15.174101 125217097647680 checkpoint.py:228] Read Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776347385509550486, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} from gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_16_zarr3_false/checkpoints/0/_CHECKPOINT_METADATA I0416 13:50:15.357284 125217097647680 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 13:50:15.572813 125217078773312 checkpoint.py:247] Updated Metadata={'item_handlers': {'items': 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler'}, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776347385509550486, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} to gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_16_zarr3_false/checkpoints/0/_CHECKPOINT_METADATA I0416 13:50:15.818209 125217097647680 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.603556s. use_zarr3=False, enable_post_merge_validation=True, directory=gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_16_zarr3_false/checkpoints/0/items I0416 13:50:15.819004 125217097647680 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_16_zarr3_false/checkpoints/0/items I0416 13:50:16.069392 125217097647680 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_16_zarr3_false/checkpoints/0 I0416 13:50:16.756031 125217097647680 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_16_zarr3_false/checkpoints/0`. I0416 13:50:16.756805 125217097647680 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 32.011946s. directory=gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_16_zarr3_false/checkpoints/0 I0416 13:50:16.756881 125217097647680 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 31.338152s. I0416 13:50:16.757131 125217171048000 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0416 13:50:16.757258 125217171048000 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0416 13:50:16.757327 125217171048000 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts... I0416 13:50:16.757375 125217171048000 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts. I0416 13:50:16.757563 125324629183616 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 13:50:16.757714 125324629183616 checkpoint_manager.py:1441] Waiting for previous save to complete took 30.531219 seconds. If this number is high, consider checkpointing less frequently. I0416 13:50:16.758586 125324629183616 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 5 I0416 13:50:16.758892 125324629183616 async_checkpointer.py:452] [process=0] Started async saving checkpoint to gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_16_zarr3_false/checkpoints/5. I0416 13:50:16.938378 125324629183616 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array. I0416 13:50:16.938529 125217171048000 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_16_zarr3_false/checkpoints/5 I0416 13:50:16.938611 125324629183616 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 13:50:17.667384 125217026344512 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_16_zarr3_false/checkpoints/5/items I0416 13:50:27.619291 125324629183616 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 10.681767s I0416 13:50:27.619869 125324629183616 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 1.146 GiB/s (total gbytes: 12.3 GiB) (time elapsed: 10 seconds) (per-host) I0416 13:50:27.619921 125324629183616 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 10.770389s (batch_requests_ready=0.084339s, total_serialization_initiated=10.685598s, others=0.000453s) I0416 13:50:27.619988 125324629183616 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 10.770983s (all_items=0.000017s, per_item={'items': '0.00001693'}, temp_paths=10.770966) I0416 13:50:27.624669 125217097647680 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started. I0416 13:50:27.624782 125324629183616 async_checkpointer.py:561] Finished blocking save. Time taken: 10.866147s. Continuing background save to gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_16_zarr3_false/checkpoints/5. I0416 13:50:27.624923 125324629183616 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=5] Starting CheckpointManager Save Finalize thread=save_finalize I0416 13:50:27.625120 125217108133440 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0416 13:50:27.625239 125324629183616 standard_logger.py:34] {'step': 5, 'event_type': 'save', 'directory': 'gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_16_zarr3_false/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776347386.2264714, 'wait_for_prev_duration_secs': 30.531218767166138, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776347416.7586133, 'checkpointer_blocking_duration_secs': 10.866228342056274, 'get_old_steps_start_time': 1776347427.624858, 'get_old_steps_duration_secs': 2.765655517578125e-05, 'checkpoint_manager_blocking_start_time': 1776347386.2264023, 'checkpoint_manager_blocking_duration_secs': 41.398813009262085} I0416 13:50:27.625382 125324629183616 checkpointing.py:408] Started an asynchronous checkpoint save for step 5 I0416 13:50:27.625909 125324629183616 metric_logger.py:185] completed step: 4, seconds: 0.118, TFLOP/s/device: 114.751, Tokens/s/device: 17296.567, total_weights: 16384, loss: 8.327 I0416 13:50:27.633117 125324629183616 metric_logger.py:185] completed step: 5, seconds: 0.119, TFLOP/s/device: 114.181, Tokens/s/device: 17210.663, total_weights: 16384, loss: 7.367 I0416 13:50:27.749529 125324629183616 metric_logger.py:185] completed step: 6, seconds: 41.637, TFLOP/s/device: 0.326, Tokens/s/device: 49.187, total_weights: 16384, loss: 6.459 I0416 13:50:27.867876 125324629183616 metric_logger.py:185] completed step: 7, seconds: 0.005, TFLOP/s/device: 2556.855, Tokens/s/device: 385397.064, total_weights: 16384, loss: 5.622 I0416 13:50:27.986404 125324629183616 metric_logger.py:185] completed step: 8, seconds: 0.117, TFLOP/s/device: 115.828, Tokens/s/device: 17458.910, total_weights: 16384, loss: 4.864 I0416 13:50:28.104552 125324629183616 checkpointing.py:772] Waiting for step 9 to finish before checkpoint... I0416 13:50:28.105423 125324629183616 checkpointing.py:776] Waited 0.0009045600891113281 seconds for step 9 to finish before starting checkpointing. I0416 13:50:28.105670 125324629183616 checkpoint_manager.py:1994] [process=0][thread=MainThread][step=5][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0416 13:50:28.574601 125217118619200 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_16_zarr3_false/checkpoints/5/items/array_metadatas/process_0 I0416 13:50:59.216112 125217026344512 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 31.592098s (commit=31.161695s, array_metadata_write=0.430403s) I0416 13:50:59.217302 125217097647680 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 298.265 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 42 seconds) (per-host) I0416 13:50:59.217418 125217097647680 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 31.592665s. I0416 13:50:59.615976 125217097647680 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 13:51:00.157719 125217097647680 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.677035s. use_zarr3=False, enable_post_merge_validation=True, directory=gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_16_zarr3_false/checkpoints/5/items I0416 13:51:00.158599 125217097647680 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_16_zarr3_false/checkpoints/5/items I0416 13:51:00.418962 125217097647680 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_16_zarr3_false/checkpoints/5 I0416 13:51:01.089627 125217097647680 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_16_zarr3_false/checkpoints/5`. I0416 13:51:01.090374 125217097647680 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 44.331746s. directory=gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_16_zarr3_false/checkpoints/5 I0416 13:51:01.090462 125217097647680 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 33.465710s. I0416 13:51:01.090692 125217108133440 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0416 13:51:01.090816 125217108133440 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0416 13:51:01.090882 125217108133440 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=5] CheckpointManager Save Finalize is syncing with other hosts... I0416 13:51:01.090929 125217108133440 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=5] CheckpointManager Save Finalize is done on all hosts. I0416 13:51:01.091129 125324629183616 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. W0416 13:51:01.091264 125324629183616 checkpoint_manager.py:1441] Waiting for previous save to complete took 32.985597 seconds. If this number is high, consider checkpointing less frequently. I0416 13:51:01.091485 125324629183616 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 9 I0416 13:51:01.091811 125324629183616 async_checkpointer.py:452] [process=0] Started async saving checkpoint to gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_16_zarr3_false/checkpoints/9. I0416 13:51:01.258913 125324629183616 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array. I0416 13:51:01.259014 125324629183616 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 13:51:01.262696 125217108133440 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_16_zarr3_false/checkpoints/9 I0416 13:51:02.018627 125217057801792 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_16_zarr3_false/checkpoints/9/items I0416 13:51:09.413874 125324629183616 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 8.155382s I0416 13:51:09.419101 125324629183616 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 1.496 GiB/s (total gbytes: 12.3 GiB) (time elapsed: 8 seconds) (per-host) I0416 13:51:09.419181 125324629183616 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 8.249728s (batch_requests_ready=0.085914s, total_serialization_initiated=8.159013s, others=0.004801s) I0416 13:51:09.419281 125324629183616 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 8.250305s (all_items=0.000015s, per_item={'items': '0.00001550'}, temp_paths=8.250290) I0416 13:51:09.421058 125217026344512 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started. I0416 13:51:09.421182 125324629183616 async_checkpointer.py:561] Finished blocking save. Time taken: 8.329649s. Continuing background save to gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_16_zarr3_false/checkpoints/9. I0416 13:51:09.421421 125324629183616 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize I0416 13:51:09.421619 125217097647680 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0416 13:51:09.421795 125324629183616 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_16_zarr3_false/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776347428.105646, 'wait_for_prev_duration_secs': 32.985596895217896, 'time_between_consecutive_saves_sec': 11.348256826400757, 'checkpointer_blocking_start_time': 1776347461.0915084, 'checkpointer_blocking_duration_secs': 8.329810619354248, 'get_old_steps_start_time': 1776347469.4213383, 'get_old_steps_duration_secs': 3.4332275390625e-05, 'checkpoint_manager_blocking_start_time': 1776347428.1056135, 'checkpoint_manager_blocking_duration_secs': 41.316115379333496} I0416 13:51:09.421961 125324629183616 checkpointing.py:408] Started an asynchronous checkpoint save for step 9 I0416 13:51:09.421997 125324629183616 checkpoint_manager.py:1994] [process=0][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0416 13:51:10.187798 125217171048000 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_16_zarr3_false/checkpoints/9/items/array_metadatas/process_0 I0416 13:51:40.839353 125217057801792 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 31.419391s (commit=30.988675s, array_metadata_write=0.430716s) I0416 13:51:40.840881 125217026344512 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 318.537 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 39 seconds) (per-host) I0416 13:51:40.840953 125217026344512 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 31.419718s. I0416 13:51:41.241929 125217026344512 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 13:51:41.748424 125217026344512 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.641015s. use_zarr3=False, enable_post_merge_validation=True, directory=gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_16_zarr3_false/checkpoints/9/items I0416 13:51:41.749347 125217026344512 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_16_zarr3_false/checkpoints/9/items I0416 13:51:41.994182 125217026344512 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_16_zarr3_false/checkpoints/9 I0416 13:51:42.681522 125217026344512 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_16_zarr3_false/checkpoints/9`. I0416 13:51:42.682321 125217026344512 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 41.590800s. directory=gs://wanglance-maxtext/linen_ckpt_main_20260416_123638/linen_main_20260416_123638_16_zarr3_false/checkpoints/9 I0416 13:51:42.682404 125217026344512 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 33.261170s. I0416 13:51:42.682607 125217097647680 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0416 13:51:42.682739 125217097647680 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0416 13:51:42.682832 125217097647680 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts... I0416 13:51:42.682890 125217097647680 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts. I0416 13:51:42.683160 125324629183616 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 13:51:42.683347 125324629183616 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0416 13:51:42.684602 125324629183616 metric_logger.py:185] completed step: 9, seconds: 0.119, TFLOP/s/device: 114.586, Tokens/s/device: 17271.623, total_weights: 16384, loss: 4.189 Per train step: Total TFLOPs: 13.59 split as 93.93% learnable weight flops and 6.07% attention flops
2026-04-16 19:31:17.614202: 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 19:31:17.731330 125430950521984 max_utils.py:238] Skipping jax distributed system due to skip_jax_distributed_system=True flag. I0416 19:32:14.206234 125430950521984 max_utils.py:800] System Information: Jax Version: 0.9.2 I0416 19:32:14.206345 125430950521984 max_utils.py:801] System Information: Jaxlib Version: 0.9.2 I0416 19:32:14.206379 125430950521984 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 19:32:14.206404 125430950521984 train_utils.py:364] WARNING: 'dataset_path' might be pointing your local file system I0416 19:32:14.206428 125430950521984 train_utils.py:377] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing. I0416 19:32:14.206503 125430950521984 train.py:811] [DECOUPLED NO-OP] skipping cloud diagnostics wrapper. W0416 19:32:14.298918 4039547 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0416 19:32:14.707970 125430950521984 maxtext_utils.py:1687] Num_devices: 8, shape (1, 1, 1, 8, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0416 19:32:14.708155 125430950521984 maxtext_utils.py:1687] Num_devices: 8, shape (1, 1, 1, 8, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0416 19:32:14.708326 125430950521984 checkpointing.py:688] Setting up checkpoint logger... I0416 19:32:14.708366 125430950521984 checkpointing.py:234] Creating checkpoint manager with ocdbt=True and zarr3=False I0416 19:32:14.708405 125430950521984 pytree_checkpoint_handler.py:577] save_device_host_concurrent_bytes=None I0416 19:32:14.708886 125430950521984 base_pytree_checkpoint_handler.py:411] Created BasePyTreeCheckpointHandler: use_ocdbt=True, use_zarr3=False, pytree_metadata_options=PyTreeMetadataOptions(support_rich_types=False), array_metadata_store=<orbax.checkpoint._src.metadata.array_metadata_store.Store object at 0x7213951fcd40>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0416 19:32:17.125876 125430950521984 checkpointing.py:266] Enabling policy for fixed interval checkpointing. I0416 19:32:17.126066 125430950521984 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 0x720d1976a420>}, handler_registry=None I0416 19:32:17.126312 125430950521984 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x720d1976a420>` 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 19:32:17.126352 125430950521984 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x720d1893cd10>` 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 19:32:17.126383 125430950521984 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 0x720d1976a420>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x720d1976a420>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x720d1893cd10>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x720d1893cd10>}). I0416 19:32:17.126675 125430950521984 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.28 I0416 19:32:17.126724 125430950521984 async_checkpointer.py:177] [process=0][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>.<lambda> at 0x720d18935ee0> timeout: 600 secs and primary_host=0 for async checkpoint writes I0416 19:32:17.260174 125430950521984 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_16_zarr3_false/checkpoints I0416 19:32:17.260391 125430950521984 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_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_16_zarr3_false/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x720d1893c260> I0416 19:32:17.260468 125430950521984 checkpointing.py:302] Checkpoint manager created! W0416 19:32:17.276553 4039547 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0416 19:32:17.459953 125430950521984 nnx_wrappers.py:437] Unknown Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0416 19:32:17.460138 125430950521984 nnx_wrappers.py:437] Unknown Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0416 19:32:17.548539 125430950521984 attentions.py:1088] attentions/inputs_q Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0416 19:32:17.548631 125430950521984 attentions.py:1088] attentions/inputs_q Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0416 19:32:17.562819 125430950521984 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0416 19:32:17.562866 125430950521984 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0416 19:32:17.584064 125430950521984 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 19:32:17.584119 125430950521984 attentions.py:1154] attentions/query Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0416 19:32:17.598305 125430950521984 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 19:32:17.598355 125430950521984 attentions.py:1155] attentions/key Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0416 19:32:17.612506 125430950521984 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 19:32:17.612555 125430950521984 attentions.py:1156] attentions/value Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0416 19:32:17.634157 125430950521984 attentions.py:1197] attentions/out Logical: bfloat16[8,2048,16,128]..................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv'). I0416 19:32:17.634215 125430950521984 attentions.py:1197] attentions/out Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0416 19:32:17.652076 125430950521984 linears.py:525] linears/x Logical: bfloat16[8,2048,7168]....................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0416 19:32:17.652125 125430950521984 linears.py:525] linears/x Physical: bfloat16[8,2048,7168]....................................... ('fsdp', None, None). I0416 19:32:17.839072 125430950521984 checkpointing.py:578] checkpoint manager exists so trying to load this run's existing checkpoint I0416 19:32:17.839163 125430950521984 checkpointing.py:676] No existing checkpoints found, not restoring checkpoint. W0416 19:32:17.889536 4039547 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 19:32:18.264469 125430950521984 maxtext_utils.py:1796] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: P('norm',) Physical: (None,) I0416 19:32:18.264573 125430950521984 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 19:32:18.264620 125430950521984 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 19:32:18.264663 125430950521984 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 19:32:18.264701 125430950521984 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 19:32:18.264730 125430950521984 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 19:32:18.264769 125430950521984 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 19:32:18.264807 125430950521984 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 19:32:18.264834 125430950521984 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 19:32:18.264859 125430950521984 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 19:32:18.264892 125430950521984 maxtext_utils.py:1796] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: P('embed_vocab', 'vocab') Physical: ('fsdp', None) I0416 19:32:18.264924 125430950521984 maxtext_utils.py:1796] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: P('vocab', 'embed_vocab') Physical: (None, 'fsdp') I0416 19:32:18.684189 125430950521984 train.py:157] train/xent Logical: float32[8,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length'). I0416 19:32:18.684271 125430950521984 train.py:157] train/xent Physical: float32[8,2048]............................................. ('fsdp', None). I0416 19:32:18.698226 125430950521984 train.py:164] train/z_loss Logical: float32[8,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length'). I0416 19:32:18.698278 125430950521984 train.py:164] train/z_loss Physical: float32[8,2048]............................................. ('fsdp', None). W0416 19:32:19.334936 4039547 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0416 19:32:19.461928 125430950521984 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 19:32:19.462411 125430950521984 max_utils.py:194] tensorboardX not available; using no-op SummaryWriter. I0416 19:32:19.462551 125430950521984 metric_logger.py:289] number parameters: 1.104 billion W0416 19:32:20.336562 4039547 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0416 19:32:20.435902 125430950521984 checkpointing.py:794] Waiting for step 0 to finish before checkpoint... I0416 19:32:20.553225 125430950521984 checkpointing.py:798] Waited 0.11734509468078613 seconds for step 0 to finish before starting checkpointing. I0416 19:32:20.553719 125430950521984 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0416 19:32:20.553874 125430950521984 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 0 I0416 19:32:20.554328 125430950521984 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_16_zarr3_false/checkpoints/0. I0416 19:32:20.640543 125430950521984 signaling_client.py:373] Using ThreadSafeKeyValueSignalingClient I0416 19:32:20.727765 125323442128448 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_16_zarr3_false/checkpoints/0 I0416 19:32:20.736258 125430950521984 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array. I0416 19:32:20.736355 125430950521984 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 19:32:20.950804 4039547 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 19:32:20.961629 4039547 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 19:32:20.974437 4039547 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 19:32:20.987040 4039547 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 19:32:21.000168 4039547 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 19:32:21.007775 4039547 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 19:32:21.014258 4039547 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 19:32:21.018589 4039547 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0416 19:32:21.201521 125430950521984 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.465664s I0416 19:32:21.202220 125430950521984 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 22.027 GiB/s (total gbytes: 12.3 GiB) (time elapsed: 560 milliseconds) (per-host) I0416 19:32:21.202371 125430950521984 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 0.560587s (batch_requests_ready=0.090355s, total_serialization_initiated=0.469626s, others=0.000605s) I0416 19:32:21.202485 125430950521984 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.561468s (all_items=0.000020s, per_item={'items': '0.00002003'}, temp_paths=0.561448) I0416 19:32:21.203388 125323379213888 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started. I0416 19:32:21.203486 125430950521984 async_checkpointer.py:561] Finished blocking save. Time taken: 0.649572s. Continuing background save to gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_16_zarr3_false/checkpoints/0. I0416 19:32:21.203665 125430950521984 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize I0416 19:32:21.203841 125323452614208 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0416 19:32:21.203958 125430950521984 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_16_zarr3_false/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776367940.5537052, 'wait_for_prev_duration_secs': 4.1961669921875e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776367940.5538943, 'checkpointer_blocking_duration_secs': 0.6496949195861816, 'get_old_steps_start_time': 1776367941.2036066, 'get_old_steps_duration_secs': 2.2649765014648438e-05, 'checkpoint_manager_blocking_start_time': 1776367940.5536087, 'checkpoint_manager_blocking_duration_secs': 0.6503114700317383} I0416 19:32:21.204089 125430950521984 checkpointing.py:409] Started an asynchronous checkpoint save for step 0 I0416 19:32:21.204134 125430950521984 max_utils.py:750] Memstats: After params initialized: I0416 19:32:21.204242 125430950521984 max_utils.py:756] Using (GB) 1.6 / 31.25 (5.120000%) on TPU_0(process=0,(0,0,0,0)) I0416 19:32:21.204271 125430950521984 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_1(process=0,(1,0,0,0)) I0416 19:32:21.204292 125430950521984 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_2(process=0,(0,1,0,0)) I0416 19:32:21.204312 125430950521984 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_3(process=0,(1,1,0,0)) I0416 19:32:21.204330 125430950521984 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_4(process=0,(0,2,0,0)) I0416 19:32:21.204349 125430950521984 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_5(process=0,(1,2,0,0)) I0416 19:32:21.204367 125430950521984 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_6(process=0,(0,3,0,0)) I0416 19:32:21.204384 125430950521984 max_utils.py:756] Using (GB) 1.59 / 31.25 (5.088000%) on TPU_7(process=0,(1,3,0,0)) W0416 19:32:21.210386 4039547 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 19:32:21.216302 4039547 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 19:32:21.221127 4039547 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0416 19:32:21.418868 125430950521984 metric_logger.py:185] completed step: 0, seconds: 0.973, TFLOP/s/device: 13.967, Tokens/s/device: 2105.183, total_weights: 16384, loss: 10.874 I0416 19:32:21.419688 125430950521984 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_16_zarr3_false/tensorboard/' I0416 19:32:21.444741 125323431642688 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_16_zarr3_false/checkpoints/0/items I0416 19:32:21.534868 125430950521984 metric_logger.py:185] completed step: 1, seconds: 0.981, TFLOP/s/device: 13.845, Tokens/s/device: 2086.893, total_weights: 16384, loss: 10.874 I0416 19:32:21.577322 125323360339520 checkpoint.py:188] Wrote Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776367941334910214, 'commit_timestamp_nsecs': None, 'custom_metadata': {}}, json={"item_handlers": null, "metrics": {}, "performance_metrics": {}, "init_timestamp_nsecs": 1776367941334910214, "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_16_zarr3_false/checkpoints/0/_CHECKPOINT_METADATA I0416 19:32:21.653111 125430950521984 metric_logger.py:185] completed step: 2, seconds: 0.005, TFLOP/s/device: 2525.958, Tokens/s/device: 380739.914, total_weights: 16384, loss: 10.276 I0416 19:32:21.771183 125430950521984 metric_logger.py:185] completed step: 3, seconds: 0.117, TFLOP/s/device: 116.242, Tokens/s/device: 17521.196, total_weights: 16384, loss: 9.282 I0416 19:32:21.775698 125430950521984 checkpointing.py:794] Waiting for step 5 to finish before checkpoint... I0416 19:32:21.909703 4041757 google_auth_provider.cc:149] Using credentials at ~/.config/gcloud/application_default_credentials.json I0416 19:32:21.909768 4041757 google_auth_provider.cc:156] Using OAuth2 AuthProvider I0416 19:32:22.007375 125430950521984 checkpointing.py:798] Waited 0.2316443920135498 seconds for step 5 to finish before starting checkpointing. I0416 19:32:22.007770 125430950521984 checkpoint_manager.py:1994] [process=0][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0416 19:32:22.641090 125323400185408 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_16_zarr3_false/checkpoints/0/items/array_metadatas/process_0 I0416 19:33:09.817519 125323389699648 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 47.904860s (commit=47.482935s, array_metadata_write=0.421926s) I0416 19:33:09.818560 125323379213888 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 256.967 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 49 seconds) (per-host) I0416 19:33:09.818676 125323379213888 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 48.615145s. I0416 19:33:10.077764 125323379213888 checkpoint.py:228] Read Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776367941334910214, '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_16_zarr3_false/checkpoints/0/_CHECKPOINT_METADATA I0416 19:33:10.259136 125323379213888 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 19:33:10.480545 125323360339520 checkpoint.py:247] Updated Metadata={'item_handlers': {'items': 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler'}, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776367941334910214, '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_16_zarr3_false/checkpoints/0/_CHECKPOINT_METADATA I0416 19:33:10.780676 125323379213888 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.660975s. use_zarr3=False, 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_16_zarr3_false/checkpoints/0/items I0416 19:33:10.781423 125323379213888 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_16_zarr3_false/checkpoints/0/items I0416 19:33:11.013454 125323379213888 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_16_zarr3_false/checkpoints/0 I0416 19:33:11.703500 125323379213888 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_16_zarr3_false/checkpoints/0`. I0416 19:33:11.704146 125323379213888 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 51.150240s. directory=gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_16_zarr3_false/checkpoints/0 I0416 19:33:11.704228 125323379213888 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 50.500700s. I0416 19:33:11.704408 125323452614208 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0416 19:33:11.704518 125323452614208 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0416 19:33:11.704590 125323452614208 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts... I0416 19:33:11.704642 125323452614208 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts. I0416 19:33:11.704805 125430950521984 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 19:33:11.704927 125430950521984 checkpoint_manager.py:1441] Waiting for previous save to complete took 49.697151 seconds. If this number is high, consider checkpointing less frequently. I0416 19:33:11.705734 125430950521984 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 5 I0416 19:33:11.706016 125430950521984 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_16_zarr3_false/checkpoints/5. I0416 19:33:11.875378 125323452614208 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_16_zarr3_false/checkpoints/5 I0416 19:33:11.893154 125430950521984 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array. I0416 19:33:11.893251 125430950521984 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 19:33:12.606639 125323307910720 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_16_zarr3_false/checkpoints/5/items I0416 19:33:24.188526 125430950521984 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 12.295805s I0416 19:33:24.192318 125430950521984 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 1019.041 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 12 seconds) (per-host) I0416 19:33:24.192377 125430950521984 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 12.400746s (batch_requests_ready=0.098150s, total_serialization_initiated=12.298910s, others=0.003686s) I0416 19:33:24.192451 125430950521984 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 12.401529s (all_items=0.000018s, per_item={'items': '0.00001812'}, temp_paths=12.401510) I0416 19:33:24.193842 125323452614208 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started. I0416 19:33:24.193923 125430950521984 async_checkpointer.py:561] Finished blocking save. Time taken: 12.488143s. Continuing background save to gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_16_zarr3_false/checkpoints/5. I0416 19:33:24.194130 125430950521984 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=5] Starting CheckpointManager Save Finalize thread=save_finalize I0416 19:33:24.194331 125323328882240 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0416 19:33:24.194441 125430950521984 standard_logger.py:34] {'step': 5, 'event_type': 'save', 'directory': 'gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_16_zarr3_false/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776367942.0077488, 'wait_for_prev_duration_secs': 49.69715094566345, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776367991.705761, 'checkpointer_blocking_duration_secs': 12.48827052116394, 'get_old_steps_start_time': 1776368004.194061, 'get_old_steps_duration_secs': 2.7418136596679688e-05, 'checkpoint_manager_blocking_start_time': 1776367942.0076823, 'checkpoint_manager_blocking_duration_secs': 62.186734676361084} I0416 19:33:24.194567 125430950521984 checkpointing.py:409] Started an asynchronous checkpoint save for step 5 I0416 19:33:24.195076 125430950521984 metric_logger.py:185] completed step: 4, seconds: 0.118, TFLOP/s/device: 115.135, Tokens/s/device: 17354.461, total_weights: 16384, loss: 8.327 I0416 19:33:24.202414 125430950521984 metric_logger.py:185] completed step: 5, seconds: 0.118, TFLOP/s/device: 115.085, Tokens/s/device: 17346.818, total_weights: 16384, loss: 7.367 I0416 19:33:24.318908 125430950521984 metric_logger.py:185] completed step: 6, seconds: 62.425, TFLOP/s/device: 0.218, Tokens/s/device: 32.807, total_weights: 16384, loss: 6.459 I0416 19:33:24.436964 125430950521984 metric_logger.py:185] completed step: 7, seconds: 0.005, TFLOP/s/device: 2569.913, Tokens/s/device: 387365.235, total_weights: 16384, loss: 5.622 I0416 19:33:24.554912 125430950521984 metric_logger.py:185] completed step: 8, seconds: 0.117, TFLOP/s/device: 115.691, Tokens/s/device: 17438.247, total_weights: 16384, loss: 4.864 I0416 19:33:24.672539 125430950521984 checkpointing.py:794] Waiting for step 9 to finish before checkpoint... I0416 19:33:24.673454 125430950521984 checkpointing.py:798] Waited 0.0009431838989257812 seconds for step 9 to finish before starting checkpointing. I0416 19:33:24.673693 125430950521984 checkpoint_manager.py:1994] [process=0][thread=MainThread][step=5][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0416 19:33:24.962129 125322143991360 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_16_zarr3_false/checkpoints/5/items/array_metadatas/process_0 I0416 19:34:00.644171 125323400185408 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 36.451076s (commit=35.999240s, array_metadata_write=0.451836s) I0416 19:34:00.645348 125323452614208 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 258.666 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 48 seconds) (per-host) I0416 19:34:00.645408 125323452614208 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 36.451440s. I0416 19:34:01.079389 125323452614208 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 19:34:01.587851 125323452614208 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.646307s. use_zarr3=False, 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_16_zarr3_false/checkpoints/5/items I0416 19:34:01.588574 125323452614208 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_16_zarr3_false/checkpoints/5/items I0416 19:34:01.823453 125323452614208 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_16_zarr3_false/checkpoints/5 I0416 19:34:02.513697 125323452614208 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_16_zarr3_false/checkpoints/5`. I0416 19:34:02.514326 125323452614208 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 50.808553s. directory=gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_16_zarr3_false/checkpoints/5 I0416 19:34:02.514397 125323452614208 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 38.320430s. I0416 19:34:02.514548 125323328882240 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0416 19:34:02.514647 125323328882240 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0416 19:34:02.514717 125323328882240 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=5] CheckpointManager Save Finalize is syncing with other hosts... I0416 19:34:02.514758 125323328882240 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=5] CheckpointManager Save Finalize is done on all hosts. I0416 19:34:02.514910 125430950521984 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. W0416 19:34:02.515025 125430950521984 checkpoint_manager.py:1441] Waiting for previous save to complete took 37.841344 seconds. If this number is high, consider checkpointing less frequently. I0416 19:34:02.515270 125430950521984 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 9 I0416 19:34:02.515560 125430950521984 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_16_zarr3_false/checkpoints/9. I0416 19:34:02.702409 125323328882240 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_16_zarr3_false/checkpoints/9 I0416 19:34:02.714186 125430950521984 jax_array_handlers.py:347] Scheduling D2H of 39 prioritized jax.Array. I0416 19:34:02.714281 125430950521984 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 19:34:03.406766 125323339368000 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_16_zarr3_false/checkpoints/9/items I0416 19:34:13.845691 125430950521984 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 11.131918s I0416 19:34:13.846220 125430950521984 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 1.098 GiB/s (total gbytes: 12.3 GiB) (time elapsed: 11 seconds) (per-host) I0416 19:34:13.846273 125430950521984 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 11.236497s (batch_requests_ready=0.100246s, total_serialization_initiated=11.135833s, others=0.000418s) I0416 19:34:13.846339 125430950521984 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 11.237001s (all_items=0.000012s, per_item={'items': '0.00001240'}, temp_paths=11.236988) I0416 19:34:13.850893 125323339368000 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started. I0416 19:34:13.851002 125430950521984 async_checkpointer.py:561] Finished blocking save. Time taken: 11.335687s. Continuing background save to gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_16_zarr3_false/checkpoints/9. I0416 19:34:13.851199 125430950521984 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize I0416 19:34:13.851353 125323452614208 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0416 19:34:13.851447 125430950521984 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_16_zarr3_false/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776368004.6736646, 'wait_for_prev_duration_secs': 37.841344356536865, 'time_between_consecutive_saves_sec': 12.969007015228271, 'checkpointer_blocking_start_time': 1776368042.5152957, 'checkpointer_blocking_duration_secs': 11.335825204849243, 'get_old_steps_start_time': 1776368053.8511412, 'get_old_steps_duration_secs': 2.2172927856445312e-05, 'checkpoint_manager_blocking_start_time': 1776368004.6736326, 'checkpoint_manager_blocking_duration_secs': 49.17779207229614} I0416 19:34:13.851572 125430950521984 checkpointing.py:409] Started an asynchronous checkpoint save for step 9 I0416 19:34:13.851603 125430950521984 checkpoint_manager.py:1994] [process=0][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0416 19:34:14.607692 125323035280960 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_16_zarr3_false/checkpoints/9/items/array_metadatas/process_0 I0416 19:34:57.199831 125323389699648 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 43.349571s (commit=42.938726s, array_metadata_write=0.410844s) I0416 19:34:57.200932 125323339368000 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 231.481 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 54 seconds) (per-host) I0416 19:34:57.201030 125323339368000 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 43.349985s. I0416 19:34:57.623945 125323339368000 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 19:34:58.093297 125323339368000 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.609403s. use_zarr3=False, 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_16_zarr3_false/checkpoints/9/items I0416 19:34:58.094066 125323339368000 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_16_zarr3_false/checkpoints/9/items I0416 19:34:58.333750 125323339368000 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_16_zarr3_false/checkpoints/9 I0416 19:34:59.029741 125323339368000 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_16_zarr3_false/checkpoints/9`. I0416 19:34:59.030391 125323339368000 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 56.515085s. directory=gs://wanglance-maxtext/linen_ckpt_feat_nnx_post_train_fixes_20260416_181521/linen_feat_nnx_post_train_fixes_20260416_181521_16_zarr3_false/checkpoints/9 I0416 19:34:59.030461 125323339368000 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 45.179416s. I0416 19:34:59.030658 125323452614208 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0416 19:34:59.030768 125323452614208 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0416 19:34:59.030821 125323452614208 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts... I0416 19:34:59.030869 125323452614208 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts. I0416 19:34:59.030999 125430950521984 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 19:34:59.031145 125430950521984 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0416 19:34:59.032264 125430950521984 metric_logger.py:185] completed step: 9, seconds: 0.118, TFLOP/s/device: 115.021, Tokens/s/device: 17337.273, total_weights: 16384, loss: 4.189 Per train step: Total TFLOPs: 13.59 split as 93.93% learnable weight flops and 6.07% attention flops