main (8a17c3d19) vs feat/nnx-set-defaults-true (73213e044)| Metric | main 8a17c3d19 | feat/nnx-set-defaults-true 73213e044 | Diff (feat/nnx-set-defaults-true − main) |
|---|---|---|---|
| Parameters | 1.104 billion | 1.104 billion | — |
| Final loss | 9.4570 | 9.4570 | 0 |
| TFLOP/s | 94.014 | 93.966 | -0.048 |
| Tok/s | 14170.9 | 14163.6 | -7.277 |
| Avg s/step | 6.818 | 5.923 | -0.895 |
| Memory % | 1.38 | 1.38 | 0 |
| JAX | 0.9.2 | 0.8.1 | — |
Diff = branch value − main value. Green = branch improved. Red = branch regressed.
XPK Start: Wed Apr 22 07:28:50 UTC 2026 PyTorch was not found. Models won't be available and only tokenizers, configuration and file/data utilities can be used. `rope_parameters`'s factor field must be a float >= 1, got 40 `rope_parameters`'s beta_fast field must be a float, got 32 `rope_parameters`'s beta_slow field must be a float, got 1 DeepseekV32Config got `key=rope_scaling` in kwargs but hasn't set it as attribute. For RoPE standardization you need to set `self.rope_parameters` in model's config. 2026-04-22 07:29:15.029127: 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) I0422 07:29:15.237967 137847962920768 max_utils.py:273] Attempting to initialize the jax distributed system... I0422 07:29:24.278456 137847962920768 distributed.py:149] Starting JAX distributed service on [::]:8482 I0422 07:29:24.280758 137847962920768 distributed.py:172] Connecting to JAX distributed service on mt-06-grad-accum-205v9-slice-job-0-0.mt-06-grad-accum-205v9:8482 I0422 07:29:26.544724 137847962920768 max_utils.py:284] Jax distributed system initialized! I0422 07:29:32.656195 137847962920768 max_utils.py:800] System Information: Jax Version: 0.9.2 I0422 07:29:32.656302 137847962920768 max_utils.py:801] System Information: Jaxlib Version: 0.9.2 I0422 07:29:32.656342 137847962920768 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 I0422 07:29:32.656378 137847962920768 train_utils.py:361] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing. I0422 07:29:33.341875 137847962920768 maxtext_utils.py:1565] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0422 07:29:33.342161 137847962920768 checkpointing.py:677] Setting up checkpoint logger... I0422 07:29:33.342215 137847962920768 checkpointing.py:233] Creating checkpoint manager with ocdbt=True and zarr3=True I0422 07:29:33.342259 137847962920768 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None I0422 07:29:33.342598 137847962920768 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 0x7d5e83248d70>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0422 07:29:36.660800 137847962920768 checkpointing.py:265] Enabling policy for fixed interval checkpointing. I0422 07:29:36.661039 137847962920768 checkpoint_manager.py:708] [process=4][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=('items',), item_handlers={'items': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7d4a6c272930>}, handler_registry=None I0422 07:29:36.661286 137847962920768 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7d4a6c272930>` 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`. I0422 07:29:36.661335 137847962920768 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7d4a542d85f0>` 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`. I0422 07:29:36.661371 137847962920768 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 0x7d4a6c272930>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7d4a6c272930>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7d4a542d85f0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7d4a542d85f0>}). I0422 07:29:36.661705 137847962920768 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34 I0422 07:29:36.661777 137847962920768 async_checkpointer.py:192] [process=4][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7d4a54251580> timeout: 1200 secs and primary_host=0 for async checkpoint writes I0422 07:29:37.371757 137847962920768 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_main_20260422_071422/linen_xpk_main_20260422_071422_06_grad_accum/checkpoints I0422 07:29:37.544448 137847962920768 checkpoint_manager.py:929] [process=4][thread=MainThread] CheckpointManager created, primary_host=0, CheckpointManagerOptions=CheckpointManagerOptions(save_interval_steps=1, max_to_keep=None, keep_time_interval=None, keep_period=None, should_keep_fn=None, best_fn=None, best_mode='max', keep_checkpoints_without_metrics=True, step_prefix=None, step_format_fixed_length=None, step_name_format=None, create=True, cleanup_tmp_directories=False, save_on_steps=frozenset(), single_host_load_and_broadcast=False, todelete_subdir=None, todelete_full_path=None, enable_background_delete=False, read_only=False, enable_async_checkpointing=True, async_options=None, multiprocessing_options=MultiprocessingOptions(primary_host=0, active_processes=None, barrier_sync_key_prefix=None), should_save_fn=None, file_options=FileOptions(path_permission_mode=None), save_root_metadata=True, temporary_path_class=None, save_decision_policy=FixedIntervalPolicy(interval=10), preservation_policy=LatestN(n=None), prevent_write_metrics=False, enable_should_save_is_saving_in_progress_check=True, enable_per_process_directory_creation=False, lightweight_initialize=False), root_directory=gs://lance-maxtext/linen_ckpt_xpk_main_20260422_071422/linen_xpk_main_20260422_071422_06_grad_accum/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7d4a542d59d0> I0422 07:29:37.544621 137847962920768 checkpointing.py:301] Checkpoint manager created! I0422 07:29:39.282500 137847962920768 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0422 07:29:39.282617 137847962920768 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0422 07:29:39.663165 137847962920768 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0422 07:29:39.663261 137847962920768 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0422 07:29:39.679896 137847962920768 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0422 07:29:39.679956 137847962920768 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0422 07:29:39.703902 137847962920768 attentions.py:1154] attentions/query Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0422 07:29:39.703973 137847962920768 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0422 07:29:39.720690 137847962920768 attentions.py:1155] attentions/key Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0422 07:29:39.720769 137847962920768 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0422 07:29:39.737361 137847962920768 attentions.py:1156] attentions/value Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0422 07:29:39.737479 137847962920768 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0422 07:29:39.762500 137847962920768 attentions.py:1198] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv'). I0422 07:29:39.762571 137847962920768 attentions.py:1198] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0422 07:29:39.783849 137847962920768 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0422 07:29:39.783916 137847962920768 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None). I0422 07:29:39.997267 137847962920768 checkpointing.py:577] checkpoint manager exists so trying to load this run's existing checkpoint I0422 07:29:39.997389 137847962920768 checkpointing.py:665] No existing checkpoints found, not restoring checkpoint. fsdp: 32 I0422 07:29:41.444055 137847962920768 maxtext_utils.py:1668] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: P('norm',) Physical: (None,) I0422 07:29:41.444179 137847962920768 maxtext_utils.py:1668] params/params/decoder/layers/mlp/wi_0/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0422 07:29:41.444232 137847962920768 maxtext_utils.py:1668] params/params/decoder/layers/mlp/wi_1/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0422 07:29:41.444290 137847962920768 maxtext_utils.py:1668] params/params/decoder/layers/mlp/wo/kernel Shape: float32[7168,16,2048] Logical: P('mlp', 'layers', 'embed') Physical: (None, None, 'fsdp') I0422 07:29:41.444339 137847962920768 maxtext_utils.py:1668] params/params/decoder/layers/post_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0422 07:29:41.444379 137847962920768 maxtext_utils.py:1668] params/params/decoder/layers/pre_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0422 07:29:41.444437 137847962920768 maxtext_utils.py:1668] 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) I0422 07:29:41.444491 137847962920768 maxtext_utils.py:1668] 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') I0422 07:29:41.444531 137847962920768 maxtext_utils.py:1668] 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) I0422 07:29:41.444568 137847962920768 maxtext_utils.py:1668] 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) I0422 07:29:41.444614 137847962920768 maxtext_utils.py:1668] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: P('embed_vocab', 'vocab') Physical: ('fsdp', None) I0422 07:29:41.444660 137847962920768 maxtext_utils.py:1668] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: P('vocab', 'embed_vocab') Physical: (None, 'fsdp') I0422 07:29:41.469741 137847962920768 gradient_accumulation.py:68] gradient_accumulation/inputs Logical: float32[2048]............................................... Unknown. I0422 07:29:41.469810 137847962920768 gradient_accumulation.py:68] gradient_accumulation/inputs Physical: float32[2048]............................................... (None,). I0422 07:29:41.484953 137847962920768 gradient_accumulation.py:68] gradient_accumulation/inputs Logical: float32[2048,16,7168]....................................... Unknown. I0422 07:29:41.485007 137847962920768 gradient_accumulation.py:68] gradient_accumulation/inputs Physical: float32[2048,16,7168]....................................... ('fsdp', None, None). I0422 07:29:41.514961 137847962920768 gradient_accumulation.py:68] gradient_accumulation/inputs Logical: float32[7168,16,2048]....................................... Unknown. I0422 07:29:41.515027 137847962920768 gradient_accumulation.py:68] gradient_accumulation/inputs Physical: float32[7168,16,2048]....................................... (None, None, 'fsdp'). I0422 07:29:41.529948 137847962920768 gradient_accumulation.py:68] gradient_accumulation/inputs Logical: float32[2048,16]............................................ Unknown. I0422 07:29:41.530004 137847962920768 gradient_accumulation.py:68] gradient_accumulation/inputs Physical: float32[2048,16]............................................ (None, None). I0422 07:29:41.559797 137847962920768 gradient_accumulation.py:68] gradient_accumulation/inputs Logical: float32[2048,16,16,128]..................................... Unknown. I0422 07:29:41.559864 137847962920768 gradient_accumulation.py:68] gradient_accumulation/inputs Physical: float32[2048,16,16,128]..................................... ('fsdp', None, None, None). I0422 07:29:41.574895 137847962920768 gradient_accumulation.py:68] gradient_accumulation/inputs Logical: float32[16,16,128,2048]..................................... Unknown. I0422 07:29:41.574949 137847962920768 gradient_accumulation.py:68] gradient_accumulation/inputs Physical: float32[16,16,128,2048]..................................... (None, None, None, 'fsdp'). I0422 07:29:41.619611 137847962920768 gradient_accumulation.py:68] gradient_accumulation/inputs Logical: float32[2048,32000]......................................... Unknown. I0422 07:29:41.619684 137847962920768 gradient_accumulation.py:68] gradient_accumulation/inputs Physical: float32[2048,32000]......................................... ('fsdp', None). I0422 07:29:41.634616 137847962920768 gradient_accumulation.py:68] gradient_accumulation/inputs Logical: float32[32000,2048]......................................... Unknown. I0422 07:29:41.634673 137847962920768 gradient_accumulation.py:68] gradient_accumulation/inputs Physical: float32[32000,2048]......................................... (None, 'fsdp'). I0422 07:29:42.310529 137847962920768 train.py:155] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0422 07:29:42.310619 137847962920768 train.py:155] train/xent Physical: float32[32,2048]............................................ ('fsdp', None). I0422 07:29:42.326449 137847962920768 train.py:162] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0422 07:29:42.326508 137847962920768 train.py:162] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None). I0422 07:29:53.809075 137847962920768 max_utils.py:791] Total memory size: 1.7 GB, Output size: 0.4 GB, Temp size: 1.3 GB, Argument size: 0.4 GB, Host temp size: 0.0 GB. I0422 07:29:53.809968 137847962920768 metric_logger.py:301] number parameters: 1.104 billion I0422 07:29:55.725687 137847962920768 checkpointing.py:772] Waiting for step 0 to finish before checkpoint... I0422 07:30:06.609395 137847962920768 checkpointing.py:776] Waited 10.883687734603882 seconds for step 0 to finish before starting checkpointing. I0422 07:30:06.611816 137847962920768 checkpoint_manager.py:2009] [process=4][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0422 07:30:06.613519 137847962920768 checkpoint_manager.py:1512] [process=4] Saving checkpoint at step 0 I0422 07:30:06.615215 137847962920768 event_tracking.py:70] [process=4] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260422_071422/linen_xpk_main_20260422_071422_06_grad_accum/checkpoints/0. I0422 07:30:07.152307 137847962920768 signaling_client.py:364] Using JaxDistributedSignalingClient I0422 07:30:07.153290 137847962920768 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0422 07:30:07.153347 137847962920768 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 I0422 07:30:07.426709 137847962920768 base_pytree_checkpoint_handler.py:154] [process=4][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.274461s I0422 07:30:07.426880 137847962920768 base_pytree_checkpoint_handler.py:130] [process=4] /jax/orbax/write/blocking_gbytes_per_sec: 5.511 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.27992725372314453 s) (per-host) I0422 07:30:07.426931 137847962920768 base_pytree_checkpoint_handler.py:768] [process=4][thread=MainThread] Initiated Pytree async_save. Time taken: 0.279989s (batch_requests_ready=0.002235s, total_serialization_initiated=0.277685s, others=0.000070s) I0422 07:30:07.427019 137847962920768 composite_checkpoint_handler.py:715] [process=4][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.284426s (all_items=0.000018s, per_item={'items': '0.00001812'}, temp_paths=0.284408) I0422 07:30:07.427850 137847962920768 event_tracking.py:125] [process=4] [async] Finished blocking save in 0.81 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260422_071422/linen_xpk_main_20260422_071422_06_grad_accum/checkpoints/0. I0422 07:30:07.428158 137719775831808 async_checkpointer.py:76] [process=4][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-22 07:50:07.428119 I0422 07:30:07.438380 137847962920768 checkpoint_manager.py:1560] [process=4][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize I0422 07:30:07.438682 137719229437696 async_checkpointer.py:280] [process=4][thread=save_finalize] Waiting for background save thread=async_save. I0422 07:30:07.438857 137847962920768 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260422_071422/linen_xpk_main_20260422_071422_06_grad_accum/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776843006.6117966, 'wait_for_prev_duration_secs': 6.151199340820312e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776843006.6135576, 'checkpointer_blocking_duration_secs': 0.8147554397583008, 'get_old_steps_start_time': 1776843007.4283395, 'get_old_steps_duration_secs': 3.0040740966796875e-05, 'checkpoint_manager_blocking_start_time': 1776843006.6099365, 'checkpoint_manager_blocking_duration_secs': 0.8288826942443848} I0422 07:30:07.438961 137847962920768 checkpointing.py:408] Started an asynchronous checkpoint save for step 0 I0422 07:30:07.439012 137847962920768 max_utils.py:750] Memstats: After params initialized: I0422 07:30:07.439059 137847962920768 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_16(process=4,(0,4,0,0)) I0422 07:30:07.439092 137847962920768 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_17(process=4,(1,4,0,0)) I0422 07:30:07.439119 137847962920768 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_20(process=4,(0,5,0,0)) I0422 07:30:07.439143 137847962920768 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_21(process=4,(1,5,0,0)) I0422 07:30:07.752342 137847962920768 metric_logger.py:196] completed step: 0, seconds: 1.916, TFLOP/s/device: 28.371, Tokens/s/device: 4276.440, total_weights: 262144, loss: 10.877, lm_loss: 10.877, perplexity: 52959.059 I0422 07:30:08.363315 137847962920768 metric_logger.py:196] completed step: 1, seconds: 12.025, TFLOP/s/device: 4.520, Tokens/s/device: 681.240, total_weights: 262144, loss: 10.877, lm_loss: 10.877, perplexity: 52959.059 I0422 07:30:08.941560 137847962920768 metric_logger.py:196] completed step: 2, seconds: 0.033, TFLOP/s/device: 1641.156, Tokens/s/device: 247372.871, total_weights: 262144, loss: 10.563, lm_loss: 10.563, perplexity: 38662.707 I0422 07:30:09.519726 137847962920768 metric_logger.py:196] completed step: 3, seconds: 0.584, TFLOP/s/device: 93.074, Tokens/s/device: 14029.175, total_weights: 262144, loss: 10.272, lm_loss: 10.272, perplexity: 28909.668 I0422 07:30:10.415733 2577 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com I0422 07:30:10.676507 137847962920768 metric_logger.py:196] completed step: 4, seconds: 0.578, TFLOP/s/device: 93.999, Tokens/s/device: 14168.598, total_weights: 262144, loss: 10.022, lm_loss: 10.022, perplexity: 22524.992 I0422 07:30:10.685266 137847962920768 metric_logger.py:196] completed step: 5, seconds: 0.578, TFLOP/s/device: 93.995, Tokens/s/device: 14167.985, total_weights: 262144, loss: 9.820, lm_loss: 9.820, perplexity: 18401.865 I0422 07:30:12.407782 137719237830400 array_metadata_store.py:203] [process=4][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260422_071422/linen_xpk_main_20260422_071422_06_grad_accum/checkpoints/0/items/array_metadatas/process_4 I0422 07:30:45.236618 137719775831808 base_pytree_checkpoint_handler.py:130] [process=4] /jax/orbax/write/gbytes_per_sec: 41.471 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 38.0896270275116 s) (per-host) I0422 07:30:45.236747 137719775831808 async_checkpointer.py:90] [process=4][thread=async_save] 3 Handler Commit operations completed. Time taken: 37.808477s. I0422 07:30:53.685003 137719775831808 async_checkpointer.py:160] [process=4][thread=async_save] Background save thread done. Time taken: 46.256718s. I0422 07:30:53.685292 137719229437696 async_checkpointer.py:288] [process=4][thread=save_finalize] Done with waiting for background save thread=async_save. I0422 07:30:53.685407 137719229437696 async_checkpointer.py:298] [process=4][thread=save_finalize] No errors found in background save thread=async_save. I0422 07:30:53.685456 137719229437696 checkpoint_manager.py:2137] [process=4][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts... I0422 07:30:53.688200 137719229437696 checkpoint_manager.py:2146] [process=4][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts. I0422 07:30:56.507108 137847962920768 metric_logger.py:196] completed step: 6, seconds: 1.159, TFLOP/s/device: 46.875, Tokens/s/device: 7065.486, total_weights: 262144, loss: 9.667, lm_loss: 9.667, perplexity: 15787.604 I0422 07:30:57.085111 137847962920768 metric_logger.py:196] completed step: 7, seconds: 45.245, TFLOP/s/device: 1.201, Tokens/s/device: 181.059, total_weights: 262144, loss: 9.561, lm_loss: 9.561, perplexity: 14203.827 I0422 07:30:57.663415 137847962920768 metric_logger.py:196] completed step: 8, seconds: 0.583, TFLOP/s/device: 93.217, Tokens/s/device: 14050.735, total_weights: 262144, loss: 9.496, lm_loss: 9.496, perplexity: 13302.920 I0422 07:30:58.240848 137847962920768 checkpointing.py:772] Waiting for step 9 to finish before checkpoint... I0422 07:30:58.241523 137847962920768 checkpointing.py:776] Waited 0.0006973743438720703 seconds for step 9 to finish before starting checkpointing. I0422 07:30:58.243679 137847962920768 checkpoint_manager.py:2009] [process=4][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0422 07:30:58.245267 137847962920768 checkpoint_manager.py:1512] [process=4] Saving checkpoint at step 9 I0422 07:30:58.246649 137847962920768 event_tracking.py:70] [process=4] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260422_071422/linen_xpk_main_20260422_071422_06_grad_accum/checkpoints/9. I0422 07:30:58.547047 137847962920768 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0422 07:30:58.547138 137847962920768 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 I0422 07:30:58.580763 137847962920768 base_pytree_checkpoint_handler.py:154] [process=4][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.034564s I0422 07:30:58.580929 137847962920768 base_pytree_checkpoint_handler.py:130] [process=4] /jax/orbax/write/blocking_gbytes_per_sec: 39.933 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.03862929344177246 s) (per-host) I0422 07:30:58.580980 137847962920768 base_pytree_checkpoint_handler.py:768] [process=4][thread=MainThread] Initiated Pytree async_save. Time taken: 0.038689s (batch_requests_ready=0.001638s, total_serialization_initiated=0.036984s, others=0.000067s) I0422 07:30:58.581070 137847962920768 composite_checkpoint_handler.py:715] [process=4][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.042968s (all_items=0.000016s, per_item={'items': '0.00001597'}, temp_paths=0.042952) I0422 07:30:58.581758 137847962920768 event_tracking.py:125] [process=4] [async] Finished blocking save in 0.34 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260422_071422/linen_xpk_main_20260422_071422_06_grad_accum/checkpoints/9. I0422 07:30:58.582071 137712816350976 async_checkpointer.py:76] [process=4][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-22 07:50:58.582035 I0422 07:30:58.583922 137847962920768 checkpoint_manager.py:1560] [process=4][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize I0422 07:30:58.584186 137719247578880 async_checkpointer.py:280] [process=4][thread=save_finalize] Waiting for background save thread=async_save. I0422 07:30:58.584320 137847962920768 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260422_071422/linen_xpk_main_20260422_071422_06_grad_accum/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776843058.2436478, 'wait_for_prev_duration_secs': 8.845329284667969e-05, 'time_between_consecutive_saves_sec': 4.55540657043457, 'checkpointer_blocking_start_time': 1776843058.2453048, 'checkpointer_blocking_duration_secs': 0.33690810203552246, 'get_old_steps_start_time': 1776843058.5822349, 'get_old_steps_duration_secs': 2.6226043701171875e-05, 'checkpoint_manager_blocking_start_time': 1776843058.2417703, 'checkpoint_manager_blocking_duration_secs': 0.3425161838531494} I0422 07:30:58.584472 137847962920768 checkpointing.py:408] Started an asynchronous checkpoint save for step 9 I0422 07:30:58.584516 137847962920768 checkpoint_manager.py:2020] [process=4][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0422 07:31:04.162814 137719255971584 array_metadata_store.py:203] [process=4][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260422_071422/linen_xpk_main_20260422_071422_06_grad_accum/checkpoints/9/items/array_metadatas/process_4 I0422 07:31:42.131255 137712816350976 base_pytree_checkpoint_handler.py:130] [process=4] /jax/orbax/write/gbytes_per_sec: 36.238 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 43.588916063308716 s) (per-host) I0422 07:31:42.131381 137712816350976 async_checkpointer.py:90] [process=4][thread=async_save] 3 Handler Commit operations completed. Time taken: 43.549201s. I0422 07:31:50.641376 137712816350976 async_checkpointer.py:160] [process=4][thread=async_save] Background save thread done. Time taken: 52.059182s. I0422 07:31:50.641664 137719247578880 async_checkpointer.py:288] [process=4][thread=save_finalize] Done with waiting for background save thread=async_save. I0422 07:31:50.641803 137719247578880 async_checkpointer.py:298] [process=4][thread=save_finalize] No errors found in background save thread=async_save. I0422 07:31:50.641851 137719247578880 checkpoint_manager.py:2137] [process=4][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts... I0422 07:31:50.643559 137719247578880 checkpoint_manager.py:2146] [process=4][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts. I0422 07:31:50.643751 137847962920768 checkpoint_manager.py:2032] [process=4][thread=MainThread][step=9][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=9. I0422 07:31:50.643906 137847962920768 checkpoint_manager.py:2009] [process=4][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0422 07:31:50.644895 137847962920768 metric_logger.py:196] completed step: 9, seconds: 0.578, TFLOP/s/device: 94.014, Tokens/s/device: 14170.853, total_weights: 262144, loss: 9.457, lm_loss: 9.457, perplexity: 12802.546 Per train step: Total TFLOPs: 54.35 split as 93.93% learnable weight flops and 6.07% attention flops XPK End: Wed Apr 22 07:32:00 UTC 2026 EXIT_CODE=0
XPK Start: Wed Apr 22 16:03:26 UTC 2026 2026-04-22 16:03:30.094540: E external/local_xla/xla/stream_executor/cuda/cuda_fft.cc:467] Unable to register cuFFT factory: Attempting to register factory for plugin cuFFT when one has already been registered WARNING: All log messages before absl::InitializeLog() is called are written to STDERR E0000 00:00:1776873810.106812 10 cuda_dnn.cc:8579] Unable to register cuDNN factory: Attempting to register factory for plugin cuDNN when one has already been registered E0000 00:00:1776873810.110698 10 cuda_blas.cc:1407] Unable to register cuBLAS factory: Attempting to register factory for plugin cuBLAS when one has already been registered W0000 00:00:1776873810.121985 10 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once. W0000 00:00:1776873810.122002 10 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once. W0000 00:00:1776873810.122004 10 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once. W0000 00:00:1776873810.122006 10 computation_placer.cc:177] computation placer already registered. Please check linkage and avoid linking the same target more than once. 2026-04-22 16:03:49.208312: 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) I0422 16:03:49.736793 138014124525376 max_utils.py:273] Attempting to initialize the jax distributed system... INFO:2026-04-22 16:03:58,777:jax._src.distributed:140: Starting JAX distributed service on [::]:8482 I0422 16:03:58.777872 138014124525376 distributed.py:140] Starting JAX distributed service on [::]:8482 INFO:2026-04-22 16:03:58,780:jax._src.distributed:157: Connecting to JAX distributed service on mt-06-grad-accum-79ahu-slice-job-0-0.mt-06-grad-accum-79ahu:8482 I0422 16:03:58.780173 138014124525376 distributed.py:157] Connecting to JAX distributed service on mt-06-grad-accum-79ahu-slice-job-0-0.mt-06-grad-accum-79ahu:8482 I0422 16:04:00.388361 138014124525376 max_utils.py:284] Jax distributed system initialized! I0422 16:04:07.359148 138014124525376 max_utils.py:800] System Information: Jax Version: 0.8.1 I0422 16:04:07.359261 138014124525376 max_utils.py:801] System Information: Jaxlib Version: 0.8.1 I0422 16:04:07.359301 138014124525376 max_utils.py:802] System Information: Jax Backend: PJRT C API TFRT TPU v6 lite Built on Nov 12 2025 14:16:36 (1762985796) cl/831091709 I0422 16:04:07.359336 138014124525376 train_utils.py:377] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing. I0422 16:04:07.976238 138014124525376 maxtext_utils.py:1631] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0422 16:04:07.976837 138014124525376 maxtext_utils.py:1631] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0422 16:04:07.977015 138014124525376 checkpointing.py:688] Setting up checkpoint logger... I0422 16:04:07.977064 138014124525376 checkpointing.py:234] Creating checkpoint manager with ocdbt=True and zarr3=True I0422 16:04:07.977106 138014124525376 pytree_checkpoint_handler.py:589] save_device_host_concurrent_bytes=None I0422 16:04:07.977494 138014124525376 base_pytree_checkpoint_handler.py:415] 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 0x7d857af754c0>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0422 16:04:11.020731 138014124525376 checkpointing.py:266] Enabling policy for fixed interval checkpointing. I0422 16:04:11.021082 138014124525376 checkpoint_manager.py:709] [process=5][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=('items',), item_handlers={'items': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7d71081b1e50>}, handler_registry=None I0422 16:04:11.021329 138014124525376 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7d71081b1e50>` 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`. I0422 16:04:11.021378 138014124525376 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7d71081b7200>` 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`. I0422 16:04:11.021417 138014124525376 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 0x7d71081b1e50>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7d71081b1e50>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7d71081b7200>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7d71081b7200>}). I0422 16:04:11.021939 138014124525376 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.33 I0422 16:04:11.022023 138014124525376 async_checkpointer.py:177] [process=5][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7d70d858c180> timeout: 600 secs and primary_host=0 for async checkpoint writes I0422 16:04:11.879248 138014124525376 checkpoint_manager.py:1818] Found 0 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_set_defaults_true_20260422_154647/linen_xpk_feat_nnx_set_defaults_true_20260422_154647_06_grad_accum/checkpoints I0422 16:04:12.329765 138014124525376 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_feat_nnx_set_defaults_true_20260422_154647/linen_xpk_feat_nnx_set_defaults_true_20260422_154647_06_grad_accum/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7d72a1d5d670> I0422 16:04:12.329961 138014124525376 checkpointing.py:302] Checkpoint manager created! I0422 16:04:13.990257 138014124525376 nnx_wrappers.py:455] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0422 16:04:13.990366 138014124525376 nnx_wrappers.py:455] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0422 16:04:14.361193 138014124525376 attentions.py:1084] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0422 16:04:14.361292 138014124525376 attentions.py:1084] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0422 16:04:14.376965 138014124525376 attentions.py:1085] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0422 16:04:14.377029 138014124525376 attentions.py:1085] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0422 16:04:14.399476 138014124525376 attentions.py:1150] attentions/query Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0422 16:04:14.399550 138014124525376 attentions.py:1150] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0422 16:04:14.415449 138014124525376 attentions.py:1151] attentions/key Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0422 16:04:14.415518 138014124525376 attentions.py:1151] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0422 16:04:14.431313 138014124525376 attentions.py:1152] attentions/value Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0422 16:04:14.431379 138014124525376 attentions.py:1152] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0422 16:04:14.455871 138014124525376 attentions.py:1193] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv'). I0422 16:04:14.455945 138014124525376 attentions.py:1193] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0422 16:04:14.475212 138014124525376 linears.py:541] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0422 16:04:14.475282 138014124525376 linears.py:541] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None). I0422 16:04:14.663936 138014124525376 checkpointing.py:578] checkpoint manager exists so trying to load this run's existing checkpoint I0422 16:04:14.664048 138014124525376 checkpointing.py:676] No existing checkpoints found, not restoring checkpoint. fsdp: 32 I0422 16:04:16.083684 138014124525376 maxtext_utils.py:1740] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: PartitionSpec('norm',) Physical: (None,) I0422 16:04:16.083820 138014124525376 maxtext_utils.py:1740] params/params/decoder/layers/mlp/wi_0/kernel Shape: float32[2048,16,7168] Logical: PartitionSpec('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0422 16:04:16.083875 138014124525376 maxtext_utils.py:1740] params/params/decoder/layers/mlp/wi_1/kernel Shape: float32[2048,16,7168] Logical: PartitionSpec('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0422 16:04:16.083937 138014124525376 maxtext_utils.py:1740] params/params/decoder/layers/mlp/wo/kernel Shape: float32[7168,16,2048] Logical: PartitionSpec('mlp', 'layers', 'embed') Physical: (None, None, 'fsdp') I0422 16:04:16.083992 138014124525376 maxtext_utils.py:1740] params/params/decoder/layers/post_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: PartitionSpec('norm', 'layers') Physical: (None, None) I0422 16:04:16.084031 138014124525376 maxtext_utils.py:1740] params/params/decoder/layers/pre_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: PartitionSpec('norm', 'layers') Physical: (None, None) I0422 16:04:16.084084 138014124525376 maxtext_utils.py:1740] params/params/decoder/layers/self_attention/key/kernel Shape: float32[2048,16,16,128] Logical: PartitionSpec('embed', 'layers', 'kv_heads', 'kv_head_dim') Physical: ('fsdp', None, None, None) I0422 16:04:16.084136 138014124525376 maxtext_utils.py:1740] params/params/decoder/layers/self_attention/out/kernel Shape: float32[16,16,128,2048] Logical: PartitionSpec('heads', 'layers', 'kv', 'embed') Physical: (None, None, None, 'fsdp') I0422 16:04:16.084178 138014124525376 maxtext_utils.py:1740] params/params/decoder/layers/self_attention/query/kernel Shape: float32[2048,16,16,128] Logical: PartitionSpec('embed', 'layers', 'q_heads', 'kv') Physical: ('fsdp', None, None, None) I0422 16:04:16.084226 138014124525376 maxtext_utils.py:1740] params/params/decoder/layers/self_attention/value/kernel Shape: float32[2048,16,16,128] Logical: PartitionSpec('embed', 'layers', 'kv_heads', 'kv_head_dim') Physical: ('fsdp', None, None, None) I0422 16:04:16.084276 138014124525376 maxtext_utils.py:1740] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: PartitionSpec('embed', 'vocab') Physical: ('fsdp', None) I0422 16:04:16.084325 138014124525376 maxtext_utils.py:1740] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: PartitionSpec('vocab', 'embed') Physical: (None, 'fsdp') I0422 16:04:16.106600 138014124525376 gradient_accumulation.py:70] gradient_accumulation/inputs Logical: float32[2048]............................................... Unknown. I0422 16:04:16.106674 138014124525376 gradient_accumulation.py:70] gradient_accumulation/inputs Physical: float32[2048]............................................... (None,). I0422 16:04:16.120894 138014124525376 gradient_accumulation.py:70] gradient_accumulation/inputs Logical: float32[2048,16,7168]....................................... Unknown. I0422 16:04:16.120949 138014124525376 gradient_accumulation.py:70] gradient_accumulation/inputs Physical: float32[2048,16,7168]....................................... ('fsdp', None, None). I0422 16:04:16.148986 138014124525376 gradient_accumulation.py:70] gradient_accumulation/inputs Logical: float32[7168,16,2048]....................................... Unknown. I0422 16:04:16.149050 138014124525376 gradient_accumulation.py:70] gradient_accumulation/inputs Physical: float32[7168,16,2048]....................................... (None, None, 'fsdp'). I0422 16:04:16.163075 138014124525376 gradient_accumulation.py:70] gradient_accumulation/inputs Logical: float32[2048,16]............................................ Unknown. I0422 16:04:16.163133 138014124525376 gradient_accumulation.py:70] gradient_accumulation/inputs Physical: float32[2048,16]............................................ (None, None). I0422 16:04:16.191334 138014124525376 gradient_accumulation.py:70] gradient_accumulation/inputs Logical: float32[2048,16,16,128]..................................... Unknown. I0422 16:04:16.191405 138014124525376 gradient_accumulation.py:70] gradient_accumulation/inputs Physical: float32[2048,16,16,128]..................................... ('fsdp', None, None, None). I0422 16:04:16.205621 138014124525376 gradient_accumulation.py:70] gradient_accumulation/inputs Logical: float32[16,16,128,2048]..................................... Unknown. I0422 16:04:16.205684 138014124525376 gradient_accumulation.py:70] gradient_accumulation/inputs Physical: float32[16,16,128,2048]..................................... (None, None, None, 'fsdp'). I0422 16:04:16.247596 138014124525376 gradient_accumulation.py:70] gradient_accumulation/inputs Logical: float32[2048,32000]......................................... Unknown. I0422 16:04:16.247679 138014124525376 gradient_accumulation.py:70] gradient_accumulation/inputs Physical: float32[2048,32000]......................................... ('fsdp', None). I0422 16:04:16.261758 138014124525376 gradient_accumulation.py:70] gradient_accumulation/inputs Logical: float32[32000,2048]......................................... Unknown. I0422 16:04:16.261814 138014124525376 gradient_accumulation.py:70] gradient_accumulation/inputs Physical: float32[32000,2048]......................................... (None, 'fsdp'). I0422 16:04:16.935576 138014124525376 train.py:158] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0422 16:04:16.935688 138014124525376 train.py:158] train/xent Physical: float32[32,2048]............................................ ('fsdp', None). I0422 16:04:16.951008 138014124525376 train.py:165] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0422 16:04:16.951072 138014124525376 train.py:165] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None). I0422 16:04:28.459981 138014124525376 max_utils.py:791] Total memory size: 1.7 GB, Output size: 0.4 GB, Temp size: 1.3 GB, Argument size: 0.4 GB, Host temp size: 0.0 GB. I0422 16:04:28.460880 138014124525376 metric_logger.py:289] number parameters: 1.104 billion I0422 16:04:30.564574 138014124525376 checkpointing.py:794] Waiting for step 0 to finish before checkpoint... I0422 16:04:40.484856 138014124525376 checkpointing.py:798] Waited 9.920260667800903 seconds for step 0 to finish before starting checkpointing. I0422 16:04:40.487525 138014124525376 checkpoint_manager.py:2013] [process=5][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0422 16:04:40.489225 138014124525376 checkpoint_manager.py:1518] [process=5] Saving checkpoint at step 0 I0422 16:04:40.492172 138014124525376 async_checkpointer.py:452] [process=5] Started async saving checkpoint to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_set_defaults_true_20260422_154647/linen_xpk_feat_nnx_set_defaults_true_20260422_154647_06_grad_accum/checkpoints/0. I0422 16:04:41.524940 138014124525376 signaling_client.py:364] Using JaxDistributedSignalingClient I0422 16:04:41.526050 138014124525376 jax_array_handlers.py:358] Scheduling D2H of 39 prioritized jax.Array. I0422 16:04:41.526109 138014124525376 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 I0422 16:04:41.771779 138014124525376 base_pytree_checkpoint_handler.py:153] [process=5][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.246917s I0422 16:04:41.771958 138014124525376 base_pytree_checkpoint_handler.py:129] [process=5] /jax/checkpoint/write/blocking_gbytes_per_sec: 2.803 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.5504209995269775 s) (per-host) I0422 16:04:41.772010 138014124525376 base_pytree_checkpoint_handler.py:737] [process=5][thread=MainThread] Initiated Pytree async_save. Time taken: 0.550482s (batch_requests_ready=0.299967s, total_serialization_initiated=0.250446s, others=0.000069s) I0422 16:04:41.772100 138014124525376 composite_checkpoint_handler.py:715] [process=5][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.554614s (all_items=0.000020s, per_item={'items': '0.00001979'}, temp_paths=0.554594) I0422 16:04:41.773052 137891619510016 async_checkpointer.py:79] [process=5][thread=async_save] Background save thread started. I0422 16:04:41.773230 138014124525376 async_checkpointer.py:561] Finished blocking save. Time taken: 1.283931s. Continuing background save to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_set_defaults_true_20260422_154647/linen_xpk_feat_nnx_set_defaults_true_20260422_154647_06_grad_accum/checkpoints/0. I0422 16:04:42.213572 138014124525376 checkpoint_manager.py:1566] [process=5][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize I0422 16:04:42.213977 137887894984448 async_checkpointer.py:265] [process=5][thread=save_finalize] Waiting for background save thread=async_save. I0422 16:04:42.214150 138014124525376 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_set_defaults_true_20260422_154647/linen_xpk_feat_nnx_set_defaults_true_20260422_154647_06_grad_accum/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776873880.487507, 'wait_for_prev_duration_secs': 6.246566772460938e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776873880.4892652, 'checkpointer_blocking_duration_secs': 1.28407621383667, 'get_old_steps_start_time': 1776873881.7733624, 'get_old_steps_duration_secs': 2.4557113647460938e-05, 'checkpoint_manager_blocking_start_time': 1776873880.485329, 'checkpoint_manager_blocking_duration_secs': 1.7287790775299072} I0422 16:04:42.214273 138014124525376 checkpointing.py:409] Started an asynchronous checkpoint save for step 0 I0422 16:04:42.214328 138014124525376 max_utils.py:750] Memstats: After params initialized: I0422 16:04:42.214385 138014124525376 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_18(process=5,(2,4,0,0)) I0422 16:04:42.214420 138014124525376 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_19(process=5,(3,4,0,0)) I0422 16:04:42.214448 138014124525376 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_22(process=5,(2,5,0,0)) I0422 16:04:42.214474 138014124525376 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_23(process=5,(3,5,0,0)) I0422 16:04:42.478670 138014124525376 metric_logger.py:185] completed step: 0, seconds: 2.104, TFLOP/s/device: 25.836, Tokens/s/device: 3894.308, total_weights: 262144, loss: 10.877 I0422 16:04:43.082641 138014124525376 metric_logger.py:185] completed step: 1, seconds: 11.913, TFLOP/s/device: 4.562, Tokens/s/device: 687.663, total_weights: 262144, loss: 10.877 I0422 16:04:43.661228 138014124525376 metric_logger.py:185] completed step: 2, seconds: 0.026, TFLOP/s/device: 2104.004, Tokens/s/device: 317138.322, total_weights: 262144, loss: 10.563 I0422 16:04:44.239331 138014124525376 metric_logger.py:185] completed step: 3, seconds: 0.584, TFLOP/s/device: 92.985, Tokens/s/device: 14015.638, total_weights: 262144, loss: 10.272 I0422 16:04:45.005341 2418 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com I0422 16:04:45.396310 138014124525376 metric_logger.py:185] completed step: 4, seconds: 0.579, TFLOP/s/device: 93.929, Tokens/s/device: 14157.946, total_weights: 262144, loss: 10.022 I0422 16:04:45.402490 138014124525376 metric_logger.py:185] completed step: 5, seconds: 0.578, TFLOP/s/device: 94.029, Tokens/s/device: 14173.059, total_weights: 262144, loss: 9.820 I0422 16:04:47.323291 137891627902720 array_metadata_store.py:203] [process=5][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_set_defaults_true_20260422_154647/linen_xpk_feat_nnx_set_defaults_true_20260422_154647_06_grad_accum/checkpoints/0/items/array_metadatas/process_5 I0422 16:05:19.820200 137891619510016 base_pytree_checkpoint_handler.py:129] [process=5] /jax/checkpoint/write/gbytes_per_sec: 40.924 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 38.59862995147705 s) (per-host) I0422 16:05:19.820304 137891619510016 async_checkpointer.py:90] [process=5][thread=async_save] 3 Handler Commit operations completed. Time taken: 38.047135s. I0422 16:05:23.290618 138014124525376 metric_logger.py:185] completed step: 6, seconds: 1.157, TFLOP/s/device: 46.959, Tokens/s/device: 7078.117, total_weights: 262144, loss: 9.667 I0422 16:05:23.869046 138014124525376 metric_logger.py:185] completed step: 7, seconds: 37.310, TFLOP/s/device: 1.457, Tokens/s/device: 219.563, total_weights: 262144, loss: 9.561 I0422 16:05:24.447412 138014124525376 metric_logger.py:185] completed step: 8, seconds: 0.584, TFLOP/s/device: 93.119, Tokens/s/device: 14035.953, total_weights: 262144, loss: 9.496 I0422 16:05:25.025564 138014124525376 checkpointing.py:794] Waiting for step 9 to finish before checkpoint... I0422 16:05:25.026235 138014124525376 checkpointing.py:798] Waited 0.000690460205078125 seconds for step 9 to finish before starting checkpointing. I0422 16:05:25.028208 138014124525376 checkpoint_manager.py:2024] [process=5][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0422 16:06:04.416134 137891619510016 async_checkpointer.py:144] [process=5][thread=async_save] Background save thread done. Time taken: 82.642949s. I0422 16:06:04.416446 137887894984448 async_checkpointer.py:273] [process=5][thread=save_finalize] Done with waiting for background save thread=async_save. I0422 16:06:04.416565 137887894984448 async_checkpointer.py:283] [process=5][thread=save_finalize] No errors found in background save thread=async_save. I0422 16:06:04.416617 137887894984448 checkpoint_manager.py:2133] [process=5][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts... I0422 16:06:04.419362 137887894984448 checkpoint_manager.py:2142] [process=5][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts. I0422 16:06:04.419527 138014124525376 checkpoint_manager.py:2036] [process=5][thread=MainThread][step=0][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=0. W0422 16:06:04.419669 138014124525376 checkpoint_manager.py:1458] Waiting for previous save to complete took 39.391454 seconds. If this number is high, consider checkpointing less frequently. I0422 16:06:04.421484 138014124525376 checkpoint_manager.py:1518] [process=5] Saving checkpoint at step 9 I0422 16:06:04.424819 138014124525376 async_checkpointer.py:452] [process=5] Started async saving checkpoint to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_set_defaults_true_20260422_154647/linen_xpk_feat_nnx_set_defaults_true_20260422_154647_06_grad_accum/checkpoints/9. I0422 16:06:05.441540 138014124525376 jax_array_handlers.py:358] Scheduling D2H of 39 prioritized jax.Array. I0422 16:06:05.441635 138014124525376 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 I0422 16:06:05.485134 138014124525376 base_pytree_checkpoint_handler.py:153] [process=5][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.044577s I0422 16:06:05.485309 138014124525376 base_pytree_checkpoint_handler.py:129] [process=5] /jax/checkpoint/write/blocking_gbytes_per_sec: 4.634 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.3329014778137207 s) (per-host) I0422 16:06:05.485363 138014124525376 base_pytree_checkpoint_handler.py:737] [process=5][thread=MainThread] Initiated Pytree async_save. Time taken: 0.332964s (batch_requests_ready=0.286426s, total_serialization_initiated=0.046468s, others=0.000070s) I0422 16:06:05.485454 138014124525376 composite_checkpoint_handler.py:715] [process=5][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.337332s (all_items=0.000015s, per_item={'items': '0.00001478'}, temp_paths=0.337317) I0422 16:06:05.486270 137887316260608 async_checkpointer.py:79] [process=5][thread=async_save] Background save thread started. I0422 16:06:05.486430 138014124525376 async_checkpointer.py:561] Finished blocking save. Time taken: 1.064875s. Continuing background save to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_set_defaults_true_20260422_154647/linen_xpk_feat_nnx_set_defaults_true_20260422_154647_06_grad_accum/checkpoints/9. I0422 16:06:05.896795 138014124525376 checkpoint_manager.py:1566] [process=5][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize I0422 16:06:05.897187 137887894984448 async_checkpointer.py:265] [process=5][thread=save_finalize] Waiting for background save thread=async_save. I0422 16:06:05.897362 138014124525376 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_set_defaults_true_20260422_154647/linen_xpk_feat_nnx_set_defaults_true_20260422_154647_06_grad_accum/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776873925.0281785, 'wait_for_prev_duration_secs': 39.391453981399536, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776873964.4215236, 'checkpointer_blocking_duration_secs': 1.0650091171264648, 'get_old_steps_start_time': 1776873965.486552, 'get_old_steps_duration_secs': 2.5510787963867188e-05, 'checkpoint_manager_blocking_start_time': 1776873925.0264513, 'checkpoint_manager_blocking_duration_secs': 40.87087321281433} I0422 16:06:05.897470 138014124525376 checkpointing.py:409] Started an asynchronous checkpoint save for step 9 I0422 16:06:05.897516 138014124525376 checkpoint_manager.py:2024] [process=5][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0422 16:06:11.014375 137887886591744 array_metadata_store.py:203] [process=5][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_set_defaults_true_20260422_154647/linen_xpk_feat_nnx_set_defaults_true_20260422_154647_06_grad_accum/checkpoints/9/items/array_metadatas/process_5 I0422 16:06:48.350872 137887316260608 base_pytree_checkpoint_handler.py:129] [process=5] /jax/checkpoint/write/gbytes_per_sec: 36.566 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 43.19842481613159 s) (per-host) I0422 16:06:48.350996 137887316260608 async_checkpointer.py:90] [process=5][thread=async_save] 3 Handler Commit operations completed. Time taken: 42.864611s. I0422 16:06:56.335738 137887316260608 async_checkpointer.py:144] [process=5][thread=async_save] Background save thread done. Time taken: 50.849339s. I0422 16:06:56.335974 137887894984448 async_checkpointer.py:273] [process=5][thread=save_finalize] Done with waiting for background save thread=async_save. I0422 16:06:56.336032 137887894984448 async_checkpointer.py:283] [process=5][thread=save_finalize] No errors found in background save thread=async_save. I0422 16:06:56.336088 137887894984448 checkpoint_manager.py:2133] [process=5][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts... I0422 16:06:56.337687 137887894984448 checkpoint_manager.py:2142] [process=5][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts. I0422 16:06:56.337867 138014124525376 checkpoint_manager.py:2036] [process=5][thread=MainThread][step=9][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=9. I0422 16:06:56.338026 138014124525376 checkpoint_manager.py:2013] [process=5][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0422 16:06:56.338691 138014124525376 metric_logger.py:185] completed step: 9, seconds: 0.578, TFLOP/s/device: 93.966, Tokens/s/device: 14163.576, total_weights: 262144, loss: 9.457 Per train step: Total TFLOPs: 54.35 split as 93.93% learnable weight flops and 6.07% attention flops XPK End: Wed Apr 22 16:07:06 UTC 2026 EXIT_CODE=0