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 | 92.135 | 92.446 | +0.311 |
| Tok/s | 13887.7 | 13934.5 | +46.867 |
| Avg s/step | 2.898 | 2.695 | -0.203 |
| 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 06:19:10 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 06:19:35.669864: 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 06:19:35.881146 138919799117632 max_utils.py:273] Attempting to initialize the jax distributed system... I0421 06:19:44.920814 138919799117632 distributed.py:149] Starting JAX distributed service on [::]:8482 I0421 06:19:44.923312 138919799117632 distributed.py:172] Connecting to JAX distributed service on mt-02-synthetic-59j93-slice-job-0-0.mt-02-synthetic-59j93:8482 I0421 06:19:46.258219 138919799117632 max_utils.py:284] Jax distributed system initialized! I0421 06:19:52.224090 138919799117632 max_utils.py:800] System Information: Jax Version: 0.9.2 I0421 06:19:52.224195 138919799117632 max_utils.py:801] System Information: Jaxlib Version: 0.9.2 I0421 06:19:52.224235 138919799117632 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 06:19:52.224270 138919799117632 train_utils.py:361] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing. I0421 06:19:52.917172 138919799117632 maxtext_utils.py:1565] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0421 06:19:52.917444 138919799117632 checkpointing.py:677] Setting up checkpoint logger... I0421 06:19:52.917500 138919799117632 checkpointing.py:233] Creating checkpoint manager with ocdbt=True and zarr3=True I0421 06:19:52.917542 138919799117632 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None I0421 06:19:52.917895 138919799117632 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 0x7e582dd1d010>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0421 06:19:56.149091 138919799117632 checkpointing.py:265] Enabling policy for fixed interval checkpointing. I0421 06:19:56.149428 138919799117632 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 0x7e44281f30e0>}, handler_registry=None I0421 06:19:56.149667 138919799117632 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7e44281f30e0>` 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 06:19:56.149715 138919799117632 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7e442828d280>` 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 06:19:56.149770 138919799117632 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 0x7e44281f30e0>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7e44281f30e0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7e442828d280>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7e442828d280>}). I0421 06:19:56.150089 138919799117632 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34 I0421 06:19:56.150156 138919799117632 async_checkpointer.py:192] [process=7][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7e44280bd580> timeout: 1200 secs and primary_host=0 for async checkpoint writes I0421 06:19:57.913616 138919799117632 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_main_20260421_061359/linen_xpk_main_20260421_061359_02_synthetic/checkpoints I0421 06:19:58.756298 138919799117632 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_02_synthetic/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7e4428288350> I0421 06:19:58.756471 138919799117632 checkpointing.py:301] Checkpoint manager created! I0421 06:19:59.687069 138919799117632 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0421 06:19:59.687189 138919799117632 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0421 06:20:00.067538 138919799117632 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0421 06:20:00.067631 138919799117632 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0421 06:20:00.084623 138919799117632 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0421 06:20:00.084682 138919799117632 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0421 06:20:00.108984 138919799117632 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 06:20:00.109052 138919799117632 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0421 06:20:00.126043 138919799117632 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 06:20:00.126108 138919799117632 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0421 06:20:00.143064 138919799117632 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 06:20:00.143123 138919799117632 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0421 06:20:00.168403 138919799117632 attentions.py:1198] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv'). I0421 06:20:00.168474 138919799117632 attentions.py:1198] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0421 06:20:00.189867 138919799117632 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0421 06:20:00.189936 138919799117632 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None). I0421 06:20:00.401957 138919799117632 checkpointing.py:577] checkpoint manager exists so trying to load this run's existing checkpoint I0421 06:20:00.402064 138919799117632 checkpointing.py:665] No existing checkpoints found, not restoring checkpoint. fsdp: 32 I0421 06:20:01.839627 138919799117632 maxtext_utils.py:1668] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: P('norm',) Physical: (None,) I0421 06:20:01.839784 138919799117632 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 06:20:01.839843 138919799117632 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 06:20:01.839905 138919799117632 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 06:20:01.839958 138919799117632 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 06:20:01.840001 138919799117632 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 06:20:01.840054 138919799117632 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 06:20:01.840105 138919799117632 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 06:20:01.840143 138919799117632 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 06:20:01.840178 138919799117632 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 06:20:01.840224 138919799117632 maxtext_utils.py:1668] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: P('embed_vocab', 'vocab') Physical: ('fsdp', None) I0421 06:20:01.840270 138919799117632 maxtext_utils.py:1668] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: P('vocab', 'embed_vocab') Physical: (None, 'fsdp') I0421 06:20:02.330982 138919799117632 train.py:155] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0421 06:20:02.331077 138919799117632 train.py:155] train/xent Physical: float32[32,2048]............................................ ('fsdp', None). I0421 06:20:02.346611 138919799117632 train.py:162] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0421 06:20:02.346673 138919799117632 train.py:162] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None). I0421 06:20:13.209961 138919799117632 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 06:20:13.210800 138919799117632 metric_logger.py:301] number parameters: 1.104 billion I0421 06:20:24.538718 138919799117632 checkpointing.py:772] Waiting for step 0 to finish before checkpoint... I0421 06:20:24.978171 138919799117632 checkpointing.py:776] Waited 0.43943262100219727 seconds for step 0 to finish before starting checkpointing. I0421 06:20:24.980644 138919799117632 checkpoint_manager.py:2009] [process=7][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0421 06:20:24.982274 138919799117632 checkpoint_manager.py:1512] [process=7] Saving checkpoint at step 0 I0421 06:20:24.983621 138919799117632 event_tracking.py:70] [process=7] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260421_061359/linen_xpk_main_20260421_061359_02_synthetic/checkpoints/0. I0421 06:20:25.718491 138919799117632 signaling_client.py:364] Using JaxDistributedSignalingClient I0421 06:20:25.719465 138919799117632 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0421 06:20:25.719523 138919799117632 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 06:20:25.999531 138919799117632 base_pytree_checkpoint_handler.py:154] [process=7][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.281114s I0421 06:20:25.999710 138919799117632 base_pytree_checkpoint_handler.py:130] [process=7] /jax/orbax/write/blocking_gbytes_per_sec: 5.382 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.28661131858825684 s) (per-host) I0421 06:20:25.999782 138919799117632 base_pytree_checkpoint_handler.py:768] [process=7][thread=MainThread] Initiated Pytree async_save. Time taken: 0.286694s (batch_requests_ready=0.002321s, total_serialization_initiated=0.284283s, others=0.000091s) I0421 06:20:25.999879 138919799117632 composite_checkpoint_handler.py:715] [process=7][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.291084s (all_items=0.000017s, per_item={'items': '0.00001717'}, temp_paths=0.291066) I0421 06:20:26.000770 138919799117632 event_tracking.py:125] [process=7] [async] Finished blocking save in 1.02 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260421_061359/linen_xpk_main_20260421_061359_02_synthetic/checkpoints/0. I0421 06:20:26.001066 138792301262592 async_checkpointer.py:76] [process=7][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-21 06:40:26.001030 I0421 06:20:26.004704 138919799117632 checkpoint_manager.py:1560] [process=7][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize I0421 06:20:26.005028 138790694479616 async_checkpointer.py:280] [process=7][thread=save_finalize] Waiting for background save thread=async_save. I0421 06:20:26.005184 138919799117632 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260421_061359/linen_xpk_main_20260421_061359_02_synthetic/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776752424.9806252, 'wait_for_prev_duration_secs': 6.175041198730469e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776752424.9823136, 'checkpointer_blocking_duration_secs': 1.0189085006713867, 'get_old_steps_start_time': 1776752426.0012474, 'get_old_steps_duration_secs': 3.0517578125e-05, 'checkpoint_manager_blocking_start_time': 1776752424.9787266, 'checkpoint_manager_blocking_duration_secs': 1.026418685913086} I0421 06:20:26.005290 138919799117632 checkpointing.py:408] Started an asynchronous checkpoint save for step 0 I0421 06:20:26.005336 138919799117632 max_utils.py:750] Memstats: After params initialized: I0421 06:20:26.005382 138919799117632 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_26(process=7,(2,6,0,0)) I0421 06:20:26.005413 138919799117632 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_27(process=7,(3,6,0,0)) I0421 06:20:26.005440 138919799117632 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_30(process=7,(2,7,0,0)) I0421 06:20:26.005465 138919799117632 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_31(process=7,(3,7,0,0)) I0421 06:20:26.318793 138919799117632 metric_logger.py:196] completed step: 0, seconds: 11.328, TFLOP/s/device: 1.199, Tokens/s/device: 180.794, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52779.875 I0421 06:20:26.495119 138919799117632 metric_logger.py:196] completed step: 1, seconds: 1.778, TFLOP/s/device: 7.640, Tokens/s/device: 1151.538, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52779.875 I0421 06:20:26.940293 138919799117632 metric_logger.py:196] completed step: 2, seconds: 0.028, TFLOP/s/device: 493.665, Tokens/s/device: 74410.493, total_weights: 65536, loss: 10.262, lm_loss: 10.262, perplexity: 28637.129 I0421 06:20:27.087517 138919799117632 metric_logger.py:196] completed step: 3, seconds: 0.447, TFLOP/s/device: 30.400, Tokens/s/device: 4582.209, total_weights: 65536, loss: 9.734, lm_loss: 9.734, perplexity: 16889.203 I0421 06:20:27.382431 138919799117632 metric_logger.py:196] completed step: 4, seconds: 0.153, TFLOP/s/device: 88.769, Tokens/s/device: 13380.286, total_weights: 65536, loss: 9.277, lm_loss: 9.277, perplexity: 10694.614 I0421 06:20:27.388252 138919799117632 metric_logger.py:196] completed step: 5, seconds: 0.147, TFLOP/s/device: 92.320, Tokens/s/device: 13915.501, total_weights: 65536, loss: 8.892, lm_loss: 8.892, perplexity: 7270.717 I0421 06:20:28.909552 2809 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com I0421 06:20:30.673407 138790702872320 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_02_synthetic/checkpoints/0/items/array_metadatas/process_7 I0421 06:20:50.467475 138919799117632 metric_logger.py:196] completed step: 6, seconds: 0.295, TFLOP/s/device: 46.004, Tokens/s/device: 6934.193, total_weights: 65536, loss: 8.592, lm_loss: 8.592, perplexity: 5390.759 I0421 06:20:50.614937 138919799117632 metric_logger.py:196] completed step: 7, seconds: 22.932, TFLOP/s/device: 0.592, Tokens/s/device: 89.307, total_weights: 65536, loss: 8.384, lm_loss: 8.384, perplexity: 4376.905 I0421 06:20:50.762076 138919799117632 metric_logger.py:196] completed step: 8, seconds: 0.152, TFLOP/s/device: 89.195, Tokens/s/device: 13444.496, total_weights: 65536, loss: 8.255, lm_loss: 8.255, perplexity: 3846.406 I0421 06:20:50.908682 138919799117632 checkpointing.py:772] Waiting for step 9 to finish before checkpoint... I0421 06:20:50.909343 138919799117632 checkpointing.py:776] Waited 0.0006792545318603516 seconds for step 9 to finish before starting checkpointing. I0421 06:20:50.912682 138919799117632 checkpoint_manager.py:2020] [process=7][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0421 06:21:02.405345 138792301262592 base_pytree_checkpoint_handler.py:130] [process=7] /jax/orbax/write/gbytes_per_sec: 43.050 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 36.69219899177551 s) (per-host) I0421 06:21:02.405464 138792301262592 async_checkpointer.py:90] [process=7][thread=async_save] 3 Handler Commit operations completed. Time taken: 36.404280s. I0421 06:21:10.940417 138792301262592 async_checkpointer.py:160] [process=7][thread=async_save] Background save thread done. Time taken: 44.939216s. I0421 06:21:10.940711 138790694479616 async_checkpointer.py:288] [process=7][thread=save_finalize] Done with waiting for background save thread=async_save. I0421 06:21:10.940895 138790694479616 async_checkpointer.py:298] [process=7][thread=save_finalize] No errors found in background save thread=async_save. I0421 06:21:10.940948 138790694479616 checkpoint_manager.py:2137] [process=7][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts... I0421 06:21:10.942213 138790694479616 checkpoint_manager.py:2146] [process=7][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts. I0421 06:21:10.942361 138919799117632 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 06:21:10.942440 138919799117632 checkpoint_manager.py:1452] Waiting for previous save to complete took 20.029761 seconds. If this number is high, consider checkpointing less frequently. I0421 06:21:10.944317 138919799117632 checkpoint_manager.py:1512] [process=7] Saving checkpoint at step 9 I0421 06:21:10.946358 138919799117632 event_tracking.py:70] [process=7] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260421_061359/linen_xpk_main_20260421_061359_02_synthetic/checkpoints/9. I0421 06:21:11.656584 138919799117632 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0421 06:21:11.656678 138919799117632 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 06:21:11.691609 138919799117632 base_pytree_checkpoint_handler.py:154] [process=7][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.036142s I0421 06:21:11.691803 138919799117632 base_pytree_checkpoint_handler.py:130] [process=7] /jax/orbax/write/blocking_gbytes_per_sec: 38.203 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.04037833213806152 s) (per-host) I0421 06:21:11.691860 138919799117632 base_pytree_checkpoint_handler.py:768] [process=7][thread=MainThread] Initiated Pytree async_save. Time taken: 0.040451s (batch_requests_ready=0.002584s, total_serialization_initiated=0.037785s, others=0.000082s) I0421 06:21:11.691954 138919799117632 composite_checkpoint_handler.py:715] [process=7][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.044595s (all_items=0.000014s, per_item={'items': '0.00001431'}, temp_paths=0.044581) I0421 06:21:11.692593 138919799117632 event_tracking.py:125] [process=7] [async] Finished blocking save in 0.75 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260421_061359/linen_xpk_main_20260421_061359_02_synthetic/checkpoints/9. I0421 06:21:11.692879 138790702872320 async_checkpointer.py:76] [process=7][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-21 06:41:11.692857 I0421 06:21:11.694651 138919799117632 checkpoint_manager.py:1560] [process=7][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize I0421 06:21:11.694893 138790694479616 async_checkpointer.py:280] [process=7][thread=save_finalize] Waiting for background save thread=async_save. I0421 06:21:11.695010 138919799117632 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260421_061359/linen_xpk_main_20260421_061359_02_synthetic/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776752450.912652, 'wait_for_prev_duration_secs': 20.02976107597351, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776752470.9443536, 'checkpointer_blocking_duration_secs': 0.7486181259155273, 'get_old_steps_start_time': 1776752471.6929939, 'get_old_steps_duration_secs': 2.9325485229492188e-05, 'checkpoint_manager_blocking_start_time': 1776752450.9095607, 'checkpoint_manager_blocking_duration_secs': 20.785415649414062} I0421 06:21:11.695112 138919799117632 checkpointing.py:408] Started an asynchronous checkpoint save for step 9 I0421 06:21:11.695155 138919799117632 checkpoint_manager.py:2020] [process=7][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0421 06:21:18.916216 138766542948096 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_02_synthetic/checkpoints/9/items/array_metadatas/process_7 I0421 06:21:55.028518 138790702872320 base_pytree_checkpoint_handler.py:130] [process=7] /jax/orbax/write/gbytes_per_sec: 36.415 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 43.37705421447754 s) (per-host) I0421 06:21:55.028644 138790702872320 async_checkpointer.py:90] [process=7][thread=async_save] 3 Handler Commit operations completed. Time taken: 43.335694s. I0421 06:22:03.121428 138790702872320 async_checkpointer.py:160] [process=7][thread=async_save] Background save thread done. Time taken: 51.428462s. I0421 06:22:03.121686 138790694479616 async_checkpointer.py:288] [process=7][thread=save_finalize] Done with waiting for background save thread=async_save. I0421 06:22:03.121760 138790694479616 async_checkpointer.py:298] [process=7][thread=save_finalize] No errors found in background save thread=async_save. I0421 06:22:03.121860 138790694479616 checkpoint_manager.py:2137] [process=7][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts... I0421 06:22:03.123291 138790694479616 checkpoint_manager.py:2146] [process=7][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts. I0421 06:22:03.123469 138919799117632 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 06:22:03.123625 138919799117632 checkpoint_manager.py:2009] [process=7][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0421 06:22:03.124601 138919799117632 metric_logger.py:196] completed step: 9, seconds: 0.147, TFLOP/s/device: 92.135, Tokens/s/device: 13887.665, 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 06:22:15 UTC 2026 EXIT_CODE=0
XPK Start: Tue Apr 21 11:46:04 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 11:46:29.146078: 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 11:46:29.686655 139494195316544 max_utils.py:273] Attempting to initialize the jax distributed system... I0421 11:46:38.729350 139494195316544 distributed.py:149] Starting JAX distributed service on [::]:8482 I0421 11:46:38.731630 139494195316544 distributed.py:172] Connecting to JAX distributed service on mt-02-synthetic-sd2ef-slice-job-0-0.mt-02-synthetic-sd2ef:8482 I0421 11:46:40.331664 139494195316544 max_utils.py:284] Jax distributed system initialized! I0421 11:46:45.560674 139494195316544 max_utils.py:800] System Information: Jax Version: 0.9.2 I0421 11:46:45.560798 139494195316544 max_utils.py:801] System Information: Jaxlib Version: 0.9.2 I0421 11:46:45.560840 139494195316544 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 11:46:45.560874 139494195316544 train_utils.py:378] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing. I0421 11:46:46.265952 139494195316544 maxtext_utils.py:1718] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0421 11:46:46.266525 139494195316544 maxtext_utils.py:1718] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0421 11:46:46.266700 139494195316544 checkpointing.py:688] Setting up checkpoint logger... I0421 11:46:46.266768 139494195316544 checkpointing.py:234] Creating checkpoint manager with ocdbt=True and zarr3=True I0421 11:46:46.266812 139494195316544 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None I0421 11:46:46.267144 139494195316544 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 0x7eddce34d130>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0421 11:46:49.577763 139494195316544 checkpointing.py:266] Enabling policy for fixed interval checkpointing. I0421 11:46:49.578000 139494195316544 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 0x7edcc057deb0>}, handler_registry=None I0421 11:46:49.578239 139494195316544 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7edcc057deb0>` 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 11:46:49.578287 139494195316544 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7eca3c0e5220>` 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 11:46:49.578322 139494195316544 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 0x7edcc057deb0>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7edcc057deb0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7eca3c0e5220>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7eca3c0e5220>}). I0421 11:46:49.578635 139494195316544 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34 I0421 11:46:49.578704 139494195316544 async_checkpointer.py:192] [process=7][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7ec9dc4e0180> timeout: 1200 secs and primary_host=0 for async checkpoint writes I0421 11:46:50.246896 139494195316544 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_02_synthetic/checkpoints I0421 11:46:50.535258 139494195316544 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_02_synthetic/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7eca3c0e5ca0> I0421 11:46:50.535426 139494195316544 checkpointing.py:302] Checkpoint manager created! I0421 11:46:51.472843 139494195316544 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0421 11:46:51.472948 139494195316544 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0421 11:46:51.855373 139494195316544 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0421 11:46:51.855461 139494195316544 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0421 11:46:51.871833 139494195316544 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0421 11:46:51.871890 139494195316544 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0421 11:46:51.895587 139494195316544 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 11:46:51.895655 139494195316544 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0421 11:46:51.912004 139494195316544 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 11:46:51.912067 139494195316544 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0421 11:46:51.928369 139494195316544 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 11:46:51.928426 139494195316544 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0421 11:46:51.953848 139494195316544 attentions.py:1197] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv'). I0421 11:46:51.953915 139494195316544 attentions.py:1197] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0421 11:46:51.974929 139494195316544 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0421 11:46:51.974997 139494195316544 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None). I0421 11:46:52.191295 139494195316544 checkpointing.py:578] checkpoint manager exists so trying to load this run's existing checkpoint I0421 11:46:52.191402 139494195316544 checkpointing.py:676] No existing checkpoints found, not restoring checkpoint. fsdp: 32 I0421 11:46:53.620628 139494195316544 maxtext_utils.py:1827] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: P('norm',) Physical: (None,) I0421 11:46:53.620793 139494195316544 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 11:46:53.621249 139494195316544 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 11:46:53.621384 139494195316544 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 11:46:53.621458 139494195316544 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 11:46:53.621506 139494195316544 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 11:46:53.621582 139494195316544 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 11:46:53.621646 139494195316544 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 11:46:53.621696 139494195316544 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 11:46:53.621756 139494195316544 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 11:46:53.621825 139494195316544 maxtext_utils.py:1827] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: P('embed_vocab', 'vocab') Physical: ('fsdp', None) I0421 11:46:53.621889 139494195316544 maxtext_utils.py:1827] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: P('vocab', 'embed_vocab') Physical: (None, 'fsdp') I0421 11:46:54.118881 139494195316544 train.py:157] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0421 11:46:54.118974 139494195316544 train.py:157] train/xent Physical: float32[32,2048]............................................ ('fsdp', None). I0421 11:46:54.134518 139494195316544 train.py:164] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0421 11:46:54.134580 139494195316544 train.py:164] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None). I0421 11:47:05.113133 139494195316544 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 11:47:05.114009 139494195316544 metric_logger.py:301] number parameters: 1.104 billion I0421 11:47:16.730636 139494195316544 checkpointing.py:794] Waiting for step 0 to finish before checkpoint... I0421 11:47:16.884751 139494195316544 checkpointing.py:798] Waited 0.15407919883728027 seconds for step 0 to finish before starting checkpointing. I0421 11:47:16.888761 139494195316544 checkpoint_manager.py:2009] [process=7][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0421 11:47:16.890222 139494195316544 checkpoint_manager.py:1512] [process=7] Saving checkpoint at step 0 I0421 11:47:16.891654 139494195316544 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_02_synthetic/checkpoints/0. I0421 11:47:17.206386 139494195316544 signaling_client.py:364] Using JaxDistributedSignalingClient I0421 11:47:17.207250 139494195316544 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0421 11:47:17.207306 139494195316544 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 11:47:17.483639 139494195316544 base_pytree_checkpoint_handler.py:154] [process=7][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.277326s I0421 11:47:17.483835 139494195316544 base_pytree_checkpoint_handler.py:130] [process=7] /jax/orbax/write/blocking_gbytes_per_sec: 5.455 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.28279638290405273 s) (per-host) I0421 11:47:17.483891 139494195316544 base_pytree_checkpoint_handler.py:768] [process=7][thread=MainThread] Initiated Pytree async_save. Time taken: 0.282869s (batch_requests_ready=0.002226s, total_serialization_initiated=0.280563s, others=0.000081s) I0421 11:47:17.483988 139494195316544 composite_checkpoint_handler.py:715] [process=7][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.287080s (all_items=0.000017s, per_item={'items': '0.00001717'}, temp_paths=0.287063) I0421 11:47:17.484769 139494195316544 event_tracking.py:125] [process=7] [async] Finished blocking save in 0.59 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_02_synthetic/checkpoints/0. I0421 11:47:17.485131 139367921489664 async_checkpointer.py:76] [process=7][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-21 12:07:17.485093 I0421 11:47:17.488432 139494195316544 checkpoint_manager.py:1560] [process=7][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize I0421 11:47:17.488686 139362561947392 async_checkpointer.py:280] [process=7][thread=save_finalize] Waiting for background save thread=async_save. I0421 11:47:17.488813 139494195316544 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_02_synthetic/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776772036.8887253, 'wait_for_prev_duration_secs': 8.225440979003906e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776772036.8902605, 'checkpointer_blocking_duration_secs': 0.5950276851654053, 'get_old_steps_start_time': 1776772037.485311, 'get_old_steps_duration_secs': 3.0040740966796875e-05, 'checkpoint_manager_blocking_start_time': 1776772036.8851478, 'checkpoint_manager_blocking_duration_secs': 0.6036262512207031} I0421 11:47:17.488922 139494195316544 checkpointing.py:409] Started an asynchronous checkpoint save for step 0 I0421 11:47:17.488971 139494195316544 max_utils.py:750] Memstats: After params initialized: I0421 11:47:17.489020 139494195316544 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_26(process=7,(2,6,0,0)) I0421 11:47:17.489059 139494195316544 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_27(process=7,(3,6,0,0)) I0421 11:47:17.489085 139494195316544 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_30(process=7,(2,7,0,0)) I0421 11:47:17.489109 139494195316544 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_31(process=7,(3,7,0,0)) I0421 11:47:17.805367 139494195316544 metric_logger.py:196] completed step: 0, seconds: 11.617, TFLOP/s/device: 1.170, Tokens/s/device: 176.301, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52779.875 I0421 11:47:17.983404 139494195316544 metric_logger.py:196] completed step: 1, seconds: 1.073, TFLOP/s/device: 12.661, Tokens/s/device: 1908.331, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52779.875 I0421 11:47:18.383726 139494195316544 metric_logger.py:196] completed step: 2, seconds: 0.031, TFLOP/s/device: 443.647, Tokens/s/device: 66871.286, total_weights: 65536, loss: 10.262, lm_loss: 10.262, perplexity: 28637.129 I0421 11:47:18.530931 139494195316544 metric_logger.py:196] completed step: 3, seconds: 0.401, TFLOP/s/device: 33.917, Tokens/s/device: 5112.408, total_weights: 65536, loss: 9.734, lm_loss: 9.734, perplexity: 16889.203 I0421 11:47:18.826101 139494195316544 metric_logger.py:196] completed step: 4, seconds: 0.153, TFLOP/s/device: 88.644, Tokens/s/device: 13361.343, total_weights: 65536, loss: 9.277, lm_loss: 9.277, perplexity: 10694.614 I0421 11:47:18.832160 139494195316544 metric_logger.py:196] completed step: 5, seconds: 0.147, TFLOP/s/device: 92.366, Tokens/s/device: 13922.407, total_weights: 65536, loss: 8.892, lm_loss: 8.892, perplexity: 7270.717 I0421 11:47:21.083806 2764 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com I0421 11:47:23.634583 139363578824448 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_02_synthetic/checkpoints/0/items/array_metadatas/process_7 I0421 11:47:40.837453 139494195316544 metric_logger.py:196] completed step: 6, seconds: 0.296, TFLOP/s/device: 45.948, Tokens/s/device: 6925.798, total_weights: 65536, loss: 8.592, lm_loss: 8.592, perplexity: 5390.759 I0421 11:47:40.984865 139494195316544 metric_logger.py:196] completed step: 7, seconds: 21.858, TFLOP/s/device: 0.622, Tokens/s/device: 93.694, total_weights: 65536, loss: 8.384, lm_loss: 8.384, perplexity: 4376.905 I0421 11:47:41.132064 139494195316544 metric_logger.py:196] completed step: 8, seconds: 0.153, TFLOP/s/device: 88.977, Tokens/s/device: 13411.655, total_weights: 65536, loss: 8.255, lm_loss: 8.255, perplexity: 3846.406 I0421 11:47:41.278796 139494195316544 checkpointing.py:794] Waiting for step 9 to finish before checkpoint... I0421 11:47:41.279437 139494195316544 checkpointing.py:798] Waited 0.0006608963012695312 seconds for step 9 to finish before starting checkpointing. I0421 11:47:41.281492 139494195316544 checkpoint_manager.py:2020] [process=7][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0421 11:47:55.147235 139367921489664 base_pytree_checkpoint_handler.py:130] [process=7] /jax/orbax/write/gbytes_per_sec: 41.627 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 37.94616365432739 s) (per-host) I0421 11:47:55.147364 139367921489664 async_checkpointer.py:90] [process=7][thread=async_save] 3 Handler Commit operations completed. Time taken: 37.662121s. I0421 11:48:03.143228 139367921489664 async_checkpointer.py:160] [process=7][thread=async_save] Background save thread done. Time taken: 45.657968s. I0421 11:48:03.143459 139362561947392 async_checkpointer.py:288] [process=7][thread=save_finalize] Done with waiting for background save thread=async_save. I0421 11:48:03.143530 139362561947392 async_checkpointer.py:298] [process=7][thread=save_finalize] No errors found in background save thread=async_save. I0421 11:48:03.143582 139362561947392 checkpoint_manager.py:2137] [process=7][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts... I0421 11:48:03.145912 139362561947392 checkpoint_manager.py:2146] [process=7][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts. I0421 11:48:03.146100 139494195316544 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 11:48:03.146245 139494195316544 checkpoint_manager.py:1452] Waiting for previous save to complete took 21.864753 seconds. If this number is high, consider checkpointing less frequently. I0421 11:48:03.147937 139494195316544 checkpoint_manager.py:1512] [process=7] Saving checkpoint at step 9 I0421 11:48:03.149916 139494195316544 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_02_synthetic/checkpoints/9. I0421 11:48:03.449974 139494195316544 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0421 11:48:03.450072 139494195316544 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 11:48:03.485295 139494195316544 base_pytree_checkpoint_handler.py:154] [process=7][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.036407s I0421 11:48:03.485461 139494195316544 base_pytree_checkpoint_handler.py:130] [process=7] /jax/orbax/write/blocking_gbytes_per_sec: 38.744 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.03981447219848633 s) (per-host) I0421 11:48:03.485512 139494195316544 base_pytree_checkpoint_handler.py:768] [process=7][thread=MainThread] Initiated Pytree async_save. Time taken: 0.039875s (batch_requests_ready=0.001761s, total_serialization_initiated=0.038046s, others=0.000068s) I0421 11:48:03.485602 139494195316544 composite_checkpoint_handler.py:715] [process=7][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.044200s (all_items=0.000015s, per_item={'items': '0.00001502'}, temp_paths=0.044185) I0421 11:48:03.487678 139494195316544 event_tracking.py:125] [process=7] [async] Finished blocking save in 0.34 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_02_synthetic/checkpoints/9. I0421 11:48:03.487987 139362561947392 async_checkpointer.py:76] [process=7][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-21 12:08:03.487954 I0421 11:48:03.489962 139494195316544 checkpoint_manager.py:1560] [process=7][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize I0421 11:48:03.490264 139346279573248 async_checkpointer.py:280] [process=7][thread=save_finalize] Waiting for background save thread=async_save. I0421 11:48:03.490396 139494195316544 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_02_synthetic/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776772061.2814627, 'wait_for_prev_duration_secs': 21.864752531051636, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776772083.1479747, 'checkpointer_blocking_duration_secs': 0.3401632308959961, 'get_old_steps_start_time': 1776772083.4881618, 'get_old_steps_duration_secs': 3.147125244140625e-05, 'checkpoint_manager_blocking_start_time': 1776772061.2796614, 'checkpoint_manager_blocking_duration_secs': 22.21070146560669} I0421 11:48:03.490549 139494195316544 checkpointing.py:409] Started an asynchronous checkpoint save for step 9 I0421 11:48:03.490598 139494195316544 checkpoint_manager.py:2020] [process=7][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0421 11:48:09.009814 139363578824448 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_02_synthetic/checkpoints/9/items/array_metadatas/process_7 I0421 11:48:45.061472 139362561947392 base_pytree_checkpoint_handler.py:130] [process=7] /jax/orbax/write/gbytes_per_sec: 37.957 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 41.61578941345215 s) (per-host) I0421 11:48:45.061582 139362561947392 async_checkpointer.py:90] [process=7][thread=async_save] 3 Handler Commit operations completed. Time taken: 41.573477s. I0421 11:48:53.682714 139362561947392 async_checkpointer.py:160] [process=7][thread=async_save] Background save thread done. Time taken: 50.194593s. I0421 11:48:53.683017 139346279573248 async_checkpointer.py:288] [process=7][thread=save_finalize] Done with waiting for background save thread=async_save. I0421 11:48:53.683148 139346279573248 async_checkpointer.py:298] [process=7][thread=save_finalize] No errors found in background save thread=async_save. I0421 11:48:53.683194 139346279573248 checkpoint_manager.py:2137] [process=7][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts... I0421 11:48:53.684659 139346279573248 checkpoint_manager.py:2146] [process=7][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts. I0421 11:48:53.684848 139494195316544 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 11:48:53.684996 139494195316544 checkpoint_manager.py:2009] [process=7][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0421 11:48:53.685896 139494195316544 metric_logger.py:196] completed step: 9, seconds: 0.147, TFLOP/s/device: 92.446, Tokens/s/device: 13934.532, 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 11:49:05 UTC 2026 EXIT_CODE=0