MaxView

‹ 06_grad_accumCase: 07_eval08_checkpoint_async_true_resume ›

Metrics: Linen vs NNX  ·  feat/nnx-post-train-fixes

MetricLinen  574ad3fb9NNX  574ad3fb9Diff (NNX − Linen)
Parameters1.104 billion1.104 billion
Final loss8.17908.1750-0.004
TFLOP/s91.92391.645-0.278
Tok/s13855.613813.8-41.868
Avg s/step2.7402.847+0.107
Memory %1.381.380
JAX0.9.20.9.2

Diff = NNX value − Linen value. Green = NNX improved. Red = NNX regressed.

Linen  ·  574ad3fb9  ·  feat_nnx_post_train_fixes_20260418_235007  ·  full log
XPK Start: Sun Apr 19 00:04:26 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 00:04:50.908198: 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 00:04:51.085818 140423904401216 max_utils.py:273] Attempting to initialize the jax distributed system...
I0419 00:05:00.126609 140423904401216 distributed.py:149] Starting JAX distributed service on [::]:8482
I0419 00:05:00.128962 140423904401216 distributed.py:172] Connecting to JAX distributed service on mt-07-eval-fz718-slice-job-0-0.mt-07-eval-fz718:8482
I0419 00:05:02.556923 140423904401216 max_utils.py:284] Jax distributed system initialized!
I0419 00:05:08.134465 140423904401216 max_utils.py:800] System Information: Jax Version: 0.9.2
I0419 00:05:08.134568 140423904401216 max_utils.py:801] System Information: Jaxlib Version: 0.9.2
I0419 00:05:08.134609 140423904401216 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 00:05:08.134642 140423904401216 train_utils.py:378] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing.
I0419 00:05:09.180892 140423904401216 maxtext_utils.py:1718] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0419 00:05:09.181474 140423904401216 maxtext_utils.py:1718] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0419 00:05:09.181651 140423904401216 checkpointing.py:688] Setting up checkpoint logger...
I0419 00:05:09.181701 140423904401216 checkpointing.py:234] Creating checkpoint manager with ocdbt=True and zarr3=True
I0419 00:05:09.181746 140423904401216 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None
I0419 00:05:09.182094 140423904401216 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 0x7fb645350710>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB)
I0419 00:05:12.130760 140423904401216 checkpointing.py:266] Enabling policy for fixed interval checkpointing.
I0419 00:05:12.131032 140423904401216 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 0x7fb5397b2d20>}, handler_registry=None
I0419 00:05:12.131264 140423904401216 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7fb5397b2d20>` 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 00:05:12.131313 140423904401216 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7fb5397b0890>` 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 00:05:12.131352 140423904401216 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 0x7fb5397b2d20>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7fb5397b2d20>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7fb5397b0890>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7fb5397b0890>}).
I0419 00:05:12.131662 140423904401216 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34
I0419 00:05:12.131732 140423904401216 async_checkpointer.py:192] [process=0][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7fa2882d2c00> timeout: 1200 secs and primary_host=0 for async checkpoint writes
I0419 00:05:13.581475 140423904401216 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_07_eval/checkpoints
I0419 00:05:14.445193 140423904401216 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_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_07_eval/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7fb539798c50>
I0419 00:05:14.445370 140423904401216 checkpointing.py:302] Checkpoint manager created!
I0419 00:05:15.376318 140423904401216 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed').
I0419 00:05:15.376424 140423904401216 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0419 00:05:15.760650 140423904401216 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0419 00:05:15.760739 140423904401216 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0419 00:05:15.777049 140423904401216 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0419 00:05:15.777111 140423904401216 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0419 00:05:15.800765 140423904401216 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 00:05:15.800845 140423904401216 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0419 00:05:15.817205 140423904401216 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 00:05:15.817266 140423904401216 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0419 00:05:15.833467 140423904401216 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 00:05:15.833529 140423904401216 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0419 00:05:15.858143 140423904401216 attentions.py:1197] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv').
I0419 00:05:15.858213 140423904401216 attentions.py:1197] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0419 00:05:15.880023 140423904401216 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp').
I0419 00:05:15.880089 140423904401216 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None).
I0419 00:05:16.093240 140423904401216 checkpointing.py:578] checkpoint manager exists so trying to load this run's existing checkpoint
I0419 00:05:16.093354 140423904401216 checkpointing.py:676] No existing checkpoints found, not restoring checkpoint.
fsdp: 32
I0419 00:05:17.522315 140423904401216 maxtext_utils.py:1827]  params/params/decoder/decoder_norm/scale
    Shape:     float32[2048]
    Logical:   P('norm',)
    Physical:  (None,)
I0419 00:05:17.522439 140423904401216 maxtext_utils.py:1827]  params/params/decoder/layers/mlp/wi_0/kernel
    Shape:     float32[2048,16,7168]
    Logical:   P('embed', 'layers', 'mlp')
    Physical:  ('fsdp', None, None)
I0419 00:05:17.522492 140423904401216 maxtext_utils.py:1827]  params/params/decoder/layers/mlp/wi_1/kernel
    Shape:     float32[2048,16,7168]
    Logical:   P('embed', 'layers', 'mlp')
    Physical:  ('fsdp', None, None)
I0419 00:05:17.522550 140423904401216 maxtext_utils.py:1827]  params/params/decoder/layers/mlp/wo/kernel
    Shape:     float32[7168,16,2048]
    Logical:   P('mlp', 'layers', 'embed')
    Physical:  (None, None, 'fsdp')
I0419 00:05:17.522633 140423904401216 maxtext_utils.py:1827]  params/params/decoder/layers/post_self_attention_layer_norm/scale
    Shape:     float32[2048,16]
    Logical:   P('norm', 'layers')
    Physical:  (None, None)
I0419 00:05:17.522702 140423904401216 maxtext_utils.py:1827]  params/params/decoder/layers/pre_self_attention_layer_norm/scale
    Shape:     float32[2048,16]
    Logical:   P('norm', 'layers')
    Physical:  (None, None)
I0419 00:05:17.522764 140423904401216 maxtext_utils.py:1827]  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 00:05:17.522850 140423904401216 maxtext_utils.py:1827]  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 00:05:17.522897 140423904401216 maxtext_utils.py:1827]  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 00:05:17.522935 140423904401216 maxtext_utils.py:1827]  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 00:05:17.522984 140423904401216 maxtext_utils.py:1827]  params/params/decoder/logits_dense/kernel
    Shape:     float32[2048,32000]
    Logical:   P('embed_vocab', 'vocab')
    Physical:  ('fsdp', None)

I0419 00:05:17.523034 140423904401216 maxtext_utils.py:1827]  params/params/token_embedder/embedding
    Shape:     float32[32000,2048]
    Logical:   P('vocab', 'embed_vocab')
    Physical:  (None, 'fsdp')
I0419 00:05:18.024748 140423904401216 train.py:157] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0419 00:05:18.024857 140423904401216 train.py:157] train/xent Physical: float32[32,2048]............................................ ('fsdp', None).
I0419 00:05:18.040288 140423904401216 train.py:164] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length').
I0419 00:05:18.040352 140423904401216 train.py:164] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None).
I0419 00:05:28.717866 140423904401216 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 00:05:28.725754 140423904401216 metric_logger.py:301] number parameters: 1.104 billion
I0419 00:05:40.307365 140423904401216 checkpointing.py:794] Waiting for step 0 to finish before checkpoint...
I0419 00:05:40.461311 140423904401216 checkpointing.py:798] Waited 0.15392684936523438 seconds for step 0 to finish before starting checkpointing.
I0419 00:05:40.463671 140423904401216 checkpoint_manager.py:2009] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0419 00:05:40.465467 140423904401216 checkpoint_manager.py:1512] [process=0] Saving checkpoint at step 0
I0419 00:05:40.466928 140423904401216 event_tracking.py:70] [process=0] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_07_eval/checkpoints/0.
I0419 00:05:41.582206 140423904401216 signaling_client.py:364] Using JaxDistributedSignalingClient
I0419 00:05:41.596915 140423904401216 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array.
I0419 00:05:41.596982 140423904401216 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 00:05:41.863335 140296172070656 atomicity.py:140] Creating tmp directory gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_07_eval/checkpoints/0
I0419 00:05:41.865537 140423904401216 base_pytree_checkpoint_handler.py:154] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.269449s
I0419 00:05:41.866366 140423904401216 base_pytree_checkpoint_handler.py:130] [process=0] /jax/orbax/write/blocking_gbytes_per_sec: 5.596 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.27565598487854004 s) (per-host)
I0419 00:05:41.866440 140423904401216 base_pytree_checkpoint_handler.py:768] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 0.275749s (batch_requests_ready=0.002079s, total_serialization_initiated=0.272919s, others=0.000752s)
I0419 00:05:41.866518 140423904401216 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.279824s (all_items=0.000018s, per_item={'items': '0.00001836'}, temp_paths=0.279806)
I0419 00:05:41.867249 140423904401216 event_tracking.py:125] [process=0] [async] Finished blocking save in 1.40 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_07_eval/checkpoints/0.
I0419 00:05:41.867587 140296180463360 async_checkpointer.py:76] [process=0][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-19 00:25:41.867549
I0419 00:05:41.884166 140423904401216 checkpoint_manager.py:1560] [process=0][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize
I0419 00:05:41.884442 140295593932544 async_checkpointer.py:280] [process=0][thread=save_finalize] Waiting for background save thread=async_save.
I0419 00:05:41.884583 140423904401216 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_07_eval/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776557140.4636521, 'wait_for_prev_duration_secs': 5.984306335449219e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776557140.465506, 'checkpointer_blocking_duration_secs': 1.4022266864776611, 'get_old_steps_start_time': 1776557141.8677568, 'get_old_steps_duration_secs': 2.6464462280273438e-05, 'checkpoint_manager_blocking_start_time': 1776557140.4617107, 'checkpoint_manager_blocking_duration_secs': 1.4228339195251465}
I0419 00:05:41.884683 140423904401216 checkpointing.py:409] Started an asynchronous checkpoint save for step 0
I0419 00:05:41.884733 140423904401216 max_utils.py:750] 
Memstats: After params initialized:
I0419 00:05:41.884784 140423904401216 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_0(process=0,(0,0,0,0))
I0419 00:05:41.884831 140423904401216 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_1(process=0,(1,0,0,0))
I0419 00:05:41.884860 140423904401216 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_4(process=0,(0,1,0,0))
I0419 00:05:41.884886 140423904401216 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_5(process=0,(1,1,0,0))
I0419 00:05:42.231190 140423904401216 metric_logger.py:196] completed step: 0, seconds: 11.281, TFLOP/s/device: 1.204, Tokens/s/device: 181.543, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52779.875
I0419 00:05:42.234689 140423904401216 metric_logger.py:281] To see full metrics 'tensorboard --logdir=gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_07_eval/tensorboard/'
I0419 00:05:42.642592 140423904401216 metric_logger.py:196] completed step: 1, seconds: 1.922, TFLOP/s/device: 7.070, Tokens/s/device: 1065.594, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52779.875
I0419 00:05:42.788592 140423904401216 metric_logger.py:196] completed step: 2, seconds: 0.412, TFLOP/s/device: 32.965, Tokens/s/device: 4968.896, total_weights: 65536, loss: 10.262, lm_loss: 10.262, perplexity: 28637.129
I0419 00:05:42.936012 140423904401216 metric_logger.py:196] completed step: 3, seconds: 0.009, TFLOP/s/device: 1501.672, Tokens/s/device: 226348.364, total_weights: 65536, loss: 9.734, lm_loss: 9.734, perplexity: 16889.203
I0419 00:05:44.075495 140296155285248 atomicity.py:140] Creating tmp directory gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_07_eval/checkpoints/0/items
I0419 00:05:44.218377 140293438052096 checkpoint.py:188] Wrote Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776557143794076602, 'commit_timestamp_nsecs': None, 'custom_metadata': {}}, json={"item_handlers": null, "metrics": {}, "performance_metrics": {}, "init_timestamp_nsecs": 1776557143794076602, "commit_timestamp_nsecs": null, "custom_metadata": {}} to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_07_eval/checkpoints/0/_CHECKPOINT_METADATA
I0419 00:05:46.096618    2837 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com
I0419 00:05:48.738622 140296126740224 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_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_07_eval/checkpoints/0/items/array_metadatas/process_0
I0419 00:06:04.655231 140423904401216 metric_logger.py:196] completed step: 4, seconds: 0.147, TFLOP/s/device: 92.516, Tokens/s/device: 13944.969, total_weights: 65536, loss: 9.277, lm_loss: 9.277, perplexity: 10694.614
I0419 00:06:04.664239 140423904401216 metric_logger.py:196] completed step: 5, seconds: 0.148, TFLOP/s/device: 92.086, Tokens/s/device: 13880.229, total_weights: 65536, loss: 8.892, lm_loss: 8.892, perplexity: 7270.717
I0419 00:06:04.810370 140423904401216 metric_logger.py:196] completed step: 6, seconds: 21.718, TFLOP/s/device: 0.626, Tokens/s/device: 94.299, total_weights: 65536, loss: 8.592, lm_loss: 8.592, perplexity: 5390.759
I0419 00:06:04.957879 140423904401216 metric_logger.py:196] completed step: 7, seconds: 0.007, TFLOP/s/device: 2023.098, Tokens/s/device: 304943.419, total_weights: 65536, loss: 8.384, lm_loss: 8.384, perplexity: 4376.905
I0419 00:06:05.105234 140423904401216 metric_logger.py:196] completed step: 8, seconds: 0.147, TFLOP/s/device: 92.156, Tokens/s/device: 13890.773, total_weights: 65536, loss: 8.255, lm_loss: 8.255, perplexity: 3846.406
I0419 00:06:05.251973 140423904401216 checkpointing.py:794] Waiting for step 9 to finish before checkpoint...
I0419 00:06:05.252662 140423904401216 checkpointing.py:798] Waited 0.0007104873657226562 seconds for step 9 to finish before starting checkpointing.
I0419 00:06:05.254698 140423904401216 checkpoint_manager.py:2020] [process=0][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0419 00:06:23.661724 140295602325248 base_pytree_checkpoint_handler.py:1282] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 37.566769s (commit=35.836670s, array_metadata_write=1.730099s)
I0419 00:06:23.662795 140296180463360 base_pytree_checkpoint_handler.py:130] [process=0] /jax/orbax/write/gbytes_per_sec: 37.545 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 42.072054862976074 s) (per-host)
I0419 00:06:23.662941 140296180463360 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 41.795241s.
I0419 00:06:24.802561 140296180463360 checkpoint.py:228] Read Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776557143794076602, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} from gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_07_eval/checkpoints/0/_CHECKPOINT_METADATA
I0419 00:06:25.989709 140296180463360 array_metadata_store.py:411] [process=0][thread=async_save] Validated ArrayMetadata from all 8 hosts. Time taken: 0.000238s.
I0419 00:06:26.005902 140293438052096 checkpoint.py:247] Updated Metadata={'item_handlers': {'items': 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler'}, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776557143794076602, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_07_eval/checkpoints/0/_CHECKPOINT_METADATA
I0419 00:06:26.179546 140296180463360 ocdbt_utils.py:49] Param validation support for Zarr3 will be added later (b/362328389).
I0419 00:06:27.109835 140296180463360 base_pytree_checkpoint_handler.py:1406] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 2.178975s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_07_eval/checkpoints/0/items
I0419 00:06:27.111416 140296180463360 atomicity.py:666] Finalizing gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_07_eval/checkpoints/0/items
I0419 00:06:27.644850 140296180463360 atomicity.py:666] Finalizing gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_07_eval/checkpoints/0
I0419 00:06:29.047071 140296180463360 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_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_07_eval/checkpoints/0`.
I0419 00:06:29.047995 140296180463360 event_tracking.py:138] [process=0] [async] Finished save (blocking + background) in 48.58 seconds @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_07_eval/checkpoints/0
I0419 00:06:29.049955 140296180463360 async_checkpointer.py:160] [process=0][thread=async_save] Background save thread done. Time taken: 47.182252s.
I0419 00:06:29.050171 140295593932544 async_checkpointer.py:288] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save.
I0419 00:06:29.050294 140295593932544 async_checkpointer.py:298] [process=0][thread=save_finalize] No errors found in background save thread=async_save.
I0419 00:06:29.050361 140295593932544 checkpoint_manager.py:2137] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts...
I0419 00:06:29.052236 140295593932544 checkpoint_manager.py:2146] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts.
I0419 00:06:29.052409 140423904401216 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.
W0419 00:06:29.052547 140423904401216 checkpoint_manager.py:1452] Waiting for previous save to complete took 23.797848 seconds. If this number is high, consider checkpointing less frequently.
I0419 00:06:29.054397 140423904401216 checkpoint_manager.py:1512] [process=0] Saving checkpoint at step 9
I0419 00:06:29.056528 140423904401216 event_tracking.py:70] [process=0] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_07_eval/checkpoints/9.
I0419 00:06:29.764226 140423904401216 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array.
I0419 00:06:29.764320 140423904401216 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 00:06:29.789711 140423904401216 base_pytree_checkpoint_handler.py:154] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.026211s
I0419 00:06:29.790503 140423904401216 base_pytree_checkpoint_handler.py:130] [process=0] /jax/orbax/write/blocking_gbytes_per_sec: 50.117 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.030779361724853516 s) (per-host)
I0419 00:06:29.790573 140423904401216 base_pytree_checkpoint_handler.py:768] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 0.030866s (batch_requests_ready=0.001845s, total_serialization_initiated=0.028286s, others=0.000735s)
I0419 00:06:29.790644 140423904401216 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.035097s (all_items=0.000018s, per_item={'items': '0.00001764'}, temp_paths=0.035079)
I0419 00:06:29.791317 140423904401216 event_tracking.py:125] [process=0] [async] Finished blocking save in 0.74 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_07_eval/checkpoints/9.
I0419 00:06:29.791602 140295593932544 async_checkpointer.py:76] [process=0][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-19 00:26:29.791569
I0419 00:06:29.805049 140423904401216 checkpoint_manager.py:1560] [process=0][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize
I0419 00:06:29.805354 140295084324608 async_checkpointer.py:280] [process=0][thread=save_finalize] Waiting for background save thread=async_save.
I0419 00:06:29.805526 140423904401216 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_07_eval/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776557165.2546656, 'wait_for_prev_duration_secs': 23.797847986221313, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776557189.054438, 'checkpointer_blocking_duration_secs': 0.7372968196868896, 'get_old_steps_start_time': 1776557189.7917576, 'get_old_steps_duration_secs': 3.123283386230469e-05, 'checkpoint_manager_blocking_start_time': 1776557165.252913, 'checkpoint_manager_blocking_duration_secs': 24.55257797241211}
I0419 00:06:29.805630 140423904401216 checkpointing.py:409] Started an asynchronous checkpoint save for step 9
I0419 00:06:29.805676 140423904401216 checkpoint_manager.py:2020] [process=0][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0419 00:06:30.001908 140296180463360 atomicity.py:140] Creating tmp directory gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_07_eval/checkpoints/9
I0419 00:06:32.161719 140296126740224 atomicity.py:140] Creating tmp directory gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_07_eval/checkpoints/9/items
I0419 00:06:35.488990 140295101110016 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_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_07_eval/checkpoints/9/items/array_metadatas/process_0
I0419 00:07:15.384865 140295092717312 base_pytree_checkpoint_handler.py:1282] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 42.118281s (commit=40.905812s, array_metadata_write=1.212469s)
I0419 00:07:15.386010 140295593932544 base_pytree_checkpoint_handler.py:130] [process=0] /jax/orbax/write/gbytes_per_sec: 34.620 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 45.626248836517334 s) (per-host)
I0419 00:07:15.386137 140295593932544 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 45.594424s.
I0419 00:07:17.280607 140295593932544 array_metadata_store.py:411] [process=0][thread=async_save] Validated ArrayMetadata from all 8 hosts. Time taken: 0.000179s.
I0419 00:07:17.648118 140295593932544 ocdbt_utils.py:49] Param validation support for Zarr3 will be added later (b/362328389).
I0419 00:07:18.758333 140295593932544 base_pytree_checkpoint_handler.py:1406] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 2.549221s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_07_eval/checkpoints/9/items
I0419 00:07:18.764682 140295593932544 atomicity.py:666] Finalizing gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_07_eval/checkpoints/9/items
I0419 00:07:19.310321 140295593932544 atomicity.py:666] Finalizing gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_07_eval/checkpoints/9
I0419 00:07:20.692918 140295593932544 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_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_07_eval/checkpoints/9`.
I0419 00:07:20.693640 140295593932544 event_tracking.py:138] [process=0] [async] Finished save (blocking + background) in 51.64 seconds @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_07_eval/checkpoints/9
I0419 00:07:20.694901 140295593932544 async_checkpointer.py:160] [process=0][thread=async_save] Background save thread done. Time taken: 50.903185s.
I0419 00:07:20.695118 140295084324608 async_checkpointer.py:288] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save.
I0419 00:07:20.695252 140295084324608 async_checkpointer.py:298] [process=0][thread=save_finalize] No errors found in background save thread=async_save.
I0419 00:07:20.695314 140295084324608 checkpoint_manager.py:2137] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts...
I0419 00:07:20.700456 140295084324608 checkpoint_manager.py:2146] [process=0][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts.
I0419 00:07:20.700652 140423904401216 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.
I0419 00:07:20.700802 140423904401216 checkpoint_manager.py:2009] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0419 00:07:20.701747 140423904401216 metric_logger.py:196] completed step: 9, seconds: 0.148, TFLOP/s/device: 91.923, Tokens/s/device: 13855.625, 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 00:07:32 UTC 2026
EXIT_CODE=0
NNX  ·  574ad3fb9  ·  feat_nnx_post_train_fixes_20260418_235007  ·  full log
XPK Start: Sun Apr 19 01:37:39 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 01:38:03.810910: 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 01:38:03.988811 138204772476736 max_utils.py:273] Attempting to initialize the jax distributed system...
I0419 01:38:13.029904 138204772476736 distributed.py:149] Starting JAX distributed service on [::]:8482
I0419 01:38:13.032201 138204772476736 distributed.py:172] Connecting to JAX distributed service on mt-07-eval-ohdxb-slice-job-0-0.mt-07-eval-ohdxb:8482
I0419 01:38:13.975704 138204772476736 max_utils.py:284] Jax distributed system initialized!
I0419 01:38:19.140773 138204772476736 max_utils.py:800] System Information: Jax Version: 0.9.2
I0419 01:38:19.140875 138204772476736 max_utils.py:801] System Information: Jaxlib Version: 0.9.2
I0419 01:38:19.140916 138204772476736 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 01:38:19.140950 138204772476736 train_utils.py:378] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing.
I0419 01:38:19.842377 138204772476736 maxtext_utils.py:1718] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1)
I0419 01:38:19.954524 138204772476736 checkpointing.py:688] Setting up checkpoint logger...
I0419 01:38:19.954646 138204772476736 checkpointing.py:234] Creating checkpoint manager with ocdbt=True and zarr3=True
I0419 01:38:19.954690 138204772476736 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None
I0419 01:38:19.954883 138204772476736 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 0x7db1b017a600>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB)
I0419 01:38:23.210165 138204772476736 checkpointing.py:266] Enabling policy for fixed interval checkpointing.
I0419 01:38:23.210422 138204772476736 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 0x7d9d90377560>}, handler_registry=None
I0419 01:38:23.210657 138204772476736 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7d9d90377560>` 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 01:38:23.210704 138204772476736 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7d9e1038cef0>` 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 01:38:23.210741 138204772476736 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 0x7d9d90377560>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7d9d90377560>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7d9e1038cef0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7d9e1038cef0>}).
I0419 01:38:23.211049 138204772476736 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34
I0419 01:38:23.211121 138204772476736 async_checkpointer.py:192] [process=6][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7d9d90698fe0> timeout: 1200 secs and primary_host=0 for async checkpoint writes
I0419 01:38:25.255542 138204772476736 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/nnx_xpk_feat_nnx_post_train_fixes_20260418_235007_07_eval/checkpoints
I0419 01:38:26.143868 138204772476736 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/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/nnx_xpk_feat_nnx_post_train_fixes_20260418_235007_07_eval/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7d9d906dd910>
I0419 01:38:26.144040 138204772476736 checkpointing.py:302] Checkpoint manager created!
I0419 01:38:27.107741 138204772476736 checkpointing.py:578] checkpoint manager exists so trying to load this run's existing checkpoint
I0419 01:38:27.107856 138204772476736 checkpointing.py:676] No existing checkpoints found, not restoring checkpoint.
fsdp: 32
I0419 01:38:28.477703 138204772476736 maxtext_utils.py:1836]  decoder/decoder_norm/scale/value
    Shape:     float32[2048]
    Physical:  (None,)
I0419 01:38:28.477815 138204772476736 maxtext_utils.py:1836]  decoder/layers/mlp/wi_0/kernel/value
    Shape:     float32[2048,16,7168]
    Physical:  ('fsdp', None, None)
I0419 01:38:28.477867 138204772476736 maxtext_utils.py:1836]  decoder/layers/mlp/wi_1/kernel/value
    Shape:     float32[2048,16,7168]
    Physical:  ('fsdp', None, None)
I0419 01:38:28.477910 138204772476736 maxtext_utils.py:1836]  decoder/layers/mlp/wo/kernel/value
    Shape:     float32[7168,16,2048]
    Physical:  (None, None, 'fsdp')
I0419 01:38:28.477946 138204772476736 maxtext_utils.py:1836]  decoder/layers/post_self_attention_layer_norm/scale/value
    Shape:     float32[2048,16]
    Physical:  (None, None)
I0419 01:38:28.478003 138204772476736 maxtext_utils.py:1836]  decoder/layers/pre_self_attention_layer_norm/scale/value
    Shape:     float32[2048,16]
    Physical:  (None, None)
I0419 01:38:28.478042 138204772476736 maxtext_utils.py:1836]  decoder/layers/self_attention/key/kernel/value
    Shape:     float32[2048,16,16,128]
    Physical:  ('fsdp', None, None, None)
I0419 01:38:28.478078 138204772476736 maxtext_utils.py:1836]  decoder/layers/self_attention/out/kernel/value
    Shape:     float32[16,16,128,2048]
    Physical:  (None, None, None, 'fsdp')
I0419 01:38:28.478112 138204772476736 maxtext_utils.py:1836]  decoder/layers/self_attention/query/kernel/value
    Shape:     float32[2048,16,16,128]
    Physical:  ('fsdp', None, None, None)
I0419 01:38:28.478145 138204772476736 maxtext_utils.py:1836]  decoder/layers/self_attention/value/kernel/value
    Shape:     float32[2048,16,16,128]
    Physical:  ('fsdp', None, None, None)
I0419 01:38:28.478176 138204772476736 maxtext_utils.py:1836]  decoder/logits_dense/kernel/value
    Shape:     float32[2048,32000]
    Physical:  ('fsdp', None)
I0419 01:38:28.478207 138204772476736 maxtext_utils.py:1836]  token_embedder/embedding/value
    Shape:     float32[32000,2048]
    Physical:  (None, 'fsdp')

I0419 01:38:28.882426 138204772476736 nnx_decoders.py:465] nnx_decoders/carry Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed').
I0419 01:38:28.882521 138204772476736 nnx_decoders.py:465] nnx_decoders/carry Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0419 01:38:28.887844 138204772476736 nnx_decoders.py:465] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed').
I0419 01:38:28.887899 138204772476736 nnx_decoders.py:465] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0419 01:38:28.904547 138204772476736 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0419 01:38:28.904608 138204772476736 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0419 01:38:28.920341 138204772476736 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed').
I0419 01:38:28.920396 138204772476736 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None).
I0419 01:38:28.945291 138204772476736 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 01:38:28.945355 138204772476736 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0419 01:38:28.961148 138204772476736 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 01:38:28.961208 138204772476736 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0419 01:38:28.983416 138204772476736 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 01:38:28.983514 138204772476736 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0419 01:38:29.014024 138204772476736 attentions.py:1197] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv').
I0419 01:38:29.014099 138204772476736 attentions.py:1197] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None).
I0419 01:38:29.034200 138204772476736 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp').
I0419 01:38:29.034287 138204772476736 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None).
I0419 01:38:40.219887 138204772476736 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 01:38:40.222292 138204772476736 metric_logger.py:301] number parameters: 1.104 billion
I0419 01:38:51.702142 138204772476736 checkpointing.py:794] Waiting for step 0 to finish before checkpoint...
I0419 01:38:51.855223 138204772476736 checkpointing.py:798] Waited 0.15306448936462402 seconds for step 0 to finish before starting checkpointing.
I0419 01:38:51.857551 138204772476736 checkpoint_manager.py:2009] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0419 01:38:51.859301 138204772476736 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 0
I0419 01:38:51.860650 138204772476736 event_tracking.py:70] [process=6] [async] Started save checkpoint @ gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/nnx_xpk_feat_nnx_post_train_fixes_20260418_235007_07_eval/checkpoints/0.
I0419 01:38:53.018806 138204772476736 signaling_client.py:364] Using JaxDistributedSignalingClient
I0419 01:38:53.019811 138204772476736 jax_array_handlers.py:360] Scheduling D2H of 69 prioritized jax.Array.
I0419 01:38:53.019869 138204772476736 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 01:38:53.295902 138204772476736 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.277868s
I0419 01:38:53.296071 138204772476736 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 5.366 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.28748536109924316 s) (per-host)
I0419 01:38:53.296123 138204772476736 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.287547s (batch_requests_ready=0.002957s, total_serialization_initiated=0.284521s, others=0.000069s)
I0419 01:38:53.296223 138204772476736 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.291623s (all_items=0.000017s, per_item={'items': '0.00001717'}, temp_paths=0.291606)
I0419 01:38:53.297063 138204772476736 event_tracking.py:125] [process=6] [async] Finished blocking save in 1.44 seconds. Continuing save @ gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/nnx_xpk_feat_nnx_post_train_fixes_20260418_235007_07_eval/checkpoints/0.
I0419 01:38:53.297378 138078025070336 async_checkpointer.py:76] [process=6][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-19 01:58:53.297346
I0419 01:38:53.305773 138204772476736 checkpoint_manager.py:1560] [process=6][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize
I0419 01:38:53.306075 138077494310656 async_checkpointer.py:280] [process=6][thread=save_finalize] Waiting for background save thread=async_save.
I0419 01:38:53.306249 138204772476736 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/nnx_xpk_feat_nnx_post_train_fixes_20260418_235007_07_eval/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776562731.8575323, 'wait_for_prev_duration_secs': 6.723403930664062e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776562731.8593395, 'checkpointer_blocking_duration_secs': 1.4381909370422363, 'get_old_steps_start_time': 1776562733.297555, 'get_old_steps_duration_secs': 3.62396240234375e-05, 'checkpoint_manager_blocking_start_time': 1776562731.8556445, 'checkpoint_manager_blocking_duration_secs': 1.4505500793457031}
I0419 01:38:53.306361 138204772476736 checkpointing.py:409] Started an asynchronous checkpoint save for step 0
I0419 01:38:53.306443 138204772476736 max_utils.py:750] 
Memstats: After params initialized:
I0419 01:38:53.306505 138204772476736 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_24(process=6,(0,6,0,0))
I0419 01:38:53.306555 138204772476736 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_25(process=6,(1,6,0,0))
I0419 01:38:53.306602 138204772476736 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_28(process=6,(0,7,0,0))
I0419 01:38:53.306646 138204772476736 max_utils.py:756] 	Using (GB) 0.43 / 31.25 (1.376000%) on TPU_29(process=6,(1,7,0,0))
I0419 01:38:53.622216 138204772476736 metric_logger.py:196] completed step: 0, seconds: 11.479, TFLOP/s/device: 1.184, Tokens/s/device: 178.416, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52786.469
I0419 01:38:53.799024 138204772476736 metric_logger.py:196] completed step: 1, seconds: 1.918, TFLOP/s/device: 7.084, Tokens/s/device: 1067.739, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52786.469
I0419 01:38:54.201670 138204772476736 metric_logger.py:196] completed step: 2, seconds: 0.031, TFLOP/s/device: 441.141, Tokens/s/device: 66493.506, total_weights: 65536, loss: 10.262, lm_loss: 10.262, perplexity: 28632.133
I0419 01:38:54.349457 138204772476736 metric_logger.py:196] completed step: 3, seconds: 0.393, TFLOP/s/device: 34.594, Tokens/s/device: 5214.407, total_weights: 65536, loss: 9.732, lm_loss: 9.732, perplexity: 16855.789
I0419 01:38:54.645549 138204772476736 metric_logger.py:196] completed step: 4, seconds: 0.164, TFLOP/s/device: 82.822, Tokens/s/device: 12483.770, total_weights: 65536, loss: 9.274, lm_loss: 9.274, perplexity: 10661.594
I0419 01:38:54.653177 138204772476736 metric_logger.py:196] completed step: 5, seconds: 0.147, TFLOP/s/device: 92.146, Tokens/s/device: 13889.266, total_weights: 65536, loss: 8.888, lm_loss: 8.888, perplexity: 7245.978
I0419 01:38:56.626717    2792 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com
I0419 01:38:58.550545 138077978920704 array_metadata_store.py:203] [process=6][thread=array_type_handler] Wrote 69 array_metadata.ArrayMetadata to gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/nnx_xpk_feat_nnx_post_train_fixes_20260418_235007_07_eval/checkpoints/0/items/array_metadatas/process_6
I0419 01:39:17.169889 138204772476736 metric_logger.py:196] completed step: 6, seconds: 0.297, TFLOP/s/device: 45.795, Tokens/s/device: 6902.758, total_weights: 65536, loss: 8.589, lm_loss: 8.589, perplexity: 5371.156
I0419 01:39:17.317361 138204772476736 metric_logger.py:196] completed step: 7, seconds: 22.370, TFLOP/s/device: 0.607, Tokens/s/device: 91.550, total_weights: 65536, loss: 8.380, lm_loss: 8.380, perplexity: 4360.465
I0419 01:39:17.464977 138204772476736 metric_logger.py:196] completed step: 8, seconds: 0.153, TFLOP/s/device: 88.517, Tokens/s/device: 13342.193, total_weights: 65536, loss: 8.251, lm_loss: 8.251, perplexity: 3831.256
I0419 01:39:17.613375 138204772476736 checkpointing.py:794] Waiting for step 9 to finish before checkpoint...
I0419 01:39:17.614022 138204772476736 checkpointing.py:798] Waited 0.0006604194641113281 seconds for step 9 to finish before starting checkpointing.
I0419 01:39:17.616105 138204772476736 checkpoint_manager.py:2020] [process=6][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0419 01:39:29.760407 138078025070336 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 42.980 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 36.75177574157715 s) (per-host)
I0419 01:39:29.760523 138078025070336 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 36.463029s.
I0419 01:39:39.165760 138078025070336 async_checkpointer.py:160] [process=6][thread=async_save] Background save thread done. Time taken: 45.868251s.
I0419 01:39:39.166031 138077494310656 async_checkpointer.py:288] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save.
I0419 01:39:39.166150 138077494310656 async_checkpointer.py:298] [process=6][thread=save_finalize] No errors found in background save thread=async_save.
I0419 01:39:39.166202 138077494310656 checkpoint_manager.py:2137] [process=6][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts...
I0419 01:39:39.168754 138077494310656 checkpoint_manager.py:2146] [process=6][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts.
I0419 01:39:39.168932 138204772476736 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 01:39:39.169063 138204772476736 checkpoint_manager.py:1452] Waiting for previous save to complete took 21.552957 seconds. If this number is high, consider checkpointing less frequently.
I0419 01:39:39.171127 138204772476736 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 9
I0419 01:39:39.173150 138204772476736 event_tracking.py:70] [process=6] [async] Started save checkpoint @ gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/nnx_xpk_feat_nnx_post_train_fixes_20260418_235007_07_eval/checkpoints/9.
I0419 01:39:39.882606 138204772476736 jax_array_handlers.py:360] Scheduling D2H of 69 prioritized jax.Array.
I0419 01:39:39.882705 138204772476736 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 01:39:39.920081 138204772476736 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.039116s
I0419 01:39:39.920252 138204772476736 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 33.452 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.0461125373840332 s) (per-host)
I0419 01:39:39.920303 138204772476736 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.046187s (batch_requests_ready=0.002638s, total_serialization_initiated=0.043467s, others=0.000082s)
I0419 01:39:39.920402 138204772476736 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.050328s (all_items=0.000016s, per_item={'items': '0.00001574'}, temp_paths=0.050312)
I0419 01:39:39.921154 138204772476736 event_tracking.py:125] [process=6] [async] Finished blocking save in 0.75 seconds. Continuing save @ gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/nnx_xpk_feat_nnx_post_train_fixes_20260418_235007_07_eval/checkpoints/9.
I0419 01:39:39.921469 138077494310656 async_checkpointer.py:76] [process=6][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-19 01:59:39.921436
I0419 01:39:39.926554 138204772476736 checkpoint_manager.py:1560] [process=6][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize
I0419 01:39:39.926787 138078016677632 async_checkpointer.py:280] [process=6][thread=save_finalize] Waiting for background save thread=async_save.
I0419 01:39:39.926909 138204772476736 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/nnx_xpk_feat_nnx_post_train_fixes_20260418_235007_07_eval/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776562757.6160746, 'wait_for_prev_duration_secs': 21.55295729637146, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776562779.1711655, 'checkpointer_blocking_duration_secs': 0.7504479885101318, 'get_old_steps_start_time': 1776562779.9216418, 'get_old_steps_duration_secs': 3.123283386230469e-05, 'checkpoint_manager_blocking_start_time': 1776562757.6142569, 'checkpoint_manager_blocking_duration_secs': 22.312620162963867}
I0419 01:39:39.927026 138204772476736 checkpointing.py:409] Started an asynchronous checkpoint save for step 9
I0419 01:39:39.927072 138204772476736 checkpoint_manager.py:2020] [process=6][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete.
I0419 01:39:46.289424 138073736214272 array_metadata_store.py:203] [process=6][thread=array_type_handler] Wrote 69 array_metadata.ArrayMetadata to gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/nnx_xpk_feat_nnx_post_train_fixes_20260418_235007_07_eval/checkpoints/9/items/array_metadatas/process_6
I0419 01:40:22.706631 138077494310656 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 36.878 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 42.8324670791626 s) (per-host)
I0419 01:40:22.706759 138077494310656 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 42.785178s.
I0419 01:40:31.392952 138077494310656 async_checkpointer.py:160] [process=6][thread=async_save] Background save thread done. Time taken: 51.471357s.
I0419 01:40:31.393208 138078016677632 async_checkpointer.py:288] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save.
I0419 01:40:31.393346 138078016677632 async_checkpointer.py:298] [process=6][thread=save_finalize] No errors found in background save thread=async_save.
I0419 01:40:31.393394 138078016677632 checkpoint_manager.py:2137] [process=6][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts...
I0419 01:40:31.395092 138078016677632 checkpoint_manager.py:2146] [process=6][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts.
I0419 01:40:31.395313 138204772476736 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 01:40:31.395632 138204772476736 checkpoint_manager.py:2009] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning.
I0419 01:40:31.396602 138204772476736 metric_logger.py:196] completed step: 9, seconds: 0.148, TFLOP/s/device: 91.645, Tokens/s/device: 13813.757, total_weights: 65536, loss: 8.175, lm_loss: 8.175, perplexity: 3551.438
Per train step:
 Total TFLOPs: 13.59 
 split as 93.93% learnable weight flops and 6.07% attention flops
XPK End: Sun Apr 19 01:40:39 UTC 2026
EXIT_CODE=0