main (e27fc1e97) vs feat/nnx-post-train-fixes (e27fc1e97)| Metric | main e27fc1e97 | feat/nnx-post-train-fixes e27fc1e97 | Diff (feat/nnx-post-train-fixes − main) |
|---|---|---|---|
| Parameters | 1.104 billion | 1.104 billion | — |
| Final loss | 9.4570 | 9.4570 | 0 |
| TFLOP/s | 93.998 | 93.968 | -0.03 |
| Tok/s | 14168.5 | 14163.9 | -4.58 |
| Avg s/step | 6.912 | 6.820 | -0.092 |
| Memory % | 1.38 | 1.38 | 0 |
| JAX | 0.9.2 | 0.9.2 | — |
Diff = branch value − main value. Green = branch improved. Red = branch regressed.
XPK Start: Tue Apr 21 06:30:22 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. 2026-04-21 06:30:46.948414: 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) I0421 06:30:47.157959 135941234521920 max_utils.py:273] Attempting to initialize the jax distributed system... I0421 06:30:56.197681 135941234521920 distributed.py:149] Starting JAX distributed service on [::]:8482 I0421 06:30:56.200051 135941234521920 distributed.py:172] Connecting to JAX distributed service on mt-06-grad-accum-oy565-slice-job-0-0.mt-06-grad-accum-oy565:8482 I0421 06:30:58.056890 135941234521920 max_utils.py:284] Jax distributed system initialized! I0421 06:31:03.344575 135941234521920 max_utils.py:800] System Information: Jax Version: 0.9.2 I0421 06:31:03.344700 135941234521920 max_utils.py:801] System Information: Jaxlib Version: 0.9.2 I0421 06:31:03.344744 135941234521920 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 I0421 06:31:03.344780 135941234521920 train_utils.py:361] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing. I0421 06:31:04.037041 135941234521920 maxtext_utils.py:1565] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0421 06:31:04.037316 135941234521920 checkpointing.py:677] Setting up checkpoint logger... I0421 06:31:04.037370 135941234521920 checkpointing.py:233] Creating checkpoint manager with ocdbt=True and zarr3=True I0421 06:31:04.037413 135941234521920 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None I0421 06:31:04.037777 135941234521920 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 0x7ba2a7f1c2f0>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0421 06:31:06.982145 135941234521920 checkpointing.py:265] Enabling policy for fixed interval checkpointing. I0421 06:31:06.982377 135941234521920 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 0x7b8f0c54d2e0>}, handler_registry=None I0421 06:31:06.982617 135941234521920 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7b8f0c54d2e0>` 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`. I0421 06:31:06.982680 135941234521920 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7b8f0c54f5c0>` 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`. I0421 06:31:06.982718 135941234521920 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 0x7b8f0c54d2e0>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7b8f0c54d2e0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7b8f0c54f5c0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7b8f0c54f5c0>}). I0421 06:31:06.983043 135941234521920 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34 I0421 06:31:06.983113 135941234521920 async_checkpointer.py:192] [process=4][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7b8f0c359800> timeout: 1200 secs and primary_host=0 for async checkpoint writes I0421 06:31:08.699543 135941234521920 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_main_20260421_061359/linen_xpk_main_20260421_061359_06_grad_accum/checkpoints I0421 06:31:08.770738 135941234521920 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_20260421_061359/linen_xpk_main_20260421_061359_06_grad_accum/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7b8f0c6a52b0> I0421 06:31:08.770902 135941234521920 checkpointing.py:301] Checkpoint manager created! I0421 06:31:10.476995 135941234521920 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0421 06:31:10.477108 135941234521920 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0421 06:31:10.857434 135941234521920 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0421 06:31:10.857526 135941234521920 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0421 06:31:10.874174 135941234521920 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0421 06:31:10.874234 135941234521920 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0421 06:31:10.898186 135941234521920 attentions.py:1154] attentions/query Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0421 06:31:10.898254 135941234521920 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0421 06:31:10.914897 135941234521920 attentions.py:1155] attentions/key Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0421 06:31:10.914959 135941234521920 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0421 06:31:10.931465 135941234521920 attentions.py:1156] attentions/value Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0421 06:31:10.931534 135941234521920 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0421 06:31:10.956271 135941234521920 attentions.py:1198] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv'). I0421 06:31:10.956340 135941234521920 attentions.py:1198] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0421 06:31:10.977402 135941234521920 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0421 06:31:10.977473 135941234521920 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None). I0421 06:31:11.192083 135941234521920 checkpointing.py:577] checkpoint manager exists so trying to load this run's existing checkpoint I0421 06:31:11.192193 135941234521920 checkpointing.py:665] No existing checkpoints found, not restoring checkpoint. fsdp: 32 I0421 06:31:12.626559 135941234521920 maxtext_utils.py:1668] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: P('norm',) Physical: (None,) I0421 06:31:12.626713 135941234521920 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) I0421 06:31:12.626785 135941234521920 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) I0421 06:31:12.626865 135941234521920 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') I0421 06:31:12.626935 135941234521920 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) I0421 06:31:12.626987 135941234521920 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) I0421 06:31:12.627057 135941234521920 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) I0421 06:31:12.627120 135941234521920 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') I0421 06:31:12.627164 135941234521920 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) I0421 06:31:12.627202 135941234521920 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) I0421 06:31:12.627256 135941234521920 maxtext_utils.py:1668] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: P('embed_vocab', 'vocab') Physical: ('fsdp', None) I0421 06:31:12.627305 135941234521920 maxtext_utils.py:1668] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: P('vocab', 'embed_vocab') Physical: (None, 'fsdp') I0421 06:31:12.650287 135941234521920 gradient_accumulation.py:68] gradient_accumulation/inputs Logical: float32[2048]............................................... Unknown. I0421 06:31:12.650348 135941234521920 gradient_accumulation.py:68] gradient_accumulation/inputs Physical: float32[2048]............................................... (None,). I0421 06:31:12.665192 135941234521920 gradient_accumulation.py:68] gradient_accumulation/inputs Logical: float32[2048,16,7168]....................................... Unknown. I0421 06:31:12.665252 135941234521920 gradient_accumulation.py:68] gradient_accumulation/inputs Physical: float32[2048,16,7168]....................................... ('fsdp', None, None). I0421 06:31:12.694791 135941234521920 gradient_accumulation.py:68] gradient_accumulation/inputs Logical: float32[7168,16,2048]....................................... Unknown. I0421 06:31:12.694857 135941234521920 gradient_accumulation.py:68] gradient_accumulation/inputs Physical: float32[7168,16,2048]....................................... (None, None, 'fsdp'). I0421 06:31:12.709995 135941234521920 gradient_accumulation.py:68] gradient_accumulation/inputs Logical: float32[2048,16]............................................ Unknown. I0421 06:31:12.710053 135941234521920 gradient_accumulation.py:68] gradient_accumulation/inputs Physical: float32[2048,16]............................................ (None, None). I0421 06:31:12.739756 135941234521920 gradient_accumulation.py:68] gradient_accumulation/inputs Logical: float32[2048,16,16,128]..................................... Unknown. I0421 06:31:12.739818 135941234521920 gradient_accumulation.py:68] gradient_accumulation/inputs Physical: float32[2048,16,16,128]..................................... ('fsdp', None, None, None). I0421 06:31:12.754739 135941234521920 gradient_accumulation.py:68] gradient_accumulation/inputs Logical: float32[16,16,128,2048]..................................... Unknown. I0421 06:31:12.754795 135941234521920 gradient_accumulation.py:68] gradient_accumulation/inputs Physical: float32[16,16,128,2048]..................................... (None, None, None, 'fsdp'). I0421 06:31:12.799152 135941234521920 gradient_accumulation.py:68] gradient_accumulation/inputs Logical: float32[2048,32000]......................................... Unknown. I0421 06:31:12.799225 135941234521920 gradient_accumulation.py:68] gradient_accumulation/inputs Physical: float32[2048,32000]......................................... ('fsdp', None). I0421 06:31:12.814071 135941234521920 gradient_accumulation.py:68] gradient_accumulation/inputs Logical: float32[32000,2048]......................................... Unknown. I0421 06:31:12.814124 135941234521920 gradient_accumulation.py:68] gradient_accumulation/inputs Physical: float32[32000,2048]......................................... (None, 'fsdp'). I0421 06:31:13.490487 135941234521920 train.py:155] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0421 06:31:13.490576 135941234521920 train.py:155] train/xent Physical: float32[32,2048]............................................ ('fsdp', None). I0421 06:31:13.506364 135941234521920 train.py:162] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0421 06:31:13.506426 135941234521920 train.py:162] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None). I0421 06:31:24.928046 135941234521920 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. I0421 06:31:24.928933 135941234521920 metric_logger.py:301] number parameters: 1.104 billion I0421 06:31:27.256673 135941234521920 checkpointing.py:772] Waiting for step 0 to finish before checkpoint... I0421 06:31:37.864723 135941234521920 checkpointing.py:776] Waited 10.60804557800293 seconds for step 0 to finish before starting checkpointing. I0421 06:31:37.867156 135941234521920 checkpoint_manager.py:2009] [process=4][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0421 06:31:37.869114 135941234521920 checkpoint_manager.py:1512] [process=4] Saving checkpoint at step 0 I0421 06:31:37.870565 135941234521920 event_tracking.py:70] [process=4] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260421_061359/linen_xpk_main_20260421_061359_06_grad_accum/checkpoints/0. I0421 06:31:38.173172 135941234521920 signaling_client.py:364] Using JaxDistributedSignalingClient I0421 06:31:38.174068 135941234521920 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0421 06:31:38.174126 135941234521920 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 I0421 06:31:38.452786 135941234521920 base_pytree_checkpoint_handler.py:154] [process=4][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.279686s I0421 06:31:38.452960 135941234521920 base_pytree_checkpoint_handler.py:130] [process=4] /jax/orbax/write/blocking_gbytes_per_sec: 5.409 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.2851707935333252 s) (per-host) I0421 06:31:38.453012 135941234521920 base_pytree_checkpoint_handler.py:768] [process=4][thread=MainThread] Initiated Pytree async_save. Time taken: 0.285233s (batch_requests_ready=0.002219s, total_serialization_initiated=0.282945s, others=0.000070s) I0421 06:31:38.453107 135941234521920 composite_checkpoint_handler.py:715] [process=4][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.289412s (all_items=0.000017s, per_item={'items': '0.00001740'}, temp_paths=0.289394) I0421 06:31:38.453904 135941234521920 event_tracking.py:125] [process=4] [async] Finished blocking save in 0.58 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260421_061359/linen_xpk_main_20260421_061359_06_grad_accum/checkpoints/0. I0421 06:31:38.454233 135816621709056 async_checkpointer.py:76] [process=4][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-21 06:51:38.454194 I0421 06:31:38.456439 135941234521920 checkpoint_manager.py:1560] [process=4][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize I0421 06:31:38.456696 135815581656832 async_checkpointer.py:280] [process=4][thread=save_finalize] Waiting for background save thread=async_save. I0421 06:31:38.456821 135941234521920 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260421_061359/linen_xpk_main_20260421_061359_06_grad_accum/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776753097.8671374, 'wait_for_prev_duration_secs': 6.103515625e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776753097.8691542, 'checkpointer_blocking_duration_secs': 0.5852401256561279, 'get_old_steps_start_time': 1776753098.454418, 'get_old_steps_duration_secs': 2.956390380859375e-05, 'checkpoint_manager_blocking_start_time': 1776753097.86527, 'checkpoint_manager_blocking_duration_secs': 0.5915131568908691} I0421 06:31:38.456935 135941234521920 checkpointing.py:408] Started an asynchronous checkpoint save for step 0 I0421 06:31:38.456985 135941234521920 max_utils.py:750] Memstats: After params initialized: I0421 06:31:38.457034 135941234521920 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_16(process=4,(0,4,0,0)) I0421 06:31:38.457068 135941234521920 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_17(process=4,(1,4,0,0)) I0421 06:31:38.457098 135941234521920 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_20(process=4,(0,5,0,0)) I0421 06:31:38.457123 135941234521920 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_21(process=4,(1,5,0,0)) I0421 06:31:38.773393 135941234521920 metric_logger.py:196] completed step: 0, seconds: 2.328, TFLOP/s/device: 23.350, Tokens/s/device: 3519.511, total_weights: 262144, loss: 10.877, lm_loss: 10.877, perplexity: 52959.059 I0421 06:31:39.374176 135941234521920 metric_logger.py:196] completed step: 1, seconds: 11.515, TFLOP/s/device: 4.720, Tokens/s/device: 711.404, total_weights: 262144, loss: 10.877, lm_loss: 10.877, perplexity: 52959.059 I0421 06:31:40.218439 135941234521920 metric_logger.py:196] completed step: 2, seconds: 0.022, TFLOP/s/device: 2433.988, Tokens/s/device: 366877.155, total_weights: 262144, loss: 10.563, lm_loss: 10.563, perplexity: 38662.707 I0421 06:31:40.796422 135941234521920 metric_logger.py:196] completed step: 3, seconds: 0.845, TFLOP/s/device: 64.329, Tokens/s/device: 9696.384, total_weights: 262144, loss: 10.272, lm_loss: 10.272, perplexity: 28909.668 I0421 06:31:41.460438 2494 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com I0421 06:31:41.953381 135941234521920 metric_logger.py:196] completed step: 4, seconds: 0.584, TFLOP/s/device: 93.089, Tokens/s/device: 14031.434, total_weights: 262144, loss: 10.022, lm_loss: 10.022, perplexity: 22524.992 I0421 06:31:41.959660 135941234521920 metric_logger.py:196] completed step: 5, seconds: 0.578, TFLOP/s/device: 94.046, Tokens/s/device: 14175.635, total_weights: 262144, loss: 9.820, lm_loss: 9.820, perplexity: 18401.865 I0421 06:31:43.933622 135815590049536 array_metadata_store.py:203] [process=4][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260421_061359/linen_xpk_main_20260421_061359_06_grad_accum/checkpoints/0/items/array_metadatas/process_4 I0421 06:32:15.445528 135816621709056 base_pytree_checkpoint_handler.py:130] [process=4] /jax/orbax/write/gbytes_per_sec: 42.374 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 37.27769994735718 s) (per-host) I0421 06:32:15.445672 135816621709056 async_checkpointer.py:90] [process=4][thread=async_save] 3 Handler Commit operations completed. Time taken: 36.991324s. I0421 06:32:25.801173 135816621709056 async_checkpointer.py:160] [process=4][thread=async_save] Background save thread done. Time taken: 47.346809s. I0421 06:32:25.801456 135815581656832 async_checkpointer.py:288] [process=4][thread=save_finalize] Done with waiting for background save thread=async_save. I0421 06:32:25.801576 135815581656832 async_checkpointer.py:298] [process=4][thread=save_finalize] No errors found in background save thread=async_save. I0421 06:32:25.801625 135815581656832 checkpoint_manager.py:2137] [process=4][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts... I0421 06:32:25.966241 135815581656832 checkpoint_manager.py:2146] [process=4][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts. I0421 06:32:28.881892 135941234521920 metric_logger.py:196] completed step: 6, seconds: 1.158, TFLOP/s/device: 46.945, Tokens/s/device: 7076.099, total_weights: 262144, loss: 9.667, lm_loss: 9.667, perplexity: 15787.604 I0421 06:32:29.460138 135941234521920 metric_logger.py:196] completed step: 7, seconds: 46.345, TFLOP/s/device: 1.173, Tokens/s/device: 176.762, total_weights: 262144, loss: 9.561, lm_loss: 9.561, perplexity: 14203.827 I0421 06:32:30.038604 135941234521920 metric_logger.py:196] completed step: 8, seconds: 0.583, TFLOP/s/device: 93.199, Tokens/s/device: 14048.036, total_weights: 262144, loss: 9.496, lm_loss: 9.496, perplexity: 13302.920 I0421 06:32:30.616393 135941234521920 checkpointing.py:772] Waiting for step 9 to finish before checkpoint... I0421 06:32:30.617103 135941234521920 checkpointing.py:776] Waited 0.0007312297821044922 seconds for step 9 to finish before starting checkpointing. I0421 06:32:30.619136 135941234521920 checkpoint_manager.py:2009] [process=4][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0421 06:32:30.620834 135941234521920 checkpoint_manager.py:1512] [process=4] Saving checkpoint at step 9 I0421 06:32:30.622290 135941234521920 event_tracking.py:70] [process=4] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260421_061359/linen_xpk_main_20260421_061359_06_grad_accum/checkpoints/9. I0421 06:32:31.335049 135941234521920 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0421 06:32:31.335143 135941234521920 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 I0421 06:32:31.369832 135941234521920 base_pytree_checkpoint_handler.py:154] [process=4][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.035715s I0421 06:32:31.369993 135941234521920 base_pytree_checkpoint_handler.py:130] [process=4] /jax/orbax/write/blocking_gbytes_per_sec: 39.372 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.039179325103759766 s) (per-host) I0421 06:32:31.370044 135941234521920 base_pytree_checkpoint_handler.py:768] [process=4][thread=MainThread] Initiated Pytree async_save. Time taken: 0.039241s (batch_requests_ready=0.001648s, total_serialization_initiated=0.037523s, others=0.000070s) I0421 06:32:31.370131 135941234521920 composite_checkpoint_handler.py:715] [process=4][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.044031s (all_items=0.000016s, per_item={'items': '0.00001597'}, temp_paths=0.044015) I0421 06:32:31.370852 135941234521920 event_tracking.py:125] [process=4] [async] Finished blocking save in 0.75 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260421_061359/linen_xpk_main_20260421_061359_06_grad_accum/checkpoints/9. I0421 06:32:31.371210 135816123553536 async_checkpointer.py:76] [process=4][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-21 06:52:31.371168 I0421 06:32:31.373405 135941234521920 checkpoint_manager.py:1560] [process=4][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize I0421 06:32:31.373727 135815581656832 async_checkpointer.py:280] [process=4][thread=save_finalize] Waiting for background save thread=async_save. I0421 06:32:31.373892 135941234521920 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260421_061359/linen_xpk_main_20260421_061359_06_grad_accum/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776753150.6191027, 'wait_for_prev_duration_secs': 7.271766662597656e-05, 'time_between_consecutive_saves_sec': 4.652761459350586, 'checkpointer_blocking_start_time': 1776753150.620874, 'checkpointer_blocking_duration_secs': 0.7504978179931641, 'get_old_steps_start_time': 1776753151.371396, 'get_old_steps_duration_secs': 3.0040740966796875e-05, 'checkpoint_manager_blocking_start_time': 1776753150.6173284, 'checkpoint_manager_blocking_duration_secs': 0.7565262317657471} I0421 06:32:31.374003 135941234521920 checkpointing.py:408] Started an asynchronous checkpoint save for step 9 I0421 06:32:31.374047 135941234521920 checkpoint_manager.py:2020] [process=4][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0421 06:32:36.460682 135816621709056 array_metadata_store.py:203] [process=4][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260421_061359/linen_xpk_main_20260421_061359_06_grad_accum/checkpoints/9/items/array_metadatas/process_4 I0421 06:33:13.288189 135816123553536 base_pytree_checkpoint_handler.py:130] [process=4] /jax/orbax/write/gbytes_per_sec: 37.648 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 41.957337379455566 s) (per-host) I0421 06:33:13.288316 135816123553536 async_checkpointer.py:90] [process=4][thread=async_save] 3 Handler Commit operations completed. Time taken: 41.916981s. I0421 06:33:21.296235 135816123553536 async_checkpointer.py:160] [process=4][thread=async_save] Background save thread done. Time taken: 49.924885s. I0421 06:33:21.296542 135815581656832 async_checkpointer.py:288] [process=4][thread=save_finalize] Done with waiting for background save thread=async_save. I0421 06:33:21.296677 135815581656832 async_checkpointer.py:298] [process=4][thread=save_finalize] No errors found in background save thread=async_save. I0421 06:33:21.296727 135815581656832 checkpoint_manager.py:2137] [process=4][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts... I0421 06:33:21.298114 135815581656832 checkpoint_manager.py:2146] [process=4][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts. I0421 06:33:21.298375 135941234521920 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. I0421 06:33:21.298532 135941234521920 checkpoint_manager.py:2009] [process=4][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0421 06:33:21.299526 135941234521920 metric_logger.py:196] completed step: 9, seconds: 0.578, TFLOP/s/device: 93.998, Tokens/s/device: 14168.475, 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: Tue Apr 21 06:33:32 UTC 2026 EXIT_CODE=0
XPK Start: Tue Apr 21 11:57:45 UTC 2026 PyTorch was not found. Models won't be available and only tokenizers, configuration and file/data utilities can be used. 2026-04-21 11:58:09.532503: 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) I0421 11:58:10.069130 138422609975104 max_utils.py:273] Attempting to initialize the jax distributed system... I0421 11:58:19.110966 138422609975104 distributed.py:149] Starting JAX distributed service on [::]:8482 I0421 11:58:19.113394 138422609975104 distributed.py:172] Connecting to JAX distributed service on mt-06-grad-accum-giv95-slice-job-0-0.mt-06-grad-accum-giv95:8482 I0421 11:58:20.494451 138422609975104 max_utils.py:284] Jax distributed system initialized! I0421 11:58:25.607234 138422609975104 max_utils.py:800] System Information: Jax Version: 0.9.2 I0421 11:58:25.607338 138422609975104 max_utils.py:801] System Information: Jaxlib Version: 0.9.2 I0421 11:58:25.607381 138422609975104 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 I0421 11:58:25.607417 138422609975104 train_utils.py:378] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing. I0421 11:58:26.306321 138422609975104 maxtext_utils.py:1718] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0421 11:58:26.306918 138422609975104 maxtext_utils.py:1718] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0421 11:58:26.307091 138422609975104 checkpointing.py:688] Setting up checkpoint logger... I0421 11:58:26.307140 138422609975104 checkpointing.py:234] Creating checkpoint manager with ocdbt=True and zarr3=True I0421 11:58:26.307183 138422609975104 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None I0421 11:58:26.307520 138422609975104 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 0x7de46e1915e0>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0421 11:58:29.728422 138422609975104 checkpointing.py:266] Enabling policy for fixed interval checkpointing. I0421 11:58:29.728669 138422609975104 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 0x7dd0ac454200>}, handler_registry=None I0421 11:58:29.728908 138422609975104 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7dd0ac454200>` 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`. I0421 11:58:29.728955 138422609975104 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7dd0ac456120>` 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`. I0421 11:58:29.728991 138422609975104 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 0x7dd0ac454200>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7dd0ac454200>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7dd0ac456120>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7dd0ac456120>}). I0421 11:58:29.729306 138422609975104 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34 I0421 11:58:29.729377 138422609975104 async_checkpointer.py:192] [process=4][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7dcfe47d8180> timeout: 1200 secs and primary_host=0 for async checkpoint writes I0421 11:58:30.431994 138422609975104 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260421_114057/linen_xpk_feat_nnx_post_train_fixes_20260421_114057_06_grad_accum/checkpoints I0421 11:58:30.597281 138422609975104 checkpoint_manager.py:929] [process=4][thread=MainThread] CheckpointManager created, primary_host=0, CheckpointManagerOptions=CheckpointManagerOptions(save_interval_steps=1, max_to_keep=None, keep_time_interval=None, keep_period=None, should_keep_fn=None, best_fn=None, best_mode='max', keep_checkpoints_without_metrics=True, step_prefix=None, step_format_fixed_length=None, step_name_format=None, create=True, cleanup_tmp_directories=False, save_on_steps=frozenset(), single_host_load_and_broadcast=False, todelete_subdir=None, todelete_full_path=None, enable_background_delete=False, read_only=False, enable_async_checkpointing=True, async_options=None, multiprocessing_options=MultiprocessingOptions(primary_host=0, active_processes=None, barrier_sync_key_prefix=None), should_save_fn=None, file_options=FileOptions(path_permission_mode=None), save_root_metadata=True, temporary_path_class=None, save_decision_policy=FixedIntervalPolicy(interval=10), preservation_policy=LatestN(n=None), prevent_write_metrics=False, enable_should_save_is_saving_in_progress_check=True, enable_per_process_directory_creation=False, lightweight_initialize=False), root_directory=gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260421_114057/linen_xpk_feat_nnx_post_train_fixes_20260421_114057_06_grad_accum/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7dd0ac457ce0> I0421 11:58:30.597448 138422609975104 checkpointing.py:302] Checkpoint manager created! I0421 11:58:32.312400 138422609975104 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0421 11:58:32.312504 138422609975104 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0421 11:58:32.694720 138422609975104 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0421 11:58:32.694805 138422609975104 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0421 11:58:32.711301 138422609975104 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0421 11:58:32.711359 138422609975104 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0421 11:58:32.735124 138422609975104 attentions.py:1154] attentions/query Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0421 11:58:32.735190 138422609975104 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0421 11:58:32.751698 138422609975104 attentions.py:1155] attentions/key Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0421 11:58:32.751755 138422609975104 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0421 11:58:32.768275 138422609975104 attentions.py:1156] attentions/value Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0421 11:58:32.768334 138422609975104 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0421 11:58:32.793752 138422609975104 attentions.py:1197] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv'). I0421 11:58:32.793820 138422609975104 attentions.py:1197] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0421 11:58:32.814820 138422609975104 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0421 11:58:32.814884 138422609975104 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None). I0421 11:58:33.029616 138422609975104 checkpointing.py:578] checkpoint manager exists so trying to load this run's existing checkpoint I0421 11:58:33.029734 138422609975104 checkpointing.py:676] No existing checkpoints found, not restoring checkpoint. fsdp: 32 I0421 11:58:34.448815 138422609975104 maxtext_utils.py:1827] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: P('norm',) Physical: (None,) I0421 11:58:34.448936 138422609975104 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) I0421 11:58:34.448989 138422609975104 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) I0421 11:58:34.449045 138422609975104 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') I0421 11:58:34.449096 138422609975104 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) I0421 11:58:34.449134 138422609975104 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) I0421 11:58:34.449190 138422609975104 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) I0421 11:58:34.449245 138422609975104 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') I0421 11:58:34.449290 138422609975104 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) I0421 11:58:34.449328 138422609975104 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) I0421 11:58:34.449376 138422609975104 maxtext_utils.py:1827] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: P('embed_vocab', 'vocab') Physical: ('fsdp', None) I0421 11:58:34.449425 138422609975104 maxtext_utils.py:1827] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: P('vocab', 'embed_vocab') Physical: (None, 'fsdp') I0421 11:58:34.474538 138422609975104 gradient_accumulation.py:70] gradient_accumulation/inputs Logical: float32[2048]............................................... Unknown. I0421 11:58:34.474607 138422609975104 gradient_accumulation.py:70] gradient_accumulation/inputs Physical: float32[2048]............................................... (None,). I0421 11:58:34.489576 138422609975104 gradient_accumulation.py:70] gradient_accumulation/inputs Logical: float32[2048,16,7168]....................................... Unknown. I0421 11:58:34.489652 138422609975104 gradient_accumulation.py:70] gradient_accumulation/inputs Physical: float32[2048,16,7168]....................................... ('fsdp', None, None). I0421 11:58:34.519383 138422609975104 gradient_accumulation.py:70] gradient_accumulation/inputs Logical: float32[7168,16,2048]....................................... Unknown. I0421 11:58:34.519447 138422609975104 gradient_accumulation.py:70] gradient_accumulation/inputs Physical: float32[7168,16,2048]....................................... (None, None, 'fsdp'). I0421 11:58:34.534215 138422609975104 gradient_accumulation.py:70] gradient_accumulation/inputs Logical: float32[2048,16]............................................ Unknown. I0421 11:58:34.534273 138422609975104 gradient_accumulation.py:70] gradient_accumulation/inputs Physical: float32[2048,16]............................................ (None, None). I0421 11:58:34.563748 138422609975104 gradient_accumulation.py:70] gradient_accumulation/inputs Logical: float32[2048,16,16,128]..................................... Unknown. I0421 11:58:34.563811 138422609975104 gradient_accumulation.py:70] gradient_accumulation/inputs Physical: float32[2048,16,16,128]..................................... ('fsdp', None, None, None). I0421 11:58:34.578627 138422609975104 gradient_accumulation.py:70] gradient_accumulation/inputs Logical: float32[16,16,128,2048]..................................... Unknown. I0421 11:58:34.578696 138422609975104 gradient_accumulation.py:70] gradient_accumulation/inputs Physical: float32[16,16,128,2048]..................................... (None, None, None, 'fsdp'). I0421 11:58:34.623164 138422609975104 gradient_accumulation.py:70] gradient_accumulation/inputs Logical: float32[2048,32000]......................................... Unknown. I0421 11:58:34.623234 138422609975104 gradient_accumulation.py:70] gradient_accumulation/inputs Physical: float32[2048,32000]......................................... ('fsdp', None). I0421 11:58:34.638045 138422609975104 gradient_accumulation.py:70] gradient_accumulation/inputs Logical: float32[32000,2048]......................................... Unknown. I0421 11:58:34.638101 138422609975104 gradient_accumulation.py:70] gradient_accumulation/inputs Physical: float32[32000,2048]......................................... (None, 'fsdp'). I0421 11:58:35.305403 138422609975104 train.py:157] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0421 11:58:35.305498 138422609975104 train.py:157] train/xent Physical: float32[32,2048]............................................ ('fsdp', None). I0421 11:58:35.321212 138422609975104 train.py:164] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0421 11:58:35.321280 138422609975104 train.py:164] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None). I0421 11:58:46.619724 138422609975104 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. I0421 11:58:46.620566 138422609975104 metric_logger.py:301] number parameters: 1.104 billion I0421 11:58:48.472834 138422609975104 checkpointing.py:794] Waiting for step 0 to finish before checkpoint... I0421 11:58:59.658997 138422609975104 checkpointing.py:798] Waited 11.186141014099121 seconds for step 0 to finish before starting checkpointing. I0421 11:58:59.661560 138422609975104 checkpoint_manager.py:2009] [process=4][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0421 11:58:59.663246 138422609975104 checkpoint_manager.py:1512] [process=4] Saving checkpoint at step 0 I0421 11:58:59.664793 138422609975104 event_tracking.py:70] [process=4] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260421_114057/linen_xpk_feat_nnx_post_train_fixes_20260421_114057_06_grad_accum/checkpoints/0. I0421 11:59:00.012786 138422609975104 signaling_client.py:364] Using JaxDistributedSignalingClient I0421 11:59:00.013822 138422609975104 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0421 11:59:00.013879 138422609975104 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 I0421 11:59:00.287029 138422609975104 base_pytree_checkpoint_handler.py:154] [process=4][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.274315s I0421 11:59:00.287199 138422609975104 base_pytree_checkpoint_handler.py:130] [process=4] /jax/orbax/write/blocking_gbytes_per_sec: 5.516 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.2796647548675537 s) (per-host) I0421 11:59:00.287251 138422609975104 base_pytree_checkpoint_handler.py:768] [process=4][thread=MainThread] Initiated Pytree async_save. Time taken: 0.279726s (batch_requests_ready=0.002196s, total_serialization_initiated=0.277461s, others=0.000069s) I0421 11:59:00.287375 138422609975104 composite_checkpoint_handler.py:715] [process=4][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.284456s (all_items=0.000018s, per_item={'items': '0.00001812'}, temp_paths=0.284438) I0421 11:59:00.288218 138422609975104 event_tracking.py:125] [process=4] [async] Finished blocking save in 0.62 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260421_114057/linen_xpk_feat_nnx_post_train_fixes_20260421_114057_06_grad_accum/checkpoints/0. I0421 11:59:00.288557 138297066120960 async_checkpointer.py:76] [process=4][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-21 12:19:00.288518 I0421 11:59:00.304176 138422609975104 checkpoint_manager.py:1560] [process=4][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize I0421 11:59:00.304455 138295463900928 async_checkpointer.py:280] [process=4][thread=save_finalize] Waiting for background save thread=async_save. I0421 11:59:00.304612 138422609975104 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260421_114057/linen_xpk_feat_nnx_post_train_fixes_20260421_114057_06_grad_accum/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776772739.6615417, 'wait_for_prev_duration_secs': 6.175041198730469e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776772739.6632853, 'checkpointer_blocking_duration_secs': 0.6253917217254639, 'get_old_steps_start_time': 1776772740.2887044, 'get_old_steps_duration_secs': 3.075599670410156e-05, 'checkpoint_manager_blocking_start_time': 1776772739.6594992, 'checkpoint_manager_blocking_duration_secs': 0.6450731754302979} I0421 11:59:00.304729 138422609975104 checkpointing.py:409] Started an asynchronous checkpoint save for step 0 I0421 11:59:00.304782 138422609975104 max_utils.py:750] Memstats: After params initialized: I0421 11:59:00.304829 138422609975104 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_16(process=4,(0,4,0,0)) I0421 11:59:00.304864 138422609975104 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_17(process=4,(1,4,0,0)) I0421 11:59:00.304891 138422609975104 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_20(process=4,(0,5,0,0)) I0421 11:59:00.304917 138422609975104 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_21(process=4,(1,5,0,0)) I0421 11:59:00.615966 138422609975104 metric_logger.py:196] completed step: 0, seconds: 1.852, TFLOP/s/device: 29.343, Tokens/s/device: 4422.939, total_weights: 262144, loss: 10.877, lm_loss: 10.877, perplexity: 52959.059 I0421 11:59:01.222781 138422609975104 metric_logger.py:196] completed step: 1, seconds: 12.142, TFLOP/s/device: 4.476, Tokens/s/device: 674.703, total_weights: 262144, loss: 10.877, lm_loss: 10.877, perplexity: 52959.059 I0421 11:59:01.800925 138422609975104 metric_logger.py:196] completed step: 2, seconds: 0.028, TFLOP/s/device: 1924.250, Tokens/s/device: 290043.903, total_weights: 262144, loss: 10.563, lm_loss: 10.563, perplexity: 38662.707 I0421 11:59:02.379273 138422609975104 metric_logger.py:196] completed step: 3, seconds: 0.585, TFLOP/s/device: 92.980, Tokens/s/device: 14014.966, total_weights: 262144, loss: 10.272, lm_loss: 10.272, perplexity: 28909.668 I0421 11:59:03.535877 138422609975104 metric_logger.py:196] completed step: 4, seconds: 0.578, TFLOP/s/device: 93.995, Tokens/s/device: 14167.912, total_weights: 262144, loss: 10.022, lm_loss: 10.022, perplexity: 22524.992 I0421 11:59:03.541805 138422609975104 metric_logger.py:196] completed step: 5, seconds: 0.578, TFLOP/s/device: 93.975, Tokens/s/device: 14164.972, total_weights: 262144, loss: 9.820, lm_loss: 9.820, perplexity: 18401.865 I0421 11:59:03.744696 2564 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com I0421 11:59:05.783726 138295946417920 array_metadata_store.py:203] [process=4][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260421_114057/linen_xpk_feat_nnx_post_train_fixes_20260421_114057_06_grad_accum/checkpoints/0/items/array_metadatas/process_4 I0421 11:59:38.065600 138297066120960 base_pytree_checkpoint_handler.py:130] [process=4] /jax/orbax/write/gbytes_per_sec: 41.505 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 38.0580267906189 s) (per-host) I0421 11:59:38.065731 138297066120960 async_checkpointer.py:90] [process=4][thread=async_save] 3 Handler Commit operations completed. Time taken: 37.776525s. I0421 11:59:46.122105 138297066120960 async_checkpointer.py:160] [process=4][thread=async_save] Background save thread done. Time taken: 45.832881s. I0421 11:59:46.122425 138295463900928 async_checkpointer.py:288] [process=4][thread=save_finalize] Done with waiting for background save thread=async_save. I0421 11:59:46.122548 138295463900928 async_checkpointer.py:298] [process=4][thread=save_finalize] No errors found in background save thread=async_save. I0421 11:59:46.122603 138295463900928 checkpoint_manager.py:2137] [process=4][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts... I0421 11:59:46.125265 138295463900928 checkpoint_manager.py:2146] [process=4][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts. I0421 11:59:49.270457 138422609975104 metric_logger.py:196] completed step: 6, seconds: 1.157, TFLOP/s/device: 46.974, Tokens/s/device: 7080.435, total_weights: 262144, loss: 9.667, lm_loss: 9.667, perplexity: 15787.604 I0421 11:59:49.848788 138422609975104 metric_logger.py:196] completed step: 7, seconds: 45.151, TFLOP/s/device: 1.204, Tokens/s/device: 181.435, total_weights: 262144, loss: 9.561, lm_loss: 9.561, perplexity: 14203.827 I0421 11:59:50.427253 138422609975104 metric_logger.py:196] completed step: 8, seconds: 0.583, TFLOP/s/device: 93.196, Tokens/s/device: 14047.530, total_weights: 262144, loss: 9.496, lm_loss: 9.496, perplexity: 13302.920 I0421 11:59:51.005079 138422609975104 checkpointing.py:794] Waiting for step 9 to finish before checkpoint... I0421 11:59:51.005765 138422609975104 checkpointing.py:798] Waited 0.0007076263427734375 seconds for step 9 to finish before starting checkpointing. I0421 11:59:51.007860 138422609975104 checkpoint_manager.py:2009] [process=4][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0421 11:59:51.009447 138422609975104 checkpoint_manager.py:1512] [process=4] Saving checkpoint at step 9 I0421 11:59:51.010929 138422609975104 event_tracking.py:70] [process=4] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260421_114057/linen_xpk_feat_nnx_post_train_fixes_20260421_114057_06_grad_accum/checkpoints/9. I0421 11:59:51.714443 138422609975104 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0421 11:59:51.714541 138422609975104 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 I0421 11:59:51.748567 138422609975104 base_pytree_checkpoint_handler.py:154] [process=4][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.035086s I0421 11:59:51.748705 138422609975104 base_pytree_checkpoint_handler.py:130] [process=4] /jax/orbax/write/blocking_gbytes_per_sec: 40.132 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.038437843322753906 s) (per-host) I0421 11:59:51.748753 138422609975104 base_pytree_checkpoint_handler.py:768] [process=4][thread=MainThread] Initiated Pytree async_save. Time taken: 0.038493s (batch_requests_ready=0.001637s, total_serialization_initiated=0.036794s, others=0.000062s) I0421 11:59:51.748825 138422609975104 composite_checkpoint_handler.py:715] [process=4][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.042634s (all_items=0.000015s, per_item={'items': '0.00001526'}, temp_paths=0.042619) I0421 11:59:51.749476 138422609975104 event_tracking.py:125] [process=4] [async] Finished blocking save in 0.74 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260421_114057/linen_xpk_feat_nnx_post_train_fixes_20260421_114057_06_grad_accum/checkpoints/9. I0421 11:59:51.749742 138301319124736 async_checkpointer.py:76] [process=4][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-21 12:19:51.749720 I0421 11:59:51.756811 138422609975104 checkpoint_manager.py:1560] [process=4][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize I0421 11:59:51.757045 138297056540416 async_checkpointer.py:280] [process=4][thread=save_finalize] Waiting for background save thread=async_save. I0421 11:59:51.757172 138422609975104 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260421_114057/linen_xpk_feat_nnx_post_train_fixes_20260421_114057_06_grad_accum/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776772791.007829, 'wait_for_prev_duration_secs': 7.343292236328125e-05, 'time_between_consecutive_saves_sec': 4.88252329826355, 'checkpointer_blocking_start_time': 1776772791.0094848, 'checkpointer_blocking_duration_secs': 0.7403607368469238, 'get_old_steps_start_time': 1776772791.7498696, 'get_old_steps_duration_secs': 2.8133392333984375e-05, 'checkpoint_manager_blocking_start_time': 1776772791.0059938, 'checkpoint_manager_blocking_duration_secs': 0.7511427402496338} I0421 11:59:51.757286 138422609975104 checkpointing.py:409] Started an asynchronous checkpoint save for step 9 I0421 11:59:51.757333 138422609975104 checkpoint_manager.py:2020] [process=4][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0421 11:59:57.044146 138297066120960 array_metadata_store.py:203] [process=4][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260421_114057/linen_xpk_feat_nnx_post_train_fixes_20260421_114057_06_grad_accum/checkpoints/9/items/array_metadatas/process_4 I0421 12:00:33.837943 138301319124736 base_pytree_checkpoint_handler.py:130] [process=4] /jax/orbax/write/gbytes_per_sec: 37.495 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 42.12763547897339 s) (per-host) I0421 12:00:33.838068 138301319124736 async_checkpointer.py:90] [process=4][thread=async_save] 3 Handler Commit operations completed. Time taken: 42.088246s. I0421 12:00:41.829301 138301319124736 async_checkpointer.py:160] [process=4][thread=async_save] Background save thread done. Time taken: 50.079465s. I0421 12:00:41.829530 138297056540416 async_checkpointer.py:288] [process=4][thread=save_finalize] Done with waiting for background save thread=async_save. I0421 12:00:41.829599 138297056540416 async_checkpointer.py:298] [process=4][thread=save_finalize] No errors found in background save thread=async_save. I0421 12:00:41.829665 138297056540416 checkpoint_manager.py:2137] [process=4][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts... I0421 12:00:41.832479 138297056540416 checkpoint_manager.py:2146] [process=4][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts. I0421 12:00:41.832676 138422609975104 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. I0421 12:00:41.832828 138422609975104 checkpoint_manager.py:2009] [process=4][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0421 12:00:41.833850 138422609975104 metric_logger.py:196] completed step: 9, seconds: 0.578, TFLOP/s/device: 93.968, Tokens/s/device: 14163.895, 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: Tue Apr 21 12:00:50 UTC 2026 EXIT_CODE=0