main (574ad3fb9) vs feat/nnx-post-train-fixes (574ad3fb9)| Metric | main 574ad3fb9 | feat/nnx-post-train-fixes 574ad3fb9 | Diff (feat/nnx-post-train-fixes − main) |
|---|---|---|---|
| Parameters | 1.105 billion | 1.105 billion | — |
| Final loss | 6.1670 | 6.1670 | 0 |
| TFLOP/s | 85.808 | 85.736 | -0.072 |
| Tok/s | 12933.9 | 12923.0 | -10.855 |
| Avg s/step | 1.712 | 1.696 | -0.016 |
| Memory % | 1.44 | 1.44 | 0 |
| JAX | 0.9.2 | 0.9.2 | — |
Diff = branch value − main value. Green = branch improved. Red = branch regressed.
XPK Start: Sat Apr 18 18:15:50 UTC 2026 PyTorch was not found. Models won't be available and only tokenizers, configuration and file/data utilities can be used. 2026-04-18 18:16:15.220328: 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) I0418 18:16:15.401217 134511891793728 max_utils.py:273] Attempting to initialize the jax distributed system... I0418 18:16:24.443118 134511891793728 distributed.py:149] Starting JAX distributed service on [::]:8482 I0418 18:16:24.445453 134511891793728 distributed.py:172] Connecting to JAX distributed service on mt-05-fp8-k1sow-slice-job-0-0.mt-05-fp8-k1sow:8482 I0418 18:16:25.863977 134511891793728 max_utils.py:284] Jax distributed system initialized! I0418 18:16:31.954729 134511891793728 max_utils.py:800] System Information: Jax Version: 0.9.2 I0418 18:16:31.954830 134511891793728 max_utils.py:801] System Information: Jaxlib Version: 0.9.2 I0418 18:16:31.954873 134511891793728 max_utils.py:802] System Information: Jax Backend: PJRT C API TFRT TPU v6 lite Built on Mar 4 2026 11:32:08 (1772652728) cl/878335365 I0418 18:16:31.954908 134511891793728 train_utils.py:348] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing. I0418 18:16:32.656683 134511891793728 maxtext_utils.py:1551] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0418 18:16:32.656968 134511891793728 checkpointing.py:677] Setting up checkpoint logger... I0418 18:16:32.657022 134511891793728 checkpointing.py:233] Creating checkpoint manager with ocdbt=True and zarr3=True I0418 18:16:32.657068 134511891793728 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None I0418 18:16:32.657423 134511891793728 base_pytree_checkpoint_handler.py:441] Created BasePyTreeCheckpointHandler: use_ocdbt=True, use_zarr3=True, pytree_metadata_options=PyTreeMetadataOptions(support_rich_types=False), array_metadata_store=<orbax.checkpoint._src.metadata.array_metadata_store.Store object at 0x7a55e21181a0>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0418 18:16:36.039294 134511891793728 checkpointing.py:265] Enabling policy for fixed interval checkpointing. I0418 18:16:36.039529 134511891793728 checkpoint_manager.py:708] [process=7][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=('items',), item_handlers={'items': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7a41a011b3e0>}, handler_registry=None I0418 18:16:36.039763 134511891793728 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7a41a011b3e0>` 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`. I0418 18:16:36.039810 134511891793728 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7a41a011b500>` 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`. I0418 18:16:36.039845 134511891793728 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 0x7a41a011b3e0>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7a41a011b3e0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7a41a011b500>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7a41a011b500>}). I0418 18:16:36.040189 134511891793728 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34 I0418 18:16:36.040256 134511891793728 async_checkpointer.py:192] [process=7][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7a40a840bc40> timeout: 1200 secs and primary_host=0 for async checkpoint writes I0418 18:16:37.969893 134511891793728 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_main_20260418_180002/linen_xpk_main_20260418_180002_05_fp8/checkpoints I0418 18:16:38.806761 134511891793728 checkpoint_manager.py:929] [process=7][thread=MainThread] CheckpointManager created, primary_host=0, CheckpointManagerOptions=CheckpointManagerOptions(save_interval_steps=1, max_to_keep=None, keep_time_interval=None, keep_period=None, should_keep_fn=None, best_fn=None, best_mode='max', keep_checkpoints_without_metrics=True, step_prefix=None, step_format_fixed_length=None, step_name_format=None, create=True, cleanup_tmp_directories=False, save_on_steps=frozenset(), single_host_load_and_broadcast=False, todelete_subdir=None, todelete_full_path=None, enable_background_delete=False, read_only=False, enable_async_checkpointing=True, async_options=None, multiprocessing_options=MultiprocessingOptions(primary_host=0, active_processes=None, barrier_sync_key_prefix=None), should_save_fn=None, file_options=FileOptions(path_permission_mode=None), save_root_metadata=True, temporary_path_class=None, save_decision_policy=FixedIntervalPolicy(interval=10), preservation_policy=LatestN(n=None), prevent_write_metrics=False, enable_should_save_is_saving_in_progress_check=True, enable_per_process_directory_creation=False, lightweight_initialize=False), root_directory=gs://lance-maxtext/linen_ckpt_xpk_main_20260418_180002/linen_xpk_main_20260418_180002_05_fp8/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7a41a011bda0> I0418 18:16:38.806934 134511891793728 checkpointing.py:301] Checkpoint manager created! I0418 18:16:39.896669 134511891793728 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0418 18:16:39.896785 134511891793728 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0418 18:16:40.282787 134511891793728 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0418 18:16:40.282877 134511891793728 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0418 18:16:40.299135 134511891793728 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0418 18:16:40.299195 134511891793728 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0418 18:16:40.350816 134511891793728 attentions.py:1154] attentions/query Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0418 18:16:40.350898 134511891793728 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0418 18:16:40.367596 134511891793728 attentions.py:1155] attentions/key Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0418 18:16:40.367654 134511891793728 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0418 18:16:40.384371 134511891793728 attentions.py:1156] attentions/value Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0418 18:16:40.384431 134511891793728 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0418 18:16:40.409204 134511891793728 attentions.py:1197] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv'). I0418 18:16:40.409275 134511891793728 attentions.py:1197] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0418 18:16:40.460380 134511891793728 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0418 18:16:40.460458 134511891793728 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None). I0418 18:16:40.911127 134511891793728 checkpointing.py:577] checkpoint manager exists so trying to load this run's existing checkpoint I0418 18:16:40.911241 134511891793728 checkpointing.py:665] No existing checkpoints found, not restoring checkpoint. fsdp: 32 I0418 18:16:42.823501 134511891793728 maxtext_utils.py:1654] params/_overwrite_with_gradient/decoder/layers/mlp/wi_0/Fp8DirectDotGeneralOp_0/input_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0418 18:16:42.823621 134511891793728 maxtext_utils.py:1654] params/_overwrite_with_gradient/decoder/layers/mlp/wi_0/Fp8DirectDotGeneralOp_0/input_scale Shape: float32[16,1] Logical: P() Physical: () I0418 18:16:42.823669 134511891793728 maxtext_utils.py:1654] params/_overwrite_with_gradient/decoder/layers/mlp/wi_0/Fp8DirectDotGeneralOp_0/kernel_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0418 18:16:42.823708 134511891793728 maxtext_utils.py:1654] params/_overwrite_with_gradient/decoder/layers/mlp/wi_0/Fp8DirectDotGeneralOp_0/kernel_scale Shape: float32[16,1] Logical: P() Physical: () I0418 18:16:42.823744 134511891793728 maxtext_utils.py:1654] params/_overwrite_with_gradient/decoder/layers/mlp/wi_0/Fp8DirectDotGeneralOp_0/output_grad_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0418 18:16:42.823778 134511891793728 maxtext_utils.py:1654] params/_overwrite_with_gradient/decoder/layers/mlp/wi_0/Fp8DirectDotGeneralOp_0/output_grad_scale Shape: float32[16,1] Logical: P() Physical: () I0418 18:16:42.823812 134511891793728 maxtext_utils.py:1654] params/_overwrite_with_gradient/decoder/layers/mlp/wi_1/Fp8DirectDotGeneralOp_0/input_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0418 18:16:42.823842 134511891793728 maxtext_utils.py:1654] params/_overwrite_with_gradient/decoder/layers/mlp/wi_1/Fp8DirectDotGeneralOp_0/input_scale Shape: float32[16,1] Logical: P() Physical: () I0418 18:16:42.823872 134511891793728 maxtext_utils.py:1654] params/_overwrite_with_gradient/decoder/layers/mlp/wi_1/Fp8DirectDotGeneralOp_0/kernel_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0418 18:16:42.823902 134511891793728 maxtext_utils.py:1654] params/_overwrite_with_gradient/decoder/layers/mlp/wi_1/Fp8DirectDotGeneralOp_0/kernel_scale Shape: float32[16,1] Logical: P() Physical: () I0418 18:16:42.823931 134511891793728 maxtext_utils.py:1654] params/_overwrite_with_gradient/decoder/layers/mlp/wi_1/Fp8DirectDotGeneralOp_0/output_grad_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0418 18:16:42.823960 134511891793728 maxtext_utils.py:1654] params/_overwrite_with_gradient/decoder/layers/mlp/wi_1/Fp8DirectDotGeneralOp_0/output_grad_scale Shape: float32[16,1] Logical: P() Physical: () I0418 18:16:42.823989 134511891793728 maxtext_utils.py:1654] params/_overwrite_with_gradient/decoder/layers/mlp/wo/Fp8DirectDotGeneralOp_0/input_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0418 18:16:42.824016 134511891793728 maxtext_utils.py:1654] params/_overwrite_with_gradient/decoder/layers/mlp/wo/Fp8DirectDotGeneralOp_0/input_scale Shape: float32[16,1] Logical: P() Physical: () I0418 18:16:42.824044 134511891793728 maxtext_utils.py:1654] params/_overwrite_with_gradient/decoder/layers/mlp/wo/Fp8DirectDotGeneralOp_0/kernel_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0418 18:16:42.824088 134511891793728 maxtext_utils.py:1654] params/_overwrite_with_gradient/decoder/layers/mlp/wo/Fp8DirectDotGeneralOp_0/kernel_scale Shape: float32[16,1] Logical: P() Physical: () I0418 18:16:42.824123 134511891793728 maxtext_utils.py:1654] params/_overwrite_with_gradient/decoder/layers/mlp/wo/Fp8DirectDotGeneralOp_0/output_grad_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0418 18:16:42.824171 134511891793728 maxtext_utils.py:1654] params/_overwrite_with_gradient/decoder/layers/mlp/wo/Fp8DirectDotGeneralOp_0/output_grad_scale Shape: float32[16,1] Logical: P() Physical: () I0418 18:16:42.824203 134511891793728 maxtext_utils.py:1654] params/_overwrite_with_gradient/decoder/layers/self_attention/key/Fp8DirectDotGeneralOp_0/input_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0418 18:16:42.824231 134511891793728 maxtext_utils.py:1654] params/_overwrite_with_gradient/decoder/layers/self_attention/key/Fp8DirectDotGeneralOp_0/input_scale Shape: float32[16,1] Logical: P() Physical: () I0418 18:16:42.824260 134511891793728 maxtext_utils.py:1654] params/_overwrite_with_gradient/decoder/layers/self_attention/key/Fp8DirectDotGeneralOp_0/kernel_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0418 18:16:42.824287 134511891793728 maxtext_utils.py:1654] params/_overwrite_with_gradient/decoder/layers/self_attention/key/Fp8DirectDotGeneralOp_0/kernel_scale Shape: float32[16,1] Logical: P() Physical: () I0418 18:16:42.824315 134511891793728 maxtext_utils.py:1654] params/_overwrite_with_gradient/decoder/layers/self_attention/key/Fp8DirectDotGeneralOp_0/output_grad_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0418 18:16:42.824342 134511891793728 maxtext_utils.py:1654] params/_overwrite_with_gradient/decoder/layers/self_attention/key/Fp8DirectDotGeneralOp_0/output_grad_scale Shape: float32[16,1] Logical: P() Physical: () I0418 18:16:42.824370 134511891793728 maxtext_utils.py:1654] params/_overwrite_with_gradient/decoder/layers/self_attention/out/Fp8DirectDotGeneralOp_0/input_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0418 18:16:42.824397 134511891793728 maxtext_utils.py:1654] params/_overwrite_with_gradient/decoder/layers/self_attention/out/Fp8DirectDotGeneralOp_0/input_scale Shape: float32[16,1] Logical: P() Physical: () I0418 18:16:42.824424 134511891793728 maxtext_utils.py:1654] params/_overwrite_with_gradient/decoder/layers/self_attention/out/Fp8DirectDotGeneralOp_0/kernel_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0418 18:16:42.824450 134511891793728 maxtext_utils.py:1654] params/_overwrite_with_gradient/decoder/layers/self_attention/out/Fp8DirectDotGeneralOp_0/kernel_scale Shape: float32[16,1] Logical: P() Physical: () I0418 18:16:42.824476 134511891793728 maxtext_utils.py:1654] params/_overwrite_with_gradient/decoder/layers/self_attention/out/Fp8DirectDotGeneralOp_0/output_grad_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0418 18:16:42.824503 134511891793728 maxtext_utils.py:1654] params/_overwrite_with_gradient/decoder/layers/self_attention/out/Fp8DirectDotGeneralOp_0/output_grad_scale Shape: float32[16,1] Logical: P() Physical: () I0418 18:16:42.824529 134511891793728 maxtext_utils.py:1654] params/_overwrite_with_gradient/decoder/layers/self_attention/query/Fp8DirectDotGeneralOp_0/input_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0418 18:16:42.824557 134511891793728 maxtext_utils.py:1654] params/_overwrite_with_gradient/decoder/layers/self_attention/query/Fp8DirectDotGeneralOp_0/input_scale Shape: float32[16,1] Logical: P() Physical: () I0418 18:16:42.824584 134511891793728 maxtext_utils.py:1654] params/_overwrite_with_gradient/decoder/layers/self_attention/query/Fp8DirectDotGeneralOp_0/kernel_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0418 18:16:42.824610 134511891793728 maxtext_utils.py:1654] params/_overwrite_with_gradient/decoder/layers/self_attention/query/Fp8DirectDotGeneralOp_0/kernel_scale Shape: float32[16,1] Logical: P() Physical: () I0418 18:16:42.824636 134511891793728 maxtext_utils.py:1654] params/_overwrite_with_gradient/decoder/layers/self_attention/query/Fp8DirectDotGeneralOp_0/output_grad_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0418 18:16:42.824662 134511891793728 maxtext_utils.py:1654] params/_overwrite_with_gradient/decoder/layers/self_attention/query/Fp8DirectDotGeneralOp_0/output_grad_scale Shape: float32[16,1] Logical: P() Physical: () I0418 18:16:42.824688 134511891793728 maxtext_utils.py:1654] params/_overwrite_with_gradient/decoder/layers/self_attention/value/Fp8DirectDotGeneralOp_0/input_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0418 18:16:42.824714 134511891793728 maxtext_utils.py:1654] params/_overwrite_with_gradient/decoder/layers/self_attention/value/Fp8DirectDotGeneralOp_0/input_scale Shape: float32[16,1] Logical: P() Physical: () I0418 18:16:42.824741 134511891793728 maxtext_utils.py:1654] params/_overwrite_with_gradient/decoder/layers/self_attention/value/Fp8DirectDotGeneralOp_0/kernel_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0418 18:16:42.824767 134511891793728 maxtext_utils.py:1654] params/_overwrite_with_gradient/decoder/layers/self_attention/value/Fp8DirectDotGeneralOp_0/kernel_scale Shape: float32[16,1] Logical: P() Physical: () I0418 18:16:42.824793 134511891793728 maxtext_utils.py:1654] params/_overwrite_with_gradient/decoder/layers/self_attention/value/Fp8DirectDotGeneralOp_0/output_grad_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0418 18:16:42.824820 134511891793728 maxtext_utils.py:1654] params/_overwrite_with_gradient/decoder/layers/self_attention/value/Fp8DirectDotGeneralOp_0/output_grad_scale Shape: float32[16,1] Logical: P() Physical: () I0418 18:16:42.824869 134511891793728 maxtext_utils.py:1654] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: P('norm',) Physical: (None,) I0418 18:16:42.824922 134511891793728 maxtext_utils.py:1654] params/params/decoder/layers/mlp/wi_0/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0418 18:16:42.824959 134511891793728 maxtext_utils.py:1654] params/params/decoder/layers/mlp/wi_1/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0418 18:16:42.825007 134511891793728 maxtext_utils.py:1654] params/params/decoder/layers/mlp/wo/kernel Shape: float32[7168,16,2048] Logical: P('mlp', 'layers', 'embed') Physical: (None, None, 'fsdp') I0418 18:16:42.825054 134511891793728 maxtext_utils.py:1654] params/params/decoder/layers/post_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0418 18:16:42.825105 134511891793728 maxtext_utils.py:1654] params/params/decoder/layers/pre_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0418 18:16:42.825155 134511891793728 maxtext_utils.py:1654] 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) I0418 18:16:42.825203 134511891793728 maxtext_utils.py:1654] 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') I0418 18:16:42.825237 134511891793728 maxtext_utils.py:1654] 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) I0418 18:16:42.825268 134511891793728 maxtext_utils.py:1654] 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) I0418 18:16:42.825311 134511891793728 maxtext_utils.py:1654] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: P('embed_vocab', 'vocab') Physical: ('fsdp', None) I0418 18:16:42.825352 134511891793728 maxtext_utils.py:1654] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: P('vocab', 'embed_vocab') Physical: (None, 'fsdp') I0418 18:16:44.433262 134511891793728 train.py:155] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0418 18:16:44.433358 134511891793728 train.py:155] train/xent Physical: float32[32,2048]............................................ ('fsdp', None). I0418 18:16:44.448648 134511891793728 train.py:162] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0418 18:16:44.448712 134511891793728 train.py:162] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None). I0418 18:16:58.130581 134511891793728 max_utils.py:791] Total memory size: 1.5 GB, Output size: 0.4 GB, Temp size: 1.1 GB, Argument size: 0.4 GB, Host temp size: 0.0 GB. I0418 18:16:58.131484 134511891793728 metric_logger.py:301] number parameters: 1.105 billion I0418 18:17:13.271946 134511891793728 checkpointing.py:772] Waiting for step 0 to finish before checkpoint... I0418 18:17:13.442163 134511891793728 checkpointing.py:776] Waited 0.1701948642730713 seconds for step 0 to finish before starting checkpointing. I0418 18:17:13.445579 134511891793728 checkpoint_manager.py:2009] [process=7][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0418 18:17:13.447524 134511891793728 checkpoint_manager.py:1512] [process=7] Saving checkpoint at step 0 I0418 18:17:13.448919 134511891793728 event_tracking.py:70] [process=7] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260418_180002/linen_xpk_main_20260418_180002_05_fp8/checkpoints/0. I0418 18:17:14.206906 134511891793728 signaling_client.py:364] Using JaxDistributedSignalingClient I0418 18:17:14.207857 134511891793728 jax_array_handlers.py:360] Scheduling D2H of 81 prioritized jax.Array. I0418 18:17:14.207912 134511891793728 replica_slices.py:424] Transferring arrays to host memory with options: use_replica_parallel=True, min_slice_bytes_for_replica_parallel=None, max_replicas_for_replica_parallel=None, enable_pinned_host_transfer=False I0418 18:17:14.626933 134511891793728 base_pytree_checkpoint_handler.py:154] [process=7][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.420151s I0418 18:17:14.627134 134511891793728 base_pytree_checkpoint_handler.py:130] [process=7] /jax/orbax/write/blocking_gbytes_per_sec: 3.600 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.4285409450531006 s) (per-host) I0418 18:17:14.627191 134511891793728 base_pytree_checkpoint_handler.py:768] [process=7][thread=MainThread] Initiated Pytree async_save. Time taken: 0.428609s (batch_requests_ready=0.003613s, total_serialization_initiated=0.424919s, others=0.000077s) I0418 18:17:14.627296 134511891793728 composite_checkpoint_handler.py:715] [process=7][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.432819s (all_items=0.000018s, per_item={'items': '0.00001764'}, temp_paths=0.432801) I0418 18:17:14.628108 134511891793728 event_tracking.py:125] [process=7] [async] Finished blocking save in 1.18 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260418_180002/linen_xpk_main_20260418_180002_05_fp8/checkpoints/0. I0418 18:17:14.628461 134382346041088 async_checkpointer.py:76] [process=7][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-18 18:37:14.628423 I0418 18:17:14.631799 134511891793728 checkpoint_manager.py:1560] [process=7][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize I0418 18:17:14.632155 134382288758528 async_checkpointer.py:280] [process=7][thread=save_finalize] Waiting for background save thread=async_save. I0418 18:17:14.632332 134511891793728 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260418_180002/linen_xpk_main_20260418_180002_05_fp8/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776536233.445561, 'wait_for_prev_duration_secs': 6.222724914550781e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776536233.4475625, 'checkpointer_blocking_duration_secs': 1.181058645248413, 'get_old_steps_start_time': 1776536234.6286476, 'get_old_steps_duration_secs': 3.0279159545898438e-05, 'checkpoint_manager_blocking_start_time': 1776536233.4426782, 'checkpoint_manager_blocking_duration_secs': 1.1896164417266846} I0418 18:17:14.632442 134511891793728 checkpointing.py:408] Started an asynchronous checkpoint save for step 0 I0418 18:17:14.632491 134511891793728 max_utils.py:750] Memstats: After params initialized: I0418 18:17:14.632543 134511891793728 max_utils.py:756] Using (GB) 0.45 / 31.25 (1.440000%) on TPU_26(process=7,(2,6,0,0)) I0418 18:17:14.632578 134511891793728 max_utils.py:756] Using (GB) 0.45 / 31.25 (1.440000%) on TPU_27(process=7,(3,6,0,0)) I0418 18:17:14.632605 134511891793728 max_utils.py:756] Using (GB) 0.45 / 31.25 (1.440000%) on TPU_30(process=7,(2,7,0,0)) I0418 18:17:14.632628 134511891793728 max_utils.py:756] Using (GB) 0.45 / 31.25 (1.440000%) on TPU_31(process=7,(3,7,0,0)) I0418 18:17:14.947874 134511891793728 metric_logger.py:196] completed step: 0, seconds: 15.140, TFLOP/s/device: 0.897, Tokens/s/device: 135.268, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52776.805 I0418 18:17:15.143847 134511891793728 metric_logger.py:196] completed step: 1, seconds: 1.674, TFLOP/s/device: 8.117, Tokens/s/device: 1223.508, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52761.840 I0418 18:17:15.564787 134511891793728 metric_logger.py:196] completed step: 2, seconds: 0.038, TFLOP/s/device: 353.914, Tokens/s/device: 53345.836, total_weights: 65536, loss: 10.816, lm_loss: 10.816, perplexity: 49832.398 I0418 18:17:15.723206 134511891793728 metric_logger.py:196] completed step: 3, seconds: 0.421, TFLOP/s/device: 32.266, Tokens/s/device: 4863.464, total_weights: 65536, loss: 10.431, lm_loss: 10.431, perplexity: 33901.820 I0418 18:17:16.040598 134511891793728 metric_logger.py:196] completed step: 4, seconds: 0.164, TFLOP/s/device: 82.688, Tokens/s/device: 12463.562, total_weights: 65536, loss: 9.992, lm_loss: 9.992, perplexity: 21847.639 I0418 18:17:16.047224 134511891793728 metric_logger.py:196] completed step: 5, seconds: 0.158, TFLOP/s/device: 85.844, Tokens/s/device: 12939.341, total_weights: 65536, loss: 9.549, lm_loss: 9.549, perplexity: 14035.006 I0418 18:17:18.280435 2847 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com I0418 18:17:20.675175 134382297151232 array_metadata_store.py:203] [process=7][thread=array_type_handler] Wrote 81 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260418_180002/linen_xpk_main_20260418_180002_05_fp8/checkpoints/0/items/array_metadatas/process_7 I0418 18:17:34.787708 134382346041088 base_pytree_checkpoint_handler.py:130] [process=7] /jax/orbax/write/gbytes_per_sec: 76.728 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 20.589051246643066 s) (per-host) I0418 18:17:34.787839 134382346041088 async_checkpointer.py:90] [process=7][thread=async_save] 3 Handler Commit operations completed. Time taken: 20.159262s. I0418 18:17:41.159799 134511891793728 metric_logger.py:196] completed step: 6, seconds: 0.318, TFLOP/s/device: 42.723, Tokens/s/device: 6439.705, total_weights: 65536, loss: 9.111, lm_loss: 9.111, perplexity: 9058.688 I0418 18:17:41.318297 134511891793728 metric_logger.py:196] completed step: 7, seconds: 24.955, TFLOP/s/device: 0.544, Tokens/s/device: 82.067, total_weights: 65536, loss: 8.685, lm_loss: 8.685, perplexity: 5914.720 I0418 18:17:41.476956 134511891793728 metric_logger.py:196] completed step: 8, seconds: 0.163, TFLOP/s/device: 83.205, Tokens/s/device: 12541.565, total_weights: 65536, loss: 8.281, lm_loss: 8.281, perplexity: 3950.010 I0418 18:17:41.482017 134511891793728 checkpointing.py:772] Waiting for step 10 to finish before checkpoint... I0418 18:17:41.793649 134511891793728 checkpointing.py:776] Waited 0.3115971088409424 seconds for step 10 to finish before starting checkpointing. I0418 18:17:41.796318 134511891793728 checkpoint_manager.py:2020] [process=7][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0418 18:17:43.478874 134382346041088 async_checkpointer.py:160] [process=7][thread=async_save] Background save thread done. Time taken: 28.850281s. I0418 18:17:43.479170 134382288758528 async_checkpointer.py:288] [process=7][thread=save_finalize] Done with waiting for background save thread=async_save. I0418 18:17:43.479291 134382288758528 async_checkpointer.py:298] [process=7][thread=save_finalize] No errors found in background save thread=async_save. I0418 18:17:43.479341 134382288758528 checkpoint_manager.py:2137] [process=7][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts... I0418 18:17:43.480841 134382288758528 checkpoint_manager.py:2146] [process=7][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts. I0418 18:17:43.481029 134511891793728 checkpoint_manager.py:2032] [process=7][thread=MainThread][step=0][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=0. W0418 18:17:43.481188 134511891793728 checkpoint_manager.py:1452] Waiting for previous save to complete took 1.684863 seconds. If this number is high, consider checkpointing less frequently. I0418 18:17:43.483016 134511891793728 checkpoint_manager.py:1512] [process=7] Saving checkpoint at step 10 I0418 18:17:43.485008 134511891793728 event_tracking.py:70] [process=7] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260418_180002/linen_xpk_main_20260418_180002_05_fp8/checkpoints/10. I0418 18:17:44.229100 134511891793728 jax_array_handlers.py:360] Scheduling D2H of 81 prioritized jax.Array. I0418 18:17:44.229199 134511891793728 replica_slices.py:424] Transferring arrays to host memory with options: use_replica_parallel=True, min_slice_bytes_for_replica_parallel=None, max_replicas_for_replica_parallel=None, enable_pinned_host_transfer=False I0418 18:17:44.266477 134511891793728 base_pytree_checkpoint_handler.py:154] [process=7][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.038349s I0418 18:17:44.266637 134511891793728 base_pytree_checkpoint_handler.py:130] [process=7] /jax/orbax/write/blocking_gbytes_per_sec: 34.620 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.04456162452697754 s) (per-host) I0418 18:17:44.266689 134511891793728 base_pytree_checkpoint_handler.py:768] [process=7][thread=MainThread] Initiated Pytree async_save. Time taken: 0.044622s (batch_requests_ready=0.003103s, total_serialization_initiated=0.041452s, others=0.000068s) I0418 18:17:44.266795 134511891793728 composite_checkpoint_handler.py:715] [process=7][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.049040s (all_items=0.000015s, per_item={'items': '0.00001502'}, temp_paths=0.049025) I0418 18:17:44.267495 134511891793728 event_tracking.py:125] [process=7] [async] Finished blocking save in 0.78 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260418_180002/linen_xpk_main_20260418_180002_05_fp8/checkpoints/10. I0418 18:17:44.267876 134379091244800 async_checkpointer.py:76] [process=7][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-18 18:37:44.267838 I0418 18:17:44.273256 134511891793728 checkpoint_manager.py:1560] [process=7][thread=MainThread][step=10] Starting CheckpointManager Save Finalize thread=save_finalize I0418 18:17:44.273501 134382288758528 async_checkpointer.py:280] [process=7][thread=save_finalize] Waiting for background save thread=async_save. I0418 18:17:44.273635 134511891793728 standard_logger.py:34] {'step': 10, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260418_180002/linen_xpk_main_20260418_180002_05_fp8/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776536261.7962868, 'wait_for_prev_duration_secs': 1.6848626136779785, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776536263.4830546, 'checkpointer_blocking_duration_secs': 0.7849705219268799, 'get_old_steps_start_time': 1776536264.2680476, 'get_old_steps_duration_secs': 4.482269287109375e-05, 'checkpoint_manager_blocking_start_time': 1776536261.7944648, 'checkpoint_manager_blocking_duration_secs': 2.479135751724243} I0418 18:17:44.273767 134511891793728 checkpointing.py:408] Started an asynchronous checkpoint save for step 10 I0418 18:17:44.274417 134511891793728 metric_logger.py:196] completed step: 9, seconds: 0.159, TFLOP/s/device: 85.670, Tokens/s/device: 12913.151, total_weights: 65536, loss: 7.908, lm_loss: 7.908, perplexity: 2720.116 I0418 18:17:44.281920 134511891793728 metric_logger.py:196] completed step: 10, seconds: 0.159, TFLOP/s/device: 85.677, Tokens/s/device: 12914.129, total_weights: 65536, loss: 7.568, lm_loss: 7.568, perplexity: 1936.198 I0418 18:17:44.439687 134511891793728 metric_logger.py:196] completed step: 11, seconds: 2.798, TFLOP/s/device: 4.857, Tokens/s/device: 732.043, total_weights: 65536, loss: 7.288, lm_loss: 7.288, perplexity: 1463.272 I0418 18:17:44.597949 134511891793728 metric_logger.py:196] completed step: 12, seconds: 0.006, TFLOP/s/device: 2103.922, Tokens/s/device: 317126.045, total_weights: 65536, loss: 7.032, lm_loss: 7.032, perplexity: 1132.061 I0418 18:17:45.155454 134511891793728 metric_logger.py:196] completed step: 13, seconds: 0.159, TFLOP/s/device: 85.361, Tokens/s/device: 12866.504, total_weights: 65536, loss: 6.815, lm_loss: 6.815, perplexity: 911.031 I0418 18:17:45.314314 134511891793728 metric_logger.py:196] completed step: 14, seconds: 0.552, TFLOP/s/device: 24.612, Tokens/s/device: 3709.748, total_weights: 65536, loss: 6.635, lm_loss: 6.635, perplexity: 761.080 I0418 18:17:45.472795 134511891793728 metric_logger.py:196] completed step: 15, seconds: 0.163, TFLOP/s/device: 83.261, Tokens/s/device: 12550.019, total_weights: 65536, loss: 6.492, lm_loss: 6.492, perplexity: 659.915 I0418 18:17:45.631316 134511891793728 metric_logger.py:196] completed step: 16, seconds: 0.159, TFLOP/s/device: 85.572, Tokens/s/device: 12898.350, total_weights: 65536, loss: 6.380, lm_loss: 6.380, perplexity: 589.948 I0418 18:17:45.789793 134511891793728 metric_logger.py:196] completed step: 17, seconds: 0.158, TFLOP/s/device: 85.789, Tokens/s/device: 12931.007, total_weights: 65536, loss: 6.293, lm_loss: 6.293, perplexity: 540.941 I0418 18:17:45.948246 134511891793728 metric_logger.py:196] completed step: 18, seconds: 0.159, TFLOP/s/device: 85.649, Tokens/s/device: 12909.895, total_weights: 65536, loss: 6.223, lm_loss: 6.223, perplexity: 504.108 I0418 18:17:46.106167 134511891793728 checkpointing.py:772] Waiting for step 19 to finish before checkpoint... I0418 18:17:46.107165 134511891793728 checkpointing.py:776] Waited 0.001016855239868164 seconds for step 19 to finish before starting checkpointing. I0418 18:17:46.109327 134511891793728 checkpoint_manager.py:2020] [process=7][thread=MainThread][step=10][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0418 18:17:52.134623 134379116422912 array_metadata_store.py:203] [process=7][thread=array_type_handler] Wrote 81 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260418_180002/linen_xpk_main_20260418_180002_05_fp8/checkpoints/10/items/array_metadatas/process_7 I0418 18:18:27.760964 134379091244800 base_pytree_checkpoint_handler.py:130] [process=7] /jax/orbax/write/gbytes_per_sec: 36.284 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 43.53883337974548 s) (per-host) I0418 18:18:27.761122 134379091244800 async_checkpointer.py:90] [process=7][thread=async_save] 3 Handler Commit operations completed. Time taken: 43.493122s. I0418 18:18:35.815789 134379091244800 async_checkpointer.py:160] [process=7][thread=async_save] Background save thread done. Time taken: 51.547785s. I0418 18:18:35.816120 134382288758528 async_checkpointer.py:288] [process=7][thread=save_finalize] Done with waiting for background save thread=async_save. I0418 18:18:35.816241 134382288758528 async_checkpointer.py:298] [process=7][thread=save_finalize] No errors found in background save thread=async_save. I0418 18:18:35.816288 134382288758528 checkpoint_manager.py:2137] [process=7][thread=save_finalize][step=10] CheckpointManager Save Finalize is syncing with other hosts... I0418 18:18:35.817670 134382288758528 checkpoint_manager.py:2146] [process=7][thread=save_finalize][step=10] CheckpointManager Save Finalize is done on all hosts. I0418 18:18:35.817838 134511891793728 checkpoint_manager.py:2032] [process=7][thread=MainThread][step=10][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=10. W0418 18:18:35.817974 134511891793728 checkpoint_manager.py:1452] Waiting for previous save to complete took 49.708653 seconds. If this number is high, consider checkpointing less frequently. I0418 18:18:35.819561 134511891793728 checkpoint_manager.py:1512] [process=7] Saving checkpoint at step 19 I0418 18:18:35.821613 134511891793728 event_tracking.py:70] [process=7] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_main_20260418_180002/linen_xpk_main_20260418_180002_05_fp8/checkpoints/19. I0418 18:18:36.537542 134511891793728 jax_array_handlers.py:360] Scheduling D2H of 81 prioritized jax.Array. I0418 18:18:36.537636 134511891793728 replica_slices.py:424] Transferring arrays to host memory with options: use_replica_parallel=True, min_slice_bytes_for_replica_parallel=None, max_replicas_for_replica_parallel=None, enable_pinned_host_transfer=False I0418 18:18:36.574451 134511891793728 base_pytree_checkpoint_handler.py:154] [process=7][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.037921s I0418 18:18:36.574618 134511891793728 base_pytree_checkpoint_handler.py:130] [process=7] /jax/orbax/write/blocking_gbytes_per_sec: 35.066 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.04399561882019043 s) (per-host) I0418 18:18:36.574666 134511891793728 base_pytree_checkpoint_handler.py:768] [process=7][thread=MainThread] Initiated Pytree async_save. Time taken: 0.044055s (batch_requests_ready=0.003006s, total_serialization_initiated=0.040984s, others=0.000065s) I0418 18:18:36.574798 134511891793728 composite_checkpoint_handler.py:715] [process=7][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.048144s (all_items=0.000011s, per_item={'items': '0.00001097'}, temp_paths=0.048133) I0418 18:18:36.575600 134511891793728 event_tracking.py:125] [process=7] [async] Finished blocking save in 0.76 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_main_20260418_180002/linen_xpk_main_20260418_180002_05_fp8/checkpoints/19. I0418 18:18:36.575882 134379116422912 async_checkpointer.py:76] [process=7][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-18 18:38:36.575847 I0418 18:18:36.579324 134511891793728 checkpoint_manager.py:1560] [process=7][thread=MainThread][step=19] Starting CheckpointManager Save Finalize thread=save_finalize I0418 18:18:36.579679 134382288758528 async_checkpointer.py:280] [process=7][thread=save_finalize] Waiting for background save thread=async_save. I0418 18:18:36.579843 134511891793728 standard_logger.py:34] {'step': 19, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_main_20260418_180002/linen_xpk_main_20260418_180002_05_fp8/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776536266.1092987, 'wait_for_prev_duration_secs': 49.70865297317505, 'time_between_consecutive_saves_sec': 2.62841796875, 'checkpointer_blocking_start_time': 1776536315.8195994, 'checkpointer_blocking_duration_secs': 0.7564229965209961, 'get_old_steps_start_time': 1776536316.576041, 'get_old_steps_duration_secs': 2.4318695068359375e-05, 'checkpoint_manager_blocking_start_time': 1776536266.1074467, 'checkpoint_manager_blocking_duration_secs': 50.47236251831055} I0418 18:18:36.579961 134511891793728 checkpointing.py:408] Started an asynchronous checkpoint save for step 19 I0418 18:18:36.580003 134511891793728 checkpoint_manager.py:2020] [process=7][thread=MainThread][step=19][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0418 18:18:43.387330 134379653293824 array_metadata_store.py:203] [process=7][thread=array_type_handler] Wrote 81 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_main_20260418_180002/linen_xpk_main_20260418_180002_05_fp8/checkpoints/19/items/array_metadatas/process_7 I0418 18:19:20.097202 134379116422912 base_pytree_checkpoint_handler.py:130] [process=7] /jax/orbax/write/gbytes_per_sec: 36.261 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 43.56653714179993 s) (per-host) I0418 18:19:20.097328 134379116422912 async_checkpointer.py:90] [process=7][thread=async_save] 3 Handler Commit operations completed. Time taken: 43.521339s. I0418 18:19:27.734118 134379116422912 async_checkpointer.py:160] [process=7][thread=async_save] Background save thread done. Time taken: 51.158081s. I0418 18:19:27.734410 134382288758528 async_checkpointer.py:288] [process=7][thread=save_finalize] Done with waiting for background save thread=async_save. I0418 18:19:27.734549 134382288758528 async_checkpointer.py:298] [process=7][thread=save_finalize] No errors found in background save thread=async_save. I0418 18:19:27.734597 134382288758528 checkpoint_manager.py:2137] [process=7][thread=save_finalize][step=19] CheckpointManager Save Finalize is syncing with other hosts... I0418 18:19:27.736280 134382288758528 checkpoint_manager.py:2146] [process=7][thread=save_finalize][step=19] CheckpointManager Save Finalize is done on all hosts. I0418 18:19:27.736471 134511891793728 checkpoint_manager.py:2032] [process=7][thread=MainThread][step=19][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=19. I0418 18:19:27.736625 134511891793728 checkpoint_manager.py:2009] [process=7][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0418 18:19:27.737592 134511891793728 metric_logger.py:196] completed step: 19, seconds: 0.158, TFLOP/s/device: 85.808, Tokens/s/device: 12933.866, total_weights: 65536, loss: 6.167, lm_loss: 6.167, perplexity: 476.947 Per train step: Total TFLOPs: 13.59 split as 93.93% learnable weight flops and 6.07% attention flops XPK End: Sat Apr 18 18:19:40 UTC 2026 EXIT_CODE=0
XPK Start: Sat Apr 18 23:59:27 UTC 2026 PyTorch was not found. Models won't be available and only tokenizers, configuration and file/data utilities can be used. 2026-04-18 23:59:51.729027: 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) I0418 23:59:51.907961 136124056586048 max_utils.py:273] Attempting to initialize the jax distributed system... I0419 00:00:00.949488 136124056586048 distributed.py:149] Starting JAX distributed service on [::]:8482 I0419 00:00:00.951760 136124056586048 distributed.py:172] Connecting to JAX distributed service on mt-05-fp8-4xgy0-slice-job-0-0.mt-05-fp8-4xgy0:8482 I0419 00:00:03.269298 136124056586048 max_utils.py:284] Jax distributed system initialized! I0419 00:00:09.382912 136124056586048 max_utils.py:800] System Information: Jax Version: 0.9.2 I0419 00:00:09.383014 136124056586048 max_utils.py:801] System Information: Jaxlib Version: 0.9.2 I0419 00:00:09.383054 136124056586048 max_utils.py:802] System Information: Jax Backend: PJRT C API TFRT TPU v6 lite Built on Mar 4 2026 11:32:08 (1772652728) cl/878335365 I0419 00:00:09.383090 136124056586048 train_utils.py:378] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing. I0419 00:00:10.434977 136124056586048 maxtext_utils.py:1718] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0419 00:00:10.435557 136124056586048 maxtext_utils.py:1718] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0419 00:00:10.435739 136124056586048 checkpointing.py:688] Setting up checkpoint logger... I0419 00:00:10.435791 136124056586048 checkpointing.py:234] Creating checkpoint manager with ocdbt=True and zarr3=True I0419 00:00:10.435848 136124056586048 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None I0419 00:00:10.436174 136124056586048 base_pytree_checkpoint_handler.py:441] Created BasePyTreeCheckpointHandler: use_ocdbt=True, use_zarr3=True, pytree_metadata_options=PyTreeMetadataOptions(support_rich_types=False), array_metadata_store=<orbax.checkpoint._src.metadata.array_metadata_store.Store object at 0x7bcd3cbe2600>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0419 00:00:14.025879 136124056586048 checkpointing.py:266] Enabling policy for fixed interval checkpointing. I0419 00:00:14.026122 136124056586048 checkpoint_manager.py:708] [process=0][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=('items',), item_handlers={'items': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7bcc16192780>}, handler_registry=None I0419 00:00:14.026361 136124056586048 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7bcc16192780>` 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`. I0419 00:00:14.026410 136124056586048 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7bcc1627bbc0>` 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`. I0419 00:00:14.026445 136124056586048 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 0x7bcc16192780>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7bcc16192780>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7bcc1627bbc0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7bcc1627bbc0>}). I0419 00:00:14.026755 136124056586048 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34 I0419 00:00:14.026838 136124056586048 async_checkpointer.py:192] [process=0][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7bb8dc33ec00> timeout: 1200 secs and primary_host=0 for async checkpoint writes I0419 00:00:16.157521 136124056586048 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_05_fp8/checkpoints I0419 00:00:16.967186 136124056586048 checkpoint_manager.py:929] [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_background_delete=False, read_only=False, enable_async_checkpointing=True, async_options=None, multiprocessing_options=MultiprocessingOptions(primary_host=0, active_processes=None, barrier_sync_key_prefix=None), should_save_fn=None, file_options=FileOptions(path_permission_mode=None), save_root_metadata=True, temporary_path_class=None, save_decision_policy=FixedIntervalPolicy(interval=10), preservation_policy=LatestN(n=None), prevent_write_metrics=False, enable_should_save_is_saving_in_progress_check=True, enable_per_process_directory_creation=False, lightweight_initialize=False), root_directory=gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_05_fp8/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7bcc162cc8c0> I0419 00:00:16.967356 136124056586048 checkpointing.py:302] Checkpoint manager created! I0419 00:00:18.033827 136124056586048 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0419 00:00:18.033930 136124056586048 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0419 00:00:18.420051 136124056586048 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0419 00:00:18.420144 136124056586048 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0419 00:00:18.436259 136124056586048 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0419 00:00:18.436317 136124056586048 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0419 00:00:18.488678 136124056586048 attentions.py:1154] attentions/query Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0419 00:00:18.488763 136124056586048 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0419 00:00:18.504971 136124056586048 attentions.py:1155] attentions/key Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0419 00:00:18.505032 136124056586048 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0419 00:00:18.521183 136124056586048 attentions.py:1156] attentions/value Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0419 00:00:18.521241 136124056586048 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0419 00:00:18.545689 136124056586048 attentions.py:1197] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv'). I0419 00:00:18.545758 136124056586048 attentions.py:1197] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0419 00:00:18.596560 136124056586048 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0419 00:00:18.596634 136124056586048 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None). I0419 00:00:19.030870 136124056586048 checkpointing.py:578] checkpoint manager exists so trying to load this run's existing checkpoint I0419 00:00:19.030979 136124056586048 checkpointing.py:676] No existing checkpoints found, not restoring checkpoint. fsdp: 32 I0419 00:00:20.930797 136124056586048 maxtext_utils.py:1827] params/_overwrite_with_gradient/decoder/layers/mlp/wi_0/Fp8DirectDotGeneralOp_0/input_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0419 00:00:20.930926 136124056586048 maxtext_utils.py:1827] params/_overwrite_with_gradient/decoder/layers/mlp/wi_0/Fp8DirectDotGeneralOp_0/input_scale Shape: float32[16,1] Logical: P() Physical: () I0419 00:00:20.930972 136124056586048 maxtext_utils.py:1827] params/_overwrite_with_gradient/decoder/layers/mlp/wi_0/Fp8DirectDotGeneralOp_0/kernel_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0419 00:00:20.931010 136124056586048 maxtext_utils.py:1827] params/_overwrite_with_gradient/decoder/layers/mlp/wi_0/Fp8DirectDotGeneralOp_0/kernel_scale Shape: float32[16,1] Logical: P() Physical: () I0419 00:00:20.931048 136124056586048 maxtext_utils.py:1827] params/_overwrite_with_gradient/decoder/layers/mlp/wi_0/Fp8DirectDotGeneralOp_0/output_grad_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0419 00:00:20.931082 136124056586048 maxtext_utils.py:1827] params/_overwrite_with_gradient/decoder/layers/mlp/wi_0/Fp8DirectDotGeneralOp_0/output_grad_scale Shape: float32[16,1] Logical: P() Physical: () I0419 00:00:20.931119 136124056586048 maxtext_utils.py:1827] params/_overwrite_with_gradient/decoder/layers/mlp/wi_1/Fp8DirectDotGeneralOp_0/input_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0419 00:00:20.931151 136124056586048 maxtext_utils.py:1827] params/_overwrite_with_gradient/decoder/layers/mlp/wi_1/Fp8DirectDotGeneralOp_0/input_scale Shape: float32[16,1] Logical: P() Physical: () I0419 00:00:20.931181 136124056586048 maxtext_utils.py:1827] params/_overwrite_with_gradient/decoder/layers/mlp/wi_1/Fp8DirectDotGeneralOp_0/kernel_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0419 00:00:20.931210 136124056586048 maxtext_utils.py:1827] params/_overwrite_with_gradient/decoder/layers/mlp/wi_1/Fp8DirectDotGeneralOp_0/kernel_scale Shape: float32[16,1] Logical: P() Physical: () I0419 00:00:20.931239 136124056586048 maxtext_utils.py:1827] params/_overwrite_with_gradient/decoder/layers/mlp/wi_1/Fp8DirectDotGeneralOp_0/output_grad_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0419 00:00:20.931269 136124056586048 maxtext_utils.py:1827] params/_overwrite_with_gradient/decoder/layers/mlp/wi_1/Fp8DirectDotGeneralOp_0/output_grad_scale Shape: float32[16,1] Logical: P() Physical: () I0419 00:00:20.931297 136124056586048 maxtext_utils.py:1827] params/_overwrite_with_gradient/decoder/layers/mlp/wo/Fp8DirectDotGeneralOp_0/input_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0419 00:00:20.931324 136124056586048 maxtext_utils.py:1827] params/_overwrite_with_gradient/decoder/layers/mlp/wo/Fp8DirectDotGeneralOp_0/input_scale Shape: float32[16,1] Logical: P() Physical: () I0419 00:00:20.931352 136124056586048 maxtext_utils.py:1827] params/_overwrite_with_gradient/decoder/layers/mlp/wo/Fp8DirectDotGeneralOp_0/kernel_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0419 00:00:20.931381 136124056586048 maxtext_utils.py:1827] params/_overwrite_with_gradient/decoder/layers/mlp/wo/Fp8DirectDotGeneralOp_0/kernel_scale Shape: float32[16,1] Logical: P() Physical: () I0419 00:00:20.931409 136124056586048 maxtext_utils.py:1827] params/_overwrite_with_gradient/decoder/layers/mlp/wo/Fp8DirectDotGeneralOp_0/output_grad_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0419 00:00:20.931437 136124056586048 maxtext_utils.py:1827] params/_overwrite_with_gradient/decoder/layers/mlp/wo/Fp8DirectDotGeneralOp_0/output_grad_scale Shape: float32[16,1] Logical: P() Physical: () I0419 00:00:20.931465 136124056586048 maxtext_utils.py:1827] params/_overwrite_with_gradient/decoder/layers/self_attention/key/Fp8DirectDotGeneralOp_0/input_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0419 00:00:20.931492 136124056586048 maxtext_utils.py:1827] params/_overwrite_with_gradient/decoder/layers/self_attention/key/Fp8DirectDotGeneralOp_0/input_scale Shape: float32[16,1] Logical: P() Physical: () I0419 00:00:20.931520 136124056586048 maxtext_utils.py:1827] params/_overwrite_with_gradient/decoder/layers/self_attention/key/Fp8DirectDotGeneralOp_0/kernel_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0419 00:00:20.931547 136124056586048 maxtext_utils.py:1827] params/_overwrite_with_gradient/decoder/layers/self_attention/key/Fp8DirectDotGeneralOp_0/kernel_scale Shape: float32[16,1] Logical: P() Physical: () I0419 00:00:20.931573 136124056586048 maxtext_utils.py:1827] params/_overwrite_with_gradient/decoder/layers/self_attention/key/Fp8DirectDotGeneralOp_0/output_grad_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0419 00:00:20.931601 136124056586048 maxtext_utils.py:1827] params/_overwrite_with_gradient/decoder/layers/self_attention/key/Fp8DirectDotGeneralOp_0/output_grad_scale Shape: float32[16,1] Logical: P() Physical: () I0419 00:00:20.931628 136124056586048 maxtext_utils.py:1827] params/_overwrite_with_gradient/decoder/layers/self_attention/out/Fp8DirectDotGeneralOp_0/input_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0419 00:00:20.931654 136124056586048 maxtext_utils.py:1827] params/_overwrite_with_gradient/decoder/layers/self_attention/out/Fp8DirectDotGeneralOp_0/input_scale Shape: float32[16,1] Logical: P() Physical: () I0419 00:00:20.931680 136124056586048 maxtext_utils.py:1827] params/_overwrite_with_gradient/decoder/layers/self_attention/out/Fp8DirectDotGeneralOp_0/kernel_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0419 00:00:20.931707 136124056586048 maxtext_utils.py:1827] params/_overwrite_with_gradient/decoder/layers/self_attention/out/Fp8DirectDotGeneralOp_0/kernel_scale Shape: float32[16,1] Logical: P() Physical: () I0419 00:00:20.931734 136124056586048 maxtext_utils.py:1827] params/_overwrite_with_gradient/decoder/layers/self_attention/out/Fp8DirectDotGeneralOp_0/output_grad_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0419 00:00:20.931760 136124056586048 maxtext_utils.py:1827] params/_overwrite_with_gradient/decoder/layers/self_attention/out/Fp8DirectDotGeneralOp_0/output_grad_scale Shape: float32[16,1] Logical: P() Physical: () I0419 00:00:20.931787 136124056586048 maxtext_utils.py:1827] params/_overwrite_with_gradient/decoder/layers/self_attention/query/Fp8DirectDotGeneralOp_0/input_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0419 00:00:20.931830 136124056586048 maxtext_utils.py:1827] params/_overwrite_with_gradient/decoder/layers/self_attention/query/Fp8DirectDotGeneralOp_0/input_scale Shape: float32[16,1] Logical: P() Physical: () I0419 00:00:20.931858 136124056586048 maxtext_utils.py:1827] params/_overwrite_with_gradient/decoder/layers/self_attention/query/Fp8DirectDotGeneralOp_0/kernel_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0419 00:00:20.931885 136124056586048 maxtext_utils.py:1827] params/_overwrite_with_gradient/decoder/layers/self_attention/query/Fp8DirectDotGeneralOp_0/kernel_scale Shape: float32[16,1] Logical: P() Physical: () I0419 00:00:20.931912 136124056586048 maxtext_utils.py:1827] params/_overwrite_with_gradient/decoder/layers/self_attention/query/Fp8DirectDotGeneralOp_0/output_grad_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0419 00:00:20.931939 136124056586048 maxtext_utils.py:1827] params/_overwrite_with_gradient/decoder/layers/self_attention/query/Fp8DirectDotGeneralOp_0/output_grad_scale Shape: float32[16,1] Logical: P() Physical: () I0419 00:00:20.931966 136124056586048 maxtext_utils.py:1827] params/_overwrite_with_gradient/decoder/layers/self_attention/value/Fp8DirectDotGeneralOp_0/input_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0419 00:00:20.931992 136124056586048 maxtext_utils.py:1827] params/_overwrite_with_gradient/decoder/layers/self_attention/value/Fp8DirectDotGeneralOp_0/input_scale Shape: float32[16,1] Logical: P() Physical: () I0419 00:00:20.932019 136124056586048 maxtext_utils.py:1827] params/_overwrite_with_gradient/decoder/layers/self_attention/value/Fp8DirectDotGeneralOp_0/kernel_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0419 00:00:20.932046 136124056586048 maxtext_utils.py:1827] params/_overwrite_with_gradient/decoder/layers/self_attention/value/Fp8DirectDotGeneralOp_0/kernel_scale Shape: float32[16,1] Logical: P() Physical: () I0419 00:00:20.932072 136124056586048 maxtext_utils.py:1827] params/_overwrite_with_gradient/decoder/layers/self_attention/value/Fp8DirectDotGeneralOp_0/output_grad_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0419 00:00:20.932099 136124056586048 maxtext_utils.py:1827] params/_overwrite_with_gradient/decoder/layers/self_attention/value/Fp8DirectDotGeneralOp_0/output_grad_scale Shape: float32[16,1] Logical: P() Physical: () I0419 00:00:20.932154 136124056586048 maxtext_utils.py:1827] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: P('norm',) Physical: (None,) I0419 00:00:20.932209 136124056586048 maxtext_utils.py:1827] params/params/decoder/layers/mlp/wi_0/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0419 00:00:20.932245 136124056586048 maxtext_utils.py:1827] params/params/decoder/layers/mlp/wi_1/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0419 00:00:20.932291 136124056586048 maxtext_utils.py:1827] params/params/decoder/layers/mlp/wo/kernel Shape: float32[7168,16,2048] Logical: P('mlp', 'layers', 'embed') Physical: (None, None, 'fsdp') I0419 00:00:20.932363 136124056586048 maxtext_utils.py:1827] params/params/decoder/layers/post_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0419 00:00:20.932424 136124056586048 maxtext_utils.py:1827] params/params/decoder/layers/pre_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0419 00:00:20.932482 136124056586048 maxtext_utils.py:1827] params/params/decoder/layers/self_attention/key/kernel Shape: float32[2048,16,16,128] Logical: P('embed', 'layers', 'kv_heads', 'kv_head_dim') Physical: ('fsdp', None, None, None) I0419 00:00:20.932532 136124056586048 maxtext_utils.py:1827] params/params/decoder/layers/self_attention/out/kernel Shape: float32[16,16,128,2048] Logical: P('heads', 'layers', 'kv', 'embed') Physical: (None, None, None, 'fsdp') I0419 00:00:20.932568 136124056586048 maxtext_utils.py:1827] params/params/decoder/layers/self_attention/query/kernel Shape: float32[2048,16,16,128] Logical: P('embed', 'layers', 'q_heads', 'kv') Physical: ('fsdp', None, None, None) I0419 00:00:20.932601 136124056586048 maxtext_utils.py:1827] params/params/decoder/layers/self_attention/value/kernel Shape: float32[2048,16,16,128] Logical: P('embed', 'layers', 'kv_heads', 'kv_head_dim') Physical: ('fsdp', None, None, None) I0419 00:00:20.932644 136124056586048 maxtext_utils.py:1827] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: P('embed_vocab', 'vocab') Physical: ('fsdp', None) I0419 00:00:20.932688 136124056586048 maxtext_utils.py:1827] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: P('vocab', 'embed_vocab') Physical: (None, 'fsdp') I0419 00:00:22.540320 136124056586048 train.py:157] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0419 00:00:22.540416 136124056586048 train.py:157] train/xent Physical: float32[32,2048]............................................ ('fsdp', None). I0419 00:00:22.555640 136124056586048 train.py:164] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0419 00:00:22.555700 136124056586048 train.py:164] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None). I0419 00:00:36.198150 136124056586048 max_utils.py:791] Total memory size: 1.5 GB, Output size: 0.4 GB, Temp size: 1.1 GB, Argument size: 0.4 GB, Host temp size: 0.0 GB. I0419 00:00:36.205785 136124056586048 metric_logger.py:301] number parameters: 1.105 billion I0419 00:00:52.050996 136124056586048 checkpointing.py:794] Waiting for step 0 to finish before checkpoint... I0419 00:00:52.217667 136124056586048 checkpointing.py:798] Waited 0.16665244102478027 seconds for step 0 to finish before starting checkpointing. I0419 00:00:52.220015 136124056586048 checkpoint_manager.py:2009] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0419 00:00:52.222274 136124056586048 checkpoint_manager.py:1512] [process=0] Saving checkpoint at step 0 I0419 00:00:52.223957 136124056586048 event_tracking.py:70] [process=0] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_05_fp8/checkpoints/0. I0419 00:00:52.975077 136124056586048 signaling_client.py:364] Using JaxDistributedSignalingClient I0419 00:00:52.992042 136124056586048 jax_array_handlers.py:360] Scheduling D2H of 81 prioritized jax.Array. I0419 00:00:52.992113 136124056586048 replica_slices.py:424] Transferring arrays to host memory with options: use_replica_parallel=True, min_slice_bytes_for_replica_parallel=None, max_replicas_for_replica_parallel=None, enable_pinned_host_transfer=False I0419 00:00:53.223966 135993151710976 atomicity.py:140] Creating tmp directory gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_05_fp8/checkpoints/0 I0419 00:00:53.402003 136124056586048 base_pytree_checkpoint_handler.py:154] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.410723s I0419 00:00:53.402857 136124056586048 base_pytree_checkpoint_handler.py:130] [process=0] /jax/orbax/write/blocking_gbytes_per_sec: 3.670 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.42032718658447266 s) (per-host) I0419 00:00:53.402926 136124056586048 base_pytree_checkpoint_handler.py:768] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 0.420414s (batch_requests_ready=0.003462s, total_serialization_initiated=0.416178s, others=0.000774s) I0419 00:00:53.403014 136124056586048 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.424353s (all_items=0.000025s, per_item={'items': '0.00002456'}, temp_paths=0.424328) I0419 00:00:53.403801 136124056586048 event_tracking.py:125] [process=0] [async] Finished blocking save in 1.18 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_05_fp8/checkpoints/0. I0419 00:00:53.404103 135993160103680 async_checkpointer.py:76] [process=0][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-19 00:20:53.404069 I0419 00:00:53.408366 136124056586048 checkpoint_manager.py:1560] [process=0][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize I0419 00:00:53.408618 135992623232768 async_checkpointer.py:280] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0419 00:00:53.408738 136124056586048 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_05_fp8/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776556852.2199967, 'wait_for_prev_duration_secs': 6.318092346191406e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776556852.2223577, 'checkpointer_blocking_duration_secs': 1.1818909645080566, 'get_old_steps_start_time': 1776556853.4042757, 'get_old_steps_duration_secs': 2.6702880859375e-05, 'checkpoint_manager_blocking_start_time': 1776556852.2182865, 'checkpoint_manager_blocking_duration_secs': 1.1904125213623047} I0419 00:00:53.408858 136124056586048 checkpointing.py:409] Started an asynchronous checkpoint save for step 0 I0419 00:00:53.408910 136124056586048 max_utils.py:750] Memstats: After params initialized: I0419 00:00:53.408962 136124056586048 max_utils.py:756] Using (GB) 0.45 / 31.25 (1.440000%) on TPU_0(process=0,(0,0,0,0)) I0419 00:00:53.408996 136124056586048 max_utils.py:756] Using (GB) 0.45 / 31.25 (1.440000%) on TPU_1(process=0,(1,0,0,0)) I0419 00:00:53.409023 136124056586048 max_utils.py:756] Using (GB) 0.45 / 31.25 (1.440000%) on TPU_4(process=0,(0,1,0,0)) I0419 00:00:53.409047 136124056586048 max_utils.py:756] Using (GB) 0.45 / 31.25 (1.440000%) on TPU_5(process=0,(1,1,0,0)) I0419 00:00:53.847918 136124056586048 metric_logger.py:196] completed step: 0, seconds: 15.402, TFLOP/s/device: 0.882, Tokens/s/device: 132.968, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52776.805 I0419 00:00:53.851363 136124056586048 metric_logger.py:281] To see full metrics 'tensorboard --logdir=gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_05_fp8/tensorboard/' I0419 00:00:54.679200 136124056586048 metric_logger.py:196] completed step: 1, seconds: 1.794, TFLOP/s/device: 7.572, Tokens/s/device: 1141.331, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52761.840 I0419 00:00:54.835923 136124056586048 metric_logger.py:196] completed step: 2, seconds: 0.832, TFLOP/s/device: 16.322, Tokens/s/device: 2460.255, total_weights: 65536, loss: 10.816, lm_loss: 10.816, perplexity: 49832.398 I0419 00:00:54.994492 136124056586048 metric_logger.py:196] completed step: 3, seconds: 0.009, TFLOP/s/device: 1474.778, Tokens/s/device: 222294.584, total_weights: 65536, loss: 10.431, lm_loss: 10.431, perplexity: 33901.820 I0419 00:00:55.484960 135993134925568 atomicity.py:140] Creating tmp directory gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_05_fp8/checkpoints/0/items I0419 00:00:55.656752 135992595384064 checkpoint.py:188] Wrote Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776556855234335681, 'commit_timestamp_nsecs': None, 'custom_metadata': {}}, json={"item_handlers": null, "metrics": {}, "performance_metrics": {}, "init_timestamp_nsecs": 1776556855234335681, "commit_timestamp_nsecs": null, "custom_metadata": {}} to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_05_fp8/checkpoints/0/_CHECKPOINT_METADATA I0419 00:00:56.677451 2850 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com I0419 00:00:58.417857 135993118140160 array_metadata_store.py:203] [process=0][thread=array_type_handler] Wrote 81 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_05_fp8/checkpoints/0/items/array_metadatas/process_0 I0419 00:01:15.772610 135993109747456 base_pytree_checkpoint_handler.py:1282] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 19.091349s (commit=18.005142s, array_metadata_write=1.086207s) I0419 00:01:15.773767 135993160103680 base_pytree_checkpoint_handler.py:130] [process=0] /jax/orbax/write/gbytes_per_sec: 69.315 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 22.79119849205017 s) (per-host) I0419 00:01:15.773916 135993160103680 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 22.369702s. I0419 00:01:16.897241 135993160103680 checkpoint.py:228] Read Metadata={'item_handlers': None, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776556855234335681, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} from gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_05_fp8/checkpoints/0/_CHECKPOINT_METADATA I0419 00:01:17.616385 135993160103680 array_metadata_store.py:411] [process=0][thread=async_save] Validated ArrayMetadata from all 8 hosts. Time taken: 0.000386s. I0419 00:01:17.643748 135992595384064 checkpoint.py:247] Updated Metadata={'item_handlers': {'items': 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler'}, 'metrics': {}, 'performance_metrics': {}, 'init_timestamp_nsecs': 1776556855234335681, 'commit_timestamp_nsecs': None, 'custom_metadata': {}} to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_05_fp8/checkpoints/0/_CHECKPOINT_METADATA I0419 00:01:18.403394 135993160103680 ocdbt_utils.py:49] Param validation support for Zarr3 will be added later (b/362328389). I0419 00:01:19.093008 136124056586048 metric_logger.py:196] completed step: 4, seconds: 0.158, TFLOP/s/device: 86.151, Tokens/s/device: 12985.613, total_weights: 65536, loss: 9.992, lm_loss: 9.992, perplexity: 21847.639 I0419 00:01:19.102474 136124056586048 metric_logger.py:196] completed step: 5, seconds: 0.159, TFLOP/s/device: 85.666, Tokens/s/device: 12912.582, total_weights: 65536, loss: 9.549, lm_loss: 9.549, perplexity: 14035.006 I0419 00:01:19.259015 136124056586048 metric_logger.py:196] completed step: 6, seconds: 24.098, TFLOP/s/device: 0.564, Tokens/s/device: 84.987, total_weights: 65536, loss: 9.111, lm_loss: 9.111, perplexity: 9058.688 I0419 00:01:19.333610 135993160103680 base_pytree_checkpoint_handler.py:1406] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 2.307844s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_05_fp8/checkpoints/0/items I0419 00:01:19.336764 135993160103680 atomicity.py:666] Finalizing gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_05_fp8/checkpoints/0/items I0419 00:01:19.417618 136124056586048 metric_logger.py:196] completed step: 7, seconds: 0.007, TFLOP/s/device: 1921.800, Tokens/s/device: 289674.682, total_weights: 65536, loss: 8.685, lm_loss: 8.685, perplexity: 5914.720 I0419 00:01:19.576157 136124056586048 metric_logger.py:196] completed step: 8, seconds: 0.158, TFLOP/s/device: 86.130, Tokens/s/device: 12982.403, total_weights: 65536, loss: 8.281, lm_loss: 8.281, perplexity: 3950.010 I0419 00:01:19.583342 136124056586048 checkpointing.py:794] Waiting for step 10 to finish before checkpoint... I0419 00:01:19.848026 135993160103680 atomicity.py:666] Finalizing gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_05_fp8/checkpoints/0 I0419 00:01:19.892530 136124056586048 checkpointing.py:798] Waited 0.30914998054504395 seconds for step 10 to finish before starting checkpointing. I0419 00:01:19.895154 136124056586048 checkpoint_manager.py:2020] [process=0][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0419 00:01:21.628617 135993160103680 atomicity.py:847] [process=0][thread=async_save] Finished saving checkpoint (finalized tmp dir) to `gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_05_fp8/checkpoints/0`. I0419 00:01:21.629465 135993160103680 event_tracking.py:138] [process=0] [async] Finished save (blocking + background) in 29.41 seconds @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_05_fp8/checkpoints/0 I0419 00:01:21.630831 135993160103680 async_checkpointer.py:160] [process=0][thread=async_save] Background save thread done. Time taken: 28.226615s. I0419 00:01:21.631029 135992623232768 async_checkpointer.py:288] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0419 00:01:21.631144 135992623232768 async_checkpointer.py:298] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0419 00:01:21.631222 135992623232768 checkpoint_manager.py:2137] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts... I0419 00:01:21.634120 135992623232768 checkpoint_manager.py:2146] [process=0][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts. I0419 00:01:21.634308 136124056586048 checkpoint_manager.py:2032] [process=0][thread=MainThread][step=0][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=0. W0419 00:01:21.634453 136124056586048 checkpoint_manager.py:1452] Waiting for previous save to complete took 1.739298 seconds. If this number is high, consider checkpointing less frequently. I0419 00:01:21.636401 136124056586048 checkpoint_manager.py:1512] [process=0] Saving checkpoint at step 10 I0419 00:01:21.638432 136124056586048 event_tracking.py:70] [process=0] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_05_fp8/checkpoints/10. I0419 00:01:22.792830 136124056586048 jax_array_handlers.py:360] Scheduling D2H of 81 prioritized jax.Array. I0419 00:01:22.792926 136124056586048 replica_slices.py:424] Transferring arrays to host memory with options: use_replica_parallel=True, min_slice_bytes_for_replica_parallel=None, max_replicas_for_replica_parallel=None, enable_pinned_host_transfer=False I0419 00:01:22.831563 136124056586048 base_pytree_checkpoint_handler.py:154] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.039493s I0419 00:01:22.832247 136124056586048 base_pytree_checkpoint_handler.py:130] [process=0] /jax/orbax/write/blocking_gbytes_per_sec: 32.878 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.04692244529724121 s) (per-host) I0419 00:01:22.832311 136124056586048 base_pytree_checkpoint_handler.py:768] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 0.047003s (batch_requests_ready=0.003130s, total_serialization_initiated=0.043247s, others=0.000626s) I0419 00:01:22.832391 136124056586048 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.051675s (all_items=0.000016s, per_item={'items': '0.00001621'}, temp_paths=0.051658) I0419 00:01:22.833041 136124056586048 event_tracking.py:125] [process=0] [async] Finished blocking save in 1.20 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_05_fp8/checkpoints/10. I0419 00:01:22.833304 135992623232768 async_checkpointer.py:76] [process=0][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-19 00:21:22.833281 I0419 00:01:22.835512 136124056586048 checkpoint_manager.py:1560] [process=0][thread=MainThread][step=10] Starting CheckpointManager Save Finalize thread=save_finalize I0419 00:01:22.835728 135991529338624 async_checkpointer.py:280] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0419 00:01:22.835857 136124056586048 standard_logger.py:34] {'step': 10, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_05_fp8/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776556879.8951218, 'wait_for_prev_duration_secs': 1.73929762840271, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1776556881.6364396, 'checkpointer_blocking_duration_secs': 1.1969621181488037, 'get_old_steps_start_time': 1776556882.8334255, 'get_old_steps_duration_secs': 3.123283386230469e-05, 'checkpoint_manager_blocking_start_time': 1776556879.893316, 'checkpoint_manager_blocking_duration_secs': 2.942505359649658} I0419 00:01:22.835968 136124056586048 checkpointing.py:409] Started an asynchronous checkpoint save for step 10 I0419 00:01:22.836628 136124056586048 metric_logger.py:196] completed step: 9, seconds: 0.159, TFLOP/s/device: 85.598, Tokens/s/device: 12902.250, total_weights: 65536, loss: 7.908, lm_loss: 7.908, perplexity: 2720.116 I0419 00:01:22.846230 136124056586048 metric_logger.py:196] completed step: 10, seconds: 0.158, TFLOP/s/device: 85.733, Tokens/s/device: 12922.685, total_weights: 65536, loss: 7.568, lm_loss: 7.568, perplexity: 1936.198 I0419 00:01:23.001709 136124056586048 metric_logger.py:196] completed step: 11, seconds: 3.261, TFLOP/s/device: 4.167, Tokens/s/device: 628.101, total_weights: 65536, loss: 7.288, lm_loss: 7.288, perplexity: 1463.272 I0419 00:01:23.079850 135993160103680 atomicity.py:140] Creating tmp directory gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_05_fp8/checkpoints/10 I0419 00:01:23.159977 136124056586048 metric_logger.py:196] completed step: 12, seconds: 0.008, TFLOP/s/device: 1638.978, Tokens/s/device: 247044.632, total_weights: 65536, loss: 7.032, lm_loss: 7.032, perplexity: 1132.061 I0419 00:01:23.318403 136124056586048 metric_logger.py:196] completed step: 13, seconds: 0.156, TFLOP/s/device: 86.959, Tokens/s/device: 13107.368, total_weights: 65536, loss: 6.815, lm_loss: 6.815, perplexity: 911.031 I0419 00:01:23.798505 136124056586048 metric_logger.py:196] completed step: 14, seconds: 0.158, TFLOP/s/device: 85.800, Tokens/s/device: 12932.640, total_weights: 65536, loss: 6.635, lm_loss: 6.635, perplexity: 761.080 I0419 00:01:23.955995 136124056586048 metric_logger.py:196] completed step: 15, seconds: 0.631, TFLOP/s/device: 21.542, Tokens/s/device: 3247.036, total_weights: 65536, loss: 6.492, lm_loss: 6.492, perplexity: 659.915 I0419 00:01:24.113746 136124056586048 metric_logger.py:196] completed step: 16, seconds: 0.007, TFLOP/s/device: 2038.579, Tokens/s/device: 307276.819, total_weights: 65536, loss: 6.380, lm_loss: 6.380, perplexity: 589.948 I0419 00:01:24.271909 136124056586048 metric_logger.py:196] completed step: 17, seconds: 0.158, TFLOP/s/device: 85.729, Tokens/s/device: 12921.951, total_weights: 65536, loss: 6.293, lm_loss: 6.293, perplexity: 540.941 I0419 00:01:24.430217 136124056586048 metric_logger.py:196] completed step: 18, seconds: 0.158, TFLOP/s/device: 86.150, Tokens/s/device: 12985.531, total_weights: 65536, loss: 6.223, lm_loss: 6.223, perplexity: 504.108 I0419 00:01:24.587871 136124056586048 checkpointing.py:794] Waiting for step 19 to finish before checkpoint... I0419 00:01:24.588822 136124056586048 checkpointing.py:798] Waited 0.0009610652923583984 seconds for step 19 to finish before starting checkpointing. I0419 00:01:24.591168 136124056586048 checkpoint_manager.py:2020] [process=0][thread=MainThread][step=10][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0419 00:01:25.355386 135993118140160 atomicity.py:140] Creating tmp directory gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_05_fp8/checkpoints/10/items I0419 00:01:29.187909 135991546124032 array_metadata_store.py:203] [process=0][thread=array_type_handler] Wrote 81 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_05_fp8/checkpoints/10/items/array_metadatas/process_0 I0419 00:02:10.818090 135991537731328 base_pytree_checkpoint_handler.py:1282] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 43.870502s (commit=41.776797s, array_metadata_write=2.093705s) I0419 00:02:10.819266 135992623232768 base_pytree_checkpoint_handler.py:130] [process=0] /jax/orbax/write/gbytes_per_sec: 32.888 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 48.033939599990845 s) (per-host) I0419 00:02:10.819339 135992623232768 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 47.985957s. I0419 00:02:12.304153 135992623232768 array_metadata_store.py:411] [process=0][thread=async_save] Validated ArrayMetadata from all 8 hosts. Time taken: 0.000443s. I0419 00:02:12.941265 135992623232768 ocdbt_utils.py:49] Param validation support for Zarr3 will be added later (b/362328389). I0419 00:02:13.873708 135992623232768 base_pytree_checkpoint_handler.py:1406] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 2.179522s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_05_fp8/checkpoints/10/items I0419 00:02:13.875302 135992623232768 atomicity.py:666] Finalizing gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_05_fp8/checkpoints/10/items I0419 00:02:14.404964 135992623232768 atomicity.py:666] Finalizing gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_05_fp8/checkpoints/10 I0419 00:02:15.809937 135992623232768 atomicity.py:847] [process=0][thread=async_save] Finished saving checkpoint (finalized tmp dir) to `gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_05_fp8/checkpoints/10`. I0419 00:02:15.810755 135992623232768 event_tracking.py:138] [process=0] [async] Finished save (blocking + background) in 54.17 seconds @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_05_fp8/checkpoints/10 I0419 00:02:15.812728 135992623232768 async_checkpointer.py:160] [process=0][thread=async_save] Background save thread done. Time taken: 52.979328s. I0419 00:02:15.813030 135991529338624 async_checkpointer.py:288] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0419 00:02:15.813156 135991529338624 async_checkpointer.py:298] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0419 00:02:15.813221 135991529338624 checkpoint_manager.py:2137] [process=0][thread=save_finalize][step=10] CheckpointManager Save Finalize is syncing with other hosts... I0419 00:02:15.814664 135991529338624 checkpoint_manager.py:2146] [process=0][thread=save_finalize][step=10] CheckpointManager Save Finalize is done on all hosts. I0419 00:02:15.814839 136124056586048 checkpoint_manager.py:2032] [process=0][thread=MainThread][step=10][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=10. W0419 00:02:15.814965 136124056586048 checkpoint_manager.py:1452] Waiting for previous save to complete took 51.223801 seconds. If this number is high, consider checkpointing less frequently. I0419 00:02:15.816491 136124056586048 checkpoint_manager.py:1512] [process=0] Saving checkpoint at step 19 I0419 00:02:15.818654 136124056586048 event_tracking.py:70] [process=0] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_05_fp8/checkpoints/19. I0419 00:02:16.524350 136124056586048 jax_array_handlers.py:360] Scheduling D2H of 81 prioritized jax.Array. I0419 00:02:16.524449 136124056586048 replica_slices.py:424] Transferring arrays to host memory with options: use_replica_parallel=True, min_slice_bytes_for_replica_parallel=None, max_replicas_for_replica_parallel=None, enable_pinned_host_transfer=False I0419 00:02:16.573215 136124056586048 base_pytree_checkpoint_handler.py:154] [process=0][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.049611s I0419 00:02:16.574038 136124056586048 base_pytree_checkpoint_handler.py:130] [process=0] /jax/orbax/write/blocking_gbytes_per_sec: 26.919 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.05730938911437988 s) (per-host) I0419 00:02:16.574113 136124056586048 base_pytree_checkpoint_handler.py:768] [process=0][thread=MainThread] Initiated Pytree async_save. Time taken: 0.057405s (batch_requests_ready=0.003264s, total_serialization_initiated=0.053393s, others=0.000747s) I0419 00:02:16.574206 136124056586048 composite_checkpoint_handler.py:715] [process=0][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.061242s (all_items=0.000010s, per_item={'items': '0.00001025'}, temp_paths=0.061232) I0419 00:02:16.574965 136124056586048 event_tracking.py:125] [process=0] [async] Finished blocking save in 0.76 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_05_fp8/checkpoints/19. I0419 00:02:16.575259 135991529338624 async_checkpointer.py:76] [process=0][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-19 00:22:16.575225 I0419 00:02:16.577251 136124056586048 checkpoint_manager.py:1560] [process=0][thread=MainThread][step=19] Starting CheckpointManager Save Finalize thread=save_finalize I0419 00:02:16.577550 135993109747456 async_checkpointer.py:280] [process=0][thread=save_finalize] Waiting for background save thread=async_save. I0419 00:02:16.577717 136124056586048 standard_logger.py:34] {'step': 19, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_05_fp8/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1776556884.5911372, 'wait_for_prev_duration_secs': 51.22380089759827, 'time_between_consecutive_saves_sec': 2.9569735527038574, 'checkpointer_blocking_start_time': 1776556935.8165295, 'checkpointer_blocking_duration_secs': 0.758864164352417, 'get_old_steps_start_time': 1776556936.5754118, 'get_old_steps_duration_secs': 2.3365020751953125e-05, 'checkpoint_manager_blocking_start_time': 1776556884.589101, 'checkpoint_manager_blocking_duration_secs': 51.988582611083984} I0419 00:02:16.577846 136124056586048 checkpointing.py:409] Started an asynchronous checkpoint save for step 19 I0419 00:02:16.577893 136124056586048 checkpoint_manager.py:2020] [process=0][thread=MainThread][step=19][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0419 00:02:16.758267 135992623232768 atomicity.py:140] Creating tmp directory gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_05_fp8/checkpoints/19 I0419 00:02:20.197003 135994180515584 atomicity.py:140] Creating tmp directory gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_05_fp8/checkpoints/19/items I0419 00:02:23.964397 135993126532864 array_metadata_store.py:203] [process=0][thread=array_type_handler] Wrote 81 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_05_fp8/checkpoints/19/items/array_metadatas/process_0 I0419 00:03:04.231483 135993118140160 base_pytree_checkpoint_handler.py:1282] [process=0][thread=write_metadata_after_commits] Commit + Array metadata written. Time taken: 42.469516s (commit=41.368706s, array_metadata_write=1.100810s) I0419 00:03:04.232668 135991529338624 base_pytree_checkpoint_handler.py:130] [process=0] /jax/orbax/write/gbytes_per_sec: 33.108 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 47.71589922904968 s) (per-host) I0419 00:03:04.232824 135991529338624 async_checkpointer.py:90] [process=0][thread=async_save] 3 Handler Commit operations completed. Time taken: 47.657440s. I0419 00:03:06.291189 135991529338624 ocdbt_utils.py:49] Param validation support for Zarr3 will be added later (b/362328389). I0419 00:03:06.447616 135991529338624 array_metadata_store.py:411] [process=0][thread=async_save] Validated ArrayMetadata from all 8 hosts. Time taken: 0.000481s. I0419 00:03:07.379381 135991529338624 base_pytree_checkpoint_handler.py:1406] [process=0][thread=async_save] Pytree save finalize (merge_ocdbt + ArrayMetadata validation) completed. Time taken: 1.919780s. use_zarr3=True, enable_post_merge_validation=True, directory=gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_05_fp8/checkpoints/19/items I0419 00:03:07.381223 135991529338624 atomicity.py:666] Finalizing gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_05_fp8/checkpoints/19/items I0419 00:03:07.901587 135991529338624 atomicity.py:666] Finalizing gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_05_fp8/checkpoints/19 I0419 00:03:10.123953 135991529338624 atomicity.py:847] [process=0][thread=async_save] Finished saving checkpoint (finalized tmp dir) to `gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_05_fp8/checkpoints/19`. I0419 00:03:10.124751 135991529338624 event_tracking.py:138] [process=0] [async] Finished save (blocking + background) in 54.31 seconds @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260418_235007/linen_xpk_feat_nnx_post_train_fixes_20260418_235007_05_fp8/checkpoints/19 I0419 00:03:10.126140 135991529338624 async_checkpointer.py:160] [process=0][thread=async_save] Background save thread done. Time taken: 53.550766s. I0419 00:03:10.126355 135993109747456 async_checkpointer.py:288] [process=0][thread=save_finalize] Done with waiting for background save thread=async_save. I0419 00:03:10.126480 135993109747456 async_checkpointer.py:298] [process=0][thread=save_finalize] No errors found in background save thread=async_save. I0419 00:03:10.126533 135993109747456 checkpoint_manager.py:2137] [process=0][thread=save_finalize][step=19] CheckpointManager Save Finalize is syncing with other hosts... I0419 00:03:10.129084 135993109747456 checkpoint_manager.py:2146] [process=0][thread=save_finalize][step=19] CheckpointManager Save Finalize is done on all hosts. I0419 00:03:10.129254 136124056586048 checkpoint_manager.py:2032] [process=0][thread=MainThread][step=19][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=19. I0419 00:03:10.129409 136124056586048 checkpoint_manager.py:2009] [process=0][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0419 00:03:10.130463 136124056586048 metric_logger.py:196] completed step: 19, seconds: 0.158, TFLOP/s/device: 85.736, Tokens/s/device: 12923.011, total_weights: 65536, loss: 6.167, lm_loss: 6.167, perplexity: 476.947 Per train step: Total TFLOPs: 13.59 split as 93.93% learnable weight flops and 6.07% attention flops XPK End: Sun Apr 19 00:03:22 UTC 2026 EXIT_CODE=0