feat/nnx-post-train-fixes| Metric | Linen 28b1e4a16 | NNX 28b1e4a16 | Diff (NNX − Linen) |
|---|---|---|---|
| Parameters | 1.105 billion | 1.104 billion | — |
| Final loss | 6.1670 | 10.5490 | +4.382 |
| TFLOP/s | 85.801 | 86.725 | +0.924 |
| Tok/s | 12932.8 | 13072.1 | +139.342 |
| Avg s/step | 1.556 | 1.712 | +0.156 |
| Memory % | 1.44 | 1.44 | 0 |
| JAX | 0.9.2 | 0.9.2 | — |
Diff = NNX value − Linen value. Green = NNX improved. Red = NNX regressed.
XPK Start: Fri Apr 24 12:16:28 UTC 2026 PyTorch was not found. Models won't be available and only tokenizers, configuration and file/data utilities can be used. `rope_parameters`'s factor field must be a float >= 1, got 40 `rope_parameters`'s beta_fast field must be a float, got 32 `rope_parameters`'s beta_slow field must be a float, got 1 DeepseekV32Config got `key=rope_scaling` in kwargs but hasn't set it as attribute. For RoPE standardization you need to set `self.rope_parameters` in model's config. 2026-04-24 12:16:53.763285: 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) I0424 12:16:53.975189 140376034367296 max_utils.py:273] Attempting to initialize the jax distributed system... I0424 12:17:03.014965 140376034367296 distributed.py:149] Starting JAX distributed service on [::]:8482 I0424 12:17:03.017277 140376034367296 distributed.py:172] Connecting to JAX distributed service on mt-05-fp8-61476-slice-job-0-0.mt-05-fp8-61476:8482 I0424 12:17:04.145262 140376034367296 max_utils.py:284] Jax distributed system initialized! I0424 12:17:10.211407 140376034367296 max_utils.py:800] System Information: Jax Version: 0.9.2 I0424 12:17:10.211511 140376034367296 max_utils.py:801] System Information: Jaxlib Version: 0.9.2 I0424 12:17:10.211553 140376034367296 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 I0424 12:17:10.211591 140376034367296 train_utils.py:391] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing. I0424 12:17:10.903529 140376034367296 maxtext_utils.py:1771] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0424 12:17:10.904124 140376034367296 maxtext_utils.py:1771] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0424 12:17:10.904312 140376034367296 checkpointing.py:688] Setting up checkpoint logger... I0424 12:17:10.904362 140376034367296 checkpointing.py:234] Creating checkpoint manager with ocdbt=True and zarr3=True I0424 12:17:10.904404 140376034367296 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None I0424 12:17:10.904754 140376034367296 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 0x7fab1fe50260>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0424 12:17:13.761442 140376034367296 checkpointing.py:266] Enabling policy for fixed interval checkpointing. I0424 12:17:13.761698 140376034367296 checkpoint_manager.py:708] [process=5][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=('items',), item_handlers={'items': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7f9714152b40>}, handler_registry=None I0424 12:17:13.761936 140376034367296 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7f9714152b40>` 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`. I0424 12:17:13.761986 140376034367296 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7f9714156660>` 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`. I0424 12:17:13.762022 140376034367296 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 0x7f9714152b40>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7f9714152b40>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7f9714156660>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7f9714156660>}). I0424 12:17:13.762349 140376034367296 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34 I0424 12:17:13.762420 140376034367296 async_checkpointer.py:192] [process=5][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7f96e46d9d00> timeout: 1200 secs and primary_host=0 for async checkpoint writes I0424 12:17:14.944969 140376034367296 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260424_120657/linen_xpk_feat_nnx_post_train_fixes_20260424_120657_05_fp8/checkpoints I0424 12:17:14.958698 140376034367296 checkpoint_manager.py:929] [process=5][thread=MainThread] CheckpointManager created, primary_host=0, CheckpointManagerOptions=CheckpointManagerOptions(save_interval_steps=1, max_to_keep=None, keep_time_interval=None, keep_period=None, should_keep_fn=None, best_fn=None, best_mode='max', keep_checkpoints_without_metrics=True, step_prefix=None, step_format_fixed_length=None, step_name_format=None, create=True, cleanup_tmp_directories=False, save_on_steps=frozenset(), single_host_load_and_broadcast=False, todelete_subdir=None, todelete_full_path=None, enable_background_delete=False, read_only=False, enable_async_checkpointing=True, async_options=None, multiprocessing_options=MultiprocessingOptions(primary_host=0, active_processes=None, barrier_sync_key_prefix=None), should_save_fn=None, file_options=FileOptions(path_permission_mode=None), save_root_metadata=True, temporary_path_class=None, save_decision_policy=FixedIntervalPolicy(interval=10), preservation_policy=LatestN(n=None), prevent_write_metrics=False, enable_should_save_is_saving_in_progress_check=True, enable_per_process_directory_creation=False, lightweight_initialize=False), root_directory=gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260424_120657/linen_xpk_feat_nnx_post_train_fixes_20260424_120657_05_fp8/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7f9714151d90> I0424 12:17:14.958819 140376034367296 checkpointing.py:302] Checkpoint manager created! I0424 12:17:16.066483 140376034367296 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0424 12:17:16.066597 140376034367296 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0424 12:17:16.458107 140376034367296 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0424 12:17:16.458204 140376034367296 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0424 12:17:16.474601 140376034367296 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0424 12:17:16.474674 140376034367296 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0424 12:17:16.528078 140376034367296 attentions.py:1154] attentions/query Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0424 12:17:16.528162 140376034367296 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0424 12:17:16.544703 140376034367296 attentions.py:1155] attentions/key Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0424 12:17:16.544764 140376034367296 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0424 12:17:16.561281 140376034367296 attentions.py:1156] attentions/value Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0424 12:17:16.561344 140376034367296 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0424 12:17:16.586983 140376034367296 attentions.py:1198] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv'). I0424 12:17:16.587057 140376034367296 attentions.py:1198] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0424 12:17:16.638693 140376034367296 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0424 12:17:16.638770 140376034367296 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None). I0424 12:17:17.094209 140376034367296 checkpointing.py:578] checkpoint manager exists so trying to load this run's existing checkpoint I0424 12:17:17.094343 140376034367296 checkpointing.py:676] No existing checkpoints found, not restoring checkpoint. fsdp: 32 I0424 12:17:19.025958 140376034367296 maxtext_utils.py:1880] params/_overwrite_with_gradient/decoder/layers/mlp/wi_0/Fp8DirectDotGeneralOp_0/input_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0424 12:17:19.026085 140376034367296 maxtext_utils.py:1880] params/_overwrite_with_gradient/decoder/layers/mlp/wi_0/Fp8DirectDotGeneralOp_0/input_scale Shape: float32[16,1] Logical: P() Physical: () I0424 12:17:19.026134 140376034367296 maxtext_utils.py:1880] params/_overwrite_with_gradient/decoder/layers/mlp/wi_0/Fp8DirectDotGeneralOp_0/kernel_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0424 12:17:19.026174 140376034367296 maxtext_utils.py:1880] params/_overwrite_with_gradient/decoder/layers/mlp/wi_0/Fp8DirectDotGeneralOp_0/kernel_scale Shape: float32[16,1] Logical: P() Physical: () I0424 12:17:19.026210 140376034367296 maxtext_utils.py:1880] params/_overwrite_with_gradient/decoder/layers/mlp/wi_0/Fp8DirectDotGeneralOp_0/output_grad_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0424 12:17:19.026244 140376034367296 maxtext_utils.py:1880] params/_overwrite_with_gradient/decoder/layers/mlp/wi_0/Fp8DirectDotGeneralOp_0/output_grad_scale Shape: float32[16,1] Logical: P() Physical: () I0424 12:17:19.026276 140376034367296 maxtext_utils.py:1880] params/_overwrite_with_gradient/decoder/layers/mlp/wi_1/Fp8DirectDotGeneralOp_0/input_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0424 12:17:19.026308 140376034367296 maxtext_utils.py:1880] params/_overwrite_with_gradient/decoder/layers/mlp/wi_1/Fp8DirectDotGeneralOp_0/input_scale Shape: float32[16,1] Logical: P() Physical: () I0424 12:17:19.026338 140376034367296 maxtext_utils.py:1880] params/_overwrite_with_gradient/decoder/layers/mlp/wi_1/Fp8DirectDotGeneralOp_0/kernel_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0424 12:17:19.026368 140376034367296 maxtext_utils.py:1880] params/_overwrite_with_gradient/decoder/layers/mlp/wi_1/Fp8DirectDotGeneralOp_0/kernel_scale Shape: float32[16,1] Logical: P() Physical: () I0424 12:17:19.026398 140376034367296 maxtext_utils.py:1880] params/_overwrite_with_gradient/decoder/layers/mlp/wi_1/Fp8DirectDotGeneralOp_0/output_grad_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0424 12:17:19.026429 140376034367296 maxtext_utils.py:1880] params/_overwrite_with_gradient/decoder/layers/mlp/wi_1/Fp8DirectDotGeneralOp_0/output_grad_scale Shape: float32[16,1] Logical: P() Physical: () I0424 12:17:19.026459 140376034367296 maxtext_utils.py:1880] params/_overwrite_with_gradient/decoder/layers/mlp/wo/Fp8DirectDotGeneralOp_0/input_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0424 12:17:19.026488 140376034367296 maxtext_utils.py:1880] params/_overwrite_with_gradient/decoder/layers/mlp/wo/Fp8DirectDotGeneralOp_0/input_scale Shape: float32[16,1] Logical: P() Physical: () I0424 12:17:19.026517 140376034367296 maxtext_utils.py:1880] params/_overwrite_with_gradient/decoder/layers/mlp/wo/Fp8DirectDotGeneralOp_0/kernel_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0424 12:17:19.026545 140376034367296 maxtext_utils.py:1880] params/_overwrite_with_gradient/decoder/layers/mlp/wo/Fp8DirectDotGeneralOp_0/kernel_scale Shape: float32[16,1] Logical: P() Physical: () I0424 12:17:19.026574 140376034367296 maxtext_utils.py:1880] params/_overwrite_with_gradient/decoder/layers/mlp/wo/Fp8DirectDotGeneralOp_0/output_grad_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0424 12:17:19.026602 140376034367296 maxtext_utils.py:1880] params/_overwrite_with_gradient/decoder/layers/mlp/wo/Fp8DirectDotGeneralOp_0/output_grad_scale Shape: float32[16,1] Logical: P() Physical: () I0424 12:17:19.026630 140376034367296 maxtext_utils.py:1880] params/_overwrite_with_gradient/decoder/layers/self_attention/key/Fp8DirectDotGeneralOp_0/input_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0424 12:17:19.026669 140376034367296 maxtext_utils.py:1880] params/_overwrite_with_gradient/decoder/layers/self_attention/key/Fp8DirectDotGeneralOp_0/input_scale Shape: float32[16,1] Logical: P() Physical: () I0424 12:17:19.026699 140376034367296 maxtext_utils.py:1880] params/_overwrite_with_gradient/decoder/layers/self_attention/key/Fp8DirectDotGeneralOp_0/kernel_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0424 12:17:19.026727 140376034367296 maxtext_utils.py:1880] params/_overwrite_with_gradient/decoder/layers/self_attention/key/Fp8DirectDotGeneralOp_0/kernel_scale Shape: float32[16,1] Logical: P() Physical: () I0424 12:17:19.026755 140376034367296 maxtext_utils.py:1880] params/_overwrite_with_gradient/decoder/layers/self_attention/key/Fp8DirectDotGeneralOp_0/output_grad_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0424 12:17:19.026782 140376034367296 maxtext_utils.py:1880] params/_overwrite_with_gradient/decoder/layers/self_attention/key/Fp8DirectDotGeneralOp_0/output_grad_scale Shape: float32[16,1] Logical: P() Physical: () I0424 12:17:19.026809 140376034367296 maxtext_utils.py:1880] params/_overwrite_with_gradient/decoder/layers/self_attention/out/Fp8DirectDotGeneralOp_0/input_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0424 12:17:19.026837 140376034367296 maxtext_utils.py:1880] params/_overwrite_with_gradient/decoder/layers/self_attention/out/Fp8DirectDotGeneralOp_0/input_scale Shape: float32[16,1] Logical: P() Physical: () I0424 12:17:19.026867 140376034367296 maxtext_utils.py:1880] params/_overwrite_with_gradient/decoder/layers/self_attention/out/Fp8DirectDotGeneralOp_0/kernel_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0424 12:17:19.026897 140376034367296 maxtext_utils.py:1880] params/_overwrite_with_gradient/decoder/layers/self_attention/out/Fp8DirectDotGeneralOp_0/kernel_scale Shape: float32[16,1] Logical: P() Physical: () I0424 12:17:19.026924 140376034367296 maxtext_utils.py:1880] params/_overwrite_with_gradient/decoder/layers/self_attention/out/Fp8DirectDotGeneralOp_0/output_grad_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0424 12:17:19.026951 140376034367296 maxtext_utils.py:1880] params/_overwrite_with_gradient/decoder/layers/self_attention/out/Fp8DirectDotGeneralOp_0/output_grad_scale Shape: float32[16,1] Logical: P() Physical: () I0424 12:17:19.026982 140376034367296 maxtext_utils.py:1880] params/_overwrite_with_gradient/decoder/layers/self_attention/query/Fp8DirectDotGeneralOp_0/input_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0424 12:17:19.027011 140376034367296 maxtext_utils.py:1880] params/_overwrite_with_gradient/decoder/layers/self_attention/query/Fp8DirectDotGeneralOp_0/input_scale Shape: float32[16,1] Logical: P() Physical: () I0424 12:17:19.027040 140376034367296 maxtext_utils.py:1880] params/_overwrite_with_gradient/decoder/layers/self_attention/query/Fp8DirectDotGeneralOp_0/kernel_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0424 12:17:19.027067 140376034367296 maxtext_utils.py:1880] params/_overwrite_with_gradient/decoder/layers/self_attention/query/Fp8DirectDotGeneralOp_0/kernel_scale Shape: float32[16,1] Logical: P() Physical: () I0424 12:17:19.027094 140376034367296 maxtext_utils.py:1880] params/_overwrite_with_gradient/decoder/layers/self_attention/query/Fp8DirectDotGeneralOp_0/output_grad_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0424 12:17:19.027122 140376034367296 maxtext_utils.py:1880] params/_overwrite_with_gradient/decoder/layers/self_attention/query/Fp8DirectDotGeneralOp_0/output_grad_scale Shape: float32[16,1] Logical: P() Physical: () I0424 12:17:19.027149 140376034367296 maxtext_utils.py:1880] params/_overwrite_with_gradient/decoder/layers/self_attention/value/Fp8DirectDotGeneralOp_0/input_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0424 12:17:19.027176 140376034367296 maxtext_utils.py:1880] params/_overwrite_with_gradient/decoder/layers/self_attention/value/Fp8DirectDotGeneralOp_0/input_scale Shape: float32[16,1] Logical: P() Physical: () I0424 12:17:19.027206 140376034367296 maxtext_utils.py:1880] params/_overwrite_with_gradient/decoder/layers/self_attention/value/Fp8DirectDotGeneralOp_0/kernel_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0424 12:17:19.027238 140376034367296 maxtext_utils.py:1880] params/_overwrite_with_gradient/decoder/layers/self_attention/value/Fp8DirectDotGeneralOp_0/kernel_scale Shape: float32[16,1] Logical: P() Physical: () I0424 12:17:19.027275 140376034367296 maxtext_utils.py:1880] params/_overwrite_with_gradient/decoder/layers/self_attention/value/Fp8DirectDotGeneralOp_0/output_grad_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0424 12:17:19.027303 140376034367296 maxtext_utils.py:1880] params/_overwrite_with_gradient/decoder/layers/self_attention/value/Fp8DirectDotGeneralOp_0/output_grad_scale Shape: float32[16,1] Logical: P() Physical: () I0424 12:17:19.027354 140376034367296 maxtext_utils.py:1880] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: P('norm',) Physical: (None,) I0424 12:17:19.027410 140376034367296 maxtext_utils.py:1880] params/params/decoder/layers/mlp/wi_0/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0424 12:17:19.027449 140376034367296 maxtext_utils.py:1880] params/params/decoder/layers/mlp/wi_1/kernel Shape: float32[2048,16,7168] Logical: P('embed', 'layers', 'mlp') Physical: ('fsdp', None, None) I0424 12:17:19.027496 140376034367296 maxtext_utils.py:1880] params/params/decoder/layers/mlp/wo/kernel Shape: float32[7168,16,2048] Logical: P('mlp', 'layers', 'embed') Physical: (None, None, 'fsdp') I0424 12:17:19.027543 140376034367296 maxtext_utils.py:1880] params/params/decoder/layers/post_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0424 12:17:19.027577 140376034367296 maxtext_utils.py:1880] params/params/decoder/layers/pre_self_attention_layer_norm/scale Shape: float32[2048,16] Logical: P('norm', 'layers') Physical: (None, None) I0424 12:17:19.027629 140376034367296 maxtext_utils.py:1880] params/params/decoder/layers/self_attention/key/kernel Shape: float32[2048,16,16,128] Logical: P('embed', 'layers', 'kv_heads', 'kv_head_dim') Physical: ('fsdp', None, None, None) I0424 12:17:19.027687 140376034367296 maxtext_utils.py:1880] params/params/decoder/layers/self_attention/out/kernel Shape: float32[16,16,128,2048] Logical: P('heads', 'layers', 'kv', 'embed') Physical: (None, None, None, 'fsdp') I0424 12:17:19.027801 140376034367296 maxtext_utils.py:1880] params/params/decoder/layers/self_attention/query/kernel Shape: float32[2048,16,16,128] Logical: P('embed', 'layers', 'q_heads', 'kv') Physical: ('fsdp', None, None, None) I0424 12:17:19.027873 140376034367296 maxtext_utils.py:1880] params/params/decoder/layers/self_attention/value/kernel Shape: float32[2048,16,16,128] Logical: P('embed', 'layers', 'kv_heads', 'kv_head_dim') Physical: ('fsdp', None, None, None) I0424 12:17:19.027938 140376034367296 maxtext_utils.py:1880] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: P('embed_vocab', 'vocab') Physical: ('fsdp', None) I0424 12:17:19.027994 140376034367296 maxtext_utils.py:1880] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: P('vocab', 'embed_vocab') Physical: (None, 'fsdp') I0424 12:17:20.678012 140376034367296 train.py:157] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0424 12:17:20.678107 140376034367296 train.py:157] train/xent Physical: float32[32,2048]............................................ ('fsdp', None). I0424 12:17:20.693700 140376034367296 train.py:164] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0424 12:17:20.693762 140376034367296 train.py:164] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None). I0424 12:17:34.585335 140376034367296 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. I0424 12:17:34.586235 140376034367296 metric_logger.py:301] number parameters: 1.105 billion I0424 12:17:49.968454 140376034367296 checkpointing.py:794] Waiting for step 0 to finish before checkpoint... I0424 12:17:50.137439 140376034367296 checkpointing.py:798] Waited 0.16896438598632812 seconds for step 0 to finish before starting checkpointing. I0424 12:17:50.139813 140376034367296 checkpoint_manager.py:2009] [process=5][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0424 12:17:50.142043 140376034367296 checkpoint_manager.py:1512] [process=5] Saving checkpoint at step 0 I0424 12:17:50.143357 140376034367296 event_tracking.py:70] [process=5] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260424_120657/linen_xpk_feat_nnx_post_train_fixes_20260424_120657_05_fp8/checkpoints/0. I0424 12:17:50.440450 140376034367296 signaling_client.py:364] Using JaxDistributedSignalingClient I0424 12:17:50.441382 140376034367296 jax_array_handlers.py:360] Scheduling D2H of 81 prioritized jax.Array. I0424 12:17:50.441443 140376034367296 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 I0424 12:17:50.859109 140376034367296 base_pytree_checkpoint_handler.py:154] [process=5][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.418779s I0424 12:17:50.859286 140376034367296 base_pytree_checkpoint_handler.py:130] [process=5] /jax/orbax/write/blocking_gbytes_per_sec: 3.611 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.42723655700683594 s) (per-host) I0424 12:17:50.859340 140376034367296 base_pytree_checkpoint_handler.py:768] [process=5][thread=MainThread] Initiated Pytree async_save. Time taken: 0.427300s (batch_requests_ready=0.003644s, total_serialization_initiated=0.423584s, others=0.000072s) I0424 12:17:50.859442 140376034367296 composite_checkpoint_handler.py:715] [process=5][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.431491s (all_items=0.000018s, per_item={'items': '0.00001836'}, temp_paths=0.431473) I0424 12:17:50.860218 140376034367296 event_tracking.py:125] [process=5] [async] Finished blocking save in 0.72 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260424_120657/linen_xpk_feat_nnx_post_train_fixes_20260424_120657_05_fp8/checkpoints/0. I0424 12:17:50.860559 140245641377536 async_checkpointer.py:76] [process=5][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-24 12:37:50.860520 I0424 12:17:50.871478 140376034367296 checkpoint_manager.py:1560] [process=5][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize I0424 12:17:50.871764 140244554131200 async_checkpointer.py:280] [process=5][thread=save_finalize] Waiting for background save thread=async_save. I0424 12:17:50.871922 140376034367296 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260424_120657/linen_xpk_feat_nnx_post_train_fixes_20260424_120657_05_fp8/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1777033070.1397955, 'wait_for_prev_duration_secs': 6.151199340820312e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1777033070.142081, 'checkpointer_blocking_duration_secs': 0.718651533126831, 'get_old_steps_start_time': 1777033070.8607578, 'get_old_steps_duration_secs': 3.0994415283203125e-05, 'checkpoint_manager_blocking_start_time': 1777033070.1380253, 'checkpoint_manager_blocking_duration_secs': 0.7338583469390869} I0424 12:17:50.872032 140376034367296 checkpointing.py:409] Started an asynchronous checkpoint save for step 0 I0424 12:17:50.872084 140376034367296 max_utils.py:750] Memstats: After params initialized: I0424 12:17:50.872144 140376034367296 max_utils.py:756] Using (GB) 0.45 / 31.25 (1.440000%) on TPU_18(process=5,(2,4,0,0)) I0424 12:17:50.872178 140376034367296 max_utils.py:756] Using (GB) 0.45 / 31.25 (1.440000%) on TPU_19(process=5,(3,4,0,0)) I0424 12:17:50.872205 140376034367296 max_utils.py:756] Using (GB) 0.45 / 31.25 (1.440000%) on TPU_22(process=5,(2,5,0,0)) I0424 12:17:50.872231 140376034367296 max_utils.py:756] Using (GB) 0.45 / 31.25 (1.440000%) on TPU_23(process=5,(3,5,0,0)) I0424 12:17:51.191282 140376034367296 metric_logger.py:196] completed step: 0, seconds: 15.382, TFLOP/s/device: 0.883, Tokens/s/device: 133.142, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52776.805 I0424 12:17:51.369549 140376034367296 metric_logger.py:196] completed step: 1, seconds: 1.221, TFLOP/s/device: 11.130, Tokens/s/device: 1677.690, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52761.840 I0424 12:17:51.792848 140376034367296 metric_logger.py:196] completed step: 2, seconds: 0.021, TFLOP/s/device: 659.697, Tokens/s/device: 99436.784, total_weights: 65536, loss: 10.816, lm_loss: 10.816, perplexity: 49832.398 I0424 12:17:51.951389 140376034367296 metric_logger.py:196] completed step: 3, seconds: 0.424, TFLOP/s/device: 32.074, Tokens/s/device: 4834.579, total_weights: 65536, loss: 10.431, lm_loss: 10.431, perplexity: 33901.820 I0424 12:17:52.269312 140376034367296 metric_logger.py:196] completed step: 4, seconds: 0.165, TFLOP/s/device: 82.429, Tokens/s/device: 12424.621, total_weights: 65536, loss: 9.992, lm_loss: 9.992, perplexity: 21847.639 I0424 12:17:52.275812 140376034367296 metric_logger.py:196] completed step: 5, seconds: 0.158, TFLOP/s/device: 86.073, Tokens/s/device: 12973.932, total_weights: 65536, loss: 9.549, lm_loss: 9.549, perplexity: 14035.006 I0424 12:17:53.586794 2873 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com I0424 12:17:55.636734 140244562523904 array_metadata_store.py:203] [process=5][thread=array_type_handler] Wrote 81 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260424_120657/linen_xpk_feat_nnx_post_train_fixes_20260424_120657_05_fp8/checkpoints/0/items/array_metadatas/process_5 I0424 12:18:10.342522 140245641377536 base_pytree_checkpoint_handler.py:130] [process=5] /jax/orbax/write/gbytes_per_sec: 79.343 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 19.910431623458862 s) (per-host) I0424 12:18:10.342647 140245641377536 async_checkpointer.py:90] [process=5][thread=async_save] 3 Handler Commit operations completed. Time taken: 19.481962s. I0424 12:18:16.237839 140376034367296 metric_logger.py:196] completed step: 6, seconds: 0.318, TFLOP/s/device: 42.671, Tokens/s/device: 6431.898, total_weights: 65536, loss: 9.111, lm_loss: 9.111, perplexity: 9058.688 I0424 12:18:16.396387 140376034367296 metric_logger.py:196] completed step: 7, seconds: 23.805, TFLOP/s/device: 0.571, Tokens/s/device: 86.032, total_weights: 65536, loss: 8.685, lm_loss: 8.685, perplexity: 5914.720 I0424 12:18:16.554781 140376034367296 metric_logger.py:196] completed step: 8, seconds: 0.163, TFLOP/s/device: 83.163, Tokens/s/device: 12535.271, total_weights: 65536, loss: 8.281, lm_loss: 8.281, perplexity: 3950.010 I0424 12:18:16.559922 140376034367296 checkpointing.py:794] Waiting for step 10 to finish before checkpoint... I0424 12:18:16.871288 140376034367296 checkpointing.py:798] Waited 0.31133246421813965 seconds for step 10 to finish before starting checkpointing. I0424 12:18:16.874119 140376034367296 checkpoint_manager.py:2020] [process=5][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0424 12:18:17.415823 140245641377536 async_checkpointer.py:160] [process=5][thread=async_save] Background save thread done. Time taken: 26.555120s. I0424 12:18:17.416113 140244554131200 async_checkpointer.py:288] [process=5][thread=save_finalize] Done with waiting for background save thread=async_save. I0424 12:18:17.416243 140244554131200 async_checkpointer.py:298] [process=5][thread=save_finalize] No errors found in background save thread=async_save. I0424 12:18:17.416296 140244554131200 checkpoint_manager.py:2137] [process=5][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts... I0424 12:18:17.418941 140244554131200 checkpoint_manager.py:2146] [process=5][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts. I0424 12:18:17.419107 140376034367296 checkpoint_manager.py:2032] [process=5][thread=MainThread][step=0][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=0. I0424 12:18:17.420604 140376034367296 checkpoint_manager.py:1512] [process=5] Saving checkpoint at step 10 I0424 12:18:17.422720 140376034367296 event_tracking.py:70] [process=5] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260424_120657/linen_xpk_feat_nnx_post_train_fixes_20260424_120657_05_fp8/checkpoints/10. I0424 12:18:17.958214 140376034367296 jax_array_handlers.py:360] Scheduling D2H of 81 prioritized jax.Array. I0424 12:18:17.958308 140376034367296 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 I0424 12:18:17.996158 140376034367296 base_pytree_checkpoint_handler.py:154] [process=5][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.038923s I0424 12:18:17.996340 140376034367296 base_pytree_checkpoint_handler.py:130] [process=5] /jax/orbax/write/blocking_gbytes_per_sec: 34.060 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.045295000076293945 s) (per-host) I0424 12:18:17.996392 140376034367296 base_pytree_checkpoint_handler.py:768] [process=5][thread=MainThread] Initiated Pytree async_save. Time taken: 0.045358s (batch_requests_ready=0.003282s, total_serialization_initiated=0.042005s, others=0.000071s) I0424 12:18:17.996499 140376034367296 composite_checkpoint_handler.py:715] [process=5][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.050257s (all_items=0.000023s, per_item={'items': '0.00002289'}, temp_paths=0.050234) I0424 12:18:17.997124 140376034367296 event_tracking.py:125] [process=5] [async] Finished blocking save in 0.58 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260424_120657/linen_xpk_feat_nnx_post_train_fixes_20260424_120657_05_fp8/checkpoints/10. I0424 12:18:17.997438 140244032386816 async_checkpointer.py:76] [process=5][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-24 12:38:17.997397 I0424 12:18:18.001996 140376034367296 checkpoint_manager.py:1560] [process=5][thread=MainThread][step=10] Starting CheckpointManager Save Finalize thread=save_finalize I0424 12:18:18.002226 140244554131200 async_checkpointer.py:280] [process=5][thread=save_finalize] Waiting for background save thread=async_save. I0424 12:18:18.002357 140376034367296 standard_logger.py:34] {'step': 10, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260424_120657/linen_xpk_feat_nnx_post_train_fixes_20260424_120657_05_fp8/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1777033096.874089, 'wait_for_prev_duration_secs': 0.5451157093048096, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1777033097.4206443, 'checkpointer_blocking_duration_secs': 0.5769684314727783, 'get_old_steps_start_time': 1777033097.997633, 'get_old_steps_duration_secs': 4.029273986816406e-05, 'checkpoint_manager_blocking_start_time': 1777033096.872131, 'checkpoint_manager_blocking_duration_secs': 1.1301915645599365} I0424 12:18:18.002461 140376034367296 checkpointing.py:409] Started an asynchronous checkpoint save for step 10 I0424 12:18:18.003113 140376034367296 metric_logger.py:196] completed step: 9, seconds: 0.158, TFLOP/s/device: 85.870, Tokens/s/device: 12943.266, total_weights: 65536, loss: 7.908, lm_loss: 7.908, perplexity: 2720.116 I0424 12:18:18.009793 140376034367296 metric_logger.py:196] completed step: 10, seconds: 0.159, TFLOP/s/device: 85.670, Tokens/s/device: 12913.070, total_weights: 65536, loss: 7.568, lm_loss: 7.568, perplexity: 1936.198 I0424 12:18:18.168431 140376034367296 metric_logger.py:196] completed step: 11, seconds: 1.448, TFLOP/s/device: 9.385, Tokens/s/device: 1414.580, total_weights: 65536, loss: 7.288, lm_loss: 7.288, perplexity: 1463.272 I0424 12:18:18.731099 140376034367296 metric_logger.py:196] completed step: 12, seconds: 0.006, TFLOP/s/device: 2182.671, Tokens/s/device: 328995.984, total_weights: 65536, loss: 7.032, lm_loss: 7.032, perplexity: 1132.061 I0424 12:18:18.889364 140376034367296 metric_logger.py:196] completed step: 13, seconds: 0.559, TFLOP/s/device: 24.304, Tokens/s/device: 3663.292, total_weights: 65536, loss: 6.815, lm_loss: 6.815, perplexity: 911.031 I0424 12:18:19.047706 140376034367296 metric_logger.py:196] completed step: 14, seconds: 0.163, TFLOP/s/device: 83.346, Tokens/s/device: 12562.876, total_weights: 65536, loss: 6.635, lm_loss: 6.635, perplexity: 761.080 I0424 12:18:19.206061 140376034367296 metric_logger.py:196] completed step: 15, seconds: 0.158, TFLOP/s/device: 85.842, Tokens/s/device: 12939.095, total_weights: 65536, loss: 6.492, lm_loss: 6.492, perplexity: 659.915 I0424 12:18:19.364402 140376034367296 metric_logger.py:196] completed step: 16, seconds: 0.158, TFLOP/s/device: 85.839, Tokens/s/device: 12938.605, total_weights: 65536, loss: 6.380, lm_loss: 6.380, perplexity: 589.948 I0424 12:18:19.522708 140376034367296 metric_logger.py:196] completed step: 17, seconds: 0.158, TFLOP/s/device: 85.756, Tokens/s/device: 12926.110, total_weights: 65536, loss: 6.293, lm_loss: 6.293, perplexity: 540.941 I0424 12:18:19.681005 140376034367296 metric_logger.py:196] completed step: 18, seconds: 0.158, TFLOP/s/device: 85.897, Tokens/s/device: 12947.357, total_weights: 65536, loss: 6.223, lm_loss: 6.223, perplexity: 504.108 I0424 12:18:19.838747 140376034367296 checkpointing.py:794] Waiting for step 19 to finish before checkpoint... I0424 12:18:19.839632 140376034367296 checkpointing.py:798] Waited 0.0009047985076904297 seconds for step 19 to finish before starting checkpointing. I0424 12:18:19.842009 140376034367296 checkpoint_manager.py:2020] [process=5][thread=MainThread][step=10][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0424 12:18:23.504797 140242392692480 array_metadata_store.py:203] [process=5][thread=array_type_handler] Wrote 81 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260424_120657/linen_xpk_feat_nnx_post_train_fixes_20260424_120657_05_fp8/checkpoints/10/items/array_metadatas/process_5 I0424 12:19:01.176046 140244032386816 base_pytree_checkpoint_handler.py:130] [process=5] /jax/orbax/write/gbytes_per_sec: 36.547 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 43.22494959831238 s) (per-host) I0424 12:19:01.176171 140244032386816 async_checkpointer.py:90] [process=5][thread=async_save] 3 Handler Commit operations completed. Time taken: 43.178603s. I0424 12:19:08.084116 140244032386816 async_checkpointer.py:160] [process=5][thread=async_save] Background save thread done. Time taken: 50.086533s. I0424 12:19:08.084332 140244554131200 async_checkpointer.py:288] [process=5][thread=save_finalize] Done with waiting for background save thread=async_save. I0424 12:19:08.084400 140244554131200 async_checkpointer.py:298] [process=5][thread=save_finalize] No errors found in background save thread=async_save. I0424 12:19:08.084460 140244554131200 checkpoint_manager.py:2137] [process=5][thread=save_finalize][step=10] CheckpointManager Save Finalize is syncing with other hosts... I0424 12:19:08.086056 140244554131200 checkpoint_manager.py:2146] [process=5][thread=save_finalize][step=10] CheckpointManager Save Finalize is done on all hosts. I0424 12:19:08.086225 140376034367296 checkpoint_manager.py:2032] [process=5][thread=MainThread][step=10][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=10. W0424 12:19:08.086362 140376034367296 checkpoint_manager.py:1452] Waiting for previous save to complete took 48.244350 seconds. If this number is high, consider checkpointing less frequently. I0424 12:19:08.088320 140376034367296 checkpoint_manager.py:1512] [process=5] Saving checkpoint at step 19 I0424 12:19:08.090448 140376034367296 event_tracking.py:70] [process=5] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260424_120657/linen_xpk_feat_nnx_post_train_fixes_20260424_120657_05_fp8/checkpoints/19. I0424 12:19:08.390252 140376034367296 jax_array_handlers.py:360] Scheduling D2H of 81 prioritized jax.Array. I0424 12:19:08.390354 140376034367296 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 I0424 12:19:08.426829 140376034367296 base_pytree_checkpoint_handler.py:154] [process=5][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.037545s I0424 12:19:08.426964 140376034367296 base_pytree_checkpoint_handler.py:130] [process=5] /jax/orbax/write/blocking_gbytes_per_sec: 35.122 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.04392528533935547 s) (per-host) I0424 12:19:08.427012 140376034367296 base_pytree_checkpoint_handler.py:768] [process=5][thread=MainThread] Initiated Pytree async_save. Time taken: 0.043984s (batch_requests_ready=0.003294s, total_serialization_initiated=0.040627s, others=0.000063s) I0424 12:19:08.427100 140376034367296 composite_checkpoint_handler.py:715] [process=5][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.048112s (all_items=0.000012s, per_item={'items': '0.00001216'}, temp_paths=0.048100) I0424 12:19:08.427774 140376034367296 event_tracking.py:125] [process=5] [async] Finished blocking save in 0.34 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260424_120657/linen_xpk_feat_nnx_post_train_fixes_20260424_120657_05_fp8/checkpoints/19. I0424 12:19:08.428157 140244554131200 async_checkpointer.py:76] [process=5][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-24 12:39:08.428112 I0424 12:19:08.433962 140376034367296 checkpoint_manager.py:1560] [process=5][thread=MainThread][step=19] Starting CheckpointManager Save Finalize thread=save_finalize I0424 12:19:08.434201 140242392692480 async_checkpointer.py:280] [process=5][thread=save_finalize] Waiting for background save thread=async_save. I0424 12:19:08.434329 140376034367296 standard_logger.py:34] {'step': 19, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260424_120657/linen_xpk_feat_nnx_post_train_fixes_20260424_120657_05_fp8/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1777033099.8419802, 'wait_for_prev_duration_secs': 48.24434971809387, 'time_between_consecutive_saves_sec': 2.422995090484619, 'checkpointer_blocking_start_time': 1777033148.0883596, 'checkpointer_blocking_duration_secs': 0.33995747566223145, 'get_old_steps_start_time': 1777033148.4283354, 'get_old_steps_duration_secs': 2.5272369384765625e-05, 'checkpoint_manager_blocking_start_time': 1777033099.839934, 'checkpoint_manager_blocking_duration_secs': 48.594361543655396} I0424 12:19:08.434466 140376034367296 checkpointing.py:409] Started an asynchronous checkpoint save for step 19 I0424 12:19:08.434510 140376034367296 checkpoint_manager.py:2020] [process=5][thread=MainThread][step=19][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0424 12:19:13.627013 140244011697920 array_metadata_store.py:203] [process=5][thread=array_type_handler] Wrote 81 array_metadata.ArrayMetadata to gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260424_120657/linen_xpk_feat_nnx_post_train_fixes_20260424_120657_05_fp8/checkpoints/19/items/array_metadatas/process_5 I0424 12:19:50.892909 140244554131200 base_pytree_checkpoint_handler.py:130] [process=5] /jax/orbax/write/gbytes_per_sec: 37.162 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 42.509827613830566 s) (per-host) I0424 12:19:50.893030 140244554131200 async_checkpointer.py:90] [process=5][thread=async_save] 3 Handler Commit operations completed. Time taken: 42.464749s. I0424 12:19:57.293829 140244554131200 async_checkpointer.py:160] [process=5][thread=async_save] Background save thread done. Time taken: 48.865531s. I0424 12:19:57.294139 140242392692480 async_checkpointer.py:288] [process=5][thread=save_finalize] Done with waiting for background save thread=async_save. I0424 12:19:57.294257 140242392692480 async_checkpointer.py:298] [process=5][thread=save_finalize] No errors found in background save thread=async_save. I0424 12:19:57.294307 140242392692480 checkpoint_manager.py:2137] [process=5][thread=save_finalize][step=19] CheckpointManager Save Finalize is syncing with other hosts... I0424 12:19:57.295592 140242392692480 checkpoint_manager.py:2146] [process=5][thread=save_finalize][step=19] CheckpointManager Save Finalize is done on all hosts. I0424 12:19:57.295790 140376034367296 checkpoint_manager.py:2032] [process=5][thread=MainThread][step=19][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=19. I0424 12:19:57.295956 140376034367296 checkpoint_manager.py:2009] [process=5][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0424 12:19:57.296850 140376034367296 metric_logger.py:196] completed step: 19, seconds: 0.158, TFLOP/s/device: 85.801, Tokens/s/device: 12932.804, 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: Fri Apr 24 12:20:09 UTC 2026 EXIT_CODE=0
XPK Start: Fri Apr 24 13:41:20 UTC 2026 PyTorch was not found. Models won't be available and only tokenizers, configuration and file/data utilities can be used. `rope_parameters`'s factor field must be a float >= 1, got 40 `rope_parameters`'s beta_fast field must be a float, got 32 `rope_parameters`'s beta_slow field must be a float, got 1 DeepseekV32Config got `key=rope_scaling` in kwargs but hasn't set it as attribute. For RoPE standardization you need to set `self.rope_parameters` in model's config. 2026-04-24 13:41:45.753061: 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) I0424 13:41:45.965963 133851429095232 max_utils.py:273] Attempting to initialize the jax distributed system... I0424 13:41:55.007399 133851429095232 distributed.py:149] Starting JAX distributed service on [::]:8482 I0424 13:41:55.009687 133851429095232 distributed.py:172] Connecting to JAX distributed service on mt-05-fp8-gyb4p-slice-job-0-0.mt-05-fp8-gyb4p:8482 I0424 13:41:56.136047 133851429095232 max_utils.py:284] Jax distributed system initialized! I0424 13:42:02.332214 133851429095232 max_utils.py:800] System Information: Jax Version: 0.9.2 I0424 13:42:02.332325 133851429095232 max_utils.py:801] System Information: Jaxlib Version: 0.9.2 I0424 13:42:02.332366 133851429095232 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 I0424 13:42:02.332402 133851429095232 train_utils.py:391] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing. I0424 13:42:03.028422 133851429095232 maxtext_utils.py:1771] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0424 13:42:03.460857 133851429095232 checkpointing.py:688] Setting up checkpoint logger... I0424 13:42:03.460980 133851429095232 checkpointing.py:234] Creating checkpoint manager with ocdbt=True and zarr3=True I0424 13:42:03.461027 133851429095232 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None I0424 13:42:03.461273 133851429095232 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 0x79bc16518dd0>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0424 13:42:06.405658 133851429095232 checkpointing.py:266] Enabling policy for fixed interval checkpointing. I0424 13:42:06.405948 133851429095232 checkpoint_manager.py:708] [process=6][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=('items',), item_handlers={'items': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x79a6d87913d0>}, handler_registry=None I0424 13:42:06.406270 133851429095232 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x79a6d87913d0>` 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`. I0424 13:42:06.406336 133851429095232 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x79a7d41b26c0>` 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`. I0424 13:42:06.406384 133851429095232 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 0x79a6d87913d0>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x79a6d87913d0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x79a7d41b26c0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x79a7d41b26c0>}). I0424 13:42:06.406806 133851429095232 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34 I0424 13:42:06.406899 133851429095232 async_checkpointer.py:192] [process=6][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x79a7d41d7d80> timeout: 1200 secs and primary_host=0 for async checkpoint writes I0424 13:42:07.147604 133851429095232 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260424_120657/nnx_xpk_feat_nnx_post_train_fixes_20260424_120657_05_fp8/checkpoints I0424 13:42:07.197192 133851429095232 checkpoint_manager.py:929] [process=6][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/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260424_120657/nnx_xpk_feat_nnx_post_train_fixes_20260424_120657_05_fp8/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x79a7d41e3500> I0424 13:42:07.197345 133851429095232 checkpointing.py:302] Checkpoint manager created! I0424 13:42:08.470385 133851429095232 checkpointing.py:578] checkpoint manager exists so trying to load this run's existing checkpoint I0424 13:42:08.470514 133851429095232 checkpointing.py:676] No existing checkpoints found, not restoring checkpoint. fsdp: 32 I0424 13:42:10.601480 133851429095232 maxtext_utils.py:1889] decoder/decoder_norm/scale/value Shape: float32[2048] Physical: (None,) I0424 13:42:10.601601 133851429095232 maxtext_utils.py:1889] decoder/layers/mlp/wi_0/kernel/value Shape: float32[2048,16,7168] Physical: ('fsdp', None, None) I0424 13:42:10.601653 133851429095232 maxtext_utils.py:1889] decoder/layers/mlp/wi_1/kernel/value Shape: float32[2048,16,7168] Physical: ('fsdp', None, None) I0424 13:42:10.601696 133851429095232 maxtext_utils.py:1889] decoder/layers/mlp/wo/kernel/value Shape: float32[7168,16,2048] Physical: (None, None, 'fsdp') I0424 13:42:10.601733 133851429095232 maxtext_utils.py:1889] decoder/layers/post_self_attention_layer_norm/scale/value Shape: float32[2048,16] Physical: (None, None) I0424 13:42:10.601767 133851429095232 maxtext_utils.py:1889] decoder/layers/pre_self_attention_layer_norm/scale/value Shape: float32[2048,16] Physical: (None, None) I0424 13:42:10.601801 133851429095232 maxtext_utils.py:1889] decoder/layers/self_attention/key/kernel/value Shape: float32[2048,16,16,128] Physical: ('fsdp', None, None, None) I0424 13:42:10.601836 133851429095232 maxtext_utils.py:1889] decoder/layers/self_attention/out/kernel/value Shape: float32[16,16,128,2048] Physical: (None, None, None, 'fsdp') I0424 13:42:10.601870 133851429095232 maxtext_utils.py:1889] decoder/layers/self_attention/query/kernel/value Shape: float32[2048,16,16,128] Physical: ('fsdp', None, None, None) I0424 13:42:10.601903 133851429095232 maxtext_utils.py:1889] decoder/layers/self_attention/value/kernel/value Shape: float32[2048,16,16,128] Physical: ('fsdp', None, None, None) I0424 13:42:10.601934 133851429095232 maxtext_utils.py:1889] decoder/logits_dense/kernel/value Shape: float32[2048,32000] Physical: ('fsdp', None) I0424 13:42:10.601965 133851429095232 maxtext_utils.py:1889] token_embedder/embedding/value Shape: float32[32000,2048] Physical: (None, 'fsdp') I0424 13:42:11.018708 133851429095232 nnx_decoders.py:465] nnx_decoders/carry Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0424 13:42:11.018815 133851429095232 nnx_decoders.py:465] nnx_decoders/carry Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0424 13:42:11.024157 133851429095232 nnx_decoders.py:465] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0424 13:42:11.024214 133851429095232 nnx_decoders.py:465] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0424 13:42:11.040802 133851429095232 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch_attn', 'activation_length_attn', 'activation_embed_attn'). I0424 13:42:11.040865 133851429095232 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0424 13:42:11.056856 133851429095232 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch_attn', 'activation_length_attn', 'activation_embed_attn'). I0424 13:42:11.056918 133851429095232 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0424 13:42:11.116523 133851429095232 attentions.py:1154] attentions/query Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_length_attn', 'activation_kv_heads', 'activation_kv_head_dim'). I0424 13:42:11.116612 133851429095232 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0424 13:42:11.132731 133851429095232 attentions.py:1155] attentions/key Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_length_attn', 'activation_kv_heads', 'activation_kv_head_dim'). I0424 13:42:11.132794 133851429095232 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0424 13:42:11.148703 133851429095232 attentions.py:1156] attentions/value Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_length_attn', 'activation_kv_heads', 'activation_kv_head_dim'). I0424 13:42:11.148766 133851429095232 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0424 13:42:11.186217 133851429095232 attentions.py:1198] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch_attn', 'activation_length_attn', 'activation_heads', 'activation_kv'). I0424 13:42:11.186333 133851429095232 attentions.py:1198] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0424 13:42:11.246810 133851429095232 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0424 13:42:11.246894 133851429095232 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None). I0424 13:42:25.819409 133851429095232 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. I0424 13:42:25.823266 133851429095232 metric_logger.py:301] number parameters: 1.104 billion I0424 13:42:39.560036 133851429095232 checkpointing.py:794] Waiting for step 0 to finish before checkpoint... I0424 13:42:39.722791 133851429095232 checkpointing.py:798] Waited 0.16274023056030273 seconds for step 0 to finish before starting checkpointing. I0424 13:42:39.725243 133851429095232 checkpoint_manager.py:2009] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0424 13:42:39.727243 133851429095232 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 0 I0424 13:42:39.728586 133851429095232 event_tracking.py:70] [process=6] [async] Started save checkpoint @ gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260424_120657/nnx_xpk_feat_nnx_post_train_fixes_20260424_120657_05_fp8/checkpoints/0. I0424 13:42:40.525894 133851429095232 signaling_client.py:364] Using JaxDistributedSignalingClient I0424 13:42:40.526904 133851429095232 jax_array_handlers.py:360] Scheduling D2H of 153 prioritized jax.Array. I0424 13:42:40.526968 133851429095232 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 I0424 13:42:40.943467 133851429095232 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.419216s I0424 13:42:40.943641 133851429095232 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 3.534 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.4365580081939697 s) (per-host) I0424 13:42:40.943696 133851429095232 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.436623s (batch_requests_ready=0.006078s, total_serialization_initiated=0.430471s, others=0.000073s) I0424 13:42:40.943816 133851429095232 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.440775s (all_items=0.000018s, per_item={'items': '0.00001836'}, temp_paths=0.440757) I0424 13:42:40.944627 133851429095232 event_tracking.py:125] [process=6] [async] Finished blocking save in 1.22 seconds. Continuing save @ gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260424_120657/nnx_xpk_feat_nnx_post_train_fixes_20260424_120657_05_fp8/checkpoints/0. I0424 13:42:40.944939 133721704994560 async_checkpointer.py:76] [process=6][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-24 14:02:40.944901 I0424 13:42:40.947540 133851429095232 checkpoint_manager.py:1560] [process=6][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize I0424 13:42:40.947840 133721175009024 async_checkpointer.py:280] [process=6][thread=save_finalize] Waiting for background save thread=async_save. I0424 13:42:40.947998 133851429095232 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260424_120657/nnx_xpk_feat_nnx_post_train_fixes_20260424_120657_05_fp8/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1777038159.7252235, 'wait_for_prev_duration_secs': 6.222724914550781e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1777038159.7272806, 'checkpointer_blocking_duration_secs': 1.2178082466125488, 'get_old_steps_start_time': 1777038160.9451306, 'get_old_steps_duration_secs': 2.8371810913085938e-05, 'checkpoint_manager_blocking_start_time': 1777038159.7232914, 'checkpoint_manager_blocking_duration_secs': 1.2246670722961426} I0424 13:42:40.948121 133851429095232 checkpointing.py:409] Started an asynchronous checkpoint save for step 0 I0424 13:42:40.948201 133851429095232 max_utils.py:750] Memstats: After params initialized: I0424 13:42:40.948251 133851429095232 max_utils.py:756] Using (GB) 0.45 / 31.25 (1.440000%) on TPU_24(process=6,(0,6,0,0)) I0424 13:42:40.948284 133851429095232 max_utils.py:756] Using (GB) 0.45 / 31.25 (1.440000%) on TPU_25(process=6,(1,6,0,0)) I0424 13:42:40.948311 133851429095232 max_utils.py:756] Using (GB) 0.45 / 31.25 (1.440000%) on TPU_28(process=6,(0,7,0,0)) I0424 13:42:40.948335 133851429095232 max_utils.py:756] Using (GB) 0.45 / 31.25 (1.440000%) on TPU_29(process=6,(1,7,0,0)) I0424 13:42:41.270878 133851429095232 metric_logger.py:196] completed step: 0, seconds: 13.735, TFLOP/s/device: 0.989, Tokens/s/device: 149.112, total_weights: 65536, loss: 10.872, lm_loss: 10.872, perplexity: 52680.742 I0424 13:42:41.446068 133851429095232 metric_logger.py:196] completed step: 1, seconds: 1.708, TFLOP/s/device: 7.955, Tokens/s/device: 1199.025, total_weights: 65536, loss: 10.872, lm_loss: 10.872, perplexity: 52680.742 I0424 13:42:41.896924 133851429095232 metric_logger.py:196] completed step: 2, seconds: 0.022, TFLOP/s/device: 611.620, Tokens/s/device: 92189.962, total_weights: 65536, loss: 10.856, lm_loss: 10.856, perplexity: 51864.762 I0424 13:42:42.054240 133851429095232 metric_logger.py:196] completed step: 3, seconds: 0.428, TFLOP/s/device: 31.733, Tokens/s/device: 4783.147, total_weights: 65536, loss: 10.824, lm_loss: 10.824, perplexity: 50225.512 I0424 13:42:42.369265 133851429095232 metric_logger.py:196] completed step: 4, seconds: 0.186, TFLOP/s/device: 72.970, Tokens/s/device: 10998.867, total_weights: 65536, loss: 10.793, lm_loss: 10.793, perplexity: 48689.156 I0424 13:42:42.379404 133851429095232 metric_logger.py:196] completed step: 5, seconds: 0.157, TFLOP/s/device: 86.509, Tokens/s/device: 13039.603, total_weights: 65536, loss: 10.763, lm_loss: 10.763, perplexity: 47229.012 I0424 13:42:43.823935 2895 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com I0424 13:42:45.865396 133721183401728 array_metadata_store.py:203] [process=6][thread=array_type_handler] Wrote 153 array_metadata.ArrayMetadata to gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260424_120657/nnx_xpk_feat_nnx_post_train_fixes_20260424_120657_05_fp8/checkpoints/0/items/array_metadatas/process_6 I0424 13:43:00.020549 133721704994560 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 80.958 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 19.513410329818726 s) (per-host) I0424 13:43:00.020675 133721704994560 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 19.075626s. I0424 13:43:06.993680 133851429095232 metric_logger.py:196] completed step: 6, seconds: 0.316, TFLOP/s/device: 42.970, Tokens/s/device: 6476.913, total_weights: 65536, loss: 10.733, lm_loss: 10.733, perplexity: 45864.258 I0424 13:43:07.150368 133851429095232 metric_logger.py:196] completed step: 7, seconds: 24.460, TFLOP/s/device: 0.555, Tokens/s/device: 83.729, total_weights: 65536, loss: 10.706, lm_loss: 10.706, perplexity: 44620.242 I0424 13:43:07.450544 133851429095232 metric_logger.py:196] completed step: 8, seconds: 0.164, TFLOP/s/device: 83.007, Tokens/s/device: 12511.684, total_weights: 65536, loss: 10.680, lm_loss: 10.680, perplexity: 43477.055 I0424 13:43:07.458784 133851429095232 checkpointing.py:794] Waiting for step 10 to finish before checkpoint... I0424 13:43:07.764784 133851429095232 checkpointing.py:798] Waited 0.3059723377227783 seconds for step 10 to finish before starting checkpointing. I0424 13:43:07.767199 133851429095232 checkpoint_manager.py:2020] [process=6][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0424 13:43:10.137982 133721704994560 async_checkpointer.py:160] [process=6][thread=async_save] Background save thread done. Time taken: 29.192916s. I0424 13:43:10.138278 133721175009024 async_checkpointer.py:288] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save. I0424 13:43:10.138406 133721175009024 async_checkpointer.py:298] [process=6][thread=save_finalize] No errors found in background save thread=async_save. I0424 13:43:10.138461 133721175009024 checkpoint_manager.py:2137] [process=6][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts... I0424 13:43:10.141166 133721175009024 checkpoint_manager.py:2146] [process=6][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts. I0424 13:43:10.141364 133851429095232 checkpoint_manager.py:2032] [process=6][thread=MainThread][step=0][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=0. W0424 13:43:10.141503 133851429095232 checkpoint_manager.py:1452] Waiting for previous save to complete took 2.374303 seconds. If this number is high, consider checkpointing less frequently. I0424 13:43:10.143600 133851429095232 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 10 I0424 13:43:10.145653 133851429095232 event_tracking.py:70] [process=6] [async] Started save checkpoint @ gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260424_120657/nnx_xpk_feat_nnx_post_train_fixes_20260424_120657_05_fp8/checkpoints/10. I0424 13:43:10.482494 133851429095232 jax_array_handlers.py:360] Scheduling D2H of 153 prioritized jax.Array. I0424 13:43:10.482603 133851429095232 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 I0424 13:43:10.522637 133851429095232 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.042623s I0424 13:43:10.522829 133851429095232 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 26.744 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.057685136795043945 s) (per-host) I0424 13:43:10.522883 133851429095232 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.057749s (batch_requests_ready=0.005597s, total_serialization_initiated=0.052080s, others=0.000072s) I0424 13:43:10.523019 133851429095232 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.062058s (all_items=0.000015s, per_item={'items': '0.00001526'}, temp_paths=0.062043) I0424 13:43:10.523699 133851429095232 event_tracking.py:125] [process=6] [async] Finished blocking save in 0.38 seconds. Continuing save @ gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260424_120657/nnx_xpk_feat_nnx_post_train_fixes_20260424_120657_05_fp8/checkpoints/10. I0424 13:43:10.524059 133721175009024 async_checkpointer.py:76] [process=6][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-24 14:03:10.524022 I0424 13:43:10.531982 133851429095232 checkpoint_manager.py:1560] [process=6][thread=MainThread][step=10] Starting CheckpointManager Save Finalize thread=save_finalize I0424 13:43:10.532244 133718505158400 async_checkpointer.py:280] [process=6][thread=save_finalize] Waiting for background save thread=async_save. I0424 13:43:10.532377 133851429095232 standard_logger.py:34] {'step': 10, 'event_type': 'save', 'directory': 'gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260424_120657/nnx_xpk_feat_nnx_post_train_fixes_20260424_120657_05_fp8/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1777038187.767169, 'wait_for_prev_duration_secs': 2.374302625656128, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1777038190.1436393, 'checkpointer_blocking_duration_secs': 0.38057565689086914, 'get_old_steps_start_time': 1777038190.5242407, 'get_old_steps_duration_secs': 2.9325485229492188e-05, 'checkpoint_manager_blocking_start_time': 1777038187.7652488, 'checkpoint_manager_blocking_duration_secs': 2.767092704772949} I0424 13:43:10.532496 133851429095232 checkpointing.py:409] Started an asynchronous checkpoint save for step 10 I0424 13:43:10.533175 133851429095232 metric_logger.py:196] completed step: 9, seconds: 0.295, TFLOP/s/device: 46.136, Tokens/s/device: 6954.112, total_weights: 65536, loss: 10.656, lm_loss: 10.656, perplexity: 42467.320 I0424 13:43:10.545180 133851429095232 metric_logger.py:196] completed step: 10, seconds: 0.162, TFLOP/s/device: 83.651, Tokens/s/device: 12608.819, total_weights: 65536, loss: 10.636, lm_loss: 10.636, perplexity: 41601.941 I0424 13:43:10.697377 133851429095232 metric_logger.py:196] completed step: 11, seconds: 3.083, TFLOP/s/device: 4.407, Tokens/s/device: 664.223, total_weights: 65536, loss: 10.618, lm_loss: 10.618, perplexity: 40849.570 I0424 13:43:11.287143 133851429095232 metric_logger.py:196] completed step: 12, seconds: 0.010, TFLOP/s/device: 1305.827, Tokens/s/device: 196828.448, total_weights: 65536, loss: 10.602, lm_loss: 10.602, perplexity: 40203.926 I0424 13:43:11.444395 133851429095232 metric_logger.py:196] completed step: 13, seconds: 0.580, TFLOP/s/device: 23.421, Tokens/s/device: 3530.249, total_weights: 65536, loss: 10.588, lm_loss: 10.588, perplexity: 39652.145 I0424 13:43:11.600970 133851429095232 metric_logger.py:196] completed step: 14, seconds: 0.163, TFLOP/s/device: 83.394, Tokens/s/device: 12570.047, total_weights: 65536, loss: 10.577, lm_loss: 10.577, perplexity: 39212.879 I0424 13:43:11.757804 133851429095232 metric_logger.py:196] completed step: 15, seconds: 0.157, TFLOP/s/device: 86.462, Tokens/s/device: 13032.550, total_weights: 65536, loss: 10.568, lm_loss: 10.568, perplexity: 38879.223 I0424 13:43:11.914503 133851429095232 metric_logger.py:196] completed step: 16, seconds: 0.157, TFLOP/s/device: 86.774, Tokens/s/device: 13079.492, total_weights: 65536, loss: 10.562, lm_loss: 10.562, perplexity: 38619.105 I0424 13:43:12.071354 133851429095232 metric_logger.py:196] completed step: 17, seconds: 0.157, TFLOP/s/device: 86.635, Tokens/s/device: 13058.643, total_weights: 65536, loss: 10.556, lm_loss: 10.556, perplexity: 38395.816 I0424 13:43:12.228075 133851429095232 metric_logger.py:196] completed step: 18, seconds: 0.157, TFLOP/s/device: 86.585, Tokens/s/device: 13051.070, total_weights: 65536, loss: 10.552, lm_loss: 10.552, perplexity: 38236.652 I0424 13:43:12.386877 133851429095232 checkpointing.py:794] Waiting for step 19 to finish before checkpoint... I0424 13:43:12.388145 133851429095232 checkpointing.py:798] Waited 0.0012798309326171875 seconds for step 19 to finish before starting checkpointing. I0424 13:43:12.390081 133851429095232 checkpoint_manager.py:2020] [process=6][thread=MainThread][step=10][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0424 13:43:16.253904 133720610678528 array_metadata_store.py:203] [process=6][thread=array_type_handler] Wrote 153 array_metadata.ArrayMetadata to gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260424_120657/nnx_xpk_feat_nnx_post_train_fixes_20260424_120657_05_fp8/checkpoints/10/items/array_metadatas/process_6 I0424 13:43:31.610904 133721175009024 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 74.708 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 21.14572310447693 s) (per-host) I0424 13:43:31.611028 133721175009024 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 21.086842s. I0424 13:43:39.442937 133721175009024 async_checkpointer.py:160] [process=6][thread=async_save] Background save thread done. Time taken: 28.918734s. I0424 13:43:39.443259 133718505158400 async_checkpointer.py:288] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save. I0424 13:43:39.443395 133718505158400 async_checkpointer.py:298] [process=6][thread=save_finalize] No errors found in background save thread=async_save. I0424 13:43:39.443443 133718505158400 checkpoint_manager.py:2137] [process=6][thread=save_finalize][step=10] CheckpointManager Save Finalize is syncing with other hosts... I0424 13:43:39.445125 133718505158400 checkpoint_manager.py:2146] [process=6][thread=save_finalize][step=10] CheckpointManager Save Finalize is done on all hosts. I0424 13:43:39.445297 133851429095232 checkpoint_manager.py:2032] [process=6][thread=MainThread][step=10][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=10. W0424 13:43:39.445448 133851429095232 checkpoint_manager.py:1452] Waiting for previous save to complete took 27.055374 seconds. If this number is high, consider checkpointing less frequently. I0424 13:43:39.447206 133851429095232 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 19 I0424 13:43:39.449234 133851429095232 event_tracking.py:70] [process=6] [async] Started save checkpoint @ gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260424_120657/nnx_xpk_feat_nnx_post_train_fixes_20260424_120657_05_fp8/checkpoints/19. I0424 13:43:40.209755 133851429095232 jax_array_handlers.py:360] Scheduling D2H of 153 prioritized jax.Array. I0424 13:43:40.209867 133851429095232 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 I0424 13:43:40.245843 133851429095232 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.038914s I0424 13:43:40.246006 133851429095232 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 3.007 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.5131120681762695 s) (per-host) I0424 13:43:40.246056 133851429095232 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.513172s (batch_requests_ready=0.465086s, total_serialization_initiated=0.048022s, others=0.000065s) I0424 13:43:40.246194 133851429095232 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.517309s (all_items=0.000011s, per_item={'items': '0.00001121'}, temp_paths=0.517298) I0424 13:43:40.246937 133851429095232 event_tracking.py:125] [process=6] [async] Finished blocking save in 0.80 seconds. Continuing save @ gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260424_120657/nnx_xpk_feat_nnx_post_train_fixes_20260424_120657_05_fp8/checkpoints/19. I0424 13:43:40.247248 133718505158400 async_checkpointer.py:76] [process=6][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-24 14:03:40.247214 I0424 13:43:40.249263 133851429095232 checkpoint_manager.py:1560] [process=6][thread=MainThread][step=19] Starting CheckpointManager Save Finalize thread=save_finalize I0424 13:43:40.249500 133719562114816 async_checkpointer.py:280] [process=6][thread=save_finalize] Waiting for background save thread=async_save. I0424 13:43:40.249623 133851429095232 standard_logger.py:34] {'step': 19, 'event_type': 'save', 'directory': 'gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260424_120657/nnx_xpk_feat_nnx_post_train_fixes_20260424_120657_05_fp8/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1777038192.3900504, 'wait_for_prev_duration_secs': 27.055373668670654, 'time_between_consecutive_saves_sec': 2.248844861984253, 'checkpointer_blocking_start_time': 1777038219.4472454, 'checkpointer_blocking_duration_secs': 0.8001492023468018, 'get_old_steps_start_time': 1777038220.247413, 'get_old_steps_duration_secs': 2.574920654296875e-05, 'checkpoint_manager_blocking_start_time': 1777038192.3884616, 'checkpoint_manager_blocking_duration_secs': 27.861127614974976} I0424 13:43:40.249740 133851429095232 checkpointing.py:409] Started an asynchronous checkpoint save for step 19 I0424 13:43:40.249781 133851429095232 checkpoint_manager.py:2020] [process=6][thread=MainThread][step=19][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0424 13:43:45.416998 133721175009024 array_metadata_store.py:203] [process=6][thread=array_type_handler] Wrote 153 array_metadata.ArrayMetadata to gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260424_120657/nnx_xpk_feat_nnx_post_train_fixes_20260424_120657_05_fp8/checkpoints/19/items/array_metadatas/process_6 I0424 13:44:00.564575 133718505158400 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 75.835 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 20.831641912460327 s) (per-host) I0424 13:44:00.564702 133718505158400 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 20.317341s. I0424 13:44:07.735205 133718505158400 async_checkpointer.py:160] [process=6][thread=async_save] Background save thread done. Time taken: 27.487827s. I0424 13:44:07.735510 133719562114816 async_checkpointer.py:288] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save. I0424 13:44:07.735629 133719562114816 async_checkpointer.py:298] [process=6][thread=save_finalize] No errors found in background save thread=async_save. I0424 13:44:07.735671 133719562114816 checkpoint_manager.py:2137] [process=6][thread=save_finalize][step=19] CheckpointManager Save Finalize is syncing with other hosts... I0424 13:44:07.737153 133719562114816 checkpoint_manager.py:2146] [process=6][thread=save_finalize][step=19] CheckpointManager Save Finalize is done on all hosts. I0424 13:44:07.737321 133851429095232 checkpoint_manager.py:2032] [process=6][thread=MainThread][step=19][wait_until_finished] Done waiting for Save Finalize thread (save_finalize) running at step=19. I0424 13:44:07.737499 133851429095232 checkpoint_manager.py:2009] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0424 13:44:07.738434 133851429095232 metric_logger.py:196] completed step: 19, seconds: 0.157, TFLOP/s/device: 86.725, Tokens/s/device: 13072.146, total_weights: 65536, loss: 10.549, lm_loss: 10.549, perplexity: 38125.074 Per train step: Total TFLOPs: 13.59 split as 93.93% learnable weight flops and 6.07% attention flops XPK End: Fri Apr 24 13:44:20 UTC 2026 EXIT_CODE=0