main (574ad3fb9) vs feat/nnx-trainstate-and-training-loop (574ad3fb9)| Metric | main 574ad3fb9 | feat/nnx-trainstate-and-training-loop 574ad3fb9 | Diff (feat/nnx-trainstate-and-training-loop − main) |
|---|---|---|---|
| Parameters | 1.104 billion | 1.104 billion | — |
| Final loss | 8.1790 | 8.1790 | 0 |
| TFLOP/s | 92.263 | 92.136 | -0.127 |
| Tok/s | 13906.9 | 13887.8 | -19.144 |
| Avg s/step | 2.871 | 3.180 | +0.309 |
| Memory % | 1.38 | 1.38 | 0 |
| JAX | 0.9.2 | 0.9.2 | — |
Diff = branch value − main value. Green = branch improved. Red = branch regressed.
XPK Start: Sun Apr 19 18:07:18 UTC 2026 PyTorch was not found. Models won't be available and only tokenizers, configuration and file/data utilities can be used. 2026-04-19 18:07:56.075514: 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) I0419 18:07:56.604995 138672836237120 max_utils.py:273] Attempting to initialize the jax distributed system... I0419 18:08:05.645052 138672836237120 distributed.py:149] Starting JAX distributed service on [::]:8482 I0419 18:08:05.647526 138672836237120 distributed.py:172] Connecting to JAX distributed service on mt-01-base-v8sur-slice-job-0-0.mt-01-base-v8sur:8482 I0419 18:08:09.926016 138672836237120 max_utils.py:284] Jax distributed system initialized! I0419 18:08:16.050664 138672836237120 max_utils.py:800] System Information: Jax Version: 0.9.2 I0419 18:08:16.050767 138672836237120 max_utils.py:801] System Information: Jaxlib Version: 0.9.2 I0419 18:08:16.050808 138672836237120 max_utils.py:802] System Information: Jax Backend: PJRT C API TFRT TPU v6 lite Built on Mar 4 2026 11:32:08 (1772652728) cl/878335365 I0419 18:08:16.050842 138672836237120 train_utils.py:348] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing. I0419 18:08:16.746917 138672836237120 maxtext_utils.py:1551] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0419 18:08:16.747186 138672836237120 checkpointing.py:677] Setting up checkpoint logger... I0419 18:08:16.747239 138672836237120 checkpointing.py:233] Creating checkpoint manager with ocdbt=True and zarr3=True I0419 18:08:16.747283 138672836237120 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None I0419 18:08:16.747636 138672836237120 base_pytree_checkpoint_handler.py:441] Created BasePyTreeCheckpointHandler: use_ocdbt=True, use_zarr3=True, pytree_metadata_options=PyTreeMetadataOptions(support_rich_types=False), array_metadata_store=<orbax.checkpoint._src.metadata.array_metadata_store.Store object at 0x7e1e91938e30>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0419 18:08:20.666183 138672836237120 checkpointing.py:265] Enabling policy for fixed interval checkpointing. I0419 18:08:20.666439 138672836237120 checkpoint_manager.py:708] [process=6][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=('items',), item_handlers={'items': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7e09605090d0>}, handler_registry=None I0419 18:08:20.666676 138672836237120 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7e09605090d0>` 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`. I0419 18:08:20.666723 138672836237120 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7e096050c0b0>` 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`. I0419 18:08:20.666758 138672836237120 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 0x7e09605090d0>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7e09605090d0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7e096050c0b0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7e096050c0b0>}). I0419 18:08:20.667066 138672836237120 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34 I0419 18:08:20.667135 138672836237120 async_checkpointer.py:192] [process=6][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7e094019b7e0> timeout: 1200 secs and primary_host=0 for async checkpoint writes I0419 18:08:21.538475 138672836237120 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_main_20260419_180002/linen_xpk_main_20260419_180002_01_base/checkpoints I0419 18:08:21.920105 138672836237120 checkpoint_manager.py:929] [process=6][thread=MainThread] CheckpointManager created, primary_host=0, CheckpointManagerOptions=CheckpointManagerOptions(save_interval_steps=1, max_to_keep=None, keep_time_interval=None, keep_period=None, should_keep_fn=None, best_fn=None, best_mode='max', keep_checkpoints_without_metrics=True, step_prefix=None, step_format_fixed_length=None, step_name_format=None, create=True, cleanup_tmp_directories=False, save_on_steps=frozenset(), single_host_load_and_broadcast=False, todelete_subdir=None, todelete_full_path=None, enable_background_delete=False, read_only=False, enable_async_checkpointing=True, async_options=None, multiprocessing_options=MultiprocessingOptions(primary_host=0, active_processes=None, barrier_sync_key_prefix=None), should_save_fn=None, file_options=FileOptions(path_permission_mode=None), save_root_metadata=True, temporary_path_class=None, save_decision_policy=FixedIntervalPolicy(interval=10), preservation_policy=LatestN(n=None), prevent_write_metrics=False, enable_should_save_is_saving_in_progress_check=True, enable_per_process_directory_creation=False, lightweight_initialize=False), root_directory=gs://lance-maxtext/linen_ckpt_xpk_main_20260419_180002/linen_xpk_main_20260419_180002_01_base/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7e0960509eb0> I0419 18:08:21.920284 138672836237120 checkpointing.py:301] Checkpoint manager created! I0419 18:08:22.853235 138672836237120 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0419 18:08:22.853349 138672836237120 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0419 18:08:23.230689 138672836237120 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0419 18:08:23.230786 138672836237120 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0419 18:08:23.246961 138672836237120 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0419 18:08:23.247020 138672836237120 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0419 18:08:23.271553 138672836237120 attentions.py:1154] attentions/query Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0419 18:08:23.271620 138672836237120 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0419 18:08:23.287872 138672836237120 attentions.py:1155] attentions/key Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0419 18:08:23.287934 138672836237120 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0419 18:08:23.304012 138672836237120 attentions.py:1156] attentions/value Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0419 18:08:23.304074 138672836237120 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0419 18:08:23.329421 138672836237120 attentions.py:1197] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv'). I0419 18:08:23.329489 138672836237120 attentions.py:1197] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0419 18:08:23.350313 138672836237120 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0419 18:08:23.350376 138672836237120 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None). I0419 18:08:23.567147 138672836237120 checkpointing.py:577] checkpoint manager exists so trying to load this run's existing checkpoint I0419 18:08:23.567251 138672836237120 checkpointing.py:665] No existing checkpoints found, not restoring checkpoint. fsdp: 32 I0419 18:08:25.013988 138672836237120 maxtext_utils.py:1654] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: P('norm',) Physical: (None,) I0419 18:08:25.014110 138672836237120 maxtext_utils.py:1654] params/params/decoder/layers/mlp/wi_0/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0419 18:08:25.014161 138672836237120 maxtext_utils.py:1654] params/params/decoder/layers/mlp/wi_1/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0419 18:08:25.014216 138672836237120 maxtext_utils.py:1654] params/params/decoder/layers/mlp/wo/kernel Shape: float32[7168,16,2048] Logical: P('mlp', 'layers', 'embed') Physical: (None, None, 'fsdp') I0419 18:08:25.014286 138672836237120 maxtext_utils.py:1654] params/params/decoder/layers/post_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0419 18:08:25.014326 138672836237120 maxtext_utils.py:1654] params/params/decoder/layers/pre_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0419 18:08:25.014376 138672836237120 maxtext_utils.py:1654] 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) I0419 18:08:25.014439 138672836237120 maxtext_utils.py:1654] 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') I0419 18:08:25.014482 138672836237120 maxtext_utils.py:1654] 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) I0419 18:08:25.014565 138672836237120 maxtext_utils.py:1654] 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) I0419 18:08:25.014641 138672836237120 maxtext_utils.py:1654] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: P('embed_vocab', 'vocab') Physical: ('fsdp', None) I0419 18:08:25.014696 138672836237120 maxtext_utils.py:1654] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: P('vocab', 'embed_vocab') Physical: (None, 'fsdp') I0419 18:08:25.500089 138672836237120 train.py:155] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0419 18:08:25.500185 138672836237120 train.py:155] train/xent Physical: float32[32,2048]............................................ ('fsdp', None). I0419 18:08:25.515475 138672836237120 train.py:162] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0419 18:08:25.515539 138672836237120 train.py:162] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None). I0419 18:08:36.526682 138672836237120 max_utils.py:791] Total memory size: 1.5 GB, Output size: 0.4 GB, Temp size: 1.1 GB, Argument size: 0.4 GB, Host temp size: 0.0 GB. I0419 18:08:36.527509 138672836237120 metric_logger.py:301] number parameters: 1.104 billion I0419 18:08:47.709027 138672836237120 checkpointing.py:772] Waiting for step 0 to finish before checkpoint... I0419 18:08:47.932739 138672836237120 checkpointing.py:776] Waited 0.2236924171447754 seconds for step 0 to finish before starting checkpointing. I0419 18:08:47.935427 138672836237120 checkpoint_manager.py:2009] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0419 18:08:47.937127 138672836237120 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 0 I0419 18:08:47.938612 138672836237120 event_tracking.py:70] [process=6] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260419_180002/linen_xpk_main_20260419_180002_01_base/checkpoints/0. I0419 18:08:48.692350 138672836237120 signaling_client.py:364] Using JaxDistributedSignalingClient I0419 18:08:48.693331 138672836237120 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0419 18:08:48.693386 138672836237120 replica_slices.py:424] Transferring arrays to host memory with options: use_replica_parallel=True, min_slice_bytes_for_replica_parallel=None, max_replicas_for_replica_parallel=None, enable_pinned_host_transfer=False I0419 18:08:48.966021 138672836237120 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.273745s I0419 18:08:48.966201 138672836237120 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 5.525 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.2792081832885742 s) (per-host) I0419 18:08:48.966256 138672836237120 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.279274s (batch_requests_ready=0.002230s, total_serialization_initiated=0.276968s, others=0.000076s) I0419 18:08:48.966347 138672836237120 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.283875s (all_items=0.000019s, per_item={'items': '0.00001931'}, temp_paths=0.283855) I0419 18:08:48.967148 138672836237120 event_tracking.py:125] [process=6] [async] Finished blocking save in 1.03 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260419_180002/linen_xpk_main_20260419_180002_01_base/checkpoints/0. I0419 18:08:48.967509 138543926757120 async_checkpointer.py:76] [process=6][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-19 18:28:48.967471 I0419 18:08:48.983151 138672836237120 checkpoint_manager.py:1560] [process=6][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize I0419 18:08:48.983458 138541812147968 async_checkpointer.py:280] [process=6][thread=save_finalize] Waiting for background save thread=async_save. I0419 18:08:48.983596 138672836237120 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260419_180002/linen_xpk_main_20260419_180002_01_base/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776622127.9353812, 'wait_for_prev_duration_secs': 0.00013113021850585938, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776622127.937174, 'checkpointer_blocking_duration_secs': 1.0304920673370361, 'get_old_steps_start_time': 1776622128.9676905, 'get_old_steps_duration_secs': 3.170967102050781e-05, 'checkpoint_manager_blocking_start_time': 1776622127.9332757, 'checkpoint_manager_blocking_duration_secs': 1.0502796173095703} I0419 18:08:48.983701 138672836237120 checkpointing.py:408] Started an asynchronous checkpoint save for step 0 I0419 18:08:48.983750 138672836237120 max_utils.py:750] Memstats: After params initialized: I0419 18:08:48.983799 138672836237120 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_24(process=6,(0,6,0,0)) I0419 18:08:48.983849 138672836237120 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_25(process=6,(1,6,0,0)) I0419 18:08:48.983877 138672836237120 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_28(process=6,(0,7,0,0)) I0419 18:08:48.983902 138672836237120 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_29(process=6,(1,7,0,0)) I0419 18:08:49.291156 138672836237120 metric_logger.py:196] completed step: 0, seconds: 11.181, TFLOP/s/device: 1.215, Tokens/s/device: 183.161, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52779.875 I0419 18:08:49.469147 138672836237120 metric_logger.py:196] completed step: 1, seconds: 1.581, TFLOP/s/device: 8.596, Tokens/s/device: 1295.737, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52779.875 I0419 18:08:50.167916 138672836237120 metric_logger.py:196] completed step: 2, seconds: 0.031, TFLOP/s/device: 442.794, Tokens/s/device: 66742.708, total_weights: 65536, loss: 10.262, lm_loss: 10.262, perplexity: 28637.129 I0419 18:08:50.315166 138672836237120 metric_logger.py:196] completed step: 3, seconds: 0.699, TFLOP/s/device: 19.439, Tokens/s/device: 2930.055, total_weights: 65536, loss: 9.734, lm_loss: 9.734, perplexity: 16889.203 I0419 18:08:50.610469 138672836237120 metric_logger.py:196] completed step: 4, seconds: 0.153, TFLOP/s/device: 88.681, Tokens/s/device: 13366.925, total_weights: 65536, loss: 9.277, lm_loss: 9.277, perplexity: 10694.614 I0419 18:08:50.616356 138672836237120 metric_logger.py:196] completed step: 5, seconds: 0.147, TFLOP/s/device: 92.342, Tokens/s/device: 13918.717, total_weights: 65536, loss: 8.892, lm_loss: 8.892, perplexity: 7270.717 I0419 18:08:52.395657 2839 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com I0419 18:08:55.364139 138541820540672 array_metadata_store.py:203] [process=6][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260419_180002/linen_xpk_main_20260419_180002_01_base/checkpoints/0/items/array_metadatas/process_6 I0419 18:09:13.393442 138672836237120 metric_logger.py:196] completed step: 6, seconds: 0.296, TFLOP/s/device: 45.951, Tokens/s/device: 6926.266, total_weights: 65536, loss: 8.592, lm_loss: 8.592, perplexity: 5390.759 I0419 18:09:13.540826 138672836237120 metric_logger.py:196] completed step: 7, seconds: 22.630, TFLOP/s/device: 0.600, Tokens/s/device: 90.498, total_weights: 65536, loss: 8.384, lm_loss: 8.384, perplexity: 4376.905 I0419 18:09:13.688110 138672836237120 metric_logger.py:196] completed step: 8, seconds: 0.152, TFLOP/s/device: 89.119, Tokens/s/device: 13433.032, total_weights: 65536, loss: 8.255, lm_loss: 8.255, perplexity: 3846.406 I0419 18:09:13.834777 138672836237120 checkpointing.py:772] Waiting for step 9 to finish before checkpoint... I0419 18:09:13.835472 138672836237120 checkpointing.py:776] Waited 0.0007104873657226562 seconds for step 9 to finish before starting checkpointing. I0419 18:09:13.837510 138672836237120 checkpoint_manager.py:2020] [process=6][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0419 18:09:25.952381 138543926757120 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 42.388 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 37.26535153388977 s) (per-host) I0419 18:09:25.952514 138543926757120 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 36.984892s. I0419 18:09:33.246073 138543926757120 async_checkpointer.py:160] [process=6][thread=async_save] Background save thread done. Time taken: 44.278436s. I0419 18:09:33.246362 138541812147968 async_checkpointer.py:288] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save. I0419 18:09:33.246503 138541812147968 async_checkpointer.py:298] [process=6][thread=save_finalize] No errors found in background save thread=async_save. I0419 18:09:33.246553 138541812147968 checkpoint_manager.py:2137] [process=6][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts... I0419 18:09:33.249221 138541812147968 checkpoint_manager.py:2146] [process=6][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts. I0419 18:09:33.249396 138672836237120 checkpoint_manager.py:2032] [process=6][thread=MainThread][step=0][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=0. W0419 18:09:33.249547 138672836237120 checkpoint_manager.py:1452] Waiting for previous save to complete took 19.412038 seconds. If this number is high, consider checkpointing less frequently. I0419 18:09:33.251334 138672836237120 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 9 I0419 18:09:33.253514 138672836237120 event_tracking.py:70] [process=6] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260419_180002/linen_xpk_main_20260419_180002_01_base/checkpoints/9. I0419 18:09:33.964988 138672836237120 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0419 18:09:33.965081 138672836237120 replica_slices.py:424] Transferring arrays to host memory with options: use_replica_parallel=True, min_slice_bytes_for_replica_parallel=None, max_replicas_for_replica_parallel=None, enable_pinned_host_transfer=False I0419 18:09:34.000187 138672836237120 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.036262s I0419 18:09:34.000343 138672836237120 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 38.840 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.039716243743896484 s) (per-host) I0419 18:09:34.000393 138672836237120 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.039776s (batch_requests_ready=0.001829s, total_serialization_initiated=0.037880s, others=0.000067s) I0419 18:09:34.000503 138672836237120 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.044299s (all_items=0.000015s, per_item={'items': '0.00001502'}, temp_paths=0.044284) I0419 18:09:34.002244 138672836237120 event_tracking.py:125] [process=6] [async] Finished blocking save in 0.75 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260419_180002/linen_xpk_main_20260419_180002_01_base/checkpoints/9. I0419 18:09:34.002584 138541812147968 async_checkpointer.py:76] [process=6][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-19 18:29:34.002546 I0419 18:09:34.005724 138672836237120 checkpoint_manager.py:1560] [process=6][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize I0419 18:09:34.006018 138543431153408 async_checkpointer.py:280] [process=6][thread=save_finalize] Waiting for background save thread=async_save. I0419 18:09:34.007565 138672836237120 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260419_180002/linen_xpk_main_20260419_180002_01_base/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776622153.8374786, 'wait_for_prev_duration_secs': 19.41203784942627, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776622173.251373, 'checkpointer_blocking_duration_secs': 0.7524449825286865, 'get_old_steps_start_time': 1776622174.0038435, 'get_old_steps_duration_secs': 3.266334533691406e-05, 'checkpoint_manager_blocking_start_time': 1776622153.835702, 'checkpoint_manager_blocking_duration_secs': 20.171825170516968} I0419 18:09:34.007688 138672836237120 checkpointing.py:408] Started an asynchronous checkpoint save for step 9 I0419 18:09:34.007738 138672836237120 checkpoint_manager.py:2020] [process=6][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0419 18:09:40.457252 138541820540672 array_metadata_store.py:203] [process=6][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260419_180002/linen_xpk_main_20260419_180002_01_base/checkpoints/9/items/array_metadatas/process_6 I0419 18:10:16.401572 138541812147968 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 37.219 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 42.44090700149536 s) (per-host) I0419 18:10:16.401701 138541812147968 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 42.399004s. I0419 18:10:23.023453 138541812147968 async_checkpointer.py:160] [process=6][thread=async_save] Background save thread done. Time taken: 49.020709s. I0419 18:10:23.023752 138543431153408 async_checkpointer.py:288] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save. I0419 18:10:23.023875 138543431153408 async_checkpointer.py:298] [process=6][thread=save_finalize] No errors found in background save thread=async_save. I0419 18:10:23.023926 138543431153408 checkpoint_manager.py:2137] [process=6][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts... I0419 18:10:23.026190 138543431153408 checkpoint_manager.py:2146] [process=6][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts. I0419 18:10:23.026360 138672836237120 checkpoint_manager.py:2032] [process=6][thread=MainThread][step=9][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=9. I0419 18:10:23.026524 138672836237120 checkpoint_manager.py:2009] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0419 18:10:23.027786 138672836237120 metric_logger.py:196] completed step: 9, seconds: 0.147, TFLOP/s/device: 92.263, Tokens/s/device: 13906.903, total_weights: 65536, loss: 8.179, lm_loss: 8.179, perplexity: 3564.635 Per train step: Total TFLOPs: 13.59 split as 93.93% learnable weight flops and 6.07% attention flops XPK End: Sun Apr 19 18:10:33 UTC 2026 EXIT_CODE=0
574ad3fb9
· feat_nnx_trainstate_and_training_loop_20260419_214525
· full log
XPK Start: Sun Apr 19 21:51:14 UTC 2026 PyTorch was not found. Models won't be available and only tokenizers, configuration and file/data utilities can be used. 2026-04-19 21:51:38.598901: 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) I0419 21:51:38.777647 137961619969856 max_utils.py:273] Attempting to initialize the jax distributed system... I0419 21:51:47.815766 137961619969856 distributed.py:149] Starting JAX distributed service on [::]:8482 I0419 21:51:47.818152 137961619969856 distributed.py:172] Connecting to JAX distributed service on mt-01-base-g0kok-slice-job-0-0.mt-01-base-g0kok:8482 I0419 21:51:51.791029 137961619969856 max_utils.py:284] Jax distributed system initialized! I0419 21:51:57.787080 137961619969856 max_utils.py:800] System Information: Jax Version: 0.9.2 I0419 21:51:57.787180 137961619969856 max_utils.py:801] System Information: Jaxlib Version: 0.9.2 I0419 21:51:57.787220 137961619969856 max_utils.py:802] System Information: Jax Backend: PJRT C API TFRT TPU v6 lite Built on Mar 4 2026 11:32:08 (1772652728) cl/878335365 I0419 21:51:57.787254 137961619969856 train_utils.py:378] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing. I0419 21:51:58.852532 137961619969856 maxtext_utils.py:1718] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0419 21:51:58.853115 137961619969856 maxtext_utils.py:1718] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0419 21:51:58.853294 137961619969856 checkpointing.py:688] Setting up checkpoint logger... I0419 21:51:58.853355 137961619969856 checkpointing.py:234] Creating checkpoint manager with ocdbt=True and zarr3=True I0419 21:51:58.853401 137961619969856 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None I0419 21:51:58.853729 137961619969856 base_pytree_checkpoint_handler.py:441] Created BasePyTreeCheckpointHandler: use_ocdbt=True, use_zarr3=True, pytree_metadata_options=PyTreeMetadataOptions(support_rich_types=False), array_metadata_store=<orbax.checkpoint._src.metadata.array_metadata_store.Store object at 0x7d791cd1c0e0>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0419 21:52:04.010976 137961619969856 checkpointing.py:266] Enabling policy for fixed interval checkpointing. I0419 21:52:04.011210 137961619969856 checkpoint_manager.py:708] [process=4][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=('items',), item_handlers={'items': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7d77ed462510>}, handler_registry=None I0419 21:52:04.011456 137961619969856 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7d77ed462510>` 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`. I0419 21:52:04.011505 137961619969856 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7d77ed4b2e70>` 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`. I0419 21:52:04.011540 137961619969856 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 0x7d77ed462510>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7d77ed462510>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7d77ed4b2e70>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7d77ed4b2e70>}). I0419 21:52:04.011858 137961619969856 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34 I0419 21:52:04.011934 137961619969856 async_checkpointer.py:192] [process=4][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7d64e868f740> timeout: 1200 secs and primary_host=0 for async checkpoint writes I0419 21:52:05.643915 137961619969856 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260419_214525/linen_xpk_feat_nnx_trainstate_and_training_loop_20260419_214525_01_base/checkpoints I0419 21:52:06.050603 137961619969856 checkpoint_manager.py:929] [process=4][thread=MainThread] CheckpointManager created, primary_host=0, CheckpointManagerOptions=CheckpointManagerOptions(save_interval_steps=1, max_to_keep=None, keep_time_interval=None, keep_period=None, should_keep_fn=None, best_fn=None, best_mode='max', keep_checkpoints_without_metrics=True, step_prefix=None, step_format_fixed_length=None, step_name_format=None, create=True, cleanup_tmp_directories=False, save_on_steps=frozenset(), single_host_load_and_broadcast=False, todelete_subdir=None, todelete_full_path=None, enable_background_delete=False, read_only=False, enable_async_checkpointing=True, async_options=None, multiprocessing_options=MultiprocessingOptions(primary_host=0, active_processes=None, barrier_sync_key_prefix=None), should_save_fn=None, file_options=FileOptions(path_permission_mode=None), save_root_metadata=True, temporary_path_class=None, save_decision_policy=FixedIntervalPolicy(interval=10), preservation_policy=LatestN(n=None), prevent_write_metrics=False, enable_should_save_is_saving_in_progress_check=True, enable_per_process_directory_creation=False, lightweight_initialize=False), root_directory=gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260419_214525/linen_xpk_feat_nnx_trainstate_and_training_loop_20260419_214525_01_base/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7d77ed463170> I0419 21:52:06.050780 137961619969856 checkpointing.py:302] Checkpoint manager created! I0419 21:52:06.980591 137961619969856 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0419 21:52:06.980696 137961619969856 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0419 21:52:07.357050 137961619969856 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0419 21:52:07.357138 137961619969856 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0419 21:52:07.373212 137961619969856 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0419 21:52:07.373272 137961619969856 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0419 21:52:07.396740 137961619969856 attentions.py:1154] attentions/query Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0419 21:52:07.396808 137961619969856 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0419 21:52:07.413008 137961619969856 attentions.py:1155] attentions/key Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0419 21:52:07.413067 137961619969856 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0419 21:52:07.429088 137961619969856 attentions.py:1156] attentions/value Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0419 21:52:07.429149 137961619969856 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0419 21:52:07.453469 137961619969856 attentions.py:1197] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv'). I0419 21:52:07.453533 137961619969856 attentions.py:1197] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0419 21:52:07.475116 137961619969856 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0419 21:52:07.475181 137961619969856 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None). I0419 21:52:07.687529 137961619969856 checkpointing.py:578] checkpoint manager exists so trying to load this run's existing checkpoint I0419 21:52:07.687631 137961619969856 checkpointing.py:676] No existing checkpoints found, not restoring checkpoint. fsdp: 32 I0419 21:52:09.106659 137961619969856 maxtext_utils.py:1821] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: P('norm',) Physical: (None,) I0419 21:52:09.106782 137961619969856 maxtext_utils.py:1821] params/params/decoder/layers/mlp/wi_0/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0419 21:52:09.106835 137961619969856 maxtext_utils.py:1821] params/params/decoder/layers/mlp/wi_1/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0419 21:52:09.106924 137961619969856 maxtext_utils.py:1821] params/params/decoder/layers/mlp/wo/kernel Shape: float32[7168,16,2048] Logical: P('mlp', 'layers', 'embed') Physical: (None, None, 'fsdp') I0419 21:52:09.106979 137961619969856 maxtext_utils.py:1821] params/params/decoder/layers/post_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0419 21:52:09.107022 137961619969856 maxtext_utils.py:1821] params/params/decoder/layers/pre_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0419 21:52:09.107075 137961619969856 maxtext_utils.py:1821] params/params/decoder/layers/self_attention/key/kernel Shape: float32[2048,16,16,128] Logical: P('embed', 'layers', 'kv_heads', 'kv_head_dim') Physical: ('fsdp', None, None, None) I0419 21:52:09.107129 137961619969856 maxtext_utils.py:1821] params/params/decoder/layers/self_attention/out/kernel Shape: float32[16,16,128,2048] Logical: P('heads', 'layers', 'kv', 'embed') Physical: (None, None, None, 'fsdp') I0419 21:52:09.107170 137961619969856 maxtext_utils.py:1821] params/params/decoder/layers/self_attention/query/kernel Shape: float32[2048,16,16,128] Logical: P('embed', 'layers', 'q_heads', 'kv') Physical: ('fsdp', None, None, None) I0419 21:52:09.107205 137961619969856 maxtext_utils.py:1821] params/params/decoder/layers/self_attention/value/kernel Shape: float32[2048,16,16,128] Logical: P('embed', 'layers', 'kv_heads', 'kv_head_dim') Physical: ('fsdp', None, None, None) I0419 21:52:09.107253 137961619969856 maxtext_utils.py:1821] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: P('embed_vocab', 'vocab') Physical: ('fsdp', None) I0419 21:52:09.107301 137961619969856 maxtext_utils.py:1821] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: P('vocab', 'embed_vocab') Physical: (None, 'fsdp') I0419 21:52:09.592656 137961619969856 train.py:157] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0419 21:52:09.592747 137961619969856 train.py:157] train/xent Physical: float32[32,2048]............................................ ('fsdp', None). I0419 21:52:09.607903 137961619969856 train.py:164] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0419 21:52:09.607960 137961619969856 train.py:164] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None). I0419 21:52:20.672772 137961619969856 max_utils.py:791] Total memory size: 1.5 GB, Output size: 0.4 GB, Temp size: 1.1 GB, Argument size: 0.4 GB, Host temp size: 0.0 GB. I0419 21:52:20.673555 137961619969856 metric_logger.py:301] number parameters: 1.104 billion I0419 21:52:31.893786 137961619969856 checkpointing.py:794] Waiting for step 0 to finish before checkpoint... I0419 21:52:32.048425 137961619969856 checkpointing.py:798] Waited 0.1546192169189453 seconds for step 0 to finish before starting checkpointing. I0419 21:52:32.050823 137961619969856 checkpoint_manager.py:2009] [process=4][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0419 21:52:32.052570 137961619969856 checkpoint_manager.py:1512] [process=4] Saving checkpoint at step 0 I0419 21:52:32.054075 137961619969856 event_tracking.py:70] [process=4] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260419_214525/linen_xpk_feat_nnx_trainstate_and_training_loop_20260419_214525_01_base/checkpoints/0. I0419 21:52:32.802135 137961619969856 signaling_client.py:364] Using JaxDistributedSignalingClient I0419 21:52:32.803116 137961619969856 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0419 21:52:32.803173 137961619969856 replica_slices.py:424] Transferring arrays to host memory with options: use_replica_parallel=True, min_slice_bytes_for_replica_parallel=None, max_replicas_for_replica_parallel=None, enable_pinned_host_transfer=False I0419 21:52:33.074609 137961619969856 base_pytree_checkpoint_handler.py:154] [process=4][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.272545s I0419 21:52:33.074784 137961619969856 base_pytree_checkpoint_handler.py:130] [process=4] /jax/orbax/write/blocking_gbytes_per_sec: 5.546 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.2781205177307129 s) (per-host) I0419 21:52:33.074844 137961619969856 base_pytree_checkpoint_handler.py:768] [process=4][thread=MainThread] Initiated Pytree async_save. Time taken: 0.278190s (batch_requests_ready=0.002315s, total_serialization_initiated=0.275799s, others=0.000076s) I0419 21:52:33.074933 137961619969856 composite_checkpoint_handler.py:715] [process=4][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.282506s (all_items=0.000018s, per_item={'items': '0.00001812'}, temp_paths=0.282488) I0419 21:52:33.075721 137961619969856 event_tracking.py:125] [process=4] [async] Finished blocking save in 1.02 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260419_214525/linen_xpk_feat_nnx_trainstate_and_training_loop_20260419_214525_01_base/checkpoints/0. I0419 21:52:33.076006 137833687807744 async_checkpointer.py:76] [process=4][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-19 22:12:33.075974 I0419 21:52:33.087162 137961619969856 checkpoint_manager.py:1560] [process=4][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize I0419 21:52:33.087428 137832034416384 async_checkpointer.py:280] [process=4][thread=save_finalize] Waiting for background save thread=async_save. I0419 21:52:33.087559 137961619969856 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260419_214525/linen_xpk_feat_nnx_trainstate_and_training_loop_20260419_214525_01_base/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776635552.0508058, 'wait_for_prev_duration_secs': 5.984306335449219e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776635552.052608, 'checkpointer_blocking_duration_secs': 1.0235517024993896, 'get_old_steps_start_time': 1776635553.076185, 'get_old_steps_duration_secs': 3.0040740966796875e-05, 'checkpoint_manager_blocking_start_time': 1776635552.0489168, 'checkpoint_manager_blocking_duration_secs': 1.0385994911193848} I0419 21:52:33.087666 137961619969856 checkpointing.py:409] Started an asynchronous checkpoint save for step 0 I0419 21:52:33.087716 137961619969856 max_utils.py:750] Memstats: After params initialized: I0419 21:52:33.087766 137961619969856 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_16(process=4,(0,4,0,0)) I0419 21:52:33.087803 137961619969856 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_17(process=4,(1,4,0,0)) I0419 21:52:33.087830 137961619969856 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_20(process=4,(0,5,0,0)) I0419 21:52:33.087853 137961619969856 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_21(process=4,(1,5,0,0)) I0419 21:52:33.397098 137961619969856 metric_logger.py:196] completed step: 0, seconds: 11.220, TFLOP/s/device: 1.211, Tokens/s/device: 182.529, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52779.875 I0419 21:52:33.579094 137961619969856 metric_logger.py:196] completed step: 1, seconds: 1.502, TFLOP/s/device: 9.047, Tokens/s/device: 1363.670, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52779.875 I0419 21:52:33.967857 137961619969856 metric_logger.py:196] completed step: 2, seconds: 0.035, TFLOP/s/device: 391.989, Tokens/s/device: 59084.877, total_weights: 65536, loss: 10.262, lm_loss: 10.262, perplexity: 28637.129 I0419 21:52:34.115384 137961619969856 metric_logger.py:196] completed step: 3, seconds: 0.389, TFLOP/s/device: 34.922, Tokens/s/device: 5263.821, total_weights: 65536, loss: 9.734, lm_loss: 9.734, perplexity: 16889.203 I0419 21:52:34.410776 137961619969856 metric_logger.py:196] completed step: 4, seconds: 0.153, TFLOP/s/device: 88.804, Tokens/s/device: 13385.533, total_weights: 65536, loss: 9.277, lm_loss: 9.277, perplexity: 10694.614 I0419 21:52:34.416518 137961619969856 metric_logger.py:196] completed step: 5, seconds: 0.148, TFLOP/s/device: 92.109, Tokens/s/device: 13883.616, total_weights: 65536, loss: 8.892, lm_loss: 8.892, perplexity: 7270.717 I0419 21:52:38.613190 2836 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com I0419 21:52:43.244442 137832074950400 array_metadata_store.py:203] [process=4][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260419_214525/linen_xpk_feat_nnx_trainstate_and_training_loop_20260419_214525_01_base/checkpoints/0/items/array_metadatas/process_4 I0419 21:53:00.359386 137961619969856 metric_logger.py:196] completed step: 6, seconds: 0.296, TFLOP/s/device: 45.962, Tokens/s/device: 6927.953, total_weights: 65536, loss: 8.592, lm_loss: 8.592, perplexity: 5390.759 I0419 21:53:00.506978 137961619969856 metric_logger.py:196] completed step: 7, seconds: 25.796, TFLOP/s/device: 0.527, Tokens/s/device: 79.392, total_weights: 65536, loss: 8.384, lm_loss: 8.384, perplexity: 4376.905 I0419 21:53:00.654398 137961619969856 metric_logger.py:196] completed step: 8, seconds: 0.152, TFLOP/s/device: 89.145, Tokens/s/device: 13436.910, total_weights: 65536, loss: 8.255, lm_loss: 8.255, perplexity: 3846.406 I0419 21:53:00.801179 137961619969856 checkpointing.py:794] Waiting for step 9 to finish before checkpoint... I0419 21:53:00.801780 137961619969856 checkpointing.py:798] Waited 0.0006191730499267578 seconds for step 9 to finish before starting checkpointing. I0419 21:53:00.804052 137961619969856 checkpoint_manager.py:2020] [process=4][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0419 21:53:11.791727 137833687807744 base_pytree_checkpoint_handler.py:130] [process=4] /jax/orbax/write/gbytes_per_sec: 40.508 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 38.99502396583557 s) (per-host) I0419 21:53:11.791853 137833687807744 async_checkpointer.py:90] [process=4][thread=async_save] 3 Handler Commit operations completed. Time taken: 38.715736s. I0419 21:53:23.544914 137833687807744 async_checkpointer.py:160] [process=4][thread=async_save] Background save thread done. Time taken: 50.468780s. I0419 21:53:23.545133 137832034416384 async_checkpointer.py:288] [process=4][thread=save_finalize] Done with waiting for background save thread=async_save. I0419 21:53:23.545196 137832034416384 async_checkpointer.py:298] [process=4][thread=save_finalize] No errors found in background save thread=async_save. I0419 21:53:23.545246 137832034416384 checkpoint_manager.py:2137] [process=4][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts... I0419 21:53:23.547196 137832034416384 checkpoint_manager.py:2146] [process=4][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts. I0419 21:53:23.547369 137961619969856 checkpoint_manager.py:2032] [process=4][thread=MainThread][step=0][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=0. W0419 21:53:23.547505 137961619969856 checkpoint_manager.py:1452] Waiting for previous save to complete took 22.743452 seconds. If this number is high, consider checkpointing less frequently. I0419 21:53:23.549378 137961619969856 checkpoint_manager.py:1512] [process=4] Saving checkpoint at step 9 I0419 21:53:23.551219 137961619969856 event_tracking.py:70] [process=4] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260419_214525/linen_xpk_feat_nnx_trainstate_and_training_loop_20260419_214525_01_base/checkpoints/9. I0419 21:53:24.710850 137961619969856 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0419 21:53:24.710945 137961619969856 replica_slices.py:424] Transferring arrays to host memory with options: use_replica_parallel=True, min_slice_bytes_for_replica_parallel=None, max_replicas_for_replica_parallel=None, enable_pinned_host_transfer=False I0419 21:53:24.750691 137961619969856 base_pytree_checkpoint_handler.py:154] [process=4][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.040999s I0419 21:53:24.750862 137961619969856 base_pytree_checkpoint_handler.py:130] [process=4] /jax/orbax/write/blocking_gbytes_per_sec: 34.608 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.04457235336303711 s) (per-host) I0419 21:53:24.750910 137961619969856 base_pytree_checkpoint_handler.py:768] [process=4][thread=MainThread] Initiated Pytree async_save. Time taken: 0.044631s (batch_requests_ready=0.001911s, total_serialization_initiated=0.042655s, others=0.000065s) I0419 21:53:24.750998 137961619969856 composite_checkpoint_handler.py:715] [process=4][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.048843s (all_items=0.000016s, per_item={'items': '0.00001597'}, temp_paths=0.048827) I0419 21:53:24.751723 137961619969856 event_tracking.py:125] [process=4] [async] Finished blocking save in 1.20 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260419_214525/linen_xpk_feat_nnx_trainstate_and_training_loop_20260419_214525_01_base/checkpoints/9. I0419 21:53:24.752012 137832034416384 async_checkpointer.py:76] [process=4][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-19 22:13:24.751980 I0419 21:53:24.753971 137961619969856 checkpoint_manager.py:1560] [process=4][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize I0419 21:53:24.754243 137811315345152 async_checkpointer.py:280] [process=4][thread=save_finalize] Waiting for background save thread=async_save. I0419 21:53:24.754407 137961619969856 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260419_214525/linen_xpk_feat_nnx_trainstate_and_training_loop_20260419_214525_01_base/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776635580.804022, 'wait_for_prev_duration_secs': 22.743452310562134, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776635603.5494173, 'checkpointer_blocking_duration_secs': 1.202737808227539, 'get_old_steps_start_time': 1776635604.752178, 'get_old_steps_duration_secs': 2.9802322387695312e-05, 'checkpoint_manager_blocking_start_time': 1776635580.8020105, 'checkpoint_manager_blocking_duration_secs': 23.95236325263977} I0419 21:53:24.754514 137961619969856 checkpointing.py:409] Started an asynchronous checkpoint save for step 9 I0419 21:53:24.754569 137961619969856 checkpoint_manager.py:2020] [process=4][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0419 21:53:33.421659 137832074950400 array_metadata_store.py:203] [process=4][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_trainstate_and_training_loop_20260419_214525/linen_xpk_feat_nnx_trainstate_and_training_loop_20260419_214525_01_base/checkpoints/9/items/array_metadatas/process_4 I0419 21:54:09.305633 137832034416384 base_pytree_checkpoint_handler.py:130] [process=4] /jax/orbax/write/gbytes_per_sec: 35.418 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 44.59930467605591 s) (per-host) I0419 21:54:09.305763 137832034416384 async_checkpointer.py:90] [process=4][thread=async_save] 3 Handler Commit operations completed. Time taken: 44.553639s. I0419 21:54:19.436594 137832034416384 async_checkpointer.py:160] [process=4][thread=async_save] Background save thread done. Time taken: 54.684454s. I0419 21:54:19.436888 137811315345152 async_checkpointer.py:288] [process=4][thread=save_finalize] Done with waiting for background save thread=async_save. I0419 21:54:19.437014 137811315345152 async_checkpointer.py:298] [process=4][thread=save_finalize] No errors found in background save thread=async_save. I0419 21:54:19.437078 137811315345152 checkpoint_manager.py:2137] [process=4][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts... I0419 21:54:19.439039 137811315345152 checkpoint_manager.py:2146] [process=4][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts. I0419 21:54:19.439194 137961619969856 checkpoint_manager.py:2032] [process=4][thread=MainThread][step=9][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=9. I0419 21:54:19.439357 137961619969856 checkpoint_manager.py:2009] [process=4][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0419 21:54:19.440303 137961619969856 metric_logger.py:196] completed step: 9, seconds: 0.147, TFLOP/s/device: 92.136, Tokens/s/device: 13887.759, total_weights: 65536, loss: 8.179, lm_loss: 8.179, perplexity: 3564.635 Per train step: Total TFLOPs: 13.59 split as 93.93% learnable weight flops and 6.07% attention flops XPK End: Sun Apr 19 21:54:29 UTC 2026 EXIT_CODE=0