main (7f7822822) vs feat/nnx-post-train-fixes (2fbc8c2f1)| Metric | main 7f7822822 | feat/nnx-post-train-fixes 2fbc8c2f1 | Diff (feat/nnx-post-train-fixes − main) |
|---|---|---|---|
| Parameters | 1.104 billion | 1.104 billion | — |
| Final loss | 8.1790 | 8.1790 | 0 |
| TFLOP/s | 92.441 | 92.475 | +0.034 |
| Tok/s | 13933.8 | 13938.9 | +5.121 |
| Avg s/step | 3.450 | 2.703 | -0.747 |
| 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: Fri Apr 17 13:02:34 UTC 2026 PyTorch was not found. Models won't be available and only tokenizers, configuration and file/data utilities can be used. 2026-04-17 13:03:13.074350: E external/local_xla/xla/stream_executor/cuda/cuda_platform.cc:51] failed call to cuInit: INTERNAL: CUDA error: Failed call to cuInit: UNKNOWN ERROR (303) I0417 13:03:13.253283 139061378029376 max_utils.py:273] Attempting to initialize the jax distributed system... I0417 13:03:22.295288 139061378029376 distributed.py:172] Connecting to JAX distributed service on mt-01-base-dw36o-slice-job-0-0.mt-01-base-dw36o:8482 I0417 13:03:34.081101 139061378029376 max_utils.py:284] Jax distributed system initialized! I0417 13:03:40.307551 139061378029376 max_utils.py:800] System Information: Jax Version: 0.9.2 I0417 13:03:40.307657 139061378029376 max_utils.py:801] System Information: Jaxlib Version: 0.9.2 I0417 13:03:40.307709 139061378029376 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 I0417 13:03:40.307744 139061378029376 train_utils.py:348] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing. I0417 13:03:41.349306 139061378029376 maxtext_utils.py:1548] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0417 13:03:41.349585 139061378029376 checkpointing.py:677] Setting up checkpoint logger... I0417 13:03:41.349640 139061378029376 checkpointing.py:233] Creating checkpoint manager with ocdbt=True and zarr3=True I0417 13:03:41.349696 139061378029376 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None I0417 13:03:41.350032 139061378029376 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 0x7e79295f3110>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0417 13:03:47.246954 139061378029376 checkpointing.py:265] Enabling policy for fixed interval checkpointing. I0417 13:03:47.247190 139061378029376 checkpoint_manager.py:708] [process=5][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=('items',), item_handlers={'items': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7e77fc161550>}, handler_registry=None I0417 13:03:47.247426 139061378029376 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7e77fc161550>` for item "items" and save args `<class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeSaveArgs'>` and restore args `<class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>` to `_handler_registry`. I0417 13:03:47.247475 139061378029376 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7e77fc160f80>` for item "metrics" and save args `<class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>` and restore args `<class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>` to `_handler_registry`. I0417 13:03:47.247510 139061378029376 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 0x7e77fc161550>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7e77fc161550>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7e77fc160f80>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7e77fc160f80>}). I0417 13:03:47.247847 139061378029376 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34 I0417 13:03:47.247917 139061378029376 async_checkpointer.py:192] [process=5][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7e665c7540e0> timeout: 1200 secs and primary_host=0 for async checkpoint writes I0417 13:03:50.918818 139061378029376 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_main_20260417_125726/linen_xpk_main_20260417_125726_01_base/checkpoints I0417 13:03:50.921055 139061378029376 checkpoint_manager.py:929] [process=5][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_20260417_125726/linen_xpk_main_20260417_125726_01_base/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7e77fc227cb0> I0417 13:03:50.921163 139061378029376 checkpointing.py:301] Checkpoint manager created! I0417 13:03:51.846249 139061378029376 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0417 13:03:51.846359 139061378029376 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0417 13:03:52.224160 139061378029376 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0417 13:03:52.224248 139061378029376 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0417 13:03:52.240210 139061378029376 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0417 13:03:52.240270 139061378029376 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0417 13:03:52.263463 139061378029376 attentions.py:1154] attentions/query Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0417 13:03:52.263527 139061378029376 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0417 13:03:52.279788 139061378029376 attentions.py:1155] attentions/key Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0417 13:03:52.279848 139061378029376 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0417 13:03:52.295936 139061378029376 attentions.py:1156] attentions/value Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0417 13:03:52.295996 139061378029376 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0417 13:03:52.321132 139061378029376 attentions.py:1197] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv'). I0417 13:03:52.321202 139061378029376 attentions.py:1197] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0417 13:03:52.341792 139061378029376 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0417 13:03:52.341857 139061378029376 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None). I0417 13:03:52.554485 139061378029376 checkpointing.py:577] checkpoint manager exists so trying to load this run's existing checkpoint I0417 13:03:52.554589 139061378029376 checkpointing.py:665] No existing checkpoints found, not restoring checkpoint. fsdp: 32 I0417 13:03:53.960474 139061378029376 maxtext_utils.py:1651] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: P('norm',) Physical: (None,) I0417 13:03:53.960595 139061378029376 maxtext_utils.py:1651] params/params/decoder/layers/mlp/wi_0/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0417 13:03:53.960646 139061378029376 maxtext_utils.py:1651] params/params/decoder/layers/mlp/wi_1/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0417 13:03:53.960714 139061378029376 maxtext_utils.py:1651] params/params/decoder/layers/mlp/wo/kernel Shape: float32[7168,16,2048] Logical: P('mlp', 'layers', 'embed') Physical: (None, None, 'fsdp') I0417 13:03:53.960768 139061378029376 maxtext_utils.py:1651] params/params/decoder/layers/post_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0417 13:03:53.960808 139061378029376 maxtext_utils.py:1651] params/params/decoder/layers/pre_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0417 13:03:53.960879 139061378029376 maxtext_utils.py:1651] params/params/decoder/layers/self_attention/key/kernel Shape: float32[2048,16,16,128] Logical: P('embed', 'layers', 'kv_heads', 'kv_head_dim') Physical: ('fsdp', None, None, None) I0417 13:03:53.960931 139061378029376 maxtext_utils.py:1651] params/params/decoder/layers/self_attention/out/kernel Shape: float32[16,16,128,2048] Logical: P('heads', 'layers', 'kv', 'embed') Physical: (None, None, None, 'fsdp') I0417 13:03:53.960973 139061378029376 maxtext_utils.py:1651] params/params/decoder/layers/self_attention/query/kernel Shape: float32[2048,16,16,128] Logical: P('embed', 'layers', 'q_heads', 'kv') Physical: ('fsdp', None, None, None) I0417 13:03:53.961027 139061378029376 maxtext_utils.py:1651] params/params/decoder/layers/self_attention/value/kernel Shape: float32[2048,16,16,128] Logical: P('embed', 'layers', 'kv_heads', 'kv_head_dim') Physical: ('fsdp', None, None, None) I0417 13:03:53.961097 139061378029376 maxtext_utils.py:1651] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: P('embed_vocab', 'vocab') Physical: ('fsdp', None) I0417 13:03:53.961153 139061378029376 maxtext_utils.py:1651] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: P('vocab', 'embed_vocab') Physical: (None, 'fsdp') I0417 13:03:54.445749 139061378029376 train.py:155] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0417 13:03:54.445838 139061378029376 train.py:155] train/xent Physical: float32[32,2048]............................................ ('fsdp', None). I0417 13:03:54.461082 139061378029376 train.py:162] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0417 13:03:54.461144 139061378029376 train.py:162] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None). I0417 13:04:04.700019 139061378029376 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. I0417 13:04:04.700795 139061378029376 metric_logger.py:301] number parameters: 1.104 billion I0417 13:04:15.505250 139061378029376 checkpointing.py:772] Waiting for step 0 to finish before checkpoint... I0417 13:04:16.635451 139061378029376 checkpointing.py:776] Waited 1.1301853656768799 seconds for step 0 to finish before starting checkpointing. I0417 13:04:16.638133 139061378029376 checkpoint_manager.py:2009] [process=5][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0417 13:04:16.640230 139061378029376 checkpoint_manager.py:1512] [process=5] Saving checkpoint at step 0 I0417 13:04:16.641477 139061378029376 event_tracking.py:70] [process=5] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260417_125726/linen_xpk_main_20260417_125726_01_base/checkpoints/0. I0417 13:04:17.834601 139061378029376 signaling_client.py:364] Using JaxDistributedSignalingClient I0417 13:04:17.835660 139061378029376 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0417 13:04:17.835736 139061378029376 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 I0417 13:04:18.112644 139061378029376 base_pytree_checkpoint_handler.py:154] [process=5][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.278116s I0417 13:04:18.112825 139061378029376 base_pytree_checkpoint_handler.py:130] [process=5] /jax/orbax/write/blocking_gbytes_per_sec: 5.437 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.28372764587402344 s) (per-host) I0417 13:04:18.112878 139061378029376 base_pytree_checkpoint_handler.py:768] [process=5][thread=MainThread] Initiated Pytree async_save. Time taken: 0.283791s (batch_requests_ready=0.002327s, total_serialization_initiated=0.281392s, others=0.000072s) I0417 13:04:18.112993 139061378029376 composite_checkpoint_handler.py:715] [process=5][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.287707s (all_items=0.000018s, per_item={'items': '0.00001836'}, temp_paths=0.287689) I0417 13:04:18.113808 139061378029376 event_tracking.py:125] [process=5] [async] Finished blocking save in 1.47 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260417_125726/linen_xpk_main_20260417_125726_01_base/checkpoints/0. I0417 13:04:18.114148 138922593179392 async_checkpointer.py:76] [process=5][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-17 13:24:18.114110 I0417 13:04:18.121369 139061378029376 checkpoint_manager.py:1560] [process=5][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize I0417 13:04:18.121631 138917936359168 async_checkpointer.py:280] [process=5][thread=save_finalize] Waiting for background save thread=async_save. I0417 13:04:18.121755 139061378029376 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260417_125726/linen_xpk_main_20260417_125726_01_base/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776431056.6381137, 'wait_for_prev_duration_secs': 6.008148193359375e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776431056.6402736, 'checkpointer_blocking_duration_secs': 1.4740350246429443, 'get_old_steps_start_time': 1776431058.114334, 'get_old_steps_duration_secs': 3.075599670410156e-05, 'checkpoint_manager_blocking_start_time': 1776431056.636006, 'checkpoint_manager_blocking_duration_secs': 1.485710620880127} I0417 13:04:18.121910 139061378029376 checkpointing.py:408] Started an asynchronous checkpoint save for step 0 I0417 13:04:18.121960 139061378029376 max_utils.py:750] Memstats: After params initialized: I0417 13:04:18.122009 139061378029376 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_18(process=5,(2,4,0,0)) I0417 13:04:18.122040 139061378029376 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_19(process=5,(3,4,0,0)) I0417 13:04:18.122067 139061378029376 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_22(process=5,(2,5,0,0)) I0417 13:04:18.122091 139061378029376 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_23(process=5,(3,5,0,0)) I0417 13:04:18.442850 139061378029376 metric_logger.py:196] completed step: 0, seconds: 10.804, TFLOP/s/device: 1.258, Tokens/s/device: 189.553, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52779.875 I0417 13:04:18.606886 139061378029376 metric_logger.py:196] completed step: 1, seconds: 2.936, TFLOP/s/device: 4.628, Tokens/s/device: 697.525, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52779.875 I0417 13:04:19.026113 139061378029376 metric_logger.py:196] completed step: 2, seconds: 0.017, TFLOP/s/device: 821.272, Tokens/s/device: 123791.103, total_weights: 65536, loss: 10.262, lm_loss: 10.262, perplexity: 28637.129 I0417 13:04:19.173369 139061378029376 metric_logger.py:196] completed step: 3, seconds: 0.420, TFLOP/s/device: 32.383, Tokens/s/device: 4881.048, total_weights: 65536, loss: 9.734, lm_loss: 9.734, perplexity: 16889.203 I0417 13:04:19.468361 139061378029376 metric_logger.py:196] completed step: 4, seconds: 0.153, TFLOP/s/device: 88.825, Tokens/s/device: 13388.596, total_weights: 65536, loss: 9.277, lm_loss: 9.277, perplexity: 10694.614 I0417 13:04:19.474195 139061378029376 metric_logger.py:196] completed step: 5, seconds: 0.147, TFLOP/s/device: 92.291, Tokens/s/device: 13911.153, total_weights: 65536, loss: 8.892, lm_loss: 8.892, perplexity: 7270.717 I0417 13:04:27.339842 2546 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com I0417 13:04:31.526099 138921044875008 array_metadata_store.py:203] [process=5][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260417_125726/linen_xpk_main_20260417_125726_01_base/checkpoints/0/items/array_metadatas/process_5 I0417 13:04:46.404038 139061378029376 metric_logger.py:196] completed step: 6, seconds: 0.296, TFLOP/s/device: 45.975, Tokens/s/device: 6929.876, total_weights: 65536, loss: 8.592, lm_loss: 8.592, perplexity: 5390.759 I0417 13:04:46.551280 139061378029376 metric_logger.py:196] completed step: 7, seconds: 26.783, TFLOP/s/device: 0.507, Tokens/s/device: 76.467, total_weights: 65536, loss: 8.384, lm_loss: 8.384, perplexity: 4376.905 I0417 13:04:46.698525 139061378029376 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 I0417 13:04:46.845354 139061378029376 checkpointing.py:772] Waiting for step 9 to finish before checkpoint... I0417 13:04:46.846024 139061378029376 checkpointing.py:776] Waited 0.000690460205078125 seconds for step 9 to finish before starting checkpointing. I0417 13:04:46.848272 139061378029376 checkpoint_manager.py:2020] [process=5][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0417 13:05:02.283326 138922593179392 base_pytree_checkpoint_handler.py:130] [process=5] /jax/orbax/write/gbytes_per_sec: 35.533 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 44.45418667793274 s) (per-host) I0417 13:05:02.283447 138922593179392 async_checkpointer.py:90] [process=5][thread=async_save] 3 Handler Commit operations completed. Time taken: 44.169186s. I0417 13:05:15.824184 138922593179392 async_checkpointer.py:160] [process=5][thread=async_save] Background save thread done. Time taken: 57.709906s. I0417 13:05:15.824479 138917936359168 async_checkpointer.py:288] [process=5][thread=save_finalize] Done with waiting for background save thread=async_save. I0417 13:05:15.824605 138917936359168 async_checkpointer.py:298] [process=5][thread=save_finalize] No errors found in background save thread=async_save. I0417 13:05:15.824655 138917936359168 checkpoint_manager.py:2137] [process=5][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts... I0417 13:05:15.825946 138917936359168 checkpoint_manager.py:2146] [process=5][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts. I0417 13:05:15.826111 139061378029376 checkpoint_manager.py:2032] [process=5][thread=MainThread][step=0][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=0. W0417 13:05:15.826269 139061378029376 checkpoint_manager.py:1452] Waiting for previous save to complete took 28.977986 seconds. If this number is high, consider checkpointing less frequently. I0417 13:05:15.828040 139061378029376 checkpoint_manager.py:1512] [process=5] Saving checkpoint at step 9 I0417 13:05:15.830111 139061378029376 event_tracking.py:70] [process=5] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260417_125726/linen_xpk_main_20260417_125726_01_base/checkpoints/9. I0417 13:05:16.967411 139061378029376 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0417 13:05:16.967506 139061378029376 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 I0417 13:05:17.000762 139061378029376 base_pytree_checkpoint_handler.py:154] [process=5][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.034511s I0417 13:05:17.000943 139061378029376 base_pytree_checkpoint_handler.py:130] [process=5] /jax/orbax/write/blocking_gbytes_per_sec: 40.541 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.03804945945739746 s) (per-host) I0417 13:05:17.000996 139061378029376 base_pytree_checkpoint_handler.py:768] [process=5][thread=MainThread] Initiated Pytree async_save. Time taken: 0.038113s (batch_requests_ready=0.001903s, total_serialization_initiated=0.036139s, others=0.000070s) I0417 13:05:17.001096 139061378029376 composite_checkpoint_handler.py:715] [process=5][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.041996s (all_items=0.000017s, per_item={'items': '0.00001717'}, temp_paths=0.041979) I0417 13:05:17.001812 139061378029376 event_tracking.py:125] [process=5] [async] Finished blocking save in 1.17 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260417_125726/linen_xpk_main_20260417_125726_01_base/checkpoints/9. I0417 13:05:17.002171 138917936359168 async_checkpointer.py:76] [process=5][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-17 13:25:17.002134 I0417 13:05:17.004183 139061378029376 checkpoint_manager.py:1560] [process=5][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize I0417 13:05:17.004502 138922593179392 async_checkpointer.py:280] [process=5][thread=save_finalize] Waiting for background save thread=async_save. I0417 13:05:17.004669 139061378029376 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260417_125726/linen_xpk_main_20260417_125726_01_base/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776431086.84824, 'wait_for_prev_duration_secs': 28.977985620498657, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776431115.8280845, 'checkpointer_blocking_duration_secs': 1.1742312908172607, 'get_old_steps_start_time': 1776431117.002339, 'get_old_steps_duration_secs': 3.170967102050781e-05, 'checkpoint_manager_blocking_start_time': 1776431086.8462543, 'checkpoint_manager_blocking_duration_secs': 30.158379077911377} I0417 13:05:17.004802 139061378029376 checkpointing.py:408] Started an asynchronous checkpoint save for step 9 I0417 13:05:17.004847 139061378029376 checkpoint_manager.py:2020] [process=5][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0417 13:05:27.830414 138925456738048 array_metadata_store.py:203] [process=5][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260417_125726/linen_xpk_main_20260417_125726_01_base/checkpoints/9/items/array_metadatas/process_5 I0417 13:06:00.443569 138917936359168 base_pytree_checkpoint_handler.py:130] [process=5] /jax/orbax/write/gbytes_per_sec: 36.329 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 43.480637311935425 s) (per-host) I0417 13:06:00.443705 138917936359168 async_checkpointer.py:90] [process=5][thread=async_save] 3 Handler Commit operations completed. Time taken: 43.441418s. I0417 13:06:13.064730 138917936359168 async_checkpointer.py:160] [process=5][thread=async_save] Background save thread done. Time taken: 56.062428s. I0417 13:06:13.065002 138922593179392 async_checkpointer.py:288] [process=5][thread=save_finalize] Done with waiting for background save thread=async_save. I0417 13:06:13.065119 138922593179392 async_checkpointer.py:298] [process=5][thread=save_finalize] No errors found in background save thread=async_save. I0417 13:06:13.065171 138922593179392 checkpoint_manager.py:2137] [process=5][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts... I0417 13:06:13.067596 138922593179392 checkpoint_manager.py:2146] [process=5][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts. I0417 13:06:13.067780 139061378029376 checkpoint_manager.py:2032] [process=5][thread=MainThread][step=9][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=9. I0417 13:06:13.067926 139061378029376 checkpoint_manager.py:2009] [process=5][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0417 13:06:13.068896 139061378029376 metric_logger.py:196] completed step: 9, seconds: 0.147, TFLOP/s/device: 92.441, Tokens/s/device: 13933.774, 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: Fri Apr 17 13:06:25 UTC 2026 EXIT_CODE=0
XPK Start: Fri Apr 17 02:37:42 UTC 2026 PyTorch was not found. Models won't be available and only tokenizers, configuration and file/data utilities can be used. 2026-04-17 02:38:06.491270: E external/local_xla/xla/stream_executor/cuda/cuda_platform.cc:51] failed call to cuInit: INTERNAL: CUDA error: Failed call to cuInit: UNKNOWN ERROR (303) I0417 02:38:06.669523 133052069353280 max_utils.py:273] Attempting to initialize the jax distributed system... I0417 02:38:15.709810 133052069353280 distributed.py:149] Starting JAX distributed service on [::]:8482 I0417 02:38:15.712147 133052069353280 distributed.py:172] Connecting to JAX distributed service on mt-01-base-nulk9-slice-job-0-0.mt-01-base-nulk9:8482 I0417 02:38:17.354006 133052069353280 max_utils.py:284] Jax distributed system initialized! I0417 02:38:23.417498 133052069353280 max_utils.py:800] System Information: Jax Version: 0.9.2 I0417 02:38:23.417599 133052069353280 max_utils.py:801] System Information: Jaxlib Version: 0.9.2 I0417 02:38:23.417640 133052069353280 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 I0417 02:38:23.417676 133052069353280 train_utils.py:377] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing. I0417 02:38:24.454374 133052069353280 maxtext_utils.py:1687] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0417 02:38:24.454995 133052069353280 maxtext_utils.py:1687] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0417 02:38:24.455185 133052069353280 checkpointing.py:688] Setting up checkpoint logger... I0417 02:38:24.455237 133052069353280 checkpointing.py:234] Creating checkpoint manager with ocdbt=True and zarr3=True I0417 02:38:24.455282 133052069353280 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None I0417 02:38:24.455628 133052069353280 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 0x79020cf1cbf0>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0417 02:38:27.992571 133052069353280 checkpointing.py:266] Enabling policy for fixed interval checkpointing. I0417 02:38:27.992808 133052069353280 checkpoint_manager.py:708] [process=0][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=('items',), item_handlers={'items': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7900d5d76c00>}, handler_registry=None I0417 02:38:27.993069 133052069353280 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7900d5d76c00>` for item "items" and save args `<class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeSaveArgs'>` and restore args `<class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>` to `_handler_registry`. I0417 02:38:27.993122 133052069353280 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7900d5d66e40>` for item "metrics" and save args `<class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>` and restore args `<class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>` to `_handler_registry`. I0417 02:38:27.993157 133052069353280 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 0x7900d5d76c00>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7900d5d76c00>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7900d5d66e40>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7900d5d66e40>}). I0417 02:38:27.993477 133052069353280 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34 I0417 02:38:27.993549 133052069353280 async_checkpointer.py:192] [process=0][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7900d5d6cae0> timeout: 1200 secs and primary_host=0 for async checkpoint writes I0417 02:38:30.459330 133052069353280 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_01_base/checkpoints I0417 02:38:30.491004 133052069353280 checkpoint_manager.py:929] [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_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_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_01_base/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7900d5d75a00> I0417 02:38:30.491142 133052069353280 checkpointing.py:302] Checkpoint manager created! I0417 02:38:31.411376 133052069353280 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0417 02:38:31.411482 133052069353280 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0417 02:38:31.792259 133052069353280 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0417 02:38:31.792347 133052069353280 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0417 02:38:31.808605 133052069353280 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0417 02:38:31.808661 133052069353280 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0417 02:38:31.832156 133052069353280 attentions.py:1154] attentions/query Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0417 02:38:31.832221 133052069353280 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0417 02:38:31.848527 133052069353280 attentions.py:1155] attentions/key Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0417 02:38:31.848593 133052069353280 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0417 02:38:31.864892 133052069353280 attentions.py:1156] attentions/value Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0417 02:38:31.864948 133052069353280 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0417 02:38:31.890194 133052069353280 attentions.py:1197] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv'). I0417 02:38:31.890263 133052069353280 attentions.py:1197] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0417 02:38:31.910864 133052069353280 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0417 02:38:31.910927 133052069353280 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None). I0417 02:38:32.123324 133052069353280 checkpointing.py:578] checkpoint manager exists so trying to load this run's existing checkpoint I0417 02:38:32.123433 133052069353280 checkpointing.py:676] No existing checkpoints found, not restoring checkpoint. fsdp: 32 I0417 02:38:33.542709 133052069353280 maxtext_utils.py:1796] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: P('norm',) Physical: (None,) I0417 02:38:33.542833 133052069353280 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) I0417 02:38:33.542902 133052069353280 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) I0417 02:38:33.542964 133052069353280 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') I0417 02:38:33.543018 133052069353280 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) I0417 02:38:33.543057 133052069353280 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) I0417 02:38:33.543114 133052069353280 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) I0417 02:38:33.543171 133052069353280 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') I0417 02:38:33.543213 133052069353280 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) I0417 02:38:33.543251 133052069353280 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) I0417 02:38:33.543298 133052069353280 maxtext_utils.py:1796] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: P('embed_vocab', 'vocab') Physical: ('fsdp', None) I0417 02:38:33.543345 133052069353280 maxtext_utils.py:1796] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: P('vocab', 'embed_vocab') Physical: (None, 'fsdp') I0417 02:38:34.025031 133052069353280 train.py:157] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0417 02:38:34.025126 133052069353280 train.py:157] train/xent Physical: float32[32,2048]............................................ ('fsdp', None). I0417 02:38:34.040536 133052069353280 train.py:164] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0417 02:38:34.040596 133052069353280 train.py:164] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None). I0417 02:38:44.955748 133052069353280 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. I0417 02:38:44.963504 133052069353280 metric_logger.py:289] number parameters: 1.104 billion I0417 02:38:56.627666 133052069353280 checkpointing.py:794] Waiting for step 0 to finish before checkpoint... I0417 02:38:56.780222 133052069353280 checkpointing.py:798] Waited 0.15253543853759766 seconds for step 0 to finish before starting checkpointing. I0417 02:38:56.782921 133052069353280 checkpoint_manager.py:2009] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0417 02:38:56.784658 133052069353280 checkpoint_manager.py:1512] [process=0] Saving checkpoint at step 0 I0417 02:38:56.786057 133052069353280 event_tracking.py:70] [process=0] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_01_base/checkpoints/0. I0417 02:38:57.523011 133052069353280 signaling_client.py:364] Using JaxDistributedSignalingClient I0417 02:38:57.537231 133052069353280 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0417 02:38:57.537297 133052069353280 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 I0417 02:38:57.793804 132923256727296 atomicity.py:140] Creating tmp directory gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_01_base/checkpoints/0 I0417 02:38:57.808743 133052069353280 base_pytree_checkpoint_handler.py:154] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.272183s I0417 02:38:57.809617 133052069353280 base_pytree_checkpoint_handler.py:130] [process=0] /jax/orbax/write/blocking_gbytes_per_sec: 5.540 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.27843403816223145 s) (per-host) I0417 02:38:57.809690 133052069353280 base_pytree_checkpoint_handler.py:768] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 0.278528s (batch_requests_ready=0.002048s, total_serialization_initiated=0.275706s, others=0.000775s) I0417 02:38:57.809763 133052069353280 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.282468s (all_items=0.000017s, per_item={'items': '0.00001693'}, temp_paths=0.282451) I0417 02:38:57.810584 133052069353280 event_tracking.py:125] [process=0] [async] Finished blocking save in 1.03 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_01_base/checkpoints/0. I0417 02:38:57.810890 132923265120000 async_checkpointer.py:76] [process=0][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-17 02:58:57.810856 I0417 02:38:57.831561 133052069353280 checkpoint_manager.py:1560] [process=0][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize I0417 02:38:57.831899 132921117636352 async_checkpointer.py:280] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0417 02:38:57.832059 133052069353280 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_01_base/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776393536.7828956, 'wait_for_prev_duration_secs': 6.580352783203125e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776393536.784697, 'checkpointer_blocking_duration_secs': 1.0263469219207764, 'get_old_steps_start_time': 1776393537.8110723, 'get_old_steps_duration_secs': 3.170967102050781e-05, 'checkpoint_manager_blocking_start_time': 1776393536.7807202, 'checkpoint_manager_blocking_duration_secs': 1.0512988567352295} I0417 02:38:57.832162 133052069353280 checkpointing.py:409] Started an asynchronous checkpoint save for step 0 I0417 02:38:57.832213 133052069353280 max_utils.py:750] Memstats: After params initialized: I0417 02:38:57.832263 133052069353280 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_0(process=0,(0,0,0,0)) I0417 02:38:57.832294 133052069353280 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_1(process=0,(1,0,0,0)) I0417 02:38:57.832320 133052069353280 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_4(process=0,(0,1,0,0)) I0417 02:38:57.832345 133052069353280 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_5(process=0,(1,1,0,0)) I0417 02:38:58.173625 133052069353280 metric_logger.py:185] completed step: 0, seconds: 11.366, TFLOP/s/device: 1.195, Tokens/s/device: 180.194, total_weights: 65536, loss: 10.874 I0417 02:38:58.176815 133052069353280 metric_logger.py:269] To see full metrics 'tensorboard --logdir=gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_01_base/tensorboard/' I0417 02:38:58.593903 133052069353280 metric_logger.py:185] completed step: 1, seconds: 1.544, TFLOP/s/device: 8.798, Tokens/s/device: 1326.085, total_weights: 65536, loss: 10.874 I0417 02:38:58.739986 133052069353280 metric_logger.py:185] completed step: 2, seconds: 0.421, TFLOP/s/device: 32.276, Tokens/s/device: 4865.024, total_weights: 65536, loss: 10.262 I0417 02:38:58.887223 133052069353280 metric_logger.py:185] completed step: 3, seconds: 0.009, TFLOP/s/device: 1530.599, Tokens/s/device: 230708.573, total_weights: 65536, loss: 9.734 I0417 02:39:00.068257 132923239941888 atomicity.py:140] Creating tmp directory gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_01_base/checkpoints/0/items I0417 02:39:00.678925 132921059579648 checkpoint.py:188] Wrote Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776393539822056264, 'commit_timestamp_nsecs': None, 'custom_metadata': {}}, json={"item_handlers": null, "metrics": {}, "performance_metrics": {}, "init_timestamp_nsecs": 1776393539822056264, "commit_timestamp_nsecs": null, "custom_metadata": {}} to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_01_base/checkpoints/0/_CHECKPOINT_METADATA I0417 02:39:01.629701 2859 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com I0417 02:39:03.676762 132922186888960 array_metadata_store.py:203] [process=0][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_01_base/checkpoints/0/items/array_metadatas/process_0 I0417 02:39:20.646878 133052069353280 metric_logger.py:185] completed step: 4, seconds: 0.147, TFLOP/s/device: 92.532, Tokens/s/device: 13947.438, total_weights: 65536, loss: 9.277 I0417 02:39:20.655924 133052069353280 metric_logger.py:185] completed step: 5, seconds: 0.148, TFLOP/s/device: 92.039, Tokens/s/device: 13873.083, total_weights: 65536, loss: 8.892 I0417 02:39:20.801947 133052069353280 metric_logger.py:185] completed step: 6, seconds: 21.759, TFLOP/s/device: 0.624, Tokens/s/device: 94.123, total_weights: 65536, loss: 8.592 I0417 02:39:20.949130 133052069353280 metric_logger.py:185] completed step: 7, seconds: 0.007, TFLOP/s/device: 2063.659, Tokens/s/device: 311057.108, total_weights: 65536, loss: 8.384 I0417 02:39:21.096502 133052069353280 metric_logger.py:185] completed step: 8, seconds: 0.147, TFLOP/s/device: 92.419, Tokens/s/device: 13930.362, total_weights: 65536, loss: 8.255 I0417 02:39:21.243487 133052069353280 checkpointing.py:794] Waiting for step 9 to finish before checkpoint... I0417 02:39:21.244200 133052069353280 checkpointing.py:798] Waited 0.0007331371307373047 seconds for step 9 to finish before starting checkpointing. I0417 02:39:21.246241 133052069353280 checkpoint_manager.py:2020] [process=0][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0417 02:39:37.939470 132922178496256 base_pytree_checkpoint_handler.py:1282] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 36.310302s (commit=35.118887s, array_metadata_write=1.191415s) I0417 02:39:37.940533 132923265120000 base_pytree_checkpoint_handler.py:130] [process=0] /jax/orbax/write/gbytes_per_sec: 39.090 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 40.40930986404419 s) (per-host) I0417 02:39:37.940666 132923265120000 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 40.129657s. I0417 02:39:39.114006 132923265120000 checkpoint.py:228] Read Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776393539822056264, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} from gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_01_base/checkpoints/0/_CHECKPOINT_METADATA I0417 02:39:39.861247 132921059579648 checkpoint.py:247] Updated Metadata={'item_handlers': {'items': 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler'}, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776393539822056264, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_01_base/checkpoints/0/_CHECKPOINT_METADATA I0417 02:39:40.239762 132923265120000 array_metadata_store.py:411] [process=0][thread=async_save] Validated ArrayMetadata from all 8 hosts. Time taken: 0.000246s. I0417 02:39:40.476187 132923265120000 ocdbt_utils.py:49] Param validation support for Zarr3 will be added later (b/362328389). I0417 02:39:41.414711 132923265120000 base_pytree_checkpoint_handler.py:1406] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 2.177618s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_01_base/checkpoints/0/items I0417 02:39:41.416398 132923265120000 atomicity.py:666] Finalizing gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_01_base/checkpoints/0/items I0417 02:39:42.698689 132923265120000 atomicity.py:666] Finalizing gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_01_base/checkpoints/0 I0417 02:39:44.093227 132923265120000 atomicity.py:847] [process=0][thread=async_save] Finished saving checkpoint (finalized tmp dir) to `gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_01_base/checkpoints/0`. I0417 02:39:44.093933 132923265120000 event_tracking.py:138] [process=0] [async] Finished save (blocking + background) in 47.31 seconds @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_01_base/checkpoints/0 I0417 02:39:44.095441 132923265120000 async_checkpointer.py:160] [process=0][thread=async_save] Background save thread done. Time taken: 46.284430s. I0417 02:39:44.095631 132921117636352 async_checkpointer.py:288] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0417 02:39:44.095767 132921117636352 async_checkpointer.py:298] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0417 02:39:44.095866 132921117636352 checkpoint_manager.py:2137] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts... I0417 02:39:44.098066 132921117636352 checkpoint_manager.py:2146] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts. I0417 02:39:44.098211 133052069353280 checkpoint_manager.py:2032] [process=0][thread=MainThread][step=0][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=0. W0417 02:39:44.098343 133052069353280 checkpoint_manager.py:1452] Waiting for previous save to complete took 22.852101 seconds. If this number is high, consider checkpointing less frequently. I0417 02:39:44.100271 133052069353280 checkpoint_manager.py:1512] [process=0] Saving checkpoint at step 9 I0417 02:39:44.102352 133052069353280 event_tracking.py:70] [process=0] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_01_base/checkpoints/9. I0417 02:39:44.391176 133052069353280 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0417 02:39:44.391269 133052069353280 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 I0417 02:39:44.417418 133052069353280 base_pytree_checkpoint_handler.py:154] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.027000s I0417 02:39:44.418303 133052069353280 base_pytree_checkpoint_handler.py:130] [process=0] /jax/orbax/write/blocking_gbytes_per_sec: 48.918 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.03153371810913086 s) (per-host) I0417 02:39:44.418382 133052069353280 base_pytree_checkpoint_handler.py:768] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 0.031633s (batch_requests_ready=0.001801s, total_serialization_initiated=0.029026s, others=0.000806s) I0417 02:39:44.418476 133052069353280 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.035842s (all_items=0.000017s, per_item={'items': '0.00001693'}, temp_paths=0.035825) I0417 02:39:44.419200 133052069353280 event_tracking.py:125] [process=0] [async] Finished blocking save in 0.32 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_01_base/checkpoints/9. I0417 02:39:44.419531 132921117636352 async_checkpointer.py:76] [process=0][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-17 02:59:44.419494 I0417 02:39:44.427651 133052069353280 checkpoint_manager.py:1560] [process=0][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize I0417 02:39:44.427933 132922677851904 async_checkpointer.py:280] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0417 02:39:44.428099 133052069353280 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_01_base/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776393561.2462075, 'wait_for_prev_duration_secs': 22.85210108757019, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776393584.100313, 'checkpointer_blocking_duration_secs': 0.31935882568359375, 'get_old_steps_start_time': 1776393584.419696, 'get_old_steps_duration_secs': 3.1948089599609375e-05, 'checkpoint_manager_blocking_start_time': 1776393561.2444322, 'checkpoint_manager_blocking_duration_secs': 23.183631420135498} I0417 02:39:44.428204 133052069353280 checkpointing.py:409] Started an asynchronous checkpoint save for step 9 I0417 02:39:44.428250 133052069353280 checkpoint_manager.py:2020] [process=0][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0417 02:39:44.622651 132923265120000 atomicity.py:140] Creating tmp directory gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_01_base/checkpoints/9 I0417 02:39:47.253646 132924326328064 atomicity.py:140] Creating tmp directory gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_01_base/checkpoints/9/items I0417 02:39:50.122019 132923239941888 array_metadata_store.py:203] [process=0][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_01_base/checkpoints/9/items/array_metadatas/process_0 I0417 02:40:29.714590 132922686244608 base_pytree_checkpoint_handler.py:1282] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 41.358136s (commit=40.152798s, array_metadata_write=1.205338s) I0417 02:40:29.715730 132921117636352 base_pytree_checkpoint_handler.py:130] [process=0] /jax/orbax/write/gbytes_per_sec: 34.847 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 45.32893371582031 s) (per-host) I0417 02:40:29.715875 132921117636352 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 45.296232s. I0417 02:40:31.857797 132921117636352 ocdbt_utils.py:49] Param validation support for Zarr3 will be added later (b/362328389). I0417 02:40:32.068885 132921117636352 array_metadata_store.py:411] [process=0][thread=async_save] Validated ArrayMetadata from all 8 hosts. Time taken: 0.000229s. I0417 02:40:32.955198 132921117636352 base_pytree_checkpoint_handler.py:1406] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 1.928437s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_01_base/checkpoints/9/items I0417 02:40:32.956779 132921117636352 atomicity.py:666] Finalizing gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_01_base/checkpoints/9/items I0417 02:40:33.929711 132921117636352 atomicity.py:666] Finalizing gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_01_base/checkpoints/9 I0417 02:40:35.821527 132921117636352 atomicity.py:847] [process=0][thread=async_save] Finished saving checkpoint (finalized tmp dir) to `gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_01_base/checkpoints/9`. I0417 02:40:35.822281 132921117636352 event_tracking.py:138] [process=0] [async] Finished save (blocking + background) in 51.72 seconds @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260417_023710/linen_xpk_feat_nnx_post_train_fixes_20260417_023710_01_base/checkpoints/9 I0417 02:40:35.823703 132921117636352 async_checkpointer.py:160] [process=0][thread=async_save] Background save thread done. Time taken: 51.404058s. I0417 02:40:35.823833 132922677851904 async_checkpointer.py:288] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0417 02:40:35.823904 132922677851904 async_checkpointer.py:298] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0417 02:40:35.823963 132922677851904 checkpoint_manager.py:2137] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts... I0417 02:40:35.826738 132922677851904 checkpoint_manager.py:2146] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts. I0417 02:40:35.826931 133052069353280 checkpoint_manager.py:2032] [process=0][thread=MainThread][step=9][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=9. I0417 02:40:35.827084 133052069353280 checkpoint_manager.py:2009] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0417 02:40:35.827775 133052069353280 metric_logger.py:185] completed step: 9, seconds: 0.147, TFLOP/s/device: 92.475, Tokens/s/device: 13938.895, total_weights: 65536, loss: 8.179 Per train step: Total TFLOPs: 13.59 split as 93.93% learnable weight flops and 6.07% attention flops XPK End: Fri Apr 17 02:40:48 UTC 2026 EXIT_CODE=0