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 | 8.1790 | 8.1790 | 0 |
| TFLOP/s | 92.329 | 92.319 | -0.01 |
| Tok/s | 13916.8 | 13915.3 | -1.513 |
| Avg s/step | 2.696 | 2.665 | -0.031 |
| 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 06:59:40 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 07:00:05.290505: 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 07:00:05.502642 137920969393984 max_utils.py:273] Attempting to initialize the jax distributed system... I0421 07:00:14.544601 137920969393984 distributed.py:149] Starting JAX distributed service on [::]:8482 I0421 07:00:14.546944 137920969393984 distributed.py:172] Connecting to JAX distributed service on mt-13-scan-layers-true-ob274-slice-job-0-0.mt-13-scan-layers-true-ob274:8482 I0421 07:00:16.159575 137920969393984 max_utils.py:284] Jax distributed system initialized! I0421 07:00:21.481230 137920969393984 max_utils.py:800] System Information: Jax Version: 0.9.2 I0421 07:00:21.481337 137920969393984 max_utils.py:801] System Information: Jaxlib Version: 0.9.2 I0421 07:00:21.481376 137920969393984 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 07:00:21.481412 137920969393984 train_utils.py:361] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing. I0421 07:00:22.172183 137920969393984 maxtext_utils.py:1565] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0421 07:00:22.172466 137920969393984 checkpointing.py:677] Setting up checkpoint logger... I0421 07:00:22.172520 137920969393984 checkpointing.py:233] Creating checkpoint manager with ocdbt=True and zarr3=True I0421 07:00:22.172564 137920969393984 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None I0421 07:00:22.172915 137920969393984 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 0x7d6f82d38da0>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0421 07:00:25.081208 137920969393984 checkpointing.py:265] Enabling policy for fixed interval checkpointing. I0421 07:00:25.081548 137920969393984 checkpoint_manager.py:708] [process=7][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=('items',), item_handlers={'items': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7d6fe79cf590>}, handler_registry=None I0421 07:00:25.081795 137920969393984 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7d6fe79cf590>` 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 07:00:25.081845 137920969393984 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7d5c241f1520>` 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 07:00:25.081881 137920969393984 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 0x7d6fe79cf590>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7d6fe79cf590>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7d5c241f1520>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7d5c241f1520>}). I0421 07:00:25.082207 137920969393984 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34 I0421 07:00:25.082276 137920969393984 async_checkpointer.py:192] [process=7][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7d5c046d5800> timeout: 1200 secs and primary_host=0 for async checkpoint writes I0421 07:00:27.027610 137920969393984 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_main_20260421_061359/linen_xpk_main_20260421_061359_13_scan_layers_true/checkpoints I0421 07:00:27.029875 137920969393984 checkpoint_manager.py:929] [process=7][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_13_scan_layers_true/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7d5c241ef020> I0421 07:00:27.029990 137920969393984 checkpointing.py:301] Checkpoint manager created! I0421 07:00:27.955101 137920969393984 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0421 07:00:27.955221 137920969393984 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0421 07:00:28.333887 137920969393984 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0421 07:00:28.333978 137920969393984 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0421 07:00:28.350542 137920969393984 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0421 07:00:28.350609 137920969393984 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0421 07:00:28.374325 137920969393984 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 07:00:28.374397 137920969393984 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0421 07:00:28.390700 137920969393984 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 07:00:28.390769 137920969393984 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0421 07:00:28.407211 137920969393984 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 07:00:28.407277 137920969393984 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0421 07:00:28.432036 137920969393984 attentions.py:1198] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv'). I0421 07:00:28.432106 137920969393984 attentions.py:1198] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0421 07:00:28.453241 137920969393984 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0421 07:00:28.453315 137920969393984 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None). I0421 07:00:28.669255 137920969393984 checkpointing.py:577] checkpoint manager exists so trying to load this run's existing checkpoint I0421 07:00:28.669368 137920969393984 checkpointing.py:665] No existing checkpoints found, not restoring checkpoint. fsdp: 32 I0421 07:00:30.086600 137920969393984 maxtext_utils.py:1668] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: P('norm',) Physical: (None,) I0421 07:00:30.086730 137920969393984 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 07:00:30.086804 137920969393984 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 07:00:30.086868 137920969393984 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 07:00:30.086924 137920969393984 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 07:00:30.086963 137920969393984 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 07:00:30.087015 137920969393984 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 07:00:30.087067 137920969393984 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 07:00:30.087106 137920969393984 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 07:00:30.087142 137920969393984 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 07:00:30.087189 137920969393984 maxtext_utils.py:1668] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: P('embed_vocab', 'vocab') Physical: ('fsdp', None) I0421 07:00:30.087236 137920969393984 maxtext_utils.py:1668] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: P('vocab', 'embed_vocab') Physical: (None, 'fsdp') I0421 07:00:30.576326 137920969393984 train.py:155] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0421 07:00:30.576422 137920969393984 train.py:155] train/xent Physical: float32[32,2048]............................................ ('fsdp', None). I0421 07:00:30.591936 137920969393984 train.py:162] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0421 07:00:30.591997 137920969393984 train.py:162] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None). I0421 07:00:41.432877 137920969393984 max_utils.py:791] Total memory size: 1.5 GB, Output size: 0.4 GB, Temp size: 1.1 GB, Argument size: 0.4 GB, Host temp size: 0.0 GB. I0421 07:00:41.433658 137920969393984 metric_logger.py:301] number parameters: 1.104 billion I0421 07:00:53.374255 137920969393984 checkpointing.py:772] Waiting for step 0 to finish before checkpoint... I0421 07:00:53.527117 137920969393984 checkpointing.py:776] Waited 0.15283799171447754 seconds for step 0 to finish before starting checkpointing. I0421 07:00:53.529707 137920969393984 checkpoint_manager.py:2009] [process=7][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0421 07:00:53.531343 137920969393984 checkpoint_manager.py:1512] [process=7] Saving checkpoint at step 0 I0421 07:00:53.532680 137920969393984 event_tracking.py:70] [process=7] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260421_061359/linen_xpk_main_20260421_061359_13_scan_layers_true/checkpoints/0. I0421 07:00:54.278158 137920969393984 signaling_client.py:364] Using JaxDistributedSignalingClient I0421 07:00:54.279101 137920969393984 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0421 07:00:54.279158 137920969393984 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 07:00:54.556601 137920969393984 base_pytree_checkpoint_handler.py:154] [process=7][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.278518s I0421 07:00:54.556783 137920969393984 base_pytree_checkpoint_handler.py:130] [process=7] /jax/orbax/write/blocking_gbytes_per_sec: 5.428 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.2842106819152832 s) (per-host) I0421 07:00:54.556836 137920969393984 base_pytree_checkpoint_handler.py:768] [process=7][thread=MainThread] Initiated Pytree async_save. Time taken: 0.284273s (batch_requests_ready=0.002396s, total_serialization_initiated=0.281808s, others=0.000070s) I0421 07:00:54.556930 137920969393984 composite_checkpoint_handler.py:715] [process=7][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.288462s (all_items=0.000018s, per_item={'items': '0.00001764'}, temp_paths=0.288444) I0421 07:00:54.557719 137920969393984 event_tracking.py:125] [process=7] [async] Finished blocking save in 1.03 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260421_061359/linen_xpk_main_20260421_061359_13_scan_layers_true/checkpoints/0. I0421 07:00:54.558038 137796106843904 async_checkpointer.py:76] [process=7][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-21 07:20:54.558004 I0421 07:00:54.560193 137920969393984 checkpoint_manager.py:1560] [process=7][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize I0421 07:00:54.560421 137795520313088 async_checkpointer.py:280] [process=7][thread=save_finalize] Waiting for background save thread=async_save. I0421 07:00:54.560540 137920969393984 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260421_061359/linen_xpk_main_20260421_061359_13_scan_layers_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776754853.5296893, 'wait_for_prev_duration_secs': 7.2479248046875e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776754853.5313826, 'checkpointer_blocking_duration_secs': 1.0268206596374512, 'get_old_steps_start_time': 1776754854.558228, 'get_old_steps_duration_secs': 3.0279159545898438e-05, 'checkpoint_manager_blocking_start_time': 1776754853.5276551, 'checkpoint_manager_blocking_duration_secs': 1.032846212387085} I0421 07:00:54.560644 137920969393984 checkpointing.py:408] Started an asynchronous checkpoint save for step 0 I0421 07:00:54.560693 137920969393984 max_utils.py:750] Memstats: After params initialized: I0421 07:00:54.560760 137920969393984 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_26(process=7,(2,6,0,0)) I0421 07:00:54.560795 137920969393984 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_27(process=7,(3,6,0,0)) I0421 07:00:54.560824 137920969393984 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_30(process=7,(2,7,0,0)) I0421 07:00:54.560852 137920969393984 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_31(process=7,(3,7,0,0)) I0421 07:00:54.877628 137920969393984 metric_logger.py:196] completed step: 0, seconds: 11.940, TFLOP/s/device: 1.138, Tokens/s/device: 171.517, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52779.875 I0421 07:00:55.082473 137920969393984 metric_logger.py:196] completed step: 1, seconds: 1.502, TFLOP/s/device: 9.048, Tokens/s/device: 1363.740, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52779.875 I0421 07:00:55.514052 137920969393984 metric_logger.py:196] completed step: 2, seconds: 0.058, TFLOP/s/device: 234.124, Tokens/s/device: 35289.658, total_weights: 65536, loss: 10.262, lm_loss: 10.262, perplexity: 28637.129 I0421 07:00:55.661369 137920969393984 metric_logger.py:196] completed step: 3, seconds: 0.432, TFLOP/s/device: 31.448, Tokens/s/device: 4740.159, total_weights: 65536, loss: 9.734, lm_loss: 9.734, perplexity: 16889.203 I0421 07:00:55.956420 137920969393984 metric_logger.py:196] completed step: 4, seconds: 0.153, TFLOP/s/device: 89.049, Tokens/s/device: 13422.467, total_weights: 65536, loss: 9.277, lm_loss: 9.277, perplexity: 10694.614 I0421 07:00:55.962107 137920969393984 metric_logger.py:196] completed step: 5, seconds: 0.147, TFLOP/s/device: 92.348, Tokens/s/device: 13919.663, total_weights: 65536, loss: 8.892, lm_loss: 8.892, perplexity: 7270.717 I0421 07:00:57.986673 2742 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com I0421 07:01:00.847043 137795528705792 array_metadata_store.py:203] [process=7][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260421_061359/linen_xpk_main_20260421_061359_13_scan_layers_true/checkpoints/0/items/array_metadatas/process_7 I0421 07:01:17.485280 137920969393984 metric_logger.py:196] completed step: 6, seconds: 0.295, TFLOP/s/device: 46.012, Tokens/s/device: 6935.484, total_weights: 65536, loss: 8.592, lm_loss: 8.592, perplexity: 5390.759 I0421 07:01:17.632580 137920969393984 metric_logger.py:196] completed step: 7, seconds: 21.377, TFLOP/s/device: 0.636, Tokens/s/device: 95.805, total_weights: 65536, loss: 8.384, lm_loss: 8.384, perplexity: 4376.905 I0421 07:01:17.779659 137920969393984 metric_logger.py:196] completed step: 8, seconds: 0.152, TFLOP/s/device: 89.383, Tokens/s/device: 13472.709, total_weights: 65536, loss: 8.255, lm_loss: 8.255, perplexity: 3846.406 I0421 07:01:17.926384 137920969393984 checkpointing.py:772] Waiting for step 9 to finish before checkpoint... I0421 07:01:17.927086 137920969393984 checkpointing.py:776] Waited 0.00072479248046875 seconds for step 9 to finish before starting checkpointing. I0421 07:01:17.930440 137920969393984 checkpoint_manager.py:2020] [process=7][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0421 07:01:32.477703 137796106843904 base_pytree_checkpoint_handler.py:130] [process=7] /jax/orbax/write/gbytes_per_sec: 41.345 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 38.205090045928955 s) (per-host) I0421 07:01:32.477830 137796106843904 async_checkpointer.py:90] [process=7][thread=async_save] 3 Handler Commit operations completed. Time taken: 37.919666s. I0421 07:01:40.757218 137796106843904 async_checkpointer.py:160] [process=7][thread=async_save] Background save thread done. Time taken: 46.199038s. I0421 07:01:40.757516 137795520313088 async_checkpointer.py:288] [process=7][thread=save_finalize] Done with waiting for background save thread=async_save. I0421 07:01:40.757645 137795520313088 async_checkpointer.py:298] [process=7][thread=save_finalize] No errors found in background save thread=async_save. I0421 07:01:40.757697 137795520313088 checkpoint_manager.py:2137] [process=7][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts... I0421 07:01:40.760338 137795520313088 checkpoint_manager.py:2146] [process=7][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts. I0421 07:01:40.760499 137920969393984 checkpoint_manager.py:2032] [process=7][thread=MainThread][step=0][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=0. W0421 07:01:40.760634 137920969393984 checkpoint_manager.py:1452] Waiting for previous save to complete took 22.830194 seconds. If this number is high, consider checkpointing less frequently. I0421 07:01:40.762480 137920969393984 checkpoint_manager.py:1512] [process=7] Saving checkpoint at step 9 I0421 07:01:40.764443 137920969393984 event_tracking.py:70] [process=7] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260421_061359/linen_xpk_main_20260421_061359_13_scan_layers_true/checkpoints/9. I0421 07:01:41.052575 137920969393984 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0421 07:01:41.052666 137920969393984 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 07:01:41.084080 137920969393984 base_pytree_checkpoint_handler.py:154] [process=7][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.032569s I0421 07:01:41.084254 137920969393984 base_pytree_checkpoint_handler.py:130] [process=7] /jax/orbax/write/blocking_gbytes_per_sec: 42.630 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.036185503005981445 s) (per-host) I0421 07:01:41.084306 137920969393984 base_pytree_checkpoint_handler.py:768] [process=7][thread=MainThread] Initiated Pytree async_save. Time taken: 0.036248s (batch_requests_ready=0.001943s, total_serialization_initiated=0.034235s, others=0.000071s) I0421 07:01:41.084403 137920969393984 composite_checkpoint_handler.py:715] [process=7][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.040609s (all_items=0.000015s, per_item={'items': '0.00001526'}, temp_paths=0.040594) I0421 07:01:41.085095 137920969393984 event_tracking.py:125] [process=7] [async] Finished blocking save in 0.32 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260421_061359/linen_xpk_main_20260421_061359_13_scan_layers_true/checkpoints/9. I0421 07:01:41.085425 137795520313088 async_checkpointer.py:76] [process=7][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-21 07:21:41.085388 I0421 07:01:41.089138 137920969393984 checkpoint_manager.py:1560] [process=7][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize I0421 07:01:41.089352 137767770117888 async_checkpointer.py:280] [process=7][thread=save_finalize] Waiting for background save thread=async_save. I0421 07:01:41.089477 137920969393984 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260421_061359/linen_xpk_main_20260421_061359_13_scan_layers_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776754877.9304085, 'wait_for_prev_duration_secs': 22.830194234848022, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776754900.7625191, 'checkpointer_blocking_duration_secs': 0.32305049896240234, 'get_old_steps_start_time': 1776754901.0855906, 'get_old_steps_duration_secs': 2.9325485229492188e-05, 'checkpoint_manager_blocking_start_time': 1776754877.9273207, 'checkpoint_manager_blocking_duration_secs': 23.162122011184692} I0421 07:01:41.089590 137920969393984 checkpointing.py:408] Started an asynchronous checkpoint save for step 9 I0421 07:01:41.089634 137920969393984 checkpoint_manager.py:2020] [process=7][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0421 07:01:46.986631 137795528705792 array_metadata_store.py:203] [process=7][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260421_061359/linen_xpk_main_20260421_061359_13_scan_layers_true/checkpoints/9/items/array_metadatas/process_7 I0421 07:02:23.630556 137795520313088 base_pytree_checkpoint_handler.py:130] [process=7] /jax/orbax/write/gbytes_per_sec: 37.095 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 42.582448959350586 s) (per-host) I0421 07:02:23.630691 137795520313088 async_checkpointer.py:90] [process=7][thread=async_save] 3 Handler Commit operations completed. Time taken: 42.545149s. I0421 07:02:33.480764 137795520313088 async_checkpointer.py:160] [process=7][thread=async_save] Background save thread done. Time taken: 52.395188s. I0421 07:02:33.481024 137767770117888 async_checkpointer.py:288] [process=7][thread=save_finalize] Done with waiting for background save thread=async_save. I0421 07:02:33.481096 137767770117888 async_checkpointer.py:298] [process=7][thread=save_finalize] No errors found in background save thread=async_save. I0421 07:02:33.481151 137767770117888 checkpoint_manager.py:2137] [process=7][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts... I0421 07:02:33.485964 137767770117888 checkpoint_manager.py:2146] [process=7][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts. I0421 07:02:33.486144 137920969393984 checkpoint_manager.py:2032] [process=7][thread=MainThread][step=9][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=9. I0421 07:02:33.486300 137920969393984 checkpoint_manager.py:2009] [process=7][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0421 07:02:33.487273 137920969393984 metric_logger.py:196] completed step: 9, seconds: 0.147, TFLOP/s/device: 92.329, Tokens/s/device: 13916.825, total_weights: 65536, loss: 8.179, lm_loss: 8.179, perplexity: 3564.635 Per train step: Total TFLOPs: 13.59 split as 93.93% learnable weight flops and 6.07% attention flops XPK End: Tue Apr 21 07:02:45 UTC 2026 EXIT_CODE=0
XPK Start: Tue Apr 21 12:26:23 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 12:26:47.684293: 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 12:26:47.865798 134726747903808 max_utils.py:273] Attempting to initialize the jax distributed system... I0421 12:26:56.908061 134726747903808 distributed.py:149] Starting JAX distributed service on [::]:8482 I0421 12:26:56.910314 134726747903808 distributed.py:172] Connecting to JAX distributed service on mt-13-scan-layers-true-8t7hx-slice-job-0-0.mt-13-scan-layers-true-8t7hx:8482 I0421 12:26:57.885482 134726747903808 max_utils.py:284] Jax distributed system initialized! I0421 12:27:04.047685 134726747903808 max_utils.py:800] System Information: Jax Version: 0.9.2 I0421 12:27:04.047805 134726747903808 max_utils.py:801] System Information: Jaxlib Version: 0.9.2 I0421 12:27:04.047849 134726747903808 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 12:27:04.047886 134726747903808 train_utils.py:378] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing. I0421 12:27:04.745834 134726747903808 maxtext_utils.py:1718] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0421 12:27:04.746433 134726747903808 maxtext_utils.py:1718] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0421 12:27:04.746614 134726747903808 checkpointing.py:688] Setting up checkpoint logger... I0421 12:27:04.746667 134726747903808 checkpointing.py:234] Creating checkpoint manager with ocdbt=True and zarr3=True I0421 12:27:04.746713 134726747903808 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None I0421 12:27:04.747055 134726747903808 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 0x7a87cc350560>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0421 12:27:07.593680 134726747903808 checkpointing.py:266] Enabling policy for fixed interval checkpointing. I0421 12:27:07.593944 134726747903808 checkpoint_manager.py:708] [process=7][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=('items',), item_handlers={'items': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7a86c1729eb0>}, handler_registry=None I0421 12:27:07.594186 134726747903808 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7a86c1729eb0>` 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 12:27:07.594236 134726747903808 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7a74481dce60>` 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 12:27:07.594274 134726747903808 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 0x7a86c1729eb0>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7a86c1729eb0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7a74481dce60>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7a74481dce60>}). I0421 12:27:07.594587 134726747903808 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34 I0421 12:27:07.594657 134726747903808 async_checkpointer.py:192] [process=7][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7a743067c540> timeout: 1200 secs and primary_host=0 for async checkpoint writes I0421 12:27:08.874370 134726747903808 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_13_scan_layers_true/checkpoints I0421 12:27:09.283950 134726747903808 checkpoint_manager.py:929] [process=7][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_13_scan_layers_true/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7a74481da090> I0421 12:27:09.284118 134726747903808 checkpointing.py:302] Checkpoint manager created! I0421 12:27:10.226238 134726747903808 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0421 12:27:10.226345 134726747903808 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0421 12:27:10.609709 134726747903808 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0421 12:27:10.609818 134726747903808 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0421 12:27:10.626488 134726747903808 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0421 12:27:10.626549 134726747903808 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0421 12:27:10.650551 134726747903808 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 12:27:10.650618 134726747903808 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0421 12:27:10.667205 134726747903808 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 12:27:10.667265 134726747903808 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0421 12:27:10.683761 134726747903808 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 12:27:10.683822 134726747903808 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0421 12:27:10.708723 134726747903808 attentions.py:1197] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv'). I0421 12:27:10.708806 134726747903808 attentions.py:1197] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0421 12:27:10.730325 134726747903808 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0421 12:27:10.730390 134726747903808 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None). I0421 12:27:10.948703 134726747903808 checkpointing.py:578] checkpoint manager exists so trying to load this run's existing checkpoint I0421 12:27:10.948855 134726747903808 checkpointing.py:676] No existing checkpoints found, not restoring checkpoint. fsdp: 32 I0421 12:27:12.387862 134726747903808 maxtext_utils.py:1827] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: P('norm',) Physical: (None,) I0421 12:27:12.387993 134726747903808 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 12:27:12.388068 134726747903808 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 12:27:12.388144 134726747903808 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 12:27:12.388202 134726747903808 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 12:27:12.388241 134726747903808 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 12:27:12.388295 134726747903808 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 12:27:12.388355 134726747903808 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 12:27:12.388399 134726747903808 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 12:27:12.388437 134726747903808 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 12:27:12.388485 134726747903808 maxtext_utils.py:1827] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: P('embed_vocab', 'vocab') Physical: ('fsdp', None) I0421 12:27:12.388533 134726747903808 maxtext_utils.py:1827] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: P('vocab', 'embed_vocab') Physical: (None, 'fsdp') I0421 12:27:12.897998 134726747903808 train.py:157] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0421 12:27:12.898103 134726747903808 train.py:157] train/xent Physical: float32[32,2048]............................................ ('fsdp', None). I0421 12:27:12.913723 134726747903808 train.py:164] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0421 12:27:12.913797 134726747903808 train.py:164] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None). I0421 12:27:23.808433 134726747903808 max_utils.py:791] Total memory size: 1.5 GB, Output size: 0.4 GB, Temp size: 1.1 GB, Argument size: 0.4 GB, Host temp size: 0.0 GB. I0421 12:27:23.809283 134726747903808 metric_logger.py:301] number parameters: 1.104 billion I0421 12:27:35.193598 134726747903808 checkpointing.py:794] Waiting for step 0 to finish before checkpoint... I0421 12:27:35.346117 134726747903808 checkpointing.py:798] Waited 0.15250015258789062 seconds for step 0 to finish before starting checkpointing. I0421 12:27:35.350104 134726747903808 checkpoint_manager.py:2009] [process=7][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0421 12:27:35.351617 134726747903808 checkpoint_manager.py:1512] [process=7] Saving checkpoint at step 0 I0421 12:27:35.353082 134726747903808 event_tracking.py:70] [process=7] [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_13_scan_layers_true/checkpoints/0. I0421 12:27:35.693574 134726747903808 signaling_client.py:364] Using JaxDistributedSignalingClient I0421 12:27:35.694499 134726747903808 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0421 12:27:35.694556 134726747903808 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 12:27:35.979549 134726747903808 base_pytree_checkpoint_handler.py:154] [process=7][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.286049s I0421 12:27:35.979729 134726747903808 base_pytree_checkpoint_handler.py:130] [process=7] /jax/orbax/write/blocking_gbytes_per_sec: 5.290 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.29158568382263184 s) (per-host) I0421 12:27:35.979801 134726747903808 base_pytree_checkpoint_handler.py:768] [process=7][thread=MainThread] Initiated Pytree async_save. Time taken: 0.291668s (batch_requests_ready=0.002214s, total_serialization_initiated=0.289365s, others=0.000090s) I0421 12:27:35.979904 134726747903808 composite_checkpoint_handler.py:715] [process=7][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.295683s (all_items=0.000017s, per_item={'items': '0.00001669'}, temp_paths=0.295666) I0421 12:27:35.980747 134726747903808 event_tracking.py:125] [process=7] [async] Finished blocking save in 0.63 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_13_scan_layers_true/checkpoints/0. I0421 12:27:35.981043 134602127570688 async_checkpointer.py:76] [process=7][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-21 12:47:35.981017 I0421 12:27:35.982953 134726747903808 checkpoint_manager.py:1560] [process=7][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize I0421 12:27:35.983166 134599941555968 async_checkpointer.py:280] [process=7][thread=save_finalize] Waiting for background save thread=async_save. I0421 12:27:35.983273 134726747903808 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_13_scan_layers_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776774455.3500843, 'wait_for_prev_duration_secs': 6.508827209472656e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776774455.3516562, 'checkpointer_blocking_duration_secs': 0.62949538230896, 'get_old_steps_start_time': 1776774455.9811766, 'get_old_steps_duration_secs': 3.123283386230469e-05, 'checkpoint_manager_blocking_start_time': 1776774455.34664, 'checkpoint_manager_blocking_duration_secs': 0.6365969181060791} I0421 12:27:35.983377 134726747903808 checkpointing.py:409] Started an asynchronous checkpoint save for step 0 I0421 12:27:35.983428 134726747903808 max_utils.py:750] Memstats: After params initialized: I0421 12:27:35.983475 134726747903808 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_26(process=7,(2,6,0,0)) I0421 12:27:35.983510 134726747903808 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_27(process=7,(3,6,0,0)) I0421 12:27:35.983535 134726747903808 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_30(process=7,(2,7,0,0)) I0421 12:27:35.983561 134726747903808 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_31(process=7,(3,7,0,0)) I0421 12:27:36.295651 134726747903808 metric_logger.py:196] completed step: 0, seconds: 11.384, TFLOP/s/device: 1.194, Tokens/s/device: 179.898, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52779.875 I0421 12:27:36.475488 134726747903808 metric_logger.py:196] completed step: 1, seconds: 1.101, TFLOP/s/device: 12.345, Tokens/s/device: 1860.788, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52779.875 I0421 12:27:36.894444 134726747903808 metric_logger.py:196] completed step: 2, seconds: 0.033, TFLOP/s/device: 417.334, Tokens/s/device: 62905.059, total_weights: 65536, loss: 10.262, lm_loss: 10.262, perplexity: 28637.129 I0421 12:27:37.042282 134726747903808 metric_logger.py:196] completed step: 3, seconds: 0.419, TFLOP/s/device: 32.440, Tokens/s/device: 4889.660, total_weights: 65536, loss: 9.734, lm_loss: 9.734, perplexity: 16889.203 I0421 12:27:37.337181 134726747903808 metric_logger.py:196] completed step: 4, seconds: 0.153, TFLOP/s/device: 88.548, Tokens/s/device: 13346.976, total_weights: 65536, loss: 9.277, lm_loss: 9.277, perplexity: 10694.614 I0421 12:27:37.343060 134726747903808 metric_logger.py:196] completed step: 5, seconds: 0.148, TFLOP/s/device: 91.955, Tokens/s/device: 13860.408, total_weights: 65536, loss: 8.892, lm_loss: 8.892, perplexity: 7270.717 I0421 12:27:39.585673 2760 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com I0421 12:27:41.632438 134600890246912 array_metadata_store.py:203] [process=7][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_13_scan_layers_true/checkpoints/0/items/array_metadatas/process_7 I0421 12:27:59.028232 134726747903808 metric_logger.py:196] completed step: 6, seconds: 0.295, TFLOP/s/device: 46.011, Tokens/s/device: 6935.297, total_weights: 65536, loss: 8.592, lm_loss: 8.592, perplexity: 5390.759 I0421 12:27:59.175690 134726747903808 metric_logger.py:196] completed step: 7, seconds: 21.538, TFLOP/s/device: 0.631, Tokens/s/device: 95.086, total_weights: 65536, loss: 8.384, lm_loss: 8.384, perplexity: 4376.905 I0421 12:27:59.322788 134726747903808 metric_logger.py:196] completed step: 8, seconds: 0.153, TFLOP/s/device: 89.007, Tokens/s/device: 13416.048, total_weights: 65536, loss: 8.255, lm_loss: 8.255, perplexity: 3846.406 I0421 12:27:59.469439 134726747903808 checkpointing.py:794] Waiting for step 9 to finish before checkpoint... I0421 12:27:59.469997 134726747903808 checkpointing.py:798] Waited 0.0005803108215332031 seconds for step 9 to finish before starting checkpointing. I0421 12:27:59.472163 134726747903808 checkpoint_manager.py:2020] [process=7][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0421 12:28:13.449173 134602127570688 base_pytree_checkpoint_handler.py:130] [process=7] /jax/orbax/write/gbytes_per_sec: 41.831 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 37.76098895072937 s) (per-host) I0421 12:28:13.449303 134602127570688 async_checkpointer.py:90] [process=7][thread=async_save] 3 Handler Commit operations completed. Time taken: 37.468183s. I0421 12:28:21.651653 134602127570688 async_checkpointer.py:160] [process=7][thread=async_save] Background save thread done. Time taken: 45.670516s. I0421 12:28:21.651972 134599941555968 async_checkpointer.py:288] [process=7][thread=save_finalize] Done with waiting for background save thread=async_save. I0421 12:28:21.652098 134599941555968 async_checkpointer.py:298] [process=7][thread=save_finalize] No errors found in background save thread=async_save. I0421 12:28:21.652150 134599941555968 checkpoint_manager.py:2137] [process=7][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts... I0421 12:28:21.654146 134599941555968 checkpoint_manager.py:2146] [process=7][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts. I0421 12:28:21.654309 134726747903808 checkpoint_manager.py:2032] [process=7][thread=MainThread][step=0][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=0. W0421 12:28:21.654442 134726747903808 checkpoint_manager.py:1452] Waiting for previous save to complete took 22.182280 seconds. If this number is high, consider checkpointing less frequently. I0421 12:28:21.656165 134726747903808 checkpoint_manager.py:1512] [process=7] Saving checkpoint at step 9 I0421 12:28:21.658071 134726747903808 event_tracking.py:70] [process=7] [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_13_scan_layers_true/checkpoints/9. I0421 12:28:22.383000 134726747903808 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0421 12:28:22.383096 134726747903808 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 12:28:22.424588 134726747903808 base_pytree_checkpoint_handler.py:154] [process=7][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.042743s I0421 12:28:22.424781 134726747903808 base_pytree_checkpoint_handler.py:130] [process=7] /jax/orbax/write/blocking_gbytes_per_sec: 33.322 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.04629254341125488 s) (per-host) I0421 12:28:22.424839 134726747903808 base_pytree_checkpoint_handler.py:768] [process=7][thread=MainThread] Initiated Pytree async_save. Time taken: 0.046360s (batch_requests_ready=0.001836s, total_serialization_initiated=0.044449s, others=0.000074s) I0421 12:28:22.424929 134726747903808 composite_checkpoint_handler.py:715] [process=7][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.050529s (all_items=0.000016s, per_item={'items': '0.00001574'}, temp_paths=0.050513) I0421 12:28:22.425617 134726747903808 event_tracking.py:125] [process=7] [async] Finished blocking save in 0.77 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_13_scan_layers_true/checkpoints/9. I0421 12:28:22.425962 134599941555968 async_checkpointer.py:76] [process=7][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-21 12:48:22.425924 I0421 12:28:22.427957 134726747903808 checkpoint_manager.py:1560] [process=7][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize I0421 12:28:22.428251 134578652034816 async_checkpointer.py:280] [process=7][thread=save_finalize] Waiting for background save thread=async_save. I0421 12:28:22.428413 134726747903808 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_13_scan_layers_true/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776774479.4721317, 'wait_for_prev_duration_secs': 22.18228030204773, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776774501.656204, 'checkpointer_blocking_duration_secs': 0.7699072360992432, 'get_old_steps_start_time': 1776774502.4261355, 'get_old_steps_duration_secs': 2.8371810913085938e-05, 'checkpoint_manager_blocking_start_time': 1776774479.470231, 'checkpoint_manager_blocking_duration_secs': 22.958147048950195} I0421 12:28:22.428519 134726747903808 checkpointing.py:409] Started an asynchronous checkpoint save for step 9 I0421 12:28:22.428565 134726747903808 checkpoint_manager.py:2020] [process=7][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0421 12:28:27.537143 134600890246912 array_metadata_store.py:203] [process=7][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_13_scan_layers_true/checkpoints/9/items/array_metadatas/process_7 I0421 12:29:04.247210 134599941555968 base_pytree_checkpoint_handler.py:130] [process=7] /jax/orbax/write/gbytes_per_sec: 37.727 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 41.86867570877075 s) (per-host) I0421 12:29:04.247338 134599941555968 async_checkpointer.py:90] [process=7][thread=async_save] 3 Handler Commit operations completed. Time taken: 41.821254s. I0421 12:29:13.150746 134599941555968 async_checkpointer.py:160] [process=7][thread=async_save] Background save thread done. Time taken: 50.724633s. I0421 12:29:13.151037 134578652034816 async_checkpointer.py:288] [process=7][thread=save_finalize] Done with waiting for background save thread=async_save. I0421 12:29:13.151167 134578652034816 async_checkpointer.py:298] [process=7][thread=save_finalize] No errors found in background save thread=async_save. I0421 12:29:13.151216 134578652034816 checkpoint_manager.py:2137] [process=7][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts... I0421 12:29:13.153043 134578652034816 checkpoint_manager.py:2146] [process=7][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts. I0421 12:29:13.153250 134726747903808 checkpoint_manager.py:2032] [process=7][thread=MainThread][step=9][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=9. I0421 12:29:13.153407 134726747903808 checkpoint_manager.py:2009] [process=7][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0421 12:29:13.154329 134726747903808 metric_logger.py:196] completed step: 9, seconds: 0.147, TFLOP/s/device: 92.319, Tokens/s/device: 13915.312, total_weights: 65536, loss: 8.179, lm_loss: 8.179, perplexity: 3564.635 Per train step: Total TFLOPs: 13.59 split as 93.93% learnable weight flops and 6.07% attention flops XPK End: Tue Apr 21 12:29:23 UTC 2026 EXIT_CODE=0