main (51c7f2b8e) vs feat/nnx-trainstate-and-training-loop (7c12861a4)| Metric | main 51c7f2b8e | feat/nnx-trainstate-and-training-loop 7c12861a4 | Diff (feat/nnx-trainstate-and-training-loop − main) |
|---|---|---|---|
| Parameters | 1.105 billion | 1.105 billion | — |
| Final loss | 10.8440 | 10.8510 | +0.007 |
| TFLOP/s | 2.942 | 123.386 | +120.444 |
| Tok/s | 443.5 | 18598.1 | +18154.591 |
| Avg s/step | — | 2.428 | — |
| Memory % | 5.15 | 5.15 | 0 |
| JAX | 0.9.2 | 0.9.2 | — |
Diff = branch value − main value. Green = branch improved. Red = branch regressed.
2026-04-15 21:22:12.864609: 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) I0415 21:22:12.982217 140311636520064 max_utils.py:238] Skipping jax distributed system due to skip_jax_distributed_system=True flag. I0415 21:23:04.649687 140311636520064 max_utils.py:800] System Information: Jax Version: 0.9.2 I0415 21:23:04.649809 140311636520064 max_utils.py:801] System Information: Jaxlib Version: 0.9.2 I0415 21:23:04.649846 140311636520064 max_utils.py:802] System Information: Jax Backend: PJRT C API TFRT TPU v6 lite Built on Apr 6 2026 20:48:10 (1775533690) cl/895581894 I0415 21:23:04.649870 140311636520064 train_utils.py:334] WARNING: 'dataset_path' might be pointing your local file system I0415 21:23:04.649950 140311636520064 train.py:683] [DECOUPLED NO-OP] skipping cloud diagnostics wrapper. W0415 21:23:04.743341 3236686 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0415 21:23:05.142014 140311636520064 maxtext_utils.py:1517] Num_devices: 8, shape (1, 1, 1, 8, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0415 21:23:05.142354 140311636520064 checkpointing.py:677] Setting up checkpoint logger... I0415 21:23:05.142402 140311636520064 checkpointing.py:233] Creating checkpoint manager with ocdbt=True and zarr3=True I0415 21:23:05.142463 140311636520064 pytree_checkpoint_handler.py:577] save_device_host_concurrent_bytes=None I0415 21:23:05.142978 140311636520064 base_pytree_checkpoint_handler.py:411] 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 0x7f9c42500bf0>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0415 21:23:07.596238 140311636520064 checkpointing.py:265] Enabling policy for fixed interval checkpointing. I0415 21:23:07.596463 140311636520064 checkpoint_manager.py:702] [process=0][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=('items',), item_handlers={'items': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7f95c6716810>}, handler_registry=None I0415 21:23:07.596796 140311636520064 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7f95c6716810>` 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`. I0415 21:23:07.596839 140311636520064 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7f95c6340b90>` 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`. I0415 21:23:07.596868 140311636520064 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 0x7f95c6716810>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7f95c6716810>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7f95c6340b90>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7f95c6340b90>}). I0415 21:23:07.597193 140311636520064 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.28 I0415 21:23:07.597252 140311636520064 async_checkpointer.py:177] [process=0][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>.<lambda> at 0x7f95c6338fe0> timeout: 600 secs and primary_host=0 for async checkpoint writes I0415 21:23:07.725083 140311636520064 checkpoint_manager.py:1788] Found 0 checkpoint steps in gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_05_fp8/checkpoints I0415 21:23:07.725334 140311636520064 checkpoint_manager.py:921] [process=0][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_hns=False, 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), root_directory=gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_05_fp8/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7f95c6340290> I0415 21:23:07.725427 140311636520064 checkpointing.py:301] Checkpoint manager created! I0415 21:23:07.804394 140311636520064 dataset_info.py:707] Load dataset info from tests/assets/local_datasets/c4_en_dataset_minimal/c4/en/3.1.0 I0415 21:23:07.807415 140311636520064 reader.py:262] Creating a tf.data.Dataset reading 8 files located in folders: tests/assets/local_datasets/c4_en_dataset_minimal/c4/en/3.1.0. I0415 21:23:07.861209 140311636520064 logging_logger.py:49] Constructing tf.data.Dataset __local_c4_builder for split train, from tests/assets/local_datasets/c4_en_dataset_minimal/c4/en/3.1.0 I0415 21:23:07.888165 140311636520064 tokenizer.py:245] Tokenizer path: src/maxtext/assets/tokenizers/tokenizer.llama2 I0415 21:23:07.888232 140311636520064 tokenizer.py:187] Loading sentencepiece tokenizer: src/maxtext/assets/tokenizers/tokenizer.llama2 I0415 21:23:08.994372 140311636520064 nnx_wrappers.py:437] Unknown Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0415 21:23:08.994486 140311636520064 nnx_wrappers.py:437] Unknown Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0415 21:23:09.098851 140311636520064 attentions.py:1088] attentions/inputs_q Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0415 21:23:09.098935 140311636520064 attentions.py:1088] attentions/inputs_q Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0415 21:23:09.114009 140311636520064 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0415 21:23:09.114067 140311636520064 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0415 21:23:09.160683 140311636520064 attentions.py:1154] attentions/query Logical: bfloat16[8,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0415 21:23:09.160766 140311636520064 attentions.py:1154] attentions/query Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0415 21:23:09.175873 140311636520064 attentions.py:1155] attentions/key Logical: bfloat16[8,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0415 21:23:09.175928 140311636520064 attentions.py:1155] attentions/key Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0415 21:23:09.191189 140311636520064 attentions.py:1156] attentions/value Logical: bfloat16[8,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0415 21:23:09.191239 140311636520064 attentions.py:1156] attentions/value Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0415 21:23:09.213714 140311636520064 attentions.py:1197] attentions/out Logical: bfloat16[8,2048,16,128]..................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv'). I0415 21:23:09.213785 140311636520064 attentions.py:1197] attentions/out Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0415 21:23:09.259338 140311636520064 linears.py:525] linears/x Logical: bfloat16[8,2048,7168]....................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0415 21:23:09.259407 140311636520064 linears.py:525] linears/x Physical: bfloat16[8,2048,7168]....................................... ('fsdp', None, None). I0415 21:23:09.639064 140311636520064 checkpointing.py:577] checkpoint manager exists so trying to load this run's existing checkpoint I0415 21:23:09.639168 140311636520064 checkpointing.py:665] No existing checkpoints found, not restoring checkpoint. W0415 21:23:09.707757 3236686 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. [DECOUPLED NO-OP] gcs_storage: using stubs. [DECOUPLED NO-OP] mldiagnostics: using stub. [DECOUPLED NO-OP] mldiagnostics: using stub. [DECOUPLED NO-OP] mldiagnostics: using stub. [DECOUPLED NO-OP] workload_monitor: using stub. [DECOUPLED NO-OP] vertex_tensorboard: using stub. fsdp: 8 I0415 21:23:10.486254 140311636520064 maxtext_utils.py:1620] params/_overwrite_with_gradient/decoder/layers/mlp/wi_0/Fp8DirectDotGeneralOp_0/input_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0415 21:23:10.486362 140311636520064 maxtext_utils.py:1620] params/_overwrite_with_gradient/decoder/layers/mlp/wi_0/Fp8DirectDotGeneralOp_0/input_scale Shape: float32[16,1] Logical: P() Physical: () I0415 21:23:10.486399 140311636520064 maxtext_utils.py:1620] params/_overwrite_with_gradient/decoder/layers/mlp/wi_0/Fp8DirectDotGeneralOp_0/kernel_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0415 21:23:10.486429 140311636520064 maxtext_utils.py:1620] params/_overwrite_with_gradient/decoder/layers/mlp/wi_0/Fp8DirectDotGeneralOp_0/kernel_scale Shape: float32[16,1] Logical: P() Physical: () I0415 21:23:10.486454 140311636520064 maxtext_utils.py:1620] params/_overwrite_with_gradient/decoder/layers/mlp/wi_0/Fp8DirectDotGeneralOp_0/output_grad_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0415 21:23:10.486479 140311636520064 maxtext_utils.py:1620] params/_overwrite_with_gradient/decoder/layers/mlp/wi_0/Fp8DirectDotGeneralOp_0/output_grad_scale Shape: float32[16,1] Logical: P() Physical: () I0415 21:23:10.486502 140311636520064 maxtext_utils.py:1620] params/_overwrite_with_gradient/decoder/layers/mlp/wi_1/Fp8DirectDotGeneralOp_0/input_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0415 21:23:10.486524 140311636520064 maxtext_utils.py:1620] params/_overwrite_with_gradient/decoder/layers/mlp/wi_1/Fp8DirectDotGeneralOp_0/input_scale Shape: float32[16,1] Logical: P() Physical: () I0415 21:23:10.486544 140311636520064 maxtext_utils.py:1620] params/_overwrite_with_gradient/decoder/layers/mlp/wi_1/Fp8DirectDotGeneralOp_0/kernel_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0415 21:23:10.486566 140311636520064 maxtext_utils.py:1620] params/_overwrite_with_gradient/decoder/layers/mlp/wi_1/Fp8DirectDotGeneralOp_0/kernel_scale Shape: float32[16,1] Logical: P() Physical: () I0415 21:23:10.486587 140311636520064 maxtext_utils.py:1620] params/_overwrite_with_gradient/decoder/layers/mlp/wi_1/Fp8DirectDotGeneralOp_0/output_grad_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0415 21:23:10.486607 140311636520064 maxtext_utils.py:1620] params/_overwrite_with_gradient/decoder/layers/mlp/wi_1/Fp8DirectDotGeneralOp_0/output_grad_scale Shape: float32[16,1] Logical: P() Physical: () I0415 21:23:10.486628 140311636520064 maxtext_utils.py:1620] params/_overwrite_with_gradient/decoder/layers/mlp/wo/Fp8DirectDotGeneralOp_0/input_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0415 21:23:10.486650 140311636520064 maxtext_utils.py:1620] params/_overwrite_with_gradient/decoder/layers/mlp/wo/Fp8DirectDotGeneralOp_0/input_scale Shape: float32[16,1] Logical: P() Physical: () I0415 21:23:10.486674 140311636520064 maxtext_utils.py:1620] params/_overwrite_with_gradient/decoder/layers/mlp/wo/Fp8DirectDotGeneralOp_0/kernel_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0415 21:23:10.486694 140311636520064 maxtext_utils.py:1620] params/_overwrite_with_gradient/decoder/layers/mlp/wo/Fp8DirectDotGeneralOp_0/kernel_scale Shape: float32[16,1] Logical: P() Physical: () I0415 21:23:10.486714 140311636520064 maxtext_utils.py:1620] params/_overwrite_with_gradient/decoder/layers/mlp/wo/Fp8DirectDotGeneralOp_0/output_grad_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0415 21:23:10.486734 140311636520064 maxtext_utils.py:1620] params/_overwrite_with_gradient/decoder/layers/mlp/wo/Fp8DirectDotGeneralOp_0/output_grad_scale Shape: float32[16,1] Logical: P() Physical: () I0415 21:23:10.486753 140311636520064 maxtext_utils.py:1620] params/_overwrite_with_gradient/decoder/layers/self_attention/key/Fp8DirectDotGeneralOp_0/input_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0415 21:23:10.486773 140311636520064 maxtext_utils.py:1620] params/_overwrite_with_gradient/decoder/layers/self_attention/key/Fp8DirectDotGeneralOp_0/input_scale Shape: float32[16,1] Logical: P() Physical: () I0415 21:23:10.486792 140311636520064 maxtext_utils.py:1620] params/_overwrite_with_gradient/decoder/layers/self_attention/key/Fp8DirectDotGeneralOp_0/kernel_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0415 21:23:10.486811 140311636520064 maxtext_utils.py:1620] params/_overwrite_with_gradient/decoder/layers/self_attention/key/Fp8DirectDotGeneralOp_0/kernel_scale Shape: float32[16,1] Logical: P() Physical: () I0415 21:23:10.486831 140311636520064 maxtext_utils.py:1620] params/_overwrite_with_gradient/decoder/layers/self_attention/key/Fp8DirectDotGeneralOp_0/output_grad_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0415 21:23:10.486851 140311636520064 maxtext_utils.py:1620] params/_overwrite_with_gradient/decoder/layers/self_attention/key/Fp8DirectDotGeneralOp_0/output_grad_scale Shape: float32[16,1] Logical: P() Physical: () I0415 21:23:10.486871 140311636520064 maxtext_utils.py:1620] params/_overwrite_with_gradient/decoder/layers/self_attention/out/Fp8DirectDotGeneralOp_0/input_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0415 21:23:10.486890 140311636520064 maxtext_utils.py:1620] params/_overwrite_with_gradient/decoder/layers/self_attention/out/Fp8DirectDotGeneralOp_0/input_scale Shape: float32[16,1] Logical: P() Physical: () I0415 21:23:10.486910 140311636520064 maxtext_utils.py:1620] params/_overwrite_with_gradient/decoder/layers/self_attention/out/Fp8DirectDotGeneralOp_0/kernel_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0415 21:23:10.486929 140311636520064 maxtext_utils.py:1620] params/_overwrite_with_gradient/decoder/layers/self_attention/out/Fp8DirectDotGeneralOp_0/kernel_scale Shape: float32[16,1] Logical: P() Physical: () I0415 21:23:10.486948 140311636520064 maxtext_utils.py:1620] params/_overwrite_with_gradient/decoder/layers/self_attention/out/Fp8DirectDotGeneralOp_0/output_grad_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0415 21:23:10.486967 140311636520064 maxtext_utils.py:1620] params/_overwrite_with_gradient/decoder/layers/self_attention/out/Fp8DirectDotGeneralOp_0/output_grad_scale Shape: float32[16,1] Logical: P() Physical: () I0415 21:23:10.486998 140311636520064 maxtext_utils.py:1620] params/_overwrite_with_gradient/decoder/layers/self_attention/query/Fp8DirectDotGeneralOp_0/input_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0415 21:23:10.487018 140311636520064 maxtext_utils.py:1620] params/_overwrite_with_gradient/decoder/layers/self_attention/query/Fp8DirectDotGeneralOp_0/input_scale Shape: float32[16,1] Logical: P() Physical: () I0415 21:23:10.487038 140311636520064 maxtext_utils.py:1620] params/_overwrite_with_gradient/decoder/layers/self_attention/query/Fp8DirectDotGeneralOp_0/kernel_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0415 21:23:10.487073 140311636520064 maxtext_utils.py:1620] params/_overwrite_with_gradient/decoder/layers/self_attention/query/Fp8DirectDotGeneralOp_0/kernel_scale Shape: float32[16,1] Logical: P() Physical: () I0415 21:23:10.487093 140311636520064 maxtext_utils.py:1620] params/_overwrite_with_gradient/decoder/layers/self_attention/query/Fp8DirectDotGeneralOp_0/output_grad_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0415 21:23:10.487113 140311636520064 maxtext_utils.py:1620] params/_overwrite_with_gradient/decoder/layers/self_attention/query/Fp8DirectDotGeneralOp_0/output_grad_scale Shape: float32[16,1] Logical: P() Physical: () I0415 21:23:10.487133 140311636520064 maxtext_utils.py:1620] params/_overwrite_with_gradient/decoder/layers/self_attention/value/Fp8DirectDotGeneralOp_0/input_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0415 21:23:10.487153 140311636520064 maxtext_utils.py:1620] params/_overwrite_with_gradient/decoder/layers/self_attention/value/Fp8DirectDotGeneralOp_0/input_scale Shape: float32[16,1] Logical: P() Physical: () I0415 21:23:10.487172 140311636520064 maxtext_utils.py:1620] params/_overwrite_with_gradient/decoder/layers/self_attention/value/Fp8DirectDotGeneralOp_0/kernel_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0415 21:23:10.487192 140311636520064 maxtext_utils.py:1620] params/_overwrite_with_gradient/decoder/layers/self_attention/value/Fp8DirectDotGeneralOp_0/kernel_scale Shape: float32[16,1] Logical: P() Physical: () I0415 21:23:10.487211 140311636520064 maxtext_utils.py:1620] params/_overwrite_with_gradient/decoder/layers/self_attention/value/Fp8DirectDotGeneralOp_0/output_grad_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0415 21:23:10.487230 140311636520064 maxtext_utils.py:1620] params/_overwrite_with_gradient/decoder/layers/self_attention/value/Fp8DirectDotGeneralOp_0/output_grad_scale Shape: float32[16,1] Logical: P() Physical: () I0415 21:23:10.487269 140311636520064 maxtext_utils.py:1620] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: P('norm',) Physical: (None,) I0415 21:23:10.487310 140311636520064 maxtext_utils.py:1620] params/params/decoder/layers/mlp/wi_0/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0415 21:23:10.487336 140311636520064 maxtext_utils.py:1620] params/params/decoder/layers/mlp/wi_1/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0415 21:23:10.487369 140311636520064 maxtext_utils.py:1620] params/params/decoder/layers/mlp/wo/kernel Shape: float32[7168,16,2048] Logical: P('mlp', 'layers', 'embed') Physical: (None, None, 'fsdp') I0415 21:23:10.487402 140311636520064 maxtext_utils.py:1620] params/params/decoder/layers/post_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0415 21:23:10.487426 140311636520064 maxtext_utils.py:1620] params/params/decoder/layers/pre_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0415 21:23:10.487461 140311636520064 maxtext_utils.py:1620] 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) I0415 21:23:10.487496 140311636520064 maxtext_utils.py:1620] 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') I0415 21:23:10.487521 140311636520064 maxtext_utils.py:1620] 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) I0415 21:23:10.487544 140311636520064 maxtext_utils.py:1620] 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) I0415 21:23:10.487577 140311636520064 maxtext_utils.py:1620] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: P('embed_vocab', 'vocab') Physical: ('fsdp', None) I0415 21:23:10.487609 140311636520064 maxtext_utils.py:1620] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: P('vocab', 'embed_vocab') Physical: (None, 'fsdp') I0415 21:23:11.942757 140311636520064 train.py:155] train/xent Logical: float32[8,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length'). I0415 21:23:11.942915 140311636520064 train.py:155] train/xent Physical: float32[8,2048]............................................. ('fsdp', None). I0415 21:23:11.957688 140311636520064 train.py:162] train/z_loss Logical: float32[8,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length'). I0415 21:23:11.957744 140311636520064 train.py:162] train/z_loss Physical: float32[8,2048]............................................. ('fsdp', None). W0415 21:23:12.400703 3236686 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0415 21:23:12.579716 140311636520064 max_utils.py:791] Total memory size: 3.5 GB, Output size: 1.5 GB, Temp size: 1.9 GB, Argument size: 1.5 GB, Host temp size: 0.0 GB. I0415 21:23:12.580368 140311636520064 max_utils.py:194] tensorboardX not available; using no-op SummaryWriter. I0415 21:23:12.580603 140311636520064 metric_logger.py:289] number parameters: 1.105 billion W0415 21:23:16.996983 3236686 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0415 21:23:17.199283 140311636520064 checkpointing.py:772] Waiting for step 0 to finish before checkpoint... I0415 21:23:17.312263 140311636520064 checkpointing.py:776] Waited 0.1129615306854248 seconds for step 0 to finish before starting checkpointing. I0415 21:23:17.312764 140311636520064 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0415 21:23:17.312918 140311636520064 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 0 I0415 21:23:17.313335 140311636520064 async_checkpointer.py:452] [process=0] Started async saving checkpoint to gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_05_fp8/checkpoints/0. I0415 21:23:17.401493 140311636520064 signaling_client.py:373] Using ThreadSafeKeyValueSignalingClient I0415 21:23:17.484201 140198388172352 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_05_fp8/checkpoints/0 I0415 21:23:17.508609 140311636520064 jax_array_handlers.py:347] Scheduling D2H of 81 prioritized jax.Array. I0415 21:23:17.508707 140311636520064 replica_slices.py:410] 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 I0415 21:23:18.173768 140198377686592 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_05_fp8/checkpoints/0/items W0415 21:23:18.216832 3236686 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0415 21:23:18.229578 3236686 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0415 21:23:18.238190 3236686 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0415 21:23:18.248386 3236686 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0415 21:23:18.260215 3236686 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0415 21:23:18.271386 3236686 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0415 21:23:18.277601 3236686 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0415 21:23:18.281847 3236686 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0415 21:23:18.390270 140198337840704 checkpoint.py:188] Wrote Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776288198078562506, 'commit_timestamp_nsecs': None, 'custom_metadata': {}}, json={"item_handlers": null, "metrics": {}, "performance_metrics": {}, "init_timestamp_nsecs": 1776288198078562506, "commit_timestamp_nsecs": null, "custom_metadata": {}} to gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_05_fp8/checkpoints/0/_CHECKPOINT_METADATA I0415 21:23:18.736185 3239351 google_auth_provider.cc:149] Using credentials at ~/.config/gcloud/application_default_credentials.json I0415 21:23:18.736248 3239351 google_auth_provider.cc:156] Using OAuth2 AuthProvider I0415 21:23:18.921861 140311636520064 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 1.413757s I0415 21:23:18.930513 140311636520064 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 8.081 GiB/s (total gbytes: 12.3 GiB) (time elapsed: a second) (per-host) I0415 21:23:18.930585 140311636520064 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 1.527348s (batch_requests_ready=0.098444s, total_serialization_initiated=1.420430s, others=0.008473s) I0415 21:23:18.930676 140311636520064 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 1.528642s (all_items=0.000020s, per_item={'items': '0.00002027'}, temp_paths=1.528621) I0415 21:23:18.932117 140198253954624 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started. I0415 21:23:18.932211 140311636520064 async_checkpointer.py:561] Finished blocking save. Time taken: 1.619253s. Continuing background save to gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_05_fp8/checkpoints/0. I0415 21:23:18.932435 140311636520064 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize I0415 21:23:18.932663 140198398658112 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0415 21:23:18.932782 140311636520064 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_05_fp8/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776288197.3127515, 'wait_for_prev_duration_secs': 4.100799560546875e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776288197.3129387, 'checkpointer_blocking_duration_secs': 1.6193909645080566, 'get_old_steps_start_time': 1776288198.9323606, 'get_old_steps_duration_secs': 3.361701965332031e-05, 'checkpoint_manager_blocking_start_time': 1776288197.312663, 'checkpoint_manager_blocking_duration_secs': 1.6200878620147705} I0415 21:23:18.932919 140311636520064 checkpointing.py:408] Started an asynchronous checkpoint save for step 0 I0415 21:23:18.932968 140311636520064 max_utils.py:750] Memstats: After params initialized: I0415 21:23:18.933024 140311636520064 max_utils.py:756] Using (GB) 1.61 / 31.25 (5.152000%) on TPU_0(process=0,(0,0,0,0)) I0415 21:23:18.933062 140311636520064 max_utils.py:756] Using (GB) 1.61 / 31.25 (5.152000%) on TPU_1(process=0,(1,0,0,0)) I0415 21:23:18.933085 140311636520064 max_utils.py:756] Using (GB) 1.61 / 31.25 (5.152000%) on TPU_2(process=0,(0,1,0,0)) I0415 21:23:18.933104 140311636520064 max_utils.py:756] Using (GB) 1.61 / 31.25 (5.152000%) on TPU_3(process=0,(1,1,0,0)) I0415 21:23:18.933122 140311636520064 max_utils.py:756] Using (GB) 1.61 / 31.25 (5.152000%) on TPU_4(process=0,(0,2,0,0)) I0415 21:23:18.933140 140311636520064 max_utils.py:756] Using (GB) 1.61 / 31.25 (5.152000%) on TPU_5(process=0,(1,2,0,0)) I0415 21:23:18.933156 140311636520064 max_utils.py:756] Using (GB) 1.61 / 31.25 (5.152000%) on TPU_6(process=0,(0,3,0,0)) I0415 21:23:18.933172 140311636520064 max_utils.py:756] Using (GB) 1.61 / 31.25 (5.152000%) on TPU_7(process=0,(1,3,0,0)) W0415 21:23:18.940265 3236686 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0415 21:23:18.946471 3236686 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0415 21:23:18.951534 3236686 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0415 21:23:19.274335 140311636520064 metric_logger.py:185] completed step: 0, seconds: 4.618, TFLOP/s/device: 2.942, Tokens/s/device: 443.471, total_weights: 13328, loss: 10.844 I0415 21:23:19.275230 140311636520064 metric_logger.py:269] To see full metrics 'tensorboard --logdir=gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_05_fp8/tensorboard/' I0415 21:23:19.382153 140311636520064 metric_logger.py:185] completed step: 1, seconds: 2.072, TFLOP/s/device: 6.557, Tokens/s/device: 988.341, total_weights: 12332, loss: nan I0415 21:23:19.383253 140311636520064 metric_logger.py:225] Aborting training due to NaN loss. I0415 21:23:19.383364 140311636520064 metric_logger.py:185] completed step: 1, seconds: 2.072, TFLOP/s/device: 6.557, Tokens/s/device: 988.341, total_weights: 12332, loss: nan I0415 21:23:19.383466 140311636520064 metric_logger.py:225] Aborting training due to NaN loss. E0415 21:23:19.570272 140198285411904 future.py:315] [process=0][thread=array_type_handler][operation_id=1] _SignalingThread.run() raised an exception: cannot schedule new futures after shutdown Traceback (most recent call last): File "~/maxtext_venv/lib/python3.12/site-packages/orbax/checkpoint/_src/futures/future.py", line 312, in run super().run() File "~/.local/share/uv/python/cpython-3.12.12-linux-x86_64-gnu/lib/python3.12/threading.py", line 1012, in run self._target(*self._args, **self._kwargs) File "~/maxtext_venv/lib/python3.12/site-packages/orbax/checkpoint/_src/futures/future.py", line 257, in _target_setting_result self._result = target() ^^^^^^^^ File "~/maxtext_venv/lib/python3.12/site-packages/orbax/checkpoint/_src/futures/future.py", line 391, in <lambda> target=lambda: asyncio_utils.run_sync(coro), ^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "~/maxtext_venv/lib/python3.12/site-packages/orbax/checkpoint/_src/asyncio_utils.py", line 36, in run_sync return asyncio.run(coro) ^^^^^^^^^^^^^^^^^ File "~/.local/share/uv/python/cpython-3.12.12-linux-x86_64-gnu/lib/python3.12/asyncio/runners.py", line 195, in run return runner.run(main) ^^^^^^^^^^^^^^^^ File "~/.local/share/uv/python/cpython-3.12.12-linux-x86_64-gnu/lib/python3.12/asyncio/runners.py", line 118, in run return self._loop.run_until_complete(task) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "~/.local/share/uv/python/cpython-3.12.12-linux-x86_64-gnu/lib/python3.12/asyncio/base_events.py", line 691, in run_until_complete return future.result() ^^^^^^^^^^^^^^^ File "~/maxtext_venv/lib/python3.12/site-packages/orbax/checkpoint/_src/serialization/jax_array_handlers.py", line 363, in _serialize_without_dispatcher await async_serialize_replica_slices_batch( File "~/maxtext_venv/lib/python3.12/site-packages/orbax/checkpoint/_src/serialization/jax_array_handlers.py", line 608, in _async_serialize_replica_slices await asyncio.gather(*write_coros) File "~/maxtext_venv/lib/python3.12/site-packages/orbax/checkpoint/_src/metadata/array_metadata_store.py", line 195, in write if await async_path.exists(file_path): ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "~/maxtext_venv/lib/python3.12/site-packages/orbax/checkpoint/_src/path/async_path.py", line 67, in exists return await asyncio.to_thread(path.exists) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "~/.local/share/uv/python/cpython-3.12.12-linux-x86_64-gnu/lib/python3.12/asyncio/threads.py", line 25, in to_thread return await loop.run_in_executor(None, func_call) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "~/.local/share/uv/python/cpython-3.12.12-linux-x86_64-gnu/lib/python3.12/asyncio/base_events.py", line 867, in run_in_executor executor.submit(func, *args), loop=self) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "~/.local/share/uv/python/cpython-3.12.12-linux-x86_64-gnu/lib/python3.12/concurrent/futures/thread.py", line 171, in submit raise RuntimeError('cannot schedule new futures after shutdown') RuntimeError: cannot schedule new futures after shutdown E0415 21:23:19.573523 140198274926144 future.py:315] [process=0][thread=write_metadata_after_commits][operation_id=1] _SignalingThread.run() raised an exception: cannot schedule new futures after shutdown Traceback (most recent call last): File "~/maxtext_venv/lib/python3.12/site-packages/orbax/checkpoint/_src/futures/future.py", line 312, in run super().run() File "~/.local/share/uv/python/cpython-3.12.12-linux-x86_64-gnu/lib/python3.12/threading.py", line 1012, in run self._target(*self._args, **self._kwargs) File "~/maxtext_venv/lib/python3.12/site-packages/orbax/checkpoint/_src/futures/future.py", line 257, in _target_setting_result self._result = target() ^^^^^^^^ File "~/maxtext_venv/lib/python3.12/site-packages/orbax/checkpoint/_src/futures/future.py", line 391, in <lambda> target=lambda: asyncio_utils.run_sync(coro), ^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "~/maxtext_venv/lib/python3.12/site-packages/orbax/checkpoint/_src/asyncio_utils.py", line 36, in run_sync return asyncio.run(coro) ^^^^^^^^^^^^^^^^^ File "~/.local/share/uv/python/cpython-3.12.12-linux-x86_64-gnu/lib/python3.12/asyncio/runners.py", line 195, in run return runner.run(main) ^^^^^^^^^^^^^^^^ File "~/.local/share/uv/python/cpython-3.12.12-linux-x86_64-gnu/lib/python3.12/asyncio/runners.py", line 118, in run return self._loop.run_until_complete(task) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "~/.local/share/uv/python/cpython-3.12.12-linux-x86_64-gnu/lib/python3.12/asyncio/base_events.py", line 691, in run_until_complete return future.result() ^^^^^^^^^^^^^^^ File "~/maxtext_venv/lib/python3.12/site-packages/orbax/checkpoint/_src/handlers/base_pytree_checkpoint_handler.py", line 1192, in _write_metadata_after_commits await asyncio.to_thread(commit_future.result) File "~/.local/share/uv/python/cpython-3.12.12-linux-x86_64-gnu/lib/python3.12/asyncio/threads.py", line 25, in to_thread return await loop.run_in_executor(None, func_call) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "~/.local/share/uv/python/cpython-3.12.12-linux-x86_64-gnu/lib/python3.12/concurrent/futures/thread.py", line 59, in run result = self.fn(*self.args, **self.kwargs) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "~/maxtext_venv/lib/python3.12/site-packages/orbax/checkpoint/_src/futures/future.py", line 449, in result return self._f.result(timeout=timeout) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "~/maxtext_venv/lib/python3.12/site-packages/orbax/checkpoint/_src/futures/future.py", line 398, in result return self._t.result(timeout=timeout) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "~/maxtext_venv/lib/python3.12/site-packages/orbax/checkpoint/_src/futures/future.py", line 347, in result self.join(timeout=timeout) File "~/maxtext_venv/lib/python3.12/site-packages/orbax/checkpoint/_src/futures/future.py", line 344, in join raise self._exception File "~/maxtext_venv/lib/python3.12/site-packages/orbax/checkpoint/_src/futures/future.py", line 312, in run super().run() File "~/.local/share/uv/python/cpython-3.12.12-linux-x86_64-gnu/lib/python3.12/threading.py", line 1012, in run self._target(*self._args, **self._kwargs) File "~/maxtext_venv/lib/python3.12/site-packages/orbax/checkpoint/_src/futures/future.py", line 257, in _target_setting_result self._result = target() ^^^^^^^^ File "~/maxtext_venv/lib/python3.12/site-packages/orbax/checkpoint/_src/futures/future.py", line 391, in <lambda> target=lambda: asyncio_utils.run_sync(coro), ^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "~/maxtext_venv/lib/python3.12/site-packages/orbax/checkpoint/_src/asyncio_utils.py", line 36, in run_sync return asyncio.run(coro) ^^^^^^^^^^^^^^^^^ File "~/.local/share/uv/python/cpython-3.12.12-linux-x86_64-gnu/lib/python3.12/asyncio/runners.py", line 195, in run return runner.run(main) ^^^^^^^^^^^^^^^^ File "~/.local/share/uv/python/cpython-3.12.12-linux-x86_64-gnu/lib/python3.12/asyncio/runners.py", line 118, in run return self._loop.run_until_complete(task) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "~/.local/share/uv/python/cpython-3.12.12-linux-x86_64-gnu/lib/python3.12/asyncio/base_events.py", line 691, in run_until_complete return future.result() ^^^^^^^^^^^^^^^ File "~/maxtext_venv/lib/python3.12/site-packages/orbax/checkpoint/_src/serialization/jax_array_handlers.py", line 363, in _serialize_without_dispatcher await async_serialize_replica_slices_batch( File "~/maxtext_venv/lib/python3.12/site-packages/orbax/checkpoint/_src/serialization/jax_array_handlers.py", line 608, in _async_serialize_replica_slices await asyncio.gather(*write_coros) File "~/maxtext_venv/lib/python3.12/site-packages/orbax/checkpoint/_src/metadata/array_metadata_store.py", line 195, in write if await async_path.exists(file_path): ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "~/maxtext_venv/lib/python3.12/site-packages/orbax/checkpoint/_src/path/async_path.py", line 67, in exists return await asyncio.to_thread(path.exists) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "~/.local/share/uv/python/cpython-3.12.12-linux-x86_64-gnu/lib/python3.12/asyncio/threads.py", line 25, in to_thread return await loop.run_in_executor(None, func_call) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "~/.local/share/uv/python/cpython-3.12.12-linux-x86_64-gnu/lib/python3.12/asyncio/base_events.py", line 867, in run_in_executor executor.submit(func, *args), loop=self) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "~/.local/share/uv/python/cpython-3.12.12-linux-x86_64-gnu/lib/python3.12/concurrent/futures/thread.py", line 171, in submit raise RuntimeError('cannot schedule new futures after shutdown') RuntimeError: cannot schedule new futures after shutdown E0415 21:23:19.575382 140198253954624 async_checkpointer.py:229] [process=0] Failed to run 3 Handler Commit operations or the Commit callback in background save thread, directory: gs://wanglance-maxtext/linen_ckpt_main_20260415_211126/linen_main_20260415_211126_05_fp8/checkpoints/0 Traceback (most recent call last): File "~/maxtext_venv/lib/python3.12/site-packages/orbax/checkpoint/_src/checkpointers/async_checkpointer.py", line 215, in _thread_func _background_wait_for_commit_futures( File "~/maxtext_venv/lib/python3.12/site-packages/orbax/checkpoint/_src/checkpointers/async_checkpointer.py", line 88, in _background_wait_for_commit_futures commit_future.result() File "~/maxtext_venv/lib/python3.12/site-packages/orbax/checkpoint/_src/futures/future.py", line 196, in result f.result(timeout=time_remaining) File "~/maxtext_venv/lib/python3.12/site-packages/orbax/checkpoint/_src/futures/future.py", line 449, in result return self._f.result(timeout=timeout) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "~/maxtext_venv/lib/python3.12/site-packages/orbax/checkpoint/_src/futures/future.py", line 398, in result return self._t.result(timeout=timeout) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "~/maxtext_venv/lib/python3.12/site-packages/orbax/checkpoint/_src/futures/future.py", line 347, in result self.join(timeout=timeout) File "~/maxtext_venv/lib/python3.12/site-packages/orbax/checkpoint/_src/futures/future.py", line 344, in join raise self._exception File "~/maxtext_venv/lib/python3.12/site-packages/orbax/checkpoint/_src/futures/future.py", line 312, in run super().run() File "~/.local/share/uv/python/cpython-3.12.12-linux-x86_64-gnu/lib/python3.12/threading.py", line 1012, in run self._target(*self._args, **self._kwargs) File "~/maxtext_venv/lib/python3.12/site-packages/orbax/checkpoint/_src/futures/future.py", line 257, in _target_setting_result self._result = target() ^^^^^^^^ File "~/maxtext_venv/lib/python3.12/site-packages/orbax/checkpoint/_src/futures/future.py", line 391, in <lambda> target=lambda: asyncio_utils.run_sync(coro), ^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "~/maxtext_venv/lib/python3.12/site-packages/orbax/checkpoint/_src/asyncio_utils.py", line 36, in run_sync return asyncio.run(coro) ^^^^^^^^^^^^^^^^^ File "~/.local/share/uv/python/cpython-3.12.12-linux-x86_64-gnu/lib/python3.12/asyncio/runners.py", line 195, in run return runner.run(main) ^^^^^^^^^^^^^^^^ File "~/.local/share/uv/python/cpython-3.12.12-linux-x86_64-gnu/lib/python3.12/asyncio/runners.py", line 118, in run return self._loop.run_until_complete(task) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "~/.local/share/uv/python/cpython-3.12.12-linux-x86_64-gnu/lib/python3.12/asyncio/base_events.py", line 691, in run_until_complete return future.result() ^^^^^^^^^^^^^^^ File "~/maxtext_venv/lib/python3.12/site-packages/orbax/checkpoint/_src/handlers/base_pytree_checkpoint_handler.py", line 1192, in _write_metadata_after_commits await asyncio.to_thread(commit_future.result) File "~/.local/share/uv/python/cpython-3.12.12-linux-x86_64-gnu/lib/python3.12/asyncio/threads.py", line 25, in to_thread return await loop.run_in_executor(None, func_call) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "~/.local/share/uv/python/cpython-3.12.12-linux-x86_64-gnu/lib/python3.12/concurrent/futures/thread.py", line 59, in run result = self.fn(*self.args, **self.kwargs) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "~/maxtext_venv/lib/python3.12/site-packages/orbax/checkpoint/_src/futures/future.py", line 449, in result return self._f.result(timeout=timeout) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "~/maxtext_venv/lib/python3.12/site-packages/orbax/checkpoint/_src/futures/future.py", line 398, in result return self._t.result(timeout=timeout) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "~/maxtext_venv/lib/python3.12/site-packages/orbax/checkpoint/_src/futures/future.py", line 347, in result self.join(timeout=timeout) File "~/maxtext_venv/lib/python3.12/site-packages/orbax/checkpoint/_src/futures/future.py", line 344, in join raise self._exception File "~/maxtext_venv/lib/python3.12/site-packages/orbax/checkpoint/_src/futures/future.py", line 312, in run super().run() File "~/.local/share/uv/python/cpython-3.12.12-linux-x86_64-gnu/lib/python3.12/threading.py", line 1012, in run self._target(*self._args, **self._kwargs) File "~/maxtext_venv/lib/python3.12/site-packages/orbax/checkpoint/_src/futures/future.py", line 257, in _target_setting_result self._result = target() ^^^^^^^^ File "~/maxtext_venv/lib/python3.12/site-packages/orbax/checkpoint/_src/futures/future.py", line 391, in <lambda> target=lambda: asyncio_utils.run_sync(coro), ^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "~/maxtext_venv/lib/python3.12/site-packages/orbax/checkpoint/_src/asyncio_utils.py", line 36, in run_sync return asyncio.run(coro) ^^^^^^^^^^^^^^^^^ File "~/.local/share/uv/python/cpython-3.12.12-linux-x86_64-gnu/lib/python3.12/asyncio/runners.py", line 195, in run return runner.run(main) ^^^^^^^^^^^^^^^^ File "~/.local/share/uv/python/cpython-3.12.12-linux-x86_64-gnu/lib/python3.12/asyncio/runners.py", line 118, in run return self._loop.run_until_complete(task) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "~/.local/share/uv/python/cpython-3.12.12-linux-x86_64-gnu/lib/python3.12/asyncio/base_events.py", line 691, in run_until_complete return future.result() ^^^^^^^^^^^^^^^ File "~/maxtext_venv/lib/python3.12/site-packages/orbax/checkpoint/_src/serialization/jax_array_handlers.py", line 363, in _serialize_without_dispatcher await async_serialize_replica_slices_batch( File "~/maxtext_venv/lib/python3.12/site-packages/orbax/checkpoint/_src/serialization/jax_array_handlers.py", line 608, in _async_serialize_replica_slices await asyncio.gather(*write_coros) File "~/maxtext_venv/lib/python3.12/site-packages/orbax/checkpoint/_src/metadata/array_metadata_store.py", line 195, in write if await async_path.exists(file_path): ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "~/maxtext_venv/lib/python3.12/site-packages/orbax/checkpoint/_src/path/async_path.py", line 67, in exists return await asyncio.to_thread(path.exists) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "~/.local/share/uv/python/cpython-3.12.12-linux-x86_64-gnu/lib/python3.12/asyncio/threads.py", line 25, in to_thread return await loop.run_in_executor(None, func_call) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "~/.local/share/uv/python/cpython-3.12.12-linux-x86_64-gnu/lib/python3.12/asyncio/base_events.py", line 867, in run_in_executor executor.submit(func, *args), loop=self) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "~/.local/share/uv/python/cpython-3.12.12-linux-x86_64-gnu/lib/python3.12/concurrent/futures/thread.py", line 171, in submit raise RuntimeError('cannot schedule new futures after shutdown') RuntimeError: cannot schedule new futures after shutdown I0415 21:23:19.577101 140198398658112 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0415 21:23:19.577280 140198398658112 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts... I0415 21:23:19.577396 140198398658112 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts. Per train step: Total TFLOPs: 13.59 split as 93.93% learnable weight flops and 6.07% attention flops
7c12861a4
· feat_nnx_trainstate_and_training_loop_20260416_033532
· full log
2026-04-16 03:46:28.191474: 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) I0416 03:46:28.307011 126542852578432 max_utils.py:238] Skipping jax distributed system due to skip_jax_distributed_system=True flag. I0416 03:47:25.524389 126542852578432 max_utils.py:800] System Information: Jax Version: 0.9.2 I0416 03:47:25.524495 126542852578432 max_utils.py:801] System Information: Jaxlib Version: 0.9.2 I0416 03:47:25.524532 126542852578432 max_utils.py:802] System Information: Jax Backend: PJRT C API TFRT TPU v6 lite Built on Apr 6 2026 20:48:10 (1775533690) cl/895581894 I0416 03:47:25.524557 126542852578432 train_utils.py:364] WARNING: 'dataset_path' might be pointing your local file system I0416 03:47:25.524653 126542852578432 train.py:818] [DECOUPLED NO-OP] skipping cloud diagnostics wrapper. W0416 03:47:25.619949 3636812 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0416 03:47:26.031025 126542852578432 maxtext_utils.py:1687] Num_devices: 8, shape (1, 1, 1, 8, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0416 03:47:26.031213 126542852578432 maxtext_utils.py:1687] Num_devices: 8, shape (1, 1, 1, 8, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0416 03:47:26.031467 126542852578432 checkpointing.py:688] Setting up checkpoint logger... I0416 03:47:26.031509 126542852578432 checkpointing.py:234] Creating checkpoint manager with ocdbt=True and zarr3=True I0416 03:47:26.031552 126542852578432 pytree_checkpoint_handler.py:577] save_device_host_concurrent_bytes=None I0416 03:47:26.032195 126542852578432 base_pytree_checkpoint_handler.py:411] 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 0x731676df4da0>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0416 03:47:28.392606 126542852578432 checkpointing.py:266] Enabling policy for fixed interval checkpointing. I0416 03:47:28.392776 126542852578432 checkpoint_manager.py:702] [process=0][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=('items',), item_handlers={'items': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x730ffab3afc0>}, handler_registry=None I0416 03:47:28.393060 126542852578432 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x730ffab3afc0>` 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`. I0416 03:47:28.393103 126542852578432 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x730ffab3b290>` 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`. I0416 03:47:28.393131 126542852578432 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 0x730ffab3afc0>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x730ffab3afc0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x730ffab3b290>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x730ffab3b290>}). I0416 03:47:28.393514 126542852578432 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.28 I0416 03:47:28.393563 126542852578432 async_checkpointer.py:177] [process=0][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>.<lambda> at 0x730ffab460c0> timeout: 600 secs and primary_host=0 for async checkpoint writes I0416 03:47:28.534258 126542852578432 checkpoint_manager.py:1788] Found 0 checkpoint steps in gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_033532/linen_feat_nnx_trainstate_and_training_loop_20260416_033532_05_fp8/checkpoints I0416 03:47:28.534476 126542852578432 checkpoint_manager.py:921] [process=0][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_hns=False, 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), root_directory=gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_033532/linen_feat_nnx_trainstate_and_training_loop_20260416_033532_05_fp8/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x730ffbdd74a0> I0416 03:47:28.534553 126542852578432 checkpointing.py:302] Checkpoint manager created! I0416 03:47:28.615623 126542852578432 dataset_info.py:707] Load dataset info from tests/assets/local_datasets/c4_en_dataset_minimal/c4/en/3.1.0 I0416 03:47:28.618700 126542852578432 reader.py:262] Creating a tf.data.Dataset reading 8 files located in folders: tests/assets/local_datasets/c4_en_dataset_minimal/c4/en/3.1.0. I0416 03:47:28.673073 126542852578432 logging_logger.py:49] Constructing tf.data.Dataset __local_c4_builder for split train, from tests/assets/local_datasets/c4_en_dataset_minimal/c4/en/3.1.0 I0416 03:47:28.705080 126542852578432 tokenizer.py:245] Tokenizer path: src/maxtext/assets/tokenizers/tokenizer.llama2 I0416 03:47:28.705144 126542852578432 tokenizer.py:187] Loading sentencepiece tokenizer: src/maxtext/assets/tokenizers/tokenizer.llama2 I0416 03:47:29.835335 126542852578432 nnx_wrappers.py:437] Unknown Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0416 03:47:29.835431 126542852578432 nnx_wrappers.py:437] Unknown Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0416 03:47:29.938442 126542852578432 attentions.py:1088] attentions/inputs_q Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0416 03:47:29.938522 126542852578432 attentions.py:1088] attentions/inputs_q Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0416 03:47:29.953622 126542852578432 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[8,2048,2048]....................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0416 03:47:29.953670 126542852578432 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[8,2048,2048]....................................... ('fsdp', None, None). I0416 03:47:29.999855 126542852578432 attentions.py:1154] attentions/query Logical: bfloat16[8,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0416 03:47:29.999923 126542852578432 attentions.py:1154] attentions/query Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0416 03:47:30.015229 126542852578432 attentions.py:1155] attentions/key Logical: bfloat16[8,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0416 03:47:30.015279 126542852578432 attentions.py:1155] attentions/key Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0416 03:47:30.030728 126542852578432 attentions.py:1156] attentions/value Logical: bfloat16[8,2048,16,128]..................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0416 03:47:30.030778 126542852578432 attentions.py:1156] attentions/value Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0416 03:47:30.053191 126542852578432 attentions.py:1197] attentions/out Logical: bfloat16[8,2048,16,128]..................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv'). I0416 03:47:30.053248 126542852578432 attentions.py:1197] attentions/out Physical: bfloat16[8,2048,16,128]..................................... ('fsdp', None, None, None). I0416 03:47:30.098165 126542852578432 linears.py:525] linears/x Logical: bfloat16[8,2048,7168]....................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0416 03:47:30.098229 126542852578432 linears.py:525] linears/x Physical: bfloat16[8,2048,7168]....................................... ('fsdp', None, None). I0416 03:47:30.487575 126542852578432 checkpointing.py:578] checkpoint manager exists so trying to load this run's existing checkpoint I0416 03:47:30.487676 126542852578432 checkpointing.py:676] No existing checkpoints found, not restoring checkpoint. W0416 03:47:30.554336 3636812 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. [DECOUPLED NO-OP] gcs_storage: using stubs. [DECOUPLED NO-OP] mldiagnostics: using stub. [DECOUPLED NO-OP] mldiagnostics: using stub. [DECOUPLED NO-OP] mldiagnostics: using stub. [DECOUPLED NO-OP] workload_monitor: using stub. [DECOUPLED NO-OP] vertex_tensorboard: using stub. fsdp: 8 I0416 03:47:31.333349 126542852578432 maxtext_utils.py:1790] params/_overwrite_with_gradient/decoder/layers/mlp/wi_0/Fp8DirectDotGeneralOp_0/input_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0416 03:47:31.333451 126542852578432 maxtext_utils.py:1790] params/_overwrite_with_gradient/decoder/layers/mlp/wi_0/Fp8DirectDotGeneralOp_0/input_scale Shape: float32[16,1] Logical: P() Physical: () I0416 03:47:31.333489 126542852578432 maxtext_utils.py:1790] params/_overwrite_with_gradient/decoder/layers/mlp/wi_0/Fp8DirectDotGeneralOp_0/kernel_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0416 03:47:31.333518 126542852578432 maxtext_utils.py:1790] params/_overwrite_with_gradient/decoder/layers/mlp/wi_0/Fp8DirectDotGeneralOp_0/kernel_scale Shape: float32[16,1] Logical: P() Physical: () I0416 03:47:31.333544 126542852578432 maxtext_utils.py:1790] params/_overwrite_with_gradient/decoder/layers/mlp/wi_0/Fp8DirectDotGeneralOp_0/output_grad_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0416 03:47:31.333568 126542852578432 maxtext_utils.py:1790] params/_overwrite_with_gradient/decoder/layers/mlp/wi_0/Fp8DirectDotGeneralOp_0/output_grad_scale Shape: float32[16,1] Logical: P() Physical: () I0416 03:47:31.333590 126542852578432 maxtext_utils.py:1790] params/_overwrite_with_gradient/decoder/layers/mlp/wi_1/Fp8DirectDotGeneralOp_0/input_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0416 03:47:31.333610 126542852578432 maxtext_utils.py:1790] params/_overwrite_with_gradient/decoder/layers/mlp/wi_1/Fp8DirectDotGeneralOp_0/input_scale Shape: float32[16,1] Logical: P() Physical: () I0416 03:47:31.333631 126542852578432 maxtext_utils.py:1790] params/_overwrite_with_gradient/decoder/layers/mlp/wi_1/Fp8DirectDotGeneralOp_0/kernel_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0416 03:47:31.333652 126542852578432 maxtext_utils.py:1790] params/_overwrite_with_gradient/decoder/layers/mlp/wi_1/Fp8DirectDotGeneralOp_0/kernel_scale Shape: float32[16,1] Logical: P() Physical: () I0416 03:47:31.333673 126542852578432 maxtext_utils.py:1790] params/_overwrite_with_gradient/decoder/layers/mlp/wi_1/Fp8DirectDotGeneralOp_0/output_grad_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0416 03:47:31.333694 126542852578432 maxtext_utils.py:1790] params/_overwrite_with_gradient/decoder/layers/mlp/wi_1/Fp8DirectDotGeneralOp_0/output_grad_scale Shape: float32[16,1] Logical: P() Physical: () I0416 03:47:31.333714 126542852578432 maxtext_utils.py:1790] params/_overwrite_with_gradient/decoder/layers/mlp/wo/Fp8DirectDotGeneralOp_0/input_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0416 03:47:31.333736 126542852578432 maxtext_utils.py:1790] params/_overwrite_with_gradient/decoder/layers/mlp/wo/Fp8DirectDotGeneralOp_0/input_scale Shape: float32[16,1] Logical: P() Physical: () I0416 03:47:31.333756 126542852578432 maxtext_utils.py:1790] params/_overwrite_with_gradient/decoder/layers/mlp/wo/Fp8DirectDotGeneralOp_0/kernel_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0416 03:47:31.333775 126542852578432 maxtext_utils.py:1790] params/_overwrite_with_gradient/decoder/layers/mlp/wo/Fp8DirectDotGeneralOp_0/kernel_scale Shape: float32[16,1] Logical: P() Physical: () I0416 03:47:31.333795 126542852578432 maxtext_utils.py:1790] params/_overwrite_with_gradient/decoder/layers/mlp/wo/Fp8DirectDotGeneralOp_0/output_grad_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0416 03:47:31.333815 126542852578432 maxtext_utils.py:1790] params/_overwrite_with_gradient/decoder/layers/mlp/wo/Fp8DirectDotGeneralOp_0/output_grad_scale Shape: float32[16,1] Logical: P() Physical: () I0416 03:47:31.333836 126542852578432 maxtext_utils.py:1790] params/_overwrite_with_gradient/decoder/layers/self_attention/key/Fp8DirectDotGeneralOp_0/input_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0416 03:47:31.333856 126542852578432 maxtext_utils.py:1790] params/_overwrite_with_gradient/decoder/layers/self_attention/key/Fp8DirectDotGeneralOp_0/input_scale Shape: float32[16,1] Logical: P() Physical: () I0416 03:47:31.333875 126542852578432 maxtext_utils.py:1790] params/_overwrite_with_gradient/decoder/layers/self_attention/key/Fp8DirectDotGeneralOp_0/kernel_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0416 03:47:31.333895 126542852578432 maxtext_utils.py:1790] params/_overwrite_with_gradient/decoder/layers/self_attention/key/Fp8DirectDotGeneralOp_0/kernel_scale Shape: float32[16,1] Logical: P() Physical: () I0416 03:47:31.333914 126542852578432 maxtext_utils.py:1790] params/_overwrite_with_gradient/decoder/layers/self_attention/key/Fp8DirectDotGeneralOp_0/output_grad_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0416 03:47:31.333939 126542852578432 maxtext_utils.py:1790] params/_overwrite_with_gradient/decoder/layers/self_attention/key/Fp8DirectDotGeneralOp_0/output_grad_scale Shape: float32[16,1] Logical: P() Physical: () I0416 03:47:31.333959 126542852578432 maxtext_utils.py:1790] params/_overwrite_with_gradient/decoder/layers/self_attention/out/Fp8DirectDotGeneralOp_0/input_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0416 03:47:31.333978 126542852578432 maxtext_utils.py:1790] params/_overwrite_with_gradient/decoder/layers/self_attention/out/Fp8DirectDotGeneralOp_0/input_scale Shape: float32[16,1] Logical: P() Physical: () I0416 03:47:31.333997 126542852578432 maxtext_utils.py:1790] params/_overwrite_with_gradient/decoder/layers/self_attention/out/Fp8DirectDotGeneralOp_0/kernel_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0416 03:47:31.334017 126542852578432 maxtext_utils.py:1790] params/_overwrite_with_gradient/decoder/layers/self_attention/out/Fp8DirectDotGeneralOp_0/kernel_scale Shape: float32[16,1] Logical: P() Physical: () I0416 03:47:31.334036 126542852578432 maxtext_utils.py:1790] params/_overwrite_with_gradient/decoder/layers/self_attention/out/Fp8DirectDotGeneralOp_0/output_grad_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0416 03:47:31.334085 126542852578432 maxtext_utils.py:1790] params/_overwrite_with_gradient/decoder/layers/self_attention/out/Fp8DirectDotGeneralOp_0/output_grad_scale Shape: float32[16,1] Logical: P() Physical: () I0416 03:47:31.334108 126542852578432 maxtext_utils.py:1790] params/_overwrite_with_gradient/decoder/layers/self_attention/query/Fp8DirectDotGeneralOp_0/input_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0416 03:47:31.334128 126542852578432 maxtext_utils.py:1790] params/_overwrite_with_gradient/decoder/layers/self_attention/query/Fp8DirectDotGeneralOp_0/input_scale Shape: float32[16,1] Logical: P() Physical: () I0416 03:47:31.334147 126542852578432 maxtext_utils.py:1790] params/_overwrite_with_gradient/decoder/layers/self_attention/query/Fp8DirectDotGeneralOp_0/kernel_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0416 03:47:31.334166 126542852578432 maxtext_utils.py:1790] params/_overwrite_with_gradient/decoder/layers/self_attention/query/Fp8DirectDotGeneralOp_0/kernel_scale Shape: float32[16,1] Logical: P() Physical: () I0416 03:47:31.334185 126542852578432 maxtext_utils.py:1790] params/_overwrite_with_gradient/decoder/layers/self_attention/query/Fp8DirectDotGeneralOp_0/output_grad_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0416 03:47:31.334204 126542852578432 maxtext_utils.py:1790] params/_overwrite_with_gradient/decoder/layers/self_attention/query/Fp8DirectDotGeneralOp_0/output_grad_scale Shape: float32[16,1] Logical: P() Physical: () I0416 03:47:31.334223 126542852578432 maxtext_utils.py:1790] params/_overwrite_with_gradient/decoder/layers/self_attention/value/Fp8DirectDotGeneralOp_0/input_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0416 03:47:31.334242 126542852578432 maxtext_utils.py:1790] params/_overwrite_with_gradient/decoder/layers/self_attention/value/Fp8DirectDotGeneralOp_0/input_scale Shape: float32[16,1] Logical: P() Physical: () I0416 03:47:31.334260 126542852578432 maxtext_utils.py:1790] params/_overwrite_with_gradient/decoder/layers/self_attention/value/Fp8DirectDotGeneralOp_0/kernel_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0416 03:47:31.334279 126542852578432 maxtext_utils.py:1790] params/_overwrite_with_gradient/decoder/layers/self_attention/value/Fp8DirectDotGeneralOp_0/kernel_scale Shape: float32[16,1] Logical: P() Physical: () I0416 03:47:31.334297 126542852578432 maxtext_utils.py:1790] params/_overwrite_with_gradient/decoder/layers/self_attention/value/Fp8DirectDotGeneralOp_0/output_grad_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0416 03:47:31.334315 126542852578432 maxtext_utils.py:1790] params/_overwrite_with_gradient/decoder/layers/self_attention/value/Fp8DirectDotGeneralOp_0/output_grad_scale Shape: float32[16,1] Logical: P() Physical: () I0416 03:47:31.334353 126542852578432 maxtext_utils.py:1790] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: P('norm',) Physical: (None,) I0416 03:47:31.334393 126542852578432 maxtext_utils.py:1790] params/params/decoder/layers/mlp/wi_0/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0416 03:47:31.334419 126542852578432 maxtext_utils.py:1790] params/params/decoder/layers/mlp/wi_1/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0416 03:47:31.334451 126542852578432 maxtext_utils.py:1790] params/params/decoder/layers/mlp/wo/kernel Shape: float32[7168,16,2048] Logical: P('mlp', 'layers', 'embed') Physical: (None, None, 'fsdp') I0416 03:47:31.334482 126542852578432 maxtext_utils.py:1790] params/params/decoder/layers/post_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0416 03:47:31.334505 126542852578432 maxtext_utils.py:1790] params/params/decoder/layers/pre_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0416 03:47:31.334538 126542852578432 maxtext_utils.py:1790] 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) I0416 03:47:31.334572 126542852578432 maxtext_utils.py:1790] 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') I0416 03:47:31.334597 126542852578432 maxtext_utils.py:1790] 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) I0416 03:47:31.334620 126542852578432 maxtext_utils.py:1790] 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) I0416 03:47:31.334654 126542852578432 maxtext_utils.py:1790] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: P('embed_vocab', 'vocab') Physical: ('fsdp', None) I0416 03:47:31.334686 126542852578432 maxtext_utils.py:1790] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: P('vocab', 'embed_vocab') Physical: (None, 'fsdp') I0416 03:47:32.749259 126542852578432 train.py:157] train/xent Logical: float32[8,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length'). I0416 03:47:32.749341 126542852578432 train.py:157] train/xent Physical: float32[8,2048]............................................. ('fsdp', None). I0416 03:47:32.764059 126542852578432 train.py:164] train/z_loss Logical: float32[8,2048]............................................. ('activation_embed_and_logits_batch', 'activation_length'). I0416 03:47:32.764112 126542852578432 train.py:164] train/z_loss Physical: float32[8,2048]............................................. ('fsdp', None). W0416 03:47:33.241372 3636812 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0416 03:47:33.427488 126542852578432 max_utils.py:791] Total memory size: 3.5 GB, Output size: 1.5 GB, Temp size: 1.9 GB, Argument size: 1.5 GB, Host temp size: 0.0 GB. I0416 03:47:33.427952 126542852578432 max_utils.py:194] tensorboardX not available; using no-op SummaryWriter. I0416 03:47:33.428170 126542852578432 metric_logger.py:289] number parameters: 1.105 billion W0416 03:47:37.749865 3636812 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0416 03:47:37.969245 126542852578432 checkpointing.py:794] Waiting for step 0 to finish before checkpoint... I0416 03:47:38.082657 126542852578432 checkpointing.py:798] Waited 0.11339998245239258 seconds for step 0 to finish before starting checkpointing. I0416 03:47:38.083152 126542852578432 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0416 03:47:38.083307 126542852578432 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 0 I0416 03:47:38.083676 126542852578432 async_checkpointer.py:452] [process=0] Started async saving checkpoint to gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_033532/linen_feat_nnx_trainstate_and_training_loop_20260416_033532_05_fp8/checkpoints/0. I0416 03:47:38.188722 126542852578432 signaling_client.py:373] Using ThreadSafeKeyValueSignalingClient I0416 03:47:38.277940 126429513647680 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_033532/linen_feat_nnx_trainstate_and_training_loop_20260416_033532_05_fp8/checkpoints/0 I0416 03:47:38.290486 126542852578432 jax_array_handlers.py:347] Scheduling D2H of 81 prioritized jax.Array. I0416 03:47:38.290578 126542852578432 replica_slices.py:410] 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 W0416 03:47:38.970875 3636812 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0416 03:47:38.976515 126429503161920 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_033532/linen_feat_nnx_trainstate_and_training_loop_20260416_033532_05_fp8/checkpoints/0/items W0416 03:47:38.985863 3636812 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 03:47:38.996355 3636812 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 03:47:39.005020 3636812 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 03:47:39.018088 3636812 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 03:47:39.023472 3636812 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 03:47:39.027779 3636812 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 03:47:39.032237 3636812 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0416 03:47:39.140737 126429463316032 checkpoint.py:188] Wrote Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776311258897625378, 'commit_timestamp_nsecs': None, 'custom_metadata': {}}, json={"item_handlers": null, "metrics": {}, "performance_metrics": {}, "init_timestamp_nsecs": 1776311258897625378, "commit_timestamp_nsecs": null, "custom_metadata": {}} to gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_033532/linen_feat_nnx_trainstate_and_training_loop_20260416_033532_05_fp8/checkpoints/0/_CHECKPOINT_METADATA I0416 03:47:39.308321 126542852578432 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 1.018304s I0416 03:47:39.308747 126542852578432 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 11.029 GiB/s (total gbytes: 12.3 GiB) (time elapsed: a second) (per-host) I0416 03:47:39.308795 126542852578432 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 1.119133s (batch_requests_ready=0.094131s, total_serialization_initiated=1.024671s, others=0.000332s) I0416 03:47:39.308866 126542852578432 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 1.119667s (all_items=0.000018s, per_item={'items': '0.00001836'}, temp_paths=1.119648) I0416 03:47:39.309629 126429442344512 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started. I0416 03:47:39.309723 126542852578432 async_checkpointer.py:561] Finished blocking save. Time taken: 1.226376s. Continuing background save to gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_033532/linen_feat_nnx_trainstate_and_training_loop_20260416_033532_05_fp8/checkpoints/0. I0416 03:47:39.309907 126542852578432 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize I0416 03:47:39.310132 126429524133440 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0416 03:47:39.310234 126542852578432 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_033532/linen_feat_nnx_trainstate_and_training_loop_20260416_033532_05_fp8/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776311258.083138, 'wait_for_prev_duration_secs': 4.291534423828125e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776311258.083328, 'checkpointer_blocking_duration_secs': 1.2265033721923828, 'get_old_steps_start_time': 1776311259.3098485, 'get_old_steps_duration_secs': 2.002716064453125e-05, 'checkpoint_manager_blocking_start_time': 1776311258.083039, 'checkpoint_manager_blocking_duration_secs': 1.227168083190918} I0416 03:47:39.310351 126542852578432 checkpointing.py:409] Started an asynchronous checkpoint save for step 0 I0416 03:47:39.310388 126542852578432 max_utils.py:750] Memstats: After params initialized: I0416 03:47:39.310522 126542852578432 max_utils.py:756] Using (GB) 1.61 / 31.25 (5.152000%) on TPU_0(process=0,(0,0,0,0)) I0416 03:47:39.310549 126542852578432 max_utils.py:756] Using (GB) 1.61 / 31.25 (5.152000%) on TPU_1(process=0,(1,0,0,0)) I0416 03:47:39.310568 126542852578432 max_utils.py:756] Using (GB) 1.61 / 31.25 (5.152000%) on TPU_2(process=0,(0,1,0,0)) I0416 03:47:39.310585 126542852578432 max_utils.py:756] Using (GB) 1.61 / 31.25 (5.152000%) on TPU_3(process=0,(1,1,0,0)) I0416 03:47:39.310602 126542852578432 max_utils.py:756] Using (GB) 1.61 / 31.25 (5.152000%) on TPU_4(process=0,(0,2,0,0)) I0416 03:47:39.310619 126542852578432 max_utils.py:756] Using (GB) 1.61 / 31.25 (5.152000%) on TPU_5(process=0,(1,2,0,0)) I0416 03:47:39.310635 126542852578432 max_utils.py:756] Using (GB) 1.61 / 31.25 (5.152000%) on TPU_6(process=0,(0,3,0,0)) I0416 03:47:39.310654 126542852578432 max_utils.py:756] Using (GB) 1.61 / 31.25 (5.152000%) on TPU_7(process=0,(1,3,0,0)) W0416 03:47:39.315979 3636812 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 03:47:39.324565 3636812 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. W0416 03:47:39.329982 3636812 pjrt_executable.cc:642] Assume version compatibility. PjRt-IFRT does not track XLA executable versions. I0416 03:47:39.508159 3639587 google_auth_provider.cc:149] Using credentials at ~/.config/gcloud/application_default_credentials.json I0416 03:47:39.508208 3639587 google_auth_provider.cc:156] Using OAuth2 AuthProvider I0416 03:47:39.655443 126542852578432 metric_logger.py:185] completed step: 0, seconds: 4.541, TFLOP/s/device: 2.992, Tokens/s/device: 451.041, total_weights: 13328, loss: 10.844 I0416 03:47:39.657811 126542852578432 metric_logger.py:269] To see full metrics 'tensorboard --logdir=gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_033532/linen_feat_nnx_trainstate_and_training_loop_20260416_033532_05_fp8/tensorboard/' I0416 03:47:39.746478 126542852578432 metric_logger.py:185] completed step: 1, seconds: 1.667, TFLOP/s/device: 8.152, Tokens/s/device: 1228.800, total_weights: 12332, loss: 10.871 I0416 03:47:39.856865 126542852578432 metric_logger.py:185] completed step: 2, seconds: 0.041, TFLOP/s/device: 330.025, Tokens/s/device: 49744.960, total_weights: 15161, loss: 10.871 I0416 03:47:39.967283 126542852578432 metric_logger.py:185] completed step: 3, seconds: 0.078, TFLOP/s/device: 174.453, Tokens/s/device: 26295.516, total_weights: 13327, loss: 10.857 I0416 03:47:40.500729 126429492676160 array_metadata_store.py:203] [process=0][thread=array_type_handler] Wrote 81 array_metadata.ArrayMetadata to gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_033532/linen_feat_nnx_trainstate_and_training_loop_20260416_033532_05_fp8/checkpoints/0/items/array_metadatas/process_0 I0416 03:48:03.103390 126542852578432 metric_logger.py:185] completed step: 4, seconds: 0.110, TFLOP/s/device: 123.144, Tokens/s/device: 18561.653, total_weights: 11939, loss: 10.861 I0416 03:48:03.113475 126542852578432 metric_logger.py:185] completed step: 5, seconds: 0.110, TFLOP/s/device: 123.050, Tokens/s/device: 18547.365, total_weights: 15502, loss: 10.882 I0416 03:48:03.222440 126542852578432 metric_logger.py:185] completed step: 6, seconds: 23.136, TFLOP/s/device: 0.587, Tokens/s/device: 88.519, total_weights: 13864, loss: 10.886 I0416 03:48:03.332400 126542852578432 metric_logger.py:185] completed step: 7, seconds: 0.007, TFLOP/s/device: 2006.369, Tokens/s/device: 302421.737, total_weights: 12988, loss: 10.829 I0416 03:48:03.443065 126542852578432 metric_logger.py:185] completed step: 8, seconds: 0.110, TFLOP/s/device: 123.735, Tokens/s/device: 18650.736, total_weights: 13820, loss: 10.848 I0416 03:48:03.449662 126542852578432 checkpointing.py:794] Waiting for step 10 to finish before checkpoint... I0416 03:48:03.663694 126542852578432 checkpointing.py:798] Waited 0.21399855613708496 seconds for step 10 to finish before starting checkpointing. I0416 03:48:03.664197 126542852578432 checkpoint_manager.py:1994] [process=0][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0416 03:48:11.769731 126429452830272 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 32.268293s (commit=31.813762s, array_metadata_write=0.454531s) I0416 03:48:11.771060 126429442344512 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 376.343 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 33 seconds) (per-host) I0416 03:48:11.771169 126429442344512 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 32.461400s. I0416 03:48:12.009874 126429442344512 checkpoint.py:228] Read Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776311258897625378, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} from gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_033532/linen_feat_nnx_trainstate_and_training_loop_20260416_033532_05_fp8/checkpoints/0/_CHECKPOINT_METADATA I0416 03:48:12.195457 126429442344512 array_metadata_store.py:367] [process=0][thread=async_save] Skipped cross-host ArrayMetadata validation because only one process is found: process_index=0. I0416 03:48:12.403275 126429463316032 checkpoint.py:247] Updated Metadata={'item_handlers': {'items': 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler'}, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776311258897625378, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} to gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_033532/linen_feat_nnx_trainstate_and_training_loop_20260416_033532_05_fp8/checkpoints/0/_CHECKPOINT_METADATA I0416 03:48:12.596833 126429442344512 ocdbt_utils.py:56] Param validation support for Zarr3 will be added later (b/362328389). I0416 03:48:12.597439 126429442344512 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.549009s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_033532/linen_feat_nnx_trainstate_and_training_loop_20260416_033532_05_fp8/checkpoints/0/items I0416 03:48:12.598311 126429442344512 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_033532/linen_feat_nnx_trainstate_and_training_loop_20260416_033532_05_fp8/checkpoints/0/items I0416 03:48:12.840076 126429442344512 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_033532/linen_feat_nnx_trainstate_and_training_loop_20260416_033532_05_fp8/checkpoints/0 I0416 03:48:13.474450 126429442344512 atomicity.py:794] [process=0][thread=async_save] Finished saving checkpoint (finalized tmp dir) to `gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_033532/linen_feat_nnx_trainstate_and_training_loop_20260416_033532_05_fp8/checkpoints/0`. I0416 03:48:13.475086 126429442344512 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 35.391746s. directory=gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_033532/linen_feat_nnx_trainstate_and_training_loop_20260416_033532_05_fp8/checkpoints/0 I0416 03:48:13.475165 126429442344512 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 34.165399s. I0416 03:48:13.475344 126429524133440 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0416 03:48:13.475444 126429524133440 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0416 03:48:13.475503 126429524133440 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts... I0416 03:48:13.475550 126429524133440 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts. I0416 03:48:13.475698 126542852578432 checkpoint_manager.py:2006] [process=0][thread=MainThread][step=0][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=0. W0416 03:48:13.475820 126542852578432 checkpoint_manager.py:1441] Waiting for previous save to complete took 9.811646 seconds. If this number is high, consider checkpointing less frequently. I0416 03:48:13.476667 126542852578432 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 10 I0416 03:48:13.476953 126542852578432 async_checkpointer.py:452] [process=0] Started async saving checkpoint to gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_033532/linen_feat_nnx_trainstate_and_training_loop_20260416_033532_05_fp8/checkpoints/10. I0416 03:48:13.657335 126542852578432 jax_array_handlers.py:347] Scheduling D2H of 81 prioritized jax.Array. I0416 03:48:13.657428 126542852578432 replica_slices.py:410] 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 I0416 03:48:13.664459 126429524133440 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_033532/linen_feat_nnx_trainstate_and_training_loop_20260416_033532_05_fp8/checkpoints/10 I0416 03:48:14.354766 126429503161920 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_033532/linen_feat_nnx_trainstate_and_training_loop_20260416_033532_05_fp8/checkpoints/10/items I0416 03:48:23.299217 126542852578432 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 9.642346s I0416 03:48:23.306601 126542852578432 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 1.267 GiB/s (total gbytes: 12.3 GiB) (time elapsed: 9 seconds) (per-host) I0416 03:48:23.306661 126542852578432 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 9.742709s (batch_requests_ready=0.087090s, total_serialization_initiated=9.648337s, others=0.007282s) I0416 03:48:23.306746 126542852578432 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 9.743478s (all_items=0.000018s, per_item={'items': '0.00001836'}, temp_paths=9.743460) I0416 03:48:23.308255 126429392012864 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started. I0416 03:48:23.308387 126542852578432 async_checkpointer.py:561] Finished blocking save. Time taken: 9.831679s. Continuing background save to gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_033532/linen_feat_nnx_trainstate_and_training_loop_20260416_033532_05_fp8/checkpoints/10. I0416 03:48:23.308528 126542852578432 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=10] Starting CheckpointManager Save Finalize thread=save_finalize I0416 03:48:23.308688 126429442344512 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0416 03:48:23.308793 126542852578432 standard_logger.py:34] {'step': 10, 'event_type': 'save', 'directory': 'gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_033532/linen_feat_nnx_trainstate_and_training_loop_20260416_033532_05_fp8/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776311283.6641457, 'wait_for_prev_duration_secs': 9.811646223068237, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776311293.4766898, 'checkpointer_blocking_duration_secs': 9.831753730773926, 'get_old_steps_start_time': 1776311303.308463, 'get_old_steps_duration_secs': 2.6226043701171875e-05, 'checkpoint_manager_blocking_start_time': 1776311283.6640413, 'checkpoint_manager_blocking_duration_secs': 19.644725799560547} I0416 03:48:23.308919 126542852578432 checkpointing.py:409] Started an asynchronous checkpoint save for step 10 I0416 03:48:23.309349 126542852578432 metric_logger.py:185] completed step: 9, seconds: 0.110, TFLOP/s/device: 123.606, Tokens/s/device: 18631.224, total_weights: 12300, loss: 10.857 I0416 03:48:23.318424 126542852578432 metric_logger.py:185] completed step: 10, seconds: 0.111, TFLOP/s/device: 122.705, Tokens/s/device: 18495.439, total_weights: 13436, loss: 10.876 I0416 03:48:23.425957 126542852578432 metric_logger.py:185] completed step: 11, seconds: 19.866, TFLOP/s/device: 0.684, Tokens/s/device: 103.089, total_weights: 13817, loss: 10.853 I0416 03:48:23.536270 126542852578432 metric_logger.py:185] completed step: 12, seconds: 0.008, TFLOP/s/device: 1759.763, Tokens/s/device: 265250.615, total_weights: 12076, loss: 10.852 I0416 03:48:23.646682 126542852578432 metric_logger.py:185] completed step: 13, seconds: 0.109, TFLOP/s/device: 124.648, Tokens/s/device: 18788.301, total_weights: 11048, loss: 10.864 I0416 03:48:23.757079 126542852578432 metric_logger.py:185] completed step: 14, seconds: 0.110, TFLOP/s/device: 123.027, Tokens/s/device: 18544.006, total_weights: 14267, loss: 10.834 I0416 03:48:23.867175 126542852578432 metric_logger.py:185] completed step: 15, seconds: 0.110, TFLOP/s/device: 123.377, Tokens/s/device: 18596.711, total_weights: 13004, loss: 10.832 I0416 03:48:23.977501 126542852578432 metric_logger.py:185] completed step: 16, seconds: 0.110, TFLOP/s/device: 123.293, Tokens/s/device: 18584.055, total_weights: 13811, loss: 10.867 I0416 03:48:24.074164 126429492676160 array_metadata_store.py:203] [process=0][thread=array_type_handler] Wrote 81 array_metadata.ArrayMetadata to gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_033532/linen_feat_nnx_trainstate_and_training_loop_20260416_033532_05_fp8/checkpoints/10/items/array_metadatas/process_0 I0416 03:48:24.087802 126542852578432 metric_logger.py:185] completed step: 17, seconds: 0.111, TFLOP/s/device: 122.847, Tokens/s/device: 18516.844, total_weights: 13709, loss: 10.850 I0416 03:48:24.200528 126542852578432 metric_logger.py:185] completed step: 18, seconds: 0.110, TFLOP/s/device: 123.537, Tokens/s/device: 18620.890, total_weights: 12831, loss: 10.839 I0416 03:48:24.308479 126542852578432 checkpointing.py:794] Waiting for step 19 to finish before checkpoint... I0416 03:48:24.310021 126542852578432 checkpointing.py:798] Waited 0.0015511512756347656 seconds for step 19 to finish before starting checkpointing. I0416 03:48:24.310467 126542852578432 checkpoint_manager.py:1994] [process=0][thread=MainThread][step=10][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0416 03:48:42.485607 126429524133440 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 19.178197s (commit=18.729626s, array_metadata_write=0.448571s) I0416 03:48:42.486655 126429392012864 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 436.962 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 28 seconds) (per-host) I0416 03:48:42.486768 126429392012864 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 19.178417s. I0416 03:48:42.892085 126429392012864 array_metadata_store.py:367] [process=0][thread=async_save] Skipped cross-host ArrayMetadata validation because only one process is found: process_index=0. I0416 03:48:43.318730 126429392012864 ocdbt_utils.py:56] Param validation support for Zarr3 will be added later (b/362328389). I0416 03:48:43.319439 126429392012864 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.571825s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_033532/linen_feat_nnx_trainstate_and_training_loop_20260416_033532_05_fp8/checkpoints/10/items I0416 03:48:43.320099 126429392012864 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_033532/linen_feat_nnx_trainstate_and_training_loop_20260416_033532_05_fp8/checkpoints/10/items I0416 03:48:43.563069 126429392012864 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_033532/linen_feat_nnx_trainstate_and_training_loop_20260416_033532_05_fp8/checkpoints/10 I0416 03:48:44.260707 126429392012864 atomicity.py:794] [process=0][thread=async_save] Finished saving checkpoint (finalized tmp dir) to `gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_033532/linen_feat_nnx_trainstate_and_training_loop_20260416_033532_05_fp8/checkpoints/10`. I0416 03:48:44.261347 126429392012864 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 30.784644s. directory=gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_033532/linen_feat_nnx_trainstate_and_training_loop_20260416_033532_05_fp8/checkpoints/10 I0416 03:48:44.261415 126429392012864 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 20.953066s. I0416 03:48:44.261608 126429442344512 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0416 03:48:44.261715 126429442344512 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0416 03:48:44.261770 126429442344512 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=10] CheckpointManager Save Finalize is syncing with other hosts... I0416 03:48:44.261808 126429442344512 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=10] CheckpointManager Save Finalize is done on all hosts. I0416 03:48:44.262210 126542852578432 checkpoint_manager.py:2006] [process=0][thread=MainThread][step=10][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=10. W0416 03:48:44.262263 126542852578432 checkpoint_manager.py:1441] Waiting for previous save to complete took 19.951812 seconds. If this number is high, consider checkpointing less frequently. I0416 03:48:44.262442 126542852578432 checkpoint_manager.py:1501] [process=0] Saving checkpoint at step 19 I0416 03:48:44.262723 126542852578432 async_checkpointer.py:452] [process=0] Started async saving checkpoint to gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_033532/linen_feat_nnx_trainstate_and_training_loop_20260416_033532_05_fp8/checkpoints/19. I0416 03:48:44.446239 126542852578432 jax_array_handlers.py:347] Scheduling D2H of 81 prioritized jax.Array. I0416 03:48:44.446379 126429442344512 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_033532/linen_feat_nnx_trainstate_and_training_loop_20260416_033532_05_fp8/checkpoints/19 I0416 03:48:44.446466 126542852578432 replica_slices.py:410] 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 I0416 03:48:45.142603 126429524133440 atomicity.py:137] Creating tmp directory gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_033532/linen_feat_nnx_trainstate_and_training_loop_20260416_033532_05_fp8/checkpoints/19/items I0416 03:48:46.340098 126542852578432 base_pytree_checkpoint_handler.py:153] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 1.894596s I0416 03:48:46.340622 126542852578432 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/blocking_gbytes_per_sec: 6.210 GiB/s (total gbytes: 12.3 GiB) (time elapsed: a second) (per-host) I0416 03:48:46.340680 126542852578432 base_pytree_checkpoint_handler.py:732] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 1.987674s (batch_requests_ready=0.086858s, total_serialization_initiated=1.900379s, others=0.000436s) I0416 03:48:46.340751 126542852578432 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 1.988155s (all_items=0.000013s, per_item={'items': '0.00001264'}, temp_paths=1.988143) I0416 03:48:46.348390 126429381527104 async_checkpointer.py:79] [process=0][thread=async_save] Background save thread started. I0416 03:48:46.348473 126542852578432 async_checkpointer.py:561] Finished blocking save. Time taken: 2.085990s. Continuing background save to gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_033532/linen_feat_nnx_trainstate_and_training_loop_20260416_033532_05_fp8/checkpoints/19. I0416 03:48:46.348647 126542852578432 checkpoint_manager.py:1549] [process=0][thread=MainThread][step=19] Starting CheckpointManager Save Finalize thread=save_finalize I0416 03:48:46.348824 126429392012864 async_checkpointer.py:265] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0416 03:48:46.348935 126542852578432 standard_logger.py:34] {'step': 19, 'event_type': 'save', 'directory': 'gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_033532/linen_feat_nnx_trainstate_and_training_loop_20260416_033532_05_fp8/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776311304.3104403, 'wait_for_prev_duration_secs': 19.95181155204773, 'time_between_consecutive_saves_sec': 10.834875106811523, 'checkpointer_blocking_start_time': 1776311324.2624645, 'checkpointer_blocking_duration_secs': 2.086115837097168, 'get_old_steps_start_time': 1776311326.3485935, 'get_old_steps_duration_secs': 1.5735626220703125e-05, 'checkpoint_manager_blocking_start_time': 1776311304.310404, 'checkpoint_manager_blocking_duration_secs': 22.038506984710693} I0416 03:48:46.349063 126542852578432 checkpointing.py:409] Started an asynchronous checkpoint save for step 19 I0416 03:48:46.349094 126542852578432 checkpoint_manager.py:1994] [process=0][thread=MainThread][step=19][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0416 03:48:47.076305 126429247309376 array_metadata_store.py:203] [process=0][thread=array_type_handler] Wrote 81 array_metadata.ArrayMetadata to gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_033532/linen_feat_nnx_trainstate_and_training_loop_20260416_033532_05_fp8/checkpoints/19/items/array_metadatas/process_0 I0416 03:49:08.891889 126429442344512 base_pytree_checkpoint_handler.py:1217] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 22.544780s (commit=22.108700s, array_metadata_write=0.436080s) I0416 03:49:08.892838 126429381527104 base_pytree_checkpoint_handler.py:128] [process=0] /jax/checkpoint/write/gbytes_per_sec: 515.004 MiB/s (total gbytes: 12.3 GiB) (time elapsed: 24 seconds) (per-host) I0416 03:49:08.892902 126429381527104 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 22.544381s. I0416 03:49:09.300193 126429381527104 array_metadata_store.py:367] [process=0][thread=async_save] Skipped cross-host ArrayMetadata validation because only one process is found: process_index=0. I0416 03:49:09.671223 126429381527104 ocdbt_utils.py:56] Param validation support for Zarr3 will be added later (b/362328389). I0416 03:49:09.671930 126429381527104 base_pytree_checkpoint_handler.py:1342] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 0.514992s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_033532/linen_feat_nnx_trainstate_and_training_loop_20260416_033532_05_fp8/checkpoints/19/items I0416 03:49:09.672553 126429381527104 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_033532/linen_feat_nnx_trainstate_and_training_loop_20260416_033532_05_fp8/checkpoints/19/items I0416 03:49:09.915270 126429381527104 atomicity.py:608] Finalizing gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_033532/linen_feat_nnx_trainstate_and_training_loop_20260416_033532_05_fp8/checkpoints/19 I0416 03:49:10.574767 126429381527104 atomicity.py:794] [process=0][thread=async_save] Finished saving checkpoint (finalized tmp dir) to `gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_033532/linen_feat_nnx_trainstate_and_training_loop_20260416_033532_05_fp8/checkpoints/19`. I0416 03:49:10.575420 126429381527104 async_checkpointer.py:420] Finished async_save (blocking + background). Time taken: 26.312945s. directory=gs://wanglance-maxtext/linen_ckpt_feat_nnx_trainstate_and_training_loop_20260416_033532/linen_feat_nnx_trainstate_and_training_loop_20260416_033532_05_fp8/checkpoints/19 I0416 03:49:10.575481 126429381527104 async_checkpointer.py:144] [process=0][thread=async_save] Background save thread done. Time taken: 24.226965s. I0416 03:49:10.575628 126429392012864 async_checkpointer.py:273] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0416 03:49:10.575726 126429392012864 async_checkpointer.py:283] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0416 03:49:10.575768 126429392012864 checkpoint_manager.py:2103] [process=0][thread=save_finalize][step=19] CheckpointManager Save Finalize is syncing with other hosts... I0416 03:49:10.575804 126429392012864 checkpoint_manager.py:2112] [process=0][thread=save_finalize][step=19] CheckpointManager Save Finalize is done on all hosts. I0416 03:49:10.576360 126542852578432 checkpoint_manager.py:2006] [process=0][thread=MainThread][step=19][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=19. I0416 03:49:10.576490 126542852578432 checkpoint_manager.py:1983] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0416 03:49:10.577040 126542852578432 metric_logger.py:185] completed step: 19, seconds: 0.110, TFLOP/s/device: 123.386, Tokens/s/device: 18598.062, total_weights: 13717, loss: 10.851 Per train step: Total TFLOPs: 13.59 split as 93.93% learnable weight flops and 6.07% attention flops