main (586e69205) vs feat/nnx-post-train-fixes (7f06c99ac)| Metric | main 586e69205 | feat/nnx-post-train-fixes 7f06c99ac | Diff (feat/nnx-post-train-fixes − main) |
|---|---|---|---|
| Parameters | 1.104 billion | 1.104 billion | — |
| Final loss | 9.4570 | 9.4570 | 0 |
| TFLOP/s | 93.952 | 94.022 | +0.07 |
| Tok/s | 14161.4 | 14172.0 | +10.608 |
| Avg s/step | 7.002 | 6.837 | -0.165 |
| 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: Thu Apr 23 07:34:09 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. 2026-04-23 07:34:34.860287: 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) I0423 07:34:35.071176 137687193626432 max_utils.py:273] Attempting to initialize the jax distributed system... I0423 07:34:44.110837 137687193626432 distributed.py:149] Starting JAX distributed service on [::]:8482 I0423 07:34:44.113262 137687193626432 distributed.py:172] Connecting to JAX distributed service on mt-06-grad-accum-ixzpw-slice-job-0-0.mt-06-grad-accum-ixzpw:8482 I0423 07:34:45.349770 137687193626432 max_utils.py:284] Jax distributed system initialized! I0423 07:34:51.439265 137687193626432 max_utils.py:800] System Information: Jax Version: 0.9.2 I0423 07:34:51.439371 137687193626432 max_utils.py:801] System Information: Jaxlib Version: 0.9.2 I0423 07:34:51.439411 137687193626432 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 I0423 07:34:51.439446 137687193626432 train_utils.py:361] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing. I0423 07:34:52.129706 137687193626432 maxtext_utils.py:1604] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0423 07:34:52.129978 137687193626432 checkpointing.py:677] Setting up checkpoint logger... I0423 07:34:52.130032 137687193626432 checkpointing.py:233] Creating checkpoint manager with ocdbt=True and zarr3=True I0423 07:34:52.130076 137687193626432 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None I0423 07:34:52.130415 137687193626432 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 0x7d39711c26c0>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0423 07:34:55.025825 137687193626432 checkpointing.py:265] Enabling policy for fixed interval checkpointing. I0423 07:34:55.026064 137687193626432 checkpoint_manager.py:708] [process=5][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=('items',), item_handlers={'items': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7d250c469a00>}, handler_registry=None I0423 07:34:55.026303 137687193626432 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7d250c469a00>` 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`. I0423 07:34:55.026351 137687193626432 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7d250c46d460>` 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`. I0423 07:34:55.026388 137687193626432 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 0x7d250c469a00>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7d250c469a00>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7d250c46d460>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7d250c46d460>}). I0423 07:34:55.026723 137687193626432 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34 I0423 07:34:55.026793 137687193626432 async_checkpointer.py:192] [process=5][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7d250c26da80> timeout: 1200 secs and primary_host=0 for async checkpoint writes I0423 07:34:56.628008 137687193626432 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_main_20260423_071538/linen_xpk_main_20260423_071538_06_grad_accum/checkpoints I0423 07:34:56.674486 137687193626432 checkpoint_manager.py:929] [process=5][thread=MainThread] CheckpointManager created, primary_host=0, CheckpointManagerOptions=CheckpointManagerOptions(save_interval_steps=1, max_to_keep=None, keep_time_interval=None, keep_period=None, should_keep_fn=None, best_fn=None, best_mode='max', keep_checkpoints_without_metrics=True, step_prefix=None, step_format_fixed_length=None, step_name_format=None, create=True, cleanup_tmp_directories=False, save_on_steps=frozenset(), single_host_load_and_broadcast=False, todelete_subdir=None, todelete_full_path=None, enable_background_delete=False, read_only=False, enable_async_checkpointing=True, async_options=None, multiprocessing_options=MultiprocessingOptions(primary_host=0, active_processes=None, barrier_sync_key_prefix=None), should_save_fn=None, file_options=FileOptions(path_permission_mode=None), save_root_metadata=True, temporary_path_class=None, save_decision_policy=FixedIntervalPolicy(interval=10), preservation_policy=LatestN(n=None), prevent_write_metrics=False, enable_should_save_is_saving_in_progress_check=True, enable_per_process_directory_creation=False, lightweight_initialize=False), root_directory=gs://lance-maxtext/linen_ckpt_xpk_main_20260423_071538/linen_xpk_main_20260423_071538_06_grad_accum/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7d250c46a810> I0423 07:34:56.674631 137687193626432 checkpointing.py:301] Checkpoint manager created! I0423 07:34:58.391894 137687193626432 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0423 07:34:58.392013 137687193626432 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0423 07:34:58.776692 137687193626432 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0423 07:34:58.776784 137687193626432 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0423 07:34:58.793683 137687193626432 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0423 07:34:58.793744 137687193626432 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0423 07:34:58.817907 137687193626432 attentions.py:1154] attentions/query Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0423 07:34:58.817985 137687193626432 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0423 07:34:58.834848 137687193626432 attentions.py:1155] attentions/key Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0423 07:34:58.834908 137687193626432 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0423 07:34:58.851855 137687193626432 attentions.py:1156] attentions/value Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0423 07:34:58.851913 137687193626432 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0423 07:34:58.877067 137687193626432 attentions.py:1198] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv'). I0423 07:34:58.877134 137687193626432 attentions.py:1198] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0423 07:34:58.898402 137687193626432 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0423 07:34:58.898470 137687193626432 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None). I0423 07:34:59.109830 137687193626432 checkpointing.py:577] checkpoint manager exists so trying to load this run's existing checkpoint I0423 07:34:59.109941 137687193626432 checkpointing.py:665] No existing checkpoints found, not restoring checkpoint. fsdp: 32 I0423 07:35:00.533377 137687193626432 maxtext_utils.py:1707] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: P('norm',) Physical: (None,) I0423 07:35:00.533503 137687193626432 maxtext_utils.py:1707] params/params/decoder/layers/mlp/wi_0/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0423 07:35:00.533556 137687193626432 maxtext_utils.py:1707] params/params/decoder/layers/mlp/wi_1/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0423 07:35:00.533615 137687193626432 maxtext_utils.py:1707] params/params/decoder/layers/mlp/wo/kernel Shape: float32[7168,16,2048] Logical: P('mlp', 'layers', 'embed') Physical: (None, None, 'fsdp') I0423 07:35:00.533680 137687193626432 maxtext_utils.py:1707] params/params/decoder/layers/post_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0423 07:35:00.533724 137687193626432 maxtext_utils.py:1707] params/params/decoder/layers/pre_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0423 07:35:00.533778 137687193626432 maxtext_utils.py:1707] 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) I0423 07:35:00.533831 137687193626432 maxtext_utils.py:1707] 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') I0423 07:35:00.533871 137687193626432 maxtext_utils.py:1707] 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) I0423 07:35:00.533907 137687193626432 maxtext_utils.py:1707] 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) I0423 07:35:00.533958 137687193626432 maxtext_utils.py:1707] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: P('embed_vocab', 'vocab') Physical: ('fsdp', None) I0423 07:35:00.534006 137687193626432 maxtext_utils.py:1707] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: P('vocab', 'embed_vocab') Physical: (None, 'fsdp') I0423 07:35:00.556779 137687193626432 gradient_accumulation.py:68] gradient_accumulation/inputs Logical: float32[2048]............................................... Unknown. I0423 07:35:00.556844 137687193626432 gradient_accumulation.py:68] gradient_accumulation/inputs Physical: float32[2048]............................................... (None,). I0423 07:35:00.571748 137687193626432 gradient_accumulation.py:68] gradient_accumulation/inputs Logical: float32[2048,16,7168]....................................... Unknown. I0423 07:35:00.571807 137687193626432 gradient_accumulation.py:68] gradient_accumulation/inputs Physical: float32[2048,16,7168]....................................... ('fsdp', None, None). I0423 07:35:00.601384 137687193626432 gradient_accumulation.py:68] gradient_accumulation/inputs Logical: float32[7168,16,2048]....................................... Unknown. I0423 07:35:00.601450 137687193626432 gradient_accumulation.py:68] gradient_accumulation/inputs Physical: float32[7168,16,2048]....................................... (None, None, 'fsdp'). I0423 07:35:00.616424 137687193626432 gradient_accumulation.py:68] gradient_accumulation/inputs Logical: float32[2048,16]............................................ Unknown. I0423 07:35:00.616477 137687193626432 gradient_accumulation.py:68] gradient_accumulation/inputs Physical: float32[2048,16]............................................ (None, None). I0423 07:35:00.646317 137687193626432 gradient_accumulation.py:68] gradient_accumulation/inputs Logical: float32[2048,16,16,128]..................................... Unknown. I0423 07:35:00.646381 137687193626432 gradient_accumulation.py:68] gradient_accumulation/inputs Physical: float32[2048,16,16,128]..................................... ('fsdp', None, None, None). I0423 07:35:00.661214 137687193626432 gradient_accumulation.py:68] gradient_accumulation/inputs Logical: float32[16,16,128,2048]..................................... Unknown. I0423 07:35:00.661273 137687193626432 gradient_accumulation.py:68] gradient_accumulation/inputs Physical: float32[16,16,128,2048]..................................... (None, None, None, 'fsdp'). I0423 07:35:00.705718 137687193626432 gradient_accumulation.py:68] gradient_accumulation/inputs Logical: float32[2048,32000]......................................... Unknown. I0423 07:35:00.705785 137687193626432 gradient_accumulation.py:68] gradient_accumulation/inputs Physical: float32[2048,32000]......................................... ('fsdp', None). I0423 07:35:00.720642 137687193626432 gradient_accumulation.py:68] gradient_accumulation/inputs Logical: float32[32000,2048]......................................... Unknown. I0423 07:35:00.720705 137687193626432 gradient_accumulation.py:68] gradient_accumulation/inputs Physical: float32[32000,2048]......................................... (None, 'fsdp'). I0423 07:35:01.395866 137687193626432 train.py:155] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0423 07:35:01.395960 137687193626432 train.py:155] train/xent Physical: float32[32,2048]............................................ ('fsdp', None). I0423 07:35:01.411646 137687193626432 train.py:162] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0423 07:35:01.411716 137687193626432 train.py:162] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None). I0423 07:35:12.661813 137687193626432 max_utils.py:791] Total memory size: 1.7 GB, Output size: 0.4 GB, Temp size: 1.3 GB, Argument size: 0.4 GB, Host temp size: 0.0 GB. I0423 07:35:12.662583 137687193626432 metric_logger.py:301] number parameters: 1.104 billion I0423 07:35:14.984833 137687193626432 checkpointing.py:772] Waiting for step 0 to finish before checkpoint... I0423 07:35:26.009070 137687193626432 checkpointing.py:776] Waited 11.024211645126343 seconds for step 0 to finish before starting checkpointing. I0423 07:35:26.011588 137687193626432 checkpoint_manager.py:2009] [process=5][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0423 07:35:26.013336 137687193626432 checkpoint_manager.py:1512] [process=5] Saving checkpoint at step 0 I0423 07:35:26.014776 137687193626432 event_tracking.py:70] [process=5] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260423_071538/linen_xpk_main_20260423_071538_06_grad_accum/checkpoints/0. I0423 07:35:26.777310 137687193626432 signaling_client.py:364] Using JaxDistributedSignalingClient I0423 07:35:26.778301 137687193626432 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0423 07:35:26.778358 137687193626432 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 I0423 07:35:27.057792 137687193626432 base_pytree_checkpoint_handler.py:154] [process=5][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.280558s I0423 07:35:27.057970 137687193626432 base_pytree_checkpoint_handler.py:130] [process=5] /jax/orbax/write/blocking_gbytes_per_sec: 5.389 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.2862355709075928 s) (per-host) I0423 07:35:27.058030 137687193626432 base_pytree_checkpoint_handler.py:768] [process=5][thread=MainThread] Initiated Pytree async_save. Time taken: 0.286304s (batch_requests_ready=0.002290s, total_serialization_initiated=0.283937s, others=0.000077s) I0423 07:35:27.058125 137687193626432 composite_checkpoint_handler.py:715] [process=5][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.290375s (all_items=0.000017s, per_item={'items': '0.00001740'}, temp_paths=0.290358) I0423 07:35:27.058930 137687193626432 event_tracking.py:125] [process=5] [async] Finished blocking save in 1.05 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260423_071538/linen_xpk_main_20260423_071538_06_grad_accum/checkpoints/0. I0423 07:35:27.059272 137557800212224 async_checkpointer.py:76] [process=5][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-23 07:55:27.059229 I0423 07:35:27.061184 137687193626432 checkpoint_manager.py:1560] [process=5][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize I0423 07:35:27.061425 137555605309184 async_checkpointer.py:280] [process=5][thread=save_finalize] Waiting for background save thread=async_save. I0423 07:35:27.061575 137687193626432 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260423_071538/linen_xpk_main_20260423_071538_06_grad_accum/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776929726.0115697, 'wait_for_prev_duration_secs': 6.0558319091796875e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776929726.013374, 'checkpointer_blocking_duration_secs': 1.0459988117218018, 'get_old_steps_start_time': 1776929727.059398, 'get_old_steps_duration_secs': 3.075599670410156e-05, 'checkpoint_manager_blocking_start_time': 1776929726.009592, 'checkpoint_manager_blocking_duration_secs': 1.0519473552703857} I0423 07:35:27.061693 137687193626432 checkpointing.py:408] Started an asynchronous checkpoint save for step 0 I0423 07:35:27.061744 137687193626432 max_utils.py:750] Memstats: After params initialized: I0423 07:35:27.061793 137687193626432 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_18(process=5,(2,4,0,0)) I0423 07:35:27.061826 137687193626432 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_19(process=5,(3,4,0,0)) I0423 07:35:27.061855 137687193626432 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_22(process=5,(2,5,0,0)) I0423 07:35:27.061881 137687193626432 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_23(process=5,(3,5,0,0)) I0423 07:35:27.374334 137687193626432 metric_logger.py:196] completed step: 0, seconds: 2.322, TFLOP/s/device: 23.404, Tokens/s/device: 3527.770, total_weights: 262144, loss: 10.877, lm_loss: 10.877, perplexity: 52959.059 I0423 07:35:27.978063 137687193626432 metric_logger.py:196] completed step: 1, seconds: 12.388, TFLOP/s/device: 4.387, Tokens/s/device: 661.284, total_weights: 262144, loss: 10.877, lm_loss: 10.877, perplexity: 52959.059 I0423 07:35:28.556168 137687193626432 metric_logger.py:196] completed step: 2, seconds: 0.026, TFLOP/s/device: 2115.551, Tokens/s/device: 318878.941, total_weights: 262144, loss: 10.563, lm_loss: 10.563, perplexity: 38662.707 I0423 07:35:29.134636 137687193626432 metric_logger.py:196] completed step: 3, seconds: 0.584, TFLOP/s/device: 93.070, Tokens/s/device: 14028.574, total_weights: 262144, loss: 10.272, lm_loss: 10.272, perplexity: 28909.668 I0423 07:35:29.924299 2555 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com I0423 07:35:30.291373 137687193626432 metric_logger.py:196] completed step: 4, seconds: 0.578, TFLOP/s/device: 94.008, Tokens/s/device: 14169.970, total_weights: 262144, loss: 10.022, lm_loss: 10.022, perplexity: 22524.992 I0423 07:35:30.297863 137687193626432 metric_logger.py:196] completed step: 5, seconds: 0.578, TFLOP/s/device: 93.972, Tokens/s/device: 14164.531, total_weights: 262144, loss: 9.820, lm_loss: 9.820, perplexity: 18401.865 I0423 07:35:32.828964 137555613701888 array_metadata_store.py:203] [process=5][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260423_071538/linen_xpk_main_20260423_071538_06_grad_accum/checkpoints/0/items/array_metadatas/process_5 I0423 07:36:05.385525 137557800212224 base_pytree_checkpoint_handler.py:130] [process=5] /jax/orbax/write/gbytes_per_sec: 40.908 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 38.61375308036804 s) (per-host) I0423 07:36:05.385648 137557800212224 async_checkpointer.py:90] [process=5][thread=async_save] 3 Handler Commit operations completed. Time taken: 38.325744s. I0423 07:36:12.760230 137557800212224 async_checkpointer.py:160] [process=5][thread=async_save] Background save thread done. Time taken: 45.700310s. I0423 07:36:12.760536 137555605309184 async_checkpointer.py:288] [process=5][thread=save_finalize] Done with waiting for background save thread=async_save. I0423 07:36:12.760678 137555605309184 async_checkpointer.py:298] [process=5][thread=save_finalize] No errors found in background save thread=async_save. I0423 07:36:12.760749 137555605309184 checkpoint_manager.py:2137] [process=5][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts... I0423 07:36:12.763150 137555605309184 checkpoint_manager.py:2146] [process=5][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts. I0423 07:36:17.421472 137687193626432 metric_logger.py:196] completed step: 6, seconds: 1.158, TFLOP/s/device: 46.941, Tokens/s/device: 7075.512, total_weights: 262144, loss: 9.667, lm_loss: 9.667, perplexity: 15787.604 I0423 07:36:17.999947 137687193626432 metric_logger.py:196] completed step: 7, seconds: 46.546, TFLOP/s/device: 1.168, Tokens/s/device: 175.998, total_weights: 262144, loss: 9.561, lm_loss: 9.561, perplexity: 14203.827 I0423 07:36:18.577988 137687193626432 metric_logger.py:196] completed step: 8, seconds: 0.583, TFLOP/s/device: 93.212, Tokens/s/device: 14049.988, total_weights: 262144, loss: 9.496, lm_loss: 9.496, perplexity: 13302.920 I0423 07:36:19.155470 137687193626432 checkpointing.py:772] Waiting for step 9 to finish before checkpoint... I0423 07:36:19.156188 137687193626432 checkpointing.py:776] Waited 0.0007278919219970703 seconds for step 9 to finish before starting checkpointing. I0423 07:36:19.158305 137687193626432 checkpoint_manager.py:2009] [process=5][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0423 07:36:19.160223 137687193626432 checkpoint_manager.py:1512] [process=5] Saving checkpoint at step 9 I0423 07:36:19.161946 137687193626432 event_tracking.py:70] [process=5] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260423_071538/linen_xpk_main_20260423_071538_06_grad_accum/checkpoints/9. I0423 07:36:19.866107 137687193626432 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0423 07:36:19.866204 137687193626432 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 I0423 07:36:19.899484 137687193626432 base_pytree_checkpoint_handler.py:154] [process=5][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.034268s I0423 07:36:19.899667 137687193626432 base_pytree_checkpoint_handler.py:130] [process=5] /jax/orbax/write/blocking_gbytes_per_sec: 40.763 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.037842750549316406 s) (per-host) I0423 07:36:19.899723 137687193626432 base_pytree_checkpoint_handler.py:768] [process=5][thread=MainThread] Initiated Pytree async_save. Time taken: 0.037920s (batch_requests_ready=0.001724s, total_serialization_initiated=0.036111s, others=0.000085s) I0423 07:36:19.899820 137687193626432 composite_checkpoint_handler.py:715] [process=5][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.042134s (all_items=0.000015s, per_item={'items': '0.00001526'}, temp_paths=0.042119) I0423 07:36:19.900455 137687193626432 event_tracking.py:125] [process=5] [async] Finished blocking save in 0.74 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260423_071538/linen_xpk_main_20260423_071538_06_grad_accum/checkpoints/9. I0423 07:36:19.900744 137557266396928 async_checkpointer.py:76] [process=5][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-23 07:56:19.900702 I0423 07:36:19.902689 137687193626432 checkpoint_manager.py:1560] [process=5][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize I0423 07:36:19.902934 137563700455168 async_checkpointer.py:280] [process=5][thread=save_finalize] Waiting for background save thread=async_save. I0423 07:36:19.903089 137687193626432 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260423_071538/linen_xpk_main_20260423_071538_06_grad_accum/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776929779.1582735, 'wait_for_prev_duration_secs': 7.534027099609375e-05, 'time_between_consecutive_saves_sec': 6.3950817584991455, 'checkpointer_blocking_start_time': 1776929779.1603103, 'checkpointer_blocking_duration_secs': 0.7405805587768555, 'get_old_steps_start_time': 1776929779.9009132, 'get_old_steps_duration_secs': 2.6702880859375e-05, 'checkpoint_manager_blocking_start_time': 1776929779.156493, 'checkpoint_manager_blocking_duration_secs': 0.7465634346008301} I0423 07:36:19.903190 137687193626432 checkpointing.py:408] Started an asynchronous checkpoint save for step 9 I0423 07:36:19.903239 137687193626432 checkpoint_manager.py:2020] [process=5][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0423 07:36:25.885803 137557800212224 array_metadata_store.py:203] [process=5][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260423_071538/linen_xpk_main_20260423_071538_06_grad_accum/checkpoints/9/items/array_metadatas/process_5 I0423 07:37:02.649752 137557266396928 base_pytree_checkpoint_handler.py:130] [process=5] /jax/orbax/write/gbytes_per_sec: 36.917 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 42.78790044784546 s) (per-host) I0423 07:37:02.649880 137557266396928 async_checkpointer.py:90] [process=5][thread=async_save] 3 Handler Commit operations completed. Time taken: 42.749018s. I0423 07:37:10.678960 137557266396928 async_checkpointer.py:160] [process=5][thread=async_save] Background save thread done. Time taken: 50.778084s. I0423 07:37:10.679266 137563700455168 async_checkpointer.py:288] [process=5][thread=save_finalize] Done with waiting for background save thread=async_save. I0423 07:37:10.679382 137563700455168 async_checkpointer.py:298] [process=5][thread=save_finalize] No errors found in background save thread=async_save. I0423 07:37:10.679427 137563700455168 checkpoint_manager.py:2137] [process=5][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts... I0423 07:37:10.681847 137563700455168 checkpoint_manager.py:2146] [process=5][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts. I0423 07:37:10.682046 137687193626432 checkpoint_manager.py:2032] [process=5][thread=MainThread][step=9][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=9. I0423 07:37:10.682198 137687193626432 checkpoint_manager.py:2009] [process=5][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0423 07:37:10.683190 137687193626432 metric_logger.py:196] completed step: 9, seconds: 0.578, TFLOP/s/device: 93.952, Tokens/s/device: 14161.397, total_weights: 262144, loss: 9.457, lm_loss: 9.457, perplexity: 12802.546 Per train step: Total TFLOPs: 54.35 split as 93.93% learnable weight flops and 6.07% attention flops XPK End: Thu Apr 23 07:37:20 UTC 2026 EXIT_CODE=0
XPK Start: Thu Apr 23 13:01:14 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. 2026-04-23 13:01:38.930611: 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) I0423 13:01:39.138603 139991267690304 max_utils.py:273] Attempting to initialize the jax distributed system... I0423 13:01:48.179437 139991267690304 distributed.py:149] Starting JAX distributed service on [::]:8482 I0423 13:01:48.181746 139991267690304 distributed.py:172] Connecting to JAX distributed service on mt-06-grad-accum-mrxk2-slice-job-0-0.mt-06-grad-accum-mrxk2:8482 I0423 13:01:49.588434 139991267690304 max_utils.py:284] Jax distributed system initialized! I0423 13:01:55.865886 139991267690304 max_utils.py:800] System Information: Jax Version: 0.9.2 I0423 13:01:55.865992 139991267690304 max_utils.py:801] System Information: Jaxlib Version: 0.9.2 I0423 13:01:55.866032 139991267690304 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 I0423 13:01:55.866069 139991267690304 train_utils.py:391] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing. I0423 13:01:56.583246 139991267690304 maxtext_utils.py:1771] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0423 13:01:56.583855 139991267690304 maxtext_utils.py:1771] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0423 13:01:56.584034 139991267690304 checkpointing.py:688] Setting up checkpoint logger... I0423 13:01:56.584083 139991267690304 checkpointing.py:234] Creating checkpoint manager with ocdbt=True and zarr3=True I0423 13:01:56.584128 139991267690304 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None I0423 13:01:56.584473 139991267690304 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 0x7f51bf518e30>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0423 13:01:59.575177 139991267690304 checkpointing.py:266] Enabling policy for fixed interval checkpointing. I0423 13:01:59.575520 139991267690304 checkpoint_manager.py:708] [process=5][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=('items',), item_handlers={'items': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7f3d1c5d29c0>}, handler_registry=None I0423 13:01:59.575766 139991267690304 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7f3d1c5d29c0>` 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`. I0423 13:01:59.575815 139991267690304 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7f3d1c5d4260>` 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`. I0423 13:01:59.575851 139991267690304 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 0x7f3d1c5d29c0>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7f3d1c5d29c0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7f3d1c5d4260>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7f3d1c5d4260>}). I0423 13:01:59.576173 139991267690304 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34 I0423 13:01:59.576243 139991267690304 async_checkpointer.py:192] [process=5][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7f3d1c3cdd00> timeout: 1200 secs and primary_host=0 for async checkpoint writes I0423 13:02:00.287720 139991267690304 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260423_124541/linen_xpk_feat_nnx_post_train_fixes_20260423_124541_06_grad_accum/checkpoints I0423 13:02:00.341284 139991267690304 checkpoint_manager.py:929] [process=5][thread=MainThread] CheckpointManager created, primary_host=0, CheckpointManagerOptions=CheckpointManagerOptions(save_interval_steps=1, max_to_keep=None, keep_time_interval=None, keep_period=None, should_keep_fn=None, best_fn=None, best_mode='max', keep_checkpoints_without_metrics=True, step_prefix=None, step_format_fixed_length=None, step_name_format=None, create=True, cleanup_tmp_directories=False, save_on_steps=frozenset(), single_host_load_and_broadcast=False, todelete_subdir=None, todelete_full_path=None, enable_background_delete=False, read_only=False, enable_async_checkpointing=True, async_options=None, multiprocessing_options=MultiprocessingOptions(primary_host=0, active_processes=None, barrier_sync_key_prefix=None), should_save_fn=None, file_options=FileOptions(path_permission_mode=None), save_root_metadata=True, temporary_path_class=None, save_decision_policy=FixedIntervalPolicy(interval=10), preservation_policy=LatestN(n=None), prevent_write_metrics=False, enable_should_save_is_saving_in_progress_check=True, enable_per_process_directory_creation=False, lightweight_initialize=False), root_directory=gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260423_124541/linen_xpk_feat_nnx_post_train_fixes_20260423_124541_06_grad_accum/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7f3d1c5d3e90> I0423 13:02:00.341438 139991267690304 checkpointing.py:302] Checkpoint manager created! I0423 13:02:02.059670 139991267690304 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0423 13:02:02.059776 139991267690304 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0423 13:02:02.443850 139991267690304 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0423 13:02:02.443938 139991267690304 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0423 13:02:02.460428 139991267690304 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0423 13:02:02.460487 139991267690304 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0423 13:02:02.484179 139991267690304 attentions.py:1154] attentions/query Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0423 13:02:02.484250 139991267690304 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0423 13:02:02.500769 139991267690304 attentions.py:1155] attentions/key Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0423 13:02:02.500831 139991267690304 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0423 13:02:02.517367 139991267690304 attentions.py:1156] attentions/value Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0423 13:02:02.517424 139991267690304 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0423 13:02:02.542137 139991267690304 attentions.py:1198] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv'). I0423 13:02:02.542208 139991267690304 attentions.py:1198] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0423 13:02:02.562927 139991267690304 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0423 13:02:02.562992 139991267690304 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None). I0423 13:02:02.781953 139991267690304 checkpointing.py:578] checkpoint manager exists so trying to load this run's existing checkpoint I0423 13:02:02.782058 139991267690304 checkpointing.py:676] No existing checkpoints found, not restoring checkpoint. fsdp: 32 I0423 13:02:04.220354 139991267690304 maxtext_utils.py:1880] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: P('norm',) Physical: (None,) I0423 13:02:04.220528 139991267690304 maxtext_utils.py:1880] params/params/decoder/layers/mlp/wi_0/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0423 13:02:04.220600 139991267690304 maxtext_utils.py:1880] params/params/decoder/layers/mlp/wi_1/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0423 13:02:04.220794 139991267690304 maxtext_utils.py:1880] params/params/decoder/layers/mlp/wo/kernel Shape: float32[7168,16,2048] Logical: P('mlp', 'layers', 'embed') Physical: (None, None, 'fsdp') I0423 13:02:04.221056 139991267690304 maxtext_utils.py:1880] params/params/decoder/layers/post_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0423 13:02:04.221119 139991267690304 maxtext_utils.py:1880] params/params/decoder/layers/pre_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0423 13:02:04.221193 139991267690304 maxtext_utils.py:1880] 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) I0423 13:02:04.221257 139991267690304 maxtext_utils.py:1880] 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') I0423 13:02:04.221307 139991267690304 maxtext_utils.py:1880] 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) I0423 13:02:04.221348 139991267690304 maxtext_utils.py:1880] 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) I0423 13:02:04.221403 139991267690304 maxtext_utils.py:1880] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: P('embed_vocab', 'vocab') Physical: ('fsdp', None) I0423 13:02:04.221469 139991267690304 maxtext_utils.py:1880] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: P('vocab', 'embed_vocab') Physical: (None, 'fsdp') I0423 13:02:04.247766 139991267690304 gradient_accumulation.py:70] gradient_accumulation/inputs Logical: float32[2048]............................................... Unknown. I0423 13:02:04.247834 139991267690304 gradient_accumulation.py:70] gradient_accumulation/inputs Physical: float32[2048]............................................... (None,). I0423 13:02:04.263459 139991267690304 gradient_accumulation.py:70] gradient_accumulation/inputs Logical: float32[2048,16,7168]....................................... Unknown. I0423 13:02:04.263519 139991267690304 gradient_accumulation.py:70] gradient_accumulation/inputs Physical: float32[2048,16,7168]....................................... ('fsdp', None, None). I0423 13:02:04.293753 139991267690304 gradient_accumulation.py:70] gradient_accumulation/inputs Logical: float32[7168,16,2048]....................................... Unknown. I0423 13:02:04.293814 139991267690304 gradient_accumulation.py:70] gradient_accumulation/inputs Physical: float32[7168,16,2048]....................................... (None, None, 'fsdp'). I0423 13:02:04.308774 139991267690304 gradient_accumulation.py:70] gradient_accumulation/inputs Logical: float32[2048,16]............................................ Unknown. I0423 13:02:04.308830 139991267690304 gradient_accumulation.py:70] gradient_accumulation/inputs Physical: float32[2048,16]............................................ (None, None). I0423 13:02:04.339013 139991267690304 gradient_accumulation.py:70] gradient_accumulation/inputs Logical: float32[2048,16,16,128]..................................... Unknown. I0423 13:02:04.339081 139991267690304 gradient_accumulation.py:70] gradient_accumulation/inputs Physical: float32[2048,16,16,128]..................................... ('fsdp', None, None, None). I0423 13:02:04.353963 139991267690304 gradient_accumulation.py:70] gradient_accumulation/inputs Logical: float32[16,16,128,2048]..................................... Unknown. I0423 13:02:04.354020 139991267690304 gradient_accumulation.py:70] gradient_accumulation/inputs Physical: float32[16,16,128,2048]..................................... (None, None, None, 'fsdp'). I0423 13:02:04.398633 139991267690304 gradient_accumulation.py:70] gradient_accumulation/inputs Logical: float32[2048,32000]......................................... Unknown. I0423 13:02:04.398712 139991267690304 gradient_accumulation.py:70] gradient_accumulation/inputs Physical: float32[2048,32000]......................................... ('fsdp', None). I0423 13:02:04.413584 139991267690304 gradient_accumulation.py:70] gradient_accumulation/inputs Logical: float32[32000,2048]......................................... Unknown. I0423 13:02:04.413644 139991267690304 gradient_accumulation.py:70] gradient_accumulation/inputs Physical: float32[32000,2048]......................................... (None, 'fsdp'). I0423 13:02:05.089867 139991267690304 train.py:157] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0423 13:02:05.089965 139991267690304 train.py:157] train/xent Physical: float32[32,2048]............................................ ('fsdp', None). I0423 13:02:05.111493 139991267690304 train.py:164] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0423 13:02:05.111593 139991267690304 train.py:164] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None). I0423 13:02:16.596912 139991267690304 max_utils.py:791] Total memory size: 1.7 GB, Output size: 0.4 GB, Temp size: 1.3 GB, Argument size: 0.4 GB, Host temp size: 0.0 GB. I0423 13:02:16.597730 139991267690304 metric_logger.py:301] number parameters: 1.104 billion I0423 13:02:18.511055 139991267690304 checkpointing.py:794] Waiting for step 0 to finish before checkpoint... I0423 13:02:29.257757 139991267690304 checkpointing.py:798] Waited 10.746683120727539 seconds for step 0 to finish before starting checkpointing. I0423 13:02:29.260136 139991267690304 checkpoint_manager.py:2009] [process=5][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0423 13:02:29.261833 139991267690304 checkpoint_manager.py:1512] [process=5] Saving checkpoint at step 0 I0423 13:02:29.263250 139991267690304 event_tracking.py:70] [process=5] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260423_124541/linen_xpk_feat_nnx_post_train_fixes_20260423_124541_06_grad_accum/checkpoints/0. I0423 13:02:30.028389 139991267690304 signaling_client.py:364] Using JaxDistributedSignalingClient I0423 13:02:30.029350 139991267690304 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0423 13:02:30.029404 139991267690304 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 I0423 13:02:30.301907 139991267690304 base_pytree_checkpoint_handler.py:154] [process=5][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.273590s I0423 13:02:30.302071 139991267690304 base_pytree_checkpoint_handler.py:130] [process=5] /jax/orbax/write/blocking_gbytes_per_sec: 5.527 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.279099702835083 s) (per-host) I0423 13:02:30.302123 139991267690304 base_pytree_checkpoint_handler.py:768] [process=5][thread=MainThread] Initiated Pytree async_save. Time taken: 0.279163s (batch_requests_ready=0.002151s, total_serialization_initiated=0.276942s, others=0.000070s) I0423 13:02:30.302220 139991267690304 composite_checkpoint_handler.py:715] [process=5][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.283208s (all_items=0.000018s, per_item={'items': '0.00001788'}, temp_paths=0.283190) I0423 13:02:30.302985 139991267690304 event_tracking.py:125] [process=5] [async] Finished blocking save in 1.04 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260423_124541/linen_xpk_feat_nnx_post_train_fixes_20260423_124541_06_grad_accum/checkpoints/0. I0423 13:02:30.303310 139862637360896 async_checkpointer.py:76] [process=5][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-23 13:22:30.303273 I0423 13:02:30.314283 139991267690304 checkpoint_manager.py:1560] [process=5][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize I0423 13:02:30.314583 139860988532480 async_checkpointer.py:280] [process=5][thread=save_finalize] Waiting for background save thread=async_save. I0423 13:02:30.314761 139991267690304 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260423_124541/linen_xpk_feat_nnx_post_train_fixes_20260423_124541_06_grad_accum/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776949349.2601173, 'wait_for_prev_duration_secs': 6.723403930664062e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776949349.2618732, 'checkpointer_blocking_duration_secs': 1.0415849685668945, 'get_old_steps_start_time': 1776949350.3034832, 'get_old_steps_duration_secs': 2.956390380859375e-05, 'checkpoint_manager_blocking_start_time': 1776949349.2582767, 'checkpoint_manager_blocking_duration_secs': 1.0564460754394531} I0423 13:02:30.314866 139991267690304 checkpointing.py:409] Started an asynchronous checkpoint save for step 0 I0423 13:02:30.314931 139991267690304 max_utils.py:750] Memstats: After params initialized: I0423 13:02:30.314979 139991267690304 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_18(process=5,(2,4,0,0)) I0423 13:02:30.315011 139991267690304 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_19(process=5,(3,4,0,0)) I0423 13:02:30.315037 139991267690304 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_22(process=5,(2,5,0,0)) I0423 13:02:30.315060 139991267690304 max_utils.py:756] Using (GB) 0.43 / 31.25 (1.376000%) on TPU_23(process=5,(3,5,0,0)) I0423 13:02:30.626817 139991267690304 metric_logger.py:196] completed step: 0, seconds: 1.913, TFLOP/s/device: 28.407, Tokens/s/device: 4281.811, total_weights: 262144, loss: 10.877, lm_loss: 10.877, perplexity: 52959.059 I0423 13:02:31.238284 139991267690304 metric_logger.py:196] completed step: 1, seconds: 12.114, TFLOP/s/device: 4.486, Tokens/s/device: 676.226, total_weights: 262144, loss: 10.877, lm_loss: 10.877, perplexity: 52959.059 I0423 13:02:31.816597 139991267690304 metric_logger.py:196] completed step: 2, seconds: 0.033, TFLOP/s/device: 1634.788, Tokens/s/device: 246412.994, total_weights: 262144, loss: 10.563, lm_loss: 10.563, perplexity: 38662.707 I0423 13:02:32.394687 139991267690304 metric_logger.py:196] completed step: 3, seconds: 0.584, TFLOP/s/device: 93.031, Tokens/s/device: 14022.667, total_weights: 262144, loss: 10.272, lm_loss: 10.272, perplexity: 28909.668 I0423 13:02:33.269145 2591 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com I0423 13:02:33.551435 139991267690304 metric_logger.py:196] completed step: 4, seconds: 0.578, TFLOP/s/device: 93.967, Tokens/s/device: 14163.797, total_weights: 262144, loss: 10.022, lm_loss: 10.022, perplexity: 22524.992 I0423 13:02:33.558231 139991267690304 metric_logger.py:196] completed step: 5, seconds: 0.578, TFLOP/s/device: 94.001, Tokens/s/device: 14168.843, total_weights: 262144, loss: 9.820, lm_loss: 9.820, perplexity: 18401.865 I0423 13:02:35.071279 139862065485568 array_metadata_store.py:203] [process=5][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260423_124541/linen_xpk_feat_nnx_post_train_fixes_20260423_124541_06_grad_accum/checkpoints/0/items/array_metadatas/process_5 I0423 13:03:07.528915 139862637360896 base_pytree_checkpoint_handler.py:130] [process=5] /jax/orbax/write/gbytes_per_sec: 42.116 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 37.5059072971344 s) (per-host) I0423 13:03:07.529040 139862637360896 async_checkpointer.py:90] [process=5][thread=async_save] 3 Handler Commit operations completed. Time taken: 37.225620s. I0423 13:03:16.563959 139862637360896 async_checkpointer.py:160] [process=5][thread=async_save] Background save thread done. Time taken: 46.260515s. I0423 13:03:16.564243 139860988532480 async_checkpointer.py:288] [process=5][thread=save_finalize] Done with waiting for background save thread=async_save. I0423 13:03:16.564369 139860988532480 async_checkpointer.py:298] [process=5][thread=save_finalize] No errors found in background save thread=async_save. I0423 13:03:16.564421 139860988532480 checkpoint_manager.py:2137] [process=5][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts... I0423 13:03:16.691155 139860988532480 checkpoint_manager.py:2146] [process=5][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts. I0423 13:03:19.459640 139991267690304 metric_logger.py:196] completed step: 6, seconds: 1.158, TFLOP/s/device: 46.941, Tokens/s/device: 7075.457, total_weights: 262144, loss: 9.667, lm_loss: 9.667, perplexity: 15787.604 I0423 13:03:20.037724 139991267690304 metric_logger.py:196] completed step: 7, seconds: 45.324, TFLOP/s/device: 1.199, Tokens/s/device: 180.744, total_weights: 262144, loss: 9.561, lm_loss: 9.561, perplexity: 14203.827 I0423 13:03:20.615994 139991267690304 metric_logger.py:196] completed step: 8, seconds: 0.584, TFLOP/s/device: 93.112, Tokens/s/device: 14034.871, total_weights: 262144, loss: 9.496, lm_loss: 9.496, perplexity: 13302.920 I0423 13:03:21.193672 139991267690304 checkpointing.py:794] Waiting for step 9 to finish before checkpoint... I0423 13:03:21.194436 139991267690304 checkpointing.py:798] Waited 0.0007979869842529297 seconds for step 9 to finish before starting checkpointing. I0423 13:03:21.196693 139991267690304 checkpoint_manager.py:2009] [process=5][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0423 13:03:21.198257 139991267690304 checkpoint_manager.py:1512] [process=5] Saving checkpoint at step 9 I0423 13:03:21.199719 139991267690304 event_tracking.py:70] [process=5] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260423_124541/linen_xpk_feat_nnx_post_train_fixes_20260423_124541_06_grad_accum/checkpoints/9. I0423 13:03:21.930452 139991267690304 jax_array_handlers.py:360] Scheduling D2H of 39 prioritized jax.Array. I0423 13:03:21.930543 139991267690304 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 I0423 13:03:21.965509 139991267690304 base_pytree_checkpoint_handler.py:154] [process=5][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.035937s I0423 13:03:21.965690 139991267690304 base_pytree_checkpoint_handler.py:130] [process=5] /jax/orbax/write/blocking_gbytes_per_sec: 39.174 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.039377450942993164 s) (per-host) I0423 13:03:21.965743 139991267690304 base_pytree_checkpoint_handler.py:768] [process=5][thread=MainThread] Initiated Pytree async_save. Time taken: 0.039441s (batch_requests_ready=0.001631s, total_serialization_initiated=0.037738s, others=0.000072s) I0423 13:03:21.965834 139991267690304 composite_checkpoint_handler.py:715] [process=5][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.043536s (all_items=0.000016s, per_item={'items': '0.00001621'}, temp_paths=0.043519) I0423 13:03:21.966504 139991267690304 event_tracking.py:125] [process=5] [async] Finished blocking save in 0.77 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260423_124541/linen_xpk_feat_nnx_post_train_fixes_20260423_124541_06_grad_accum/checkpoints/9. I0423 13:03:21.966857 139862120331008 async_checkpointer.py:76] [process=5][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-23 13:23:21.966815 I0423 13:03:21.970782 139991267690304 checkpoint_manager.py:1560] [process=5][thread=MainThread][step=9] Starting CheckpointManager Save Finalize thread=save_finalize I0423 13:03:21.971024 139862078367488 async_checkpointer.py:280] [process=5][thread=save_finalize] Waiting for background save thread=async_save. I0423 13:03:21.971148 139991267690304 standard_logger.py:34] {'step': 9, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260423_124541/linen_xpk_feat_nnx_post_train_fixes_20260423_124541_06_grad_accum/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776949401.1966395, 'wait_for_prev_duration_secs': 0.00010013580322265625, 'time_between_consecutive_saves_sec': 4.50538969039917, 'checkpointer_blocking_start_time': 1776949401.198296, 'checkpointer_blocking_duration_secs': 0.7687146663665771, 'get_old_steps_start_time': 1776949401.9670327, 'get_old_steps_duration_secs': 3.0040740966796875e-05, 'checkpoint_manager_blocking_start_time': 1776949401.1947443, 'checkpoint_manager_blocking_duration_secs': 0.776369571685791} I0423 13:03:21.971248 139991267690304 checkpointing.py:409] Started an asynchronous checkpoint save for step 9 I0423 13:03:21.971300 139991267690304 checkpoint_manager.py:2020] [process=5][thread=MainThread][step=9][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0423 13:03:28.055095 139862603671296 array_metadata_store.py:203] [process=5][thread=array_type_handler] Wrote 39 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260423_124541/linen_xpk_feat_nnx_post_train_fixes_20260423_124541_06_grad_accum/checkpoints/9/items/array_metadatas/process_5 I0423 13:04:04.267977 139862120331008 base_pytree_checkpoint_handler.py:130] [process=5] /jax/orbax/write/gbytes_per_sec: 37.306 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 42.34162640571594 s) (per-host) I0423 13:04:04.268106 139862120331008 async_checkpointer.py:90] [process=5][thread=async_save] 3 Handler Commit operations completed. Time taken: 42.301131s. I0423 13:04:12.336756 139862120331008 async_checkpointer.py:160] [process=5][thread=async_save] Background save thread done. Time taken: 50.369766s. I0423 13:04:12.337152 139862078367488 async_checkpointer.py:288] [process=5][thread=save_finalize] Done with waiting for background save thread=async_save. I0423 13:04:12.337284 139862078367488 async_checkpointer.py:298] [process=5][thread=save_finalize] No errors found in background save thread=async_save. I0423 13:04:12.337331 139862078367488 checkpoint_manager.py:2137] [process=5][thread=save_finalize][step=9] CheckpointManager Save Finalize is syncing with other hosts... I0423 13:04:12.338816 139862078367488 checkpoint_manager.py:2146] [process=5][thread=save_finalize][step=9] CheckpointManager Save Finalize is done on all hosts. I0423 13:04:12.339020 139991267690304 checkpoint_manager.py:2032] [process=5][thread=MainThread][step=9][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=9. I0423 13:04:12.339169 139991267690304 checkpoint_manager.py:2009] [process=5][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0423 13:04:12.340122 139991267690304 metric_logger.py:196] completed step: 9, seconds: 0.578, TFLOP/s/device: 94.022, Tokens/s/device: 14172.005, total_weights: 262144, loss: 9.457, lm_loss: 9.457, perplexity: 12802.546 Per train step: Total TFLOPs: 54.35 split as 93.93% learnable weight flops and 6.07% attention flops XPK End: Thu Apr 23 13:04:24 UTC 2026 EXIT_CODE=0