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 | 85.865 | 86.673 | +0.808 |
| Tok/s | 12942.4 | 13064.2 | +121.776 |
| Avg s/step | 1.575 | 1.667 | +0.092 |
| 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: Sat Apr 25 12:23:17 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. 2026-04-25 12:23:42.881878: 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) I0425 12:23:43.093938 133752034244416 max_utils.py:273] Attempting to initialize the jax distributed system... I0425 12:23:52.134837 133752034244416 distributed.py:149] Starting JAX distributed service on [::]:8482 I0425 12:23:52.137181 133752034244416 distributed.py:172] Connecting to JAX distributed service on mt-05-fp8-j0ddf-slice-job-0-0.mt-05-fp8-j0ddf:8482 I0425 12:23:53.312472 133752034244416 max_utils.py:284] Jax distributed system initialized! I0425 12:23:59.546159 133752034244416 max_utils.py:800] System Information: Jax Version: 0.9.2 I0425 12:23:59.546271 133752034244416 max_utils.py:801] System Information: Jaxlib Version: 0.9.2 I0425 12:23:59.546311 133752034244416 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 I0425 12:23:59.546346 133752034244416 train_utils.py:391] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing. I0425 12:24:00.241500 133752034244416 maxtext_utils.py:1771] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0425 12:24:00.242100 133752034244416 maxtext_utils.py:1771] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0425 12:24:00.242286 133752034244416 checkpointing.py:688] Setting up checkpoint logger... I0425 12:24:00.242337 133752034244416 checkpointing.py:234] Creating checkpoint manager with ocdbt=True and zarr3=True I0425 12:24:00.242380 133752034244416 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None I0425 12:24:00.242732 133752034244416 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 0x79a4daa4b7a0>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0425 12:24:03.073066 133752034244416 checkpointing.py:266] Enabling policy for fixed interval checkpointing. I0425 12:24:03.073332 133752034244416 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 0x7990cc27ecc0>}, handler_registry=None I0425 12:24:03.073579 133752034244416 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7990cc27ecc0>` 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`. I0425 12:24:03.073628 133752034244416 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7990cc282090>` 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`. I0425 12:24:03.073678 133752034244416 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 0x7990cc27ecc0>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7990cc27ecc0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7990cc282090>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7990cc282090>}). I0425 12:24:03.074004 133752034244416 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34 I0425 12:24:03.074074 133752034244416 async_checkpointer.py:192] [process=5][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7990ac7d5d00> timeout: 1200 secs and primary_host=0 for async checkpoint writes I0425 12:24:04.072622 133752034244416 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260425_121355/linen_xpk_feat_nnx_post_train_fixes_20260425_121355_05_fp8/checkpoints I0425 12:24:04.074840 133752034244416 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_20260425_121355/linen_xpk_feat_nnx_post_train_fixes_20260425_121355_05_fp8/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7990cc27ed50> I0425 12:24:04.074965 133752034244416 checkpointing.py:302] Checkpoint manager created! I0425 12:24:05.201096 133752034244416 nnx_wrappers.py:437] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0425 12:24:05.201210 133752034244416 nnx_wrappers.py:437] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0425 12:24:05.587243 133752034244416 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0425 12:24:05.587337 133752034244416 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0425 12:24:05.603719 133752034244416 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_attn_length', 'activation_attn_embed'). I0425 12:24:05.603783 133752034244416 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0425 12:24:05.655910 133752034244416 attentions.py:1154] attentions/query Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0425 12:24:05.656000 133752034244416 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0425 12:24:05.672457 133752034244416 attentions.py:1155] attentions/key Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0425 12:24:05.672518 133752034244416 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0425 12:24:05.688949 133752034244416 attentions.py:1156] attentions/value Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_attn_length', 'activation_kv_heads', 'activation_kv_head_dim'). I0425 12:24:05.689014 133752034244416 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0425 12:24:05.714084 133752034244416 attentions.py:1198] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch', 'activation_attn_length', 'activation_heads', 'activation_kv'). I0425 12:24:05.714159 133752034244416 attentions.py:1198] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0425 12:24:05.764795 133752034244416 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0425 12:24:05.764872 133752034244416 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None). I0425 12:24:06.209342 133752034244416 checkpointing.py:578] checkpoint manager exists so trying to load this run's existing checkpoint I0425 12:24:06.209464 133752034244416 checkpointing.py:676] No existing checkpoints found, not restoring checkpoint. fsdp: 32 I0425 12:24:08.122172 133752034244416 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: () I0425 12:24:08.122303 133752034244416 maxtext_utils.py:1880] params/_overwrite_with_gradient/decoder/layers/mlp/wi_0/Fp8DirectDotGeneralOp_0/input_scale Shape: float32[16,1] Logical: P() Physical: () I0425 12:24:08.122354 133752034244416 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: () I0425 12:24:08.122395 133752034244416 maxtext_utils.py:1880] params/_overwrite_with_gradient/decoder/layers/mlp/wi_0/Fp8DirectDotGeneralOp_0/kernel_scale Shape: float32[16,1] Logical: P() Physical: () I0425 12:24:08.122431 133752034244416 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: () I0425 12:24:08.122465 133752034244416 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: () I0425 12:24:08.122496 133752034244416 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: () I0425 12:24:08.122527 133752034244416 maxtext_utils.py:1880] params/_overwrite_with_gradient/decoder/layers/mlp/wi_1/Fp8DirectDotGeneralOp_0/input_scale Shape: float32[16,1] Logical: P() Physical: () I0425 12:24:08.122557 133752034244416 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: () I0425 12:24:08.122585 133752034244416 maxtext_utils.py:1880] params/_overwrite_with_gradient/decoder/layers/mlp/wi_1/Fp8DirectDotGeneralOp_0/kernel_scale Shape: float32[16,1] Logical: P() Physical: () I0425 12:24:08.122615 133752034244416 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: () I0425 12:24:08.122648 133752034244416 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: () I0425 12:24:08.122694 133752034244416 maxtext_utils.py:1880] params/_overwrite_with_gradient/decoder/layers/mlp/wo/Fp8DirectDotGeneralOp_0/input_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0425 12:24:08.122725 133752034244416 maxtext_utils.py:1880] params/_overwrite_with_gradient/decoder/layers/mlp/wo/Fp8DirectDotGeneralOp_0/input_scale Shape: float32[16,1] Logical: P() Physical: () I0425 12:24:08.122753 133752034244416 maxtext_utils.py:1880] params/_overwrite_with_gradient/decoder/layers/mlp/wo/Fp8DirectDotGeneralOp_0/kernel_amax_history Shape: float32[16,1024] Logical: P() Physical: () I0425 12:24:08.122788 133752034244416 maxtext_utils.py:1880] params/_overwrite_with_gradient/decoder/layers/mlp/wo/Fp8DirectDotGeneralOp_0/kernel_scale Shape: float32[16,1] Logical: P() Physical: () I0425 12:24:08.122818 133752034244416 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: () I0425 12:24:08.122846 133752034244416 maxtext_utils.py:1880] params/_overwrite_with_gradient/decoder/layers/mlp/wo/Fp8DirectDotGeneralOp_0/output_grad_scale Shape: float32[16,1] Logical: P() Physical: () I0425 12:24:08.122873 133752034244416 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: () I0425 12:24:08.122900 133752034244416 maxtext_utils.py:1880] params/_overwrite_with_gradient/decoder/layers/self_attention/key/Fp8DirectDotGeneralOp_0/input_scale Shape: float32[16,1] Logical: P() Physical: () I0425 12:24:08.122926 133752034244416 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: () I0425 12:24:08.122952 133752034244416 maxtext_utils.py:1880] params/_overwrite_with_gradient/decoder/layers/self_attention/key/Fp8DirectDotGeneralOp_0/kernel_scale Shape: float32[16,1] Logical: P() Physical: () I0425 12:24:08.122979 133752034244416 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: () I0425 12:24:08.123005 133752034244416 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: () I0425 12:24:08.123032 133752034244416 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: () I0425 12:24:08.123058 133752034244416 maxtext_utils.py:1880] params/_overwrite_with_gradient/decoder/layers/self_attention/out/Fp8DirectDotGeneralOp_0/input_scale Shape: float32[16,1] Logical: P() Physical: () I0425 12:24:08.123085 133752034244416 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: () I0425 12:24:08.123111 133752034244416 maxtext_utils.py:1880] params/_overwrite_with_gradient/decoder/layers/self_attention/out/Fp8DirectDotGeneralOp_0/kernel_scale Shape: float32[16,1] Logical: P() Physical: () I0425 12:24:08.123137 133752034244416 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: () I0425 12:24:08.123164 133752034244416 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: () I0425 12:24:08.123191 133752034244416 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: () I0425 12:24:08.123222 133752034244416 maxtext_utils.py:1880] params/_overwrite_with_gradient/decoder/layers/self_attention/query/Fp8DirectDotGeneralOp_0/input_scale Shape: float32[16,1] Logical: P() Physical: () I0425 12:24:08.123249 133752034244416 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: () I0425 12:24:08.123275 133752034244416 maxtext_utils.py:1880] params/_overwrite_with_gradient/decoder/layers/self_attention/query/Fp8DirectDotGeneralOp_0/kernel_scale Shape: float32[16,1] Logical: P() Physical: () I0425 12:24:08.123302 133752034244416 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: () I0425 12:24:08.123328 133752034244416 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: () I0425 12:24:08.123355 133752034244416 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: () I0425 12:24:08.123382 133752034244416 maxtext_utils.py:1880] params/_overwrite_with_gradient/decoder/layers/self_attention/value/Fp8DirectDotGeneralOp_0/input_scale Shape: float32[16,1] Logical: P() Physical: () I0425 12:24:08.123408 133752034244416 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: () I0425 12:24:08.123435 133752034244416 maxtext_utils.py:1880] params/_overwrite_with_gradient/decoder/layers/self_attention/value/Fp8DirectDotGeneralOp_0/kernel_scale Shape: float32[16,1] Logical: P() Physical: () I0425 12:24:08.123462 133752034244416 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: () I0425 12:24:08.123489 133752034244416 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: () I0425 12:24:08.123539 133752034244416 maxtext_utils.py:1880] params/params/decoder/decoder_norm/scale Shape: float32[2048] Logical: P('norm',) Physical: (None,) I0425 12:24:08.123591 133752034244416 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) I0425 12:24:08.123629 133752034244416 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) I0425 12:24:08.123709 133752034244416 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') I0425 12:24:08.123764 133752034244416 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) I0425 12:24:08.123800 133752034244416 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) I0425 12:24:08.123849 133752034244416 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) I0425 12:24:08.123897 133752034244416 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') I0425 12:24:08.123933 133752034244416 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) I0425 12:24:08.123966 133752034244416 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) I0425 12:24:08.124010 133752034244416 maxtext_utils.py:1880] params/params/decoder/logits_dense/kernel Shape: float32[2048,32000] Logical: P('embed_vocab', 'vocab') Physical: ('fsdp', None) I0425 12:24:08.124052 133752034244416 maxtext_utils.py:1880] params/params/token_embedder/embedding Shape: float32[32000,2048] Logical: P('vocab', 'embed_vocab') Physical: (None, 'fsdp') I0425 12:24:09.751559 133752034244416 train.py:157] train/xent Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0425 12:24:09.751664 133752034244416 train.py:157] train/xent Physical: float32[32,2048]............................................ ('fsdp', None). I0425 12:24:09.767220 133752034244416 train.py:164] train/z_loss Logical: float32[32,2048]............................................ ('activation_embed_and_logits_batch', 'activation_length'). I0425 12:24:09.767280 133752034244416 train.py:164] train/z_loss Physical: float32[32,2048]............................................ ('fsdp', None). I0425 12:24:23.532289 133752034244416 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. I0425 12:24:23.533165 133752034244416 metric_logger.py:301] number parameters: 1.105 billion I0425 12:24:38.614198 133752034244416 checkpointing.py:794] Waiting for step 0 to finish before checkpoint... I0425 12:24:38.855256 133752034244416 checkpointing.py:798] Waited 0.2410290241241455 seconds for step 0 to finish before starting checkpointing. I0425 12:24:38.857768 133752034244416 checkpoint_manager.py:2009] [process=5][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0425 12:24:38.859714 133752034244416 checkpoint_manager.py:1512] [process=5] Saving checkpoint at step 0 I0425 12:24:38.861085 133752034244416 event_tracking.py:70] [process=5] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260425_121355/linen_xpk_feat_nnx_post_train_fixes_20260425_121355_05_fp8/checkpoints/0. I0425 12:24:39.202111 133752034244416 signaling_client.py:364] Using JaxDistributedSignalingClient I0425 12:24:39.203149 133752034244416 jax_array_handlers.py:360] Scheduling D2H of 81 prioritized jax.Array. I0425 12:24:39.203216 133752034244416 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 I0425 12:24:39.613162 133752034244416 base_pytree_checkpoint_handler.py:154] [process=5][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.411167s I0425 12:24:39.613334 133752034244416 base_pytree_checkpoint_handler.py:130] [process=5] /jax/orbax/write/blocking_gbytes_per_sec: 3.678 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.41940736770629883 s) (per-host) I0425 12:24:39.613387 133752034244416 base_pytree_checkpoint_handler.py:768] [process=5][thread=MainThread] Initiated Pytree async_save. Time taken: 0.419471s (batch_requests_ready=0.003401s, total_serialization_initiated=0.415998s, others=0.000072s) I0425 12:24:39.613494 133752034244416 composite_checkpoint_handler.py:715] [process=5][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.423575s (all_items=0.000017s, per_item={'items': '0.00001740'}, temp_paths=0.423558) I0425 12:24:39.614335 133752034244416 event_tracking.py:125] [process=5] [async] Finished blocking save in 0.75 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260425_121355/linen_xpk_feat_nnx_post_train_fixes_20260425_121355_05_fp8/checkpoints/0. I0425 12:24:39.614697 133620538324736 async_checkpointer.py:76] [process=5][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-25 12:44:39.614647 I0425 12:24:39.629589 133752034244416 checkpoint_manager.py:1560] [process=5][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize I0425 12:24:39.629903 133619502798592 async_checkpointer.py:280] [process=5][thread=save_finalize] Waiting for background save thread=async_save. I0425 12:24:39.630047 133752034244416 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260425_121355/linen_xpk_feat_nnx_post_train_fixes_20260425_121355_05_fp8/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1777119878.8577495, 'wait_for_prev_duration_secs': 6.0558319091796875e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1777119878.859751, 'checkpointer_blocking_duration_secs': 0.755103349685669, 'get_old_steps_start_time': 1777119879.6148798, 'get_old_steps_duration_secs': 4.1961669921875e-05, 'checkpoint_manager_blocking_start_time': 1777119878.8558152, 'checkpoint_manager_blocking_duration_secs': 0.7741942405700684} I0425 12:24:39.630156 133752034244416 checkpointing.py:409] Started an asynchronous checkpoint save for step 0 I0425 12:24:39.630205 133752034244416 max_utils.py:750] Memstats: After params initialized: I0425 12:24:39.630261 133752034244416 max_utils.py:756] Using (GB) 0.45 / 31.25 (1.440000%) on TPU_18(process=5,(2,4,0,0)) I0425 12:24:39.630294 133752034244416 max_utils.py:756] Using (GB) 0.45 / 31.25 (1.440000%) on TPU_19(process=5,(3,4,0,0)) I0425 12:24:39.630321 133752034244416 max_utils.py:756] Using (GB) 0.45 / 31.25 (1.440000%) on TPU_22(process=5,(2,5,0,0)) I0425 12:24:39.630345 133752034244416 max_utils.py:756] Using (GB) 0.45 / 31.25 (1.440000%) on TPU_23(process=5,(3,5,0,0)) I0425 12:24:39.944461 133752034244416 metric_logger.py:196] completed step: 0, seconds: 15.081, TFLOP/s/device: 0.901, Tokens/s/device: 135.801, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52776.805 I0425 12:24:40.125899 133752034244416 metric_logger.py:196] completed step: 1, seconds: 1.328, TFLOP/s/device: 10.229, Tokens/s/device: 1541.842, total_weights: 65536, loss: 10.874, lm_loss: 10.874, perplexity: 52761.840 I0425 12:24:40.534285 133752034244416 metric_logger.py:196] completed step: 2, seconds: 0.024, TFLOP/s/device: 572.162, Tokens/s/device: 86242.473, total_weights: 65536, loss: 10.816, lm_loss: 10.816, perplexity: 49832.398 I0425 12:24:40.692459 133752034244416 metric_logger.py:196] completed step: 3, seconds: 0.409, TFLOP/s/device: 33.236, Tokens/s/device: 5009.650, total_weights: 65536, loss: 10.431, lm_loss: 10.431, perplexity: 33901.820 I0425 12:24:41.010110 133752034244416 metric_logger.py:196] completed step: 4, seconds: 0.164, TFLOP/s/device: 82.892, Tokens/s/device: 12494.433, total_weights: 65536, loss: 9.992, lm_loss: 9.992, perplexity: 21847.639 I0425 12:24:41.016539 133752034244416 metric_logger.py:196] completed step: 5, seconds: 0.158, TFLOP/s/device: 85.922, Tokens/s/device: 12951.042, total_weights: 65536, loss: 9.549, lm_loss: 9.549, perplexity: 14035.006 I0425 12:24:42.479512 2816 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com I0425 12:24:44.489412 133619511191296 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_20260425_121355/linen_xpk_feat_nnx_post_train_fixes_20260425_121355_05_fp8/checkpoints/0/items/array_metadatas/process_5 I0425 12:24:59.184597 133620538324736 base_pytree_checkpoint_handler.py:130] [process=5] /jax/orbax/write/gbytes_per_sec: 79.025 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 19.990628957748413 s) (per-host) I0425 12:24:59.184736 133620538324736 async_checkpointer.py:90] [process=5][thread=async_save] 3 Handler Commit operations completed. Time taken: 19.569923s. I0425 12:25:04.896769 133752034244416 metric_logger.py:196] completed step: 6, seconds: 0.318, TFLOP/s/device: 42.696, Tokens/s/device: 6435.637, total_weights: 65536, loss: 9.111, lm_loss: 9.111, perplexity: 9058.688 I0425 12:25:05.055510 133752034244416 metric_logger.py:196] completed step: 7, seconds: 23.723, TFLOP/s/device: 0.573, Tokens/s/device: 86.330, total_weights: 65536, loss: 8.685, lm_loss: 8.685, perplexity: 5914.720 I0425 12:25:05.213860 133752034244416 metric_logger.py:196] completed step: 8, seconds: 0.164, TFLOP/s/device: 83.094, Tokens/s/device: 12524.845, total_weights: 65536, loss: 8.281, lm_loss: 8.281, perplexity: 3950.010 I0425 12:25:05.218994 133752034244416 checkpointing.py:794] Waiting for step 10 to finish before checkpoint... I0425 12:25:05.530959 133752034244416 checkpointing.py:798] Waited 0.31193065643310547 seconds for step 10 to finish before starting checkpointing. I0425 12:25:05.533598 133752034244416 checkpoint_manager.py:2020] [process=5][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0425 12:25:06.571590 133620538324736 async_checkpointer.py:160] [process=5][thread=async_save] Background save thread done. Time taken: 26.956762s. I0425 12:25:06.571909 133619502798592 async_checkpointer.py:288] [process=5][thread=save_finalize] Done with waiting for background save thread=async_save. I0425 12:25:06.572034 133619502798592 async_checkpointer.py:298] [process=5][thread=save_finalize] No errors found in background save thread=async_save. I0425 12:25:06.572086 133619502798592 checkpoint_manager.py:2137] [process=5][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts... I0425 12:25:06.573635 133619502798592 checkpoint_manager.py:2146] [process=5][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts. I0425 12:25:06.573841 133752034244416 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. W0425 12:25:06.573986 133752034244416 checkpoint_manager.py:1452] Waiting for previous save to complete took 1.040376 seconds. If this number is high, consider checkpointing less frequently. I0425 12:25:06.575762 133752034244416 checkpoint_manager.py:1512] [process=5] Saving checkpoint at step 10 I0425 12:25:06.577791 133752034244416 event_tracking.py:70] [process=5] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260425_121355/linen_xpk_feat_nnx_post_train_fixes_20260425_121355_05_fp8/checkpoints/10. I0425 12:25:06.934150 133752034244416 jax_array_handlers.py:360] Scheduling D2H of 81 prioritized jax.Array. I0425 12:25:06.934245 133752034244416 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 I0425 12:25:06.988986 133752034244416 base_pytree_checkpoint_handler.py:154] [process=5][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.055744s I0425 12:25:06.989153 133752034244416 base_pytree_checkpoint_handler.py:130] [process=5] /jax/orbax/write/blocking_gbytes_per_sec: 24.929 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.06188464164733887 s) (per-host) I0425 12:25:06.989203 133752034244416 base_pytree_checkpoint_handler.py:768] [process=5][thread=MainThread] Initiated Pytree async_save. Time taken: 0.061944s (batch_requests_ready=0.003047s, total_serialization_initiated=0.058831s, others=0.000067s) I0425 12:25:06.989306 133752034244416 composite_checkpoint_handler.py:715] [process=5][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.067618s (all_items=0.000017s, per_item={'items': '0.00001717'}, temp_paths=0.067601) I0425 12:25:06.990044 133752034244416 event_tracking.py:125] [process=5] [async] Finished blocking save in 0.41 seconds. Continuing save @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260425_121355/linen_xpk_feat_nnx_post_train_fixes_20260425_121355_05_fp8/checkpoints/10. I0425 12:25:06.990389 133618949142272 async_checkpointer.py:76] [process=5][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-25 12:45:06.990366 I0425 12:25:06.992404 133752034244416 checkpoint_manager.py:1560] [process=5][thread=MainThread][step=10] Starting CheckpointManager Save Finalize thread=save_finalize I0425 12:25:06.992630 133619502798592 async_checkpointer.py:280] [process=5][thread=save_finalize] Waiting for background save thread=async_save. I0425 12:25:06.992762 133752034244416 standard_logger.py:34] {'step': 10, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260425_121355/linen_xpk_feat_nnx_post_train_fixes_20260425_121355_05_fp8/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1777119905.533569, 'wait_for_prev_duration_secs': 1.0403759479522705, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1777119906.5758007, 'checkpointer_blocking_duration_secs': 0.41469669342041016, 'get_old_steps_start_time': 1777119906.9905193, 'get_old_steps_duration_secs': 3.0279159545898438e-05, 'checkpoint_manager_blocking_start_time': 1777119905.5317109, 'checkpoint_manager_blocking_duration_secs': 1.4610161781311035} I0425 12:25:06.992874 133752034244416 checkpointing.py:409] Started an asynchronous checkpoint save for step 10 I0425 12:25:06.993505 133752034244416 metric_logger.py:196] completed step: 9, seconds: 0.159, TFLOP/s/device: 85.593, Tokens/s/device: 12901.519, total_weights: 65536, loss: 7.908, lm_loss: 7.908, perplexity: 2720.116 I0425 12:25:07.000751 133752034244416 metric_logger.py:196] completed step: 10, seconds: 0.158, TFLOP/s/device: 85.886, Tokens/s/device: 12945.720, total_weights: 65536, loss: 7.568, lm_loss: 7.568, perplexity: 1936.198 I0425 12:25:07.160566 133752034244416 metric_logger.py:196] completed step: 11, seconds: 1.780, TFLOP/s/device: 7.635, Tokens/s/device: 1150.838, total_weights: 65536, loss: 7.288, lm_loss: 7.288, perplexity: 1463.272 I0425 12:25:07.738157 133752034244416 metric_logger.py:196] completed step: 12, seconds: 0.007, TFLOP/s/device: 2088.400, Tokens/s/device: 314786.351, total_weights: 65536, loss: 7.032, lm_loss: 7.032, perplexity: 1132.061 I0425 12:25:07.896608 133752034244416 metric_logger.py:196] completed step: 13, seconds: 0.575, TFLOP/s/device: 23.629, Tokens/s/device: 3561.553, total_weights: 65536, loss: 6.815, lm_loss: 6.815, perplexity: 911.031 I0425 12:25:08.054839 133752034244416 metric_logger.py:196] completed step: 14, seconds: 0.164, TFLOP/s/device: 83.057, Tokens/s/device: 12519.256, total_weights: 65536, loss: 6.635, lm_loss: 6.635, perplexity: 761.080 I0425 12:25:08.213288 133752034244416 metric_logger.py:196] completed step: 15, seconds: 0.158, TFLOP/s/device: 85.944, Tokens/s/device: 12954.482, total_weights: 65536, loss: 6.492, lm_loss: 6.492, perplexity: 659.915 I0425 12:25:08.371714 133752034244416 metric_logger.py:196] completed step: 16, seconds: 0.158, TFLOP/s/device: 85.814, Tokens/s/device: 12934.764, total_weights: 65536, loss: 6.380, lm_loss: 6.380, perplexity: 589.948 I0425 12:25:08.529906 133752034244416 metric_logger.py:196] completed step: 17, seconds: 0.158, TFLOP/s/device: 85.778, Tokens/s/device: 12929.456, total_weights: 65536, loss: 6.293, lm_loss: 6.293, perplexity: 540.941 I0425 12:25:08.688143 133752034244416 metric_logger.py:196] completed step: 18, seconds: 0.158, TFLOP/s/device: 85.774, Tokens/s/device: 12928.722, total_weights: 65536, loss: 6.223, lm_loss: 6.223, perplexity: 504.108 I0425 12:25:08.845866 133752034244416 checkpointing.py:794] Waiting for step 19 to finish before checkpoint... I0425 12:25:08.846868 133752034244416 checkpointing.py:798] Waited 0.0010221004486083984 seconds for step 19 to finish before starting checkpointing. I0425 12:25:08.849001 133752034244416 checkpoint_manager.py:2020] [process=5][thread=MainThread][step=10][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0425 12:25:12.570511 133618400007936 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_20260425_121355/linen_xpk_feat_nnx_post_train_fixes_20260425_121355_05_fp8/checkpoints/10/items/array_metadatas/process_5 I0425 12:25:49.088015 133618949142272 base_pytree_checkpoint_handler.py:130] [process=5] /jax/orbax/write/gbytes_per_sec: 37.470 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 42.160704374313354 s) (per-host) I0425 12:25:49.088143 133618949142272 async_checkpointer.py:90] [process=5][thread=async_save] 3 Handler Commit operations completed. Time taken: 42.097671s. I0425 12:25:56.003501 133618949142272 async_checkpointer.py:160] [process=5][thread=async_save] Background save thread done. Time taken: 49.013013s. I0425 12:25:56.003779 133619502798592 async_checkpointer.py:288] [process=5][thread=save_finalize] Done with waiting for background save thread=async_save. I0425 12:25:56.003837 133619502798592 async_checkpointer.py:298] [process=5][thread=save_finalize] No errors found in background save thread=async_save. I0425 12:25:56.003906 133619502798592 checkpoint_manager.py:2137] [process=5][thread=save_finalize][step=10] CheckpointManager Save Finalize is syncing with other hosts... I0425 12:25:56.005269 133619502798592 checkpoint_manager.py:2146] [process=5][thread=save_finalize][step=10] CheckpointManager Save Finalize is done on all hosts. I0425 12:25:56.005435 133752034244416 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. W0425 12:25:56.005564 133752034244416 checkpoint_manager.py:1452] Waiting for previous save to complete took 47.156566 seconds. If this number is high, consider checkpointing less frequently. I0425 12:25:56.007309 133752034244416 checkpoint_manager.py:1512] [process=5] Saving checkpoint at step 19 I0425 12:25:56.009455 133752034244416 event_tracking.py:70] [process=5] [async] Started save checkpoint @ gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260425_121355/linen_xpk_feat_nnx_post_train_fixes_20260425_121355_05_fp8/checkpoints/19. I0425 12:25:56.287761 133752034244416 jax_array_handlers.py:360] Scheduling D2H of 81 prioritized jax.Array. I0425 12:25:56.287853 133752034244416 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 I0425 12:25:56.342075 133752034244416 base_pytree_checkpoint_handler.py:154] [process=5][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.055263s I0425 12:25:56.342235 133752034244416 base_pytree_checkpoint_handler.py:130] [process=5] /jax/orbax/write/blocking_gbytes_per_sec: 25.155 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.06132864952087402 s) (per-host) I0425 12:25:56.342289 133752034244416 base_pytree_checkpoint_handler.py:768] [process=5][thread=MainThread] Initiated Pytree async_save. Time taken: 0.061393s (batch_requests_ready=0.002978s, total_serialization_initiated=0.058347s, others=0.000068s) I0425 12:25:56.342383 133752034244416 composite_checkpoint_handler.py:715] [process=5][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.065488s (all_items=0.000011s, per_item={'items': '0.00001073'}, temp_paths=0.065477) I0425 12:25:56.343068 133752034244416 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_20260425_121355/linen_xpk_feat_nnx_post_train_fixes_20260425_121355_05_fp8/checkpoints/19. I0425 12:25:56.343394 133618400007936 async_checkpointer.py:76] [process=5][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-25 12:45:56.343356 I0425 12:25:56.345416 133752034244416 checkpoint_manager.py:1560] [process=5][thread=MainThread][step=19] Starting CheckpointManager Save Finalize thread=save_finalize I0425 12:25:56.345734 133619502798592 async_checkpointer.py:280] [process=5][thread=save_finalize] Waiting for background save thread=async_save. I0425 12:25:56.345863 133752034244416 standard_logger.py:34] {'step': 19, 'event_type': 'save', 'directory': 'gs://lance-maxtext/linen_ckpt_xpk_feat_nnx_post_train_fixes_20260425_121355/linen_xpk_feat_nnx_post_train_fixes_20260425_121355_05_fp8/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1777119908.848974, 'wait_for_prev_duration_secs': 47.15656638145447, 'time_between_consecutive_saves_sec': 2.275280714035034, 'checkpointer_blocking_start_time': 1777119956.0073488, 'checkpointer_blocking_duration_secs': 0.33619117736816406, 'get_old_steps_start_time': 1777119956.343557, 'get_old_steps_duration_secs': 2.5033950805664062e-05, 'checkpoint_manager_blocking_start_time': 1777119908.847148, 'checkpoint_manager_blocking_duration_secs': 47.498682737350464} I0425 12:25:56.345973 133752034244416 checkpointing.py:409] Started an asynchronous checkpoint save for step 19 I0425 12:25:56.346014 133752034244416 checkpoint_manager.py:2020] [process=5][thread=MainThread][step=19][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0425 12:26:01.358953 133617321744128 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_20260425_121355/linen_xpk_feat_nnx_post_train_fixes_20260425_121355_05_fp8/checkpoints/19/items/array_metadatas/process_5 I0425 12:26:38.152768 133618400007936 base_pytree_checkpoint_handler.py:130] [process=5] /jax/orbax/write/gbytes_per_sec: 37.728 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 41.87182307243347 s) (per-host) I0425 12:26:38.152887 133618400007936 async_checkpointer.py:90] [process=5][thread=async_save] 3 Handler Commit operations completed. Time taken: 41.809379s. I0425 12:26:46.673789 133618400007936 async_checkpointer.py:160] [process=5][thread=async_save] Background save thread done. Time taken: 50.330264s. I0425 12:26:46.674005 133619502798592 async_checkpointer.py:288] [process=5][thread=save_finalize] Done with waiting for background save thread=async_save. I0425 12:26:46.674096 133619502798592 async_checkpointer.py:298] [process=5][thread=save_finalize] No errors found in background save thread=async_save. I0425 12:26:46.674149 133619502798592 checkpoint_manager.py:2137] [process=5][thread=save_finalize][step=19] CheckpointManager Save Finalize is syncing with other hosts... I0425 12:26:46.675793 133619502798592 checkpoint_manager.py:2146] [process=5][thread=save_finalize][step=19] CheckpointManager Save Finalize is done on all hosts. I0425 12:26:46.675996 133752034244416 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. I0425 12:26:46.676148 133752034244416 checkpoint_manager.py:2009] [process=5][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0425 12:26:46.677117 133752034244416 metric_logger.py:196] completed step: 19, seconds: 0.158, TFLOP/s/device: 85.865, Tokens/s/device: 12942.448, 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 25 12:26:58 UTC 2026 EXIT_CODE=0
XPK Start: Sat Apr 25 13:48:02 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. 2026-04-25 13:48:27.022321: 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) I0425 13:48:27.234325 135413858600768 max_utils.py:273] Attempting to initialize the jax distributed system... I0425 13:48:36.276525 135413858600768 distributed.py:149] Starting JAX distributed service on [::]:8482 I0425 13:48:36.279026 135413858600768 distributed.py:172] Connecting to JAX distributed service on mt-05-fp8-hotgr-slice-job-0-0.mt-05-fp8-hotgr:8482 I0425 13:48:37.091077 135413858600768 max_utils.py:284] Jax distributed system initialized! I0425 13:48:43.317158 135413858600768 max_utils.py:800] System Information: Jax Version: 0.9.2 I0425 13:48:43.317269 135413858600768 max_utils.py:801] System Information: Jaxlib Version: 0.9.2 I0425 13:48:43.317310 135413858600768 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 I0425 13:48:43.317347 135413858600768 train_utils.py:391] WARNING: Sequence packing is essentially ignored for synthetic data. Please use a real dataset to use sequence packing. I0425 13:48:44.014969 135413858600768 maxtext_utils.py:1771] Num_devices: 32, shape (1, 1, 1, 32, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0425 13:48:44.441615 135413858600768 checkpointing.py:688] Setting up checkpoint logger... I0425 13:48:44.441737 135413858600768 checkpointing.py:234] Creating checkpoint manager with ocdbt=True and zarr3=True I0425 13:48:44.441784 135413858600768 pytree_checkpoint_handler.py:592] save_device_host_concurrent_bytes=None I0425 13:48:44.442003 135413858600768 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 0x7b27c72486e0>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0425 13:48:47.317631 135413858600768 checkpointing.py:266] Enabling policy for fixed interval checkpointing. I0425 13:48:47.317869 135413858600768 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 0x7b26bb05e480>}, handler_registry=None I0425 13:48:47.318129 135413858600768 composite_checkpoint_handler.py:237] Deferred registration for item: "items". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7b26bb05e480>` 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`. I0425 13:48:47.318180 135413858600768 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7b13bc2785f0>` 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`. I0425 13:48:47.318217 135413858600768 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 0x7b26bb05e480>, ('items', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7b26bb05e480>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7b13bc2785f0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7b13bc2785f0>}). I0425 13:48:47.318561 135413858600768 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.34 I0425 13:48:47.318633 135413858600768 async_checkpointer.py:192] [process=6][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7b13bc282340> timeout: 1200 secs and primary_host=0 for async checkpoint writes I0425 13:48:48.034538 135413858600768 checkpoint_manager.py:1812] Found 0 checkpoint steps in gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260425_121355/nnx_xpk_feat_nnx_post_train_fixes_20260425_121355_05_fp8/checkpoints I0425 13:48:48.036825 135413858600768 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_20260425_121355/nnx_xpk_feat_nnx_post_train_fixes_20260425_121355_05_fp8/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7b138c577950> I0425 13:48:48.036942 135413858600768 checkpointing.py:302] Checkpoint manager created! I0425 13:48:49.298832 135413858600768 checkpointing.py:578] checkpoint manager exists so trying to load this run's existing checkpoint I0425 13:48:49.298957 135413858600768 checkpointing.py:676] No existing checkpoints found, not restoring checkpoint. fsdp: 32 I0425 13:48:51.379650 135413858600768 maxtext_utils.py:1889] decoder/decoder_norm/scale/value Shape: float32[2048] Physical: (None,) I0425 13:48:51.379770 135413858600768 maxtext_utils.py:1889] decoder/layers/mlp/wi_0/kernel/value Shape: float32[2048,16,7168] Physical: ('fsdp', None, None) I0425 13:48:51.379821 135413858600768 maxtext_utils.py:1889] decoder/layers/mlp/wi_1/kernel/value Shape: float32[2048,16,7168] Physical: ('fsdp', None, None) I0425 13:48:51.379863 135413858600768 maxtext_utils.py:1889] decoder/layers/mlp/wo/kernel/value Shape: float32[7168,16,2048] Physical: (None, None, 'fsdp') I0425 13:48:51.379899 135413858600768 maxtext_utils.py:1889] decoder/layers/post_self_attention_layer_norm/scale/value Shape: float32[2048,16] Physical: (None, None) I0425 13:48:51.379934 135413858600768 maxtext_utils.py:1889] decoder/layers/pre_self_attention_layer_norm/scale/value Shape: float32[2048,16] Physical: (None, None) I0425 13:48:51.379971 135413858600768 maxtext_utils.py:1889] decoder/layers/self_attention/key/kernel/value Shape: float32[2048,16,16,128] Physical: ('fsdp', None, None, None) I0425 13:48:51.380007 135413858600768 maxtext_utils.py:1889] decoder/layers/self_attention/out/kernel/value Shape: float32[16,16,128,2048] Physical: (None, None, None, 'fsdp') I0425 13:48:51.380040 135413858600768 maxtext_utils.py:1889] decoder/layers/self_attention/query/kernel/value Shape: float32[2048,16,16,128] Physical: ('fsdp', None, None, None) I0425 13:48:51.380073 135413858600768 maxtext_utils.py:1889] decoder/layers/self_attention/value/kernel/value Shape: float32[2048,16,16,128] Physical: ('fsdp', None, None, None) I0425 13:48:51.380119 135413858600768 maxtext_utils.py:1889] decoder/logits_dense/kernel/value Shape: float32[2048,32000] Physical: ('fsdp', None) I0425 13:48:51.380157 135413858600768 maxtext_utils.py:1889] token_embedder/embedding/value Shape: float32[32000,2048] Physical: (None, 'fsdp') I0425 13:48:51.801797 135413858600768 nnx_decoders.py:465] nnx_decoders/carry Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0425 13:48:51.801900 135413858600768 nnx_decoders.py:465] nnx_decoders/carry Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0425 13:48:51.807084 135413858600768 nnx_decoders.py:465] Unknown Logical: bfloat16[32,2048,2048]...................................... ('activation_batch', 'activation_norm_length', 'activation_embed'). I0425 13:48:51.807164 135413858600768 nnx_decoders.py:465] Unknown Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0425 13:48:51.824049 135413858600768 attentions.py:1088] attentions/inputs_q Logical: bfloat16[32,2048,2048]...................................... ('activation_batch_attn', 'activation_length_attn', 'activation_embed_attn'). I0425 13:48:51.824122 135413858600768 attentions.py:1088] attentions/inputs_q Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0425 13:48:51.840251 135413858600768 attentions.py:1089] attentions/inputs_kv Logical: bfloat16[32,2048,2048]...................................... ('activation_batch_attn', 'activation_length_attn', 'activation_embed_attn'). I0425 13:48:51.840319 135413858600768 attentions.py:1089] attentions/inputs_kv Physical: bfloat16[32,2048,2048]...................................... ('fsdp', None, None). I0425 13:48:51.899392 135413858600768 attentions.py:1154] attentions/query Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_length_attn', 'activation_kv_heads', 'activation_kv_head_dim'). I0425 13:48:51.899481 135413858600768 attentions.py:1154] attentions/query Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0425 13:48:51.915908 135413858600768 attentions.py:1155] attentions/key Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_length_attn', 'activation_kv_heads', 'activation_kv_head_dim'). I0425 13:48:51.915973 135413858600768 attentions.py:1155] attentions/key Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0425 13:48:51.932199 135413858600768 attentions.py:1156] attentions/value Logical: bfloat16[32,2048,16,128].................................... ('activation_kv_batch', 'activation_length_attn', 'activation_kv_heads', 'activation_kv_head_dim'). I0425 13:48:51.932260 135413858600768 attentions.py:1156] attentions/value Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0425 13:48:51.963563 135413858600768 attentions.py:1198] attentions/out Logical: bfloat16[32,2048,16,128].................................... ('activation_batch_attn', 'activation_length_attn', 'activation_heads', 'activation_kv'). I0425 13:48:51.963637 135413858600768 attentions.py:1198] attentions/out Physical: bfloat16[32,2048,16,128].................................... ('fsdp', None, None, None). I0425 13:48:52.022627 135413858600768 linears.py:525] linears/x Logical: bfloat16[32,2048,7168]...................................... ('activation_batch', 'activation_length', 'activation_mlp'). I0425 13:48:52.022709 135413858600768 linears.py:525] linears/x Physical: bfloat16[32,2048,7168]...................................... ('fsdp', None, None). I0425 13:49:06.204895 135413858600768 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. I0425 13:49:06.208802 135413858600768 metric_logger.py:301] number parameters: 1.104 billion I0425 13:49:19.797344 135413858600768 checkpointing.py:794] Waiting for step 0 to finish before checkpoint... I0425 13:49:19.958498 135413858600768 checkpointing.py:798] Waited 0.16113972663879395 seconds for step 0 to finish before starting checkpointing. I0425 13:49:19.960937 135413858600768 checkpoint_manager.py:2009] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0425 13:49:19.963115 135413858600768 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 0 I0425 13:49:19.964555 135413858600768 event_tracking.py:70] [process=6] [async] Started save checkpoint @ gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260425_121355/nnx_xpk_feat_nnx_post_train_fixes_20260425_121355_05_fp8/checkpoints/0. I0425 13:49:20.328836 135413858600768 signaling_client.py:364] Using JaxDistributedSignalingClient I0425 13:49:20.329924 135413858600768 jax_array_handlers.py:360] Scheduling D2H of 153 prioritized jax.Array. I0425 13:49:20.329987 135413858600768 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 I0425 13:49:20.745185 135413858600768 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.417968s I0425 13:49:20.745370 135413858600768 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 3.542 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.43555712699890137 s) (per-host) I0425 13:49:20.745423 135413858600768 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.435624s (batch_requests_ready=0.006581s, total_serialization_initiated=0.428968s, others=0.000076s) I0425 13:49:20.745547 135413858600768 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.439886s (all_items=0.000019s, per_item={'items': '0.00001860'}, temp_paths=0.439867) I0425 13:49:20.746383 135413858600768 event_tracking.py:125] [process=6] [async] Finished blocking save in 0.78 seconds. Continuing save @ gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260425_121355/nnx_xpk_feat_nnx_post_train_fixes_20260425_121355_05_fp8/checkpoints/0. I0425 13:49:20.746734 135285918127872 async_checkpointer.py:76] [process=6][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-25 14:09:20.746694 I0425 13:49:20.754756 135413858600768 checkpoint_manager.py:1560] [process=6][thread=MainThread][step=0] Starting CheckpointManager Save Finalize thread=save_finalize I0425 13:49:20.754999 135285421631232 async_checkpointer.py:280] [process=6][thread=save_finalize] Waiting for background save thread=async_save. I0425 13:49:20.755170 135413858600768 standard_logger.py:34] {'step': 0, 'event_type': 'save', 'directory': 'gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260425_121355/nnx_xpk_feat_nnx_post_train_fixes_20260425_121355_05_fp8/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1777124959.9609182, 'wait_for_prev_duration_secs': 6.4849853515625e-05, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1777124959.963153, 'checkpointer_blocking_duration_secs': 0.7837369441986084, 'get_old_steps_start_time': 1777124960.7469146, 'get_old_steps_duration_secs': 3.0040740966796875e-05, 'checkpoint_manager_blocking_start_time': 1777124959.958975, 'checkpoint_manager_blocking_duration_secs': 0.7961571216583252} I0425 13:49:20.755283 135413858600768 checkpointing.py:409] Started an asynchronous checkpoint save for step 0 I0425 13:49:20.755368 135413858600768 max_utils.py:750] Memstats: After params initialized: I0425 13:49:20.755418 135413858600768 max_utils.py:756] Using (GB) 0.45 / 31.25 (1.440000%) on TPU_24(process=6,(0,6,0,0)) I0425 13:49:20.755452 135413858600768 max_utils.py:756] Using (GB) 0.45 / 31.25 (1.440000%) on TPU_25(process=6,(1,6,0,0)) I0425 13:49:20.755489 135413858600768 max_utils.py:756] Using (GB) 0.45 / 31.25 (1.440000%) on TPU_28(process=6,(0,7,0,0)) I0425 13:49:20.755528 135413858600768 max_utils.py:756] Using (GB) 0.45 / 31.25 (1.440000%) on TPU_29(process=6,(1,7,0,0)) I0425 13:49:21.071860 135413858600768 metric_logger.py:196] completed step: 0, seconds: 13.586, TFLOP/s/device: 1.000, Tokens/s/device: 150.740, total_weights: 65536, loss: 10.872, lm_loss: 10.872, perplexity: 52680.742 I0425 13:49:21.266871 135413858600768 metric_logger.py:196] completed step: 1, seconds: 1.272, TFLOP/s/device: 10.683, Tokens/s/device: 1610.306, total_weights: 65536, loss: 10.872, lm_loss: 10.872, perplexity: 52680.742 I0425 13:49:21.741384 135413858600768 metric_logger.py:196] completed step: 2, seconds: 0.042, TFLOP/s/device: 320.792, Tokens/s/device: 48353.205, total_weights: 65536, loss: 10.856, lm_loss: 10.856, perplexity: 51864.762 I0425 13:49:21.898972 135413858600768 metric_logger.py:196] completed step: 3, seconds: 0.452, TFLOP/s/device: 30.091, Tokens/s/device: 4535.629, total_weights: 65536, loss: 10.824, lm_loss: 10.824, perplexity: 50225.512 I0425 13:49:22.214495 135413858600768 metric_logger.py:196] completed step: 4, seconds: 0.186, TFLOP/s/device: 72.937, Tokens/s/device: 10993.848, total_weights: 65536, loss: 10.793, lm_loss: 10.793, perplexity: 48689.156 I0425 13:49:22.224141 135413858600768 metric_logger.py:196] completed step: 5, seconds: 0.157, TFLOP/s/device: 86.290, Tokens/s/device: 13006.560, total_weights: 65536, loss: 10.763, lm_loss: 10.763, perplexity: 47229.012 I0425 13:49:23.731269 2832 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com I0425 13:49:25.728223 135285430023936 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_20260425_121355/nnx_xpk_feat_nnx_post_train_fixes_20260425_121355_05_fp8/checkpoints/0/items/array_metadatas/process_6 I0425 13:49:39.730320 135285918127872 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 81.345 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 19.42047095298767 s) (per-host) I0425 13:49:39.730466 135285918127872 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 18.983614s. I0425 13:49:46.455287 135413858600768 metric_logger.py:196] completed step: 6, seconds: 0.316, TFLOP/s/device: 43.024, Tokens/s/device: 6485.117, total_weights: 65536, loss: 10.733, lm_loss: 10.733, perplexity: 45864.258 I0425 13:49:46.464493 135413858600768 metric_logger.py:196] completed step: 7, seconds: 24.231, TFLOP/s/device: 0.561, Tokens/s/device: 84.521, total_weights: 65536, loss: 10.706, lm_loss: 10.706, perplexity: 44620.242 I0425 13:49:46.621199 135413858600768 metric_logger.py:196] completed step: 8, seconds: 0.010, TFLOP/s/device: 1367.465, Tokens/s/device: 206119.163, total_weights: 65536, loss: 10.680, lm_loss: 10.680, perplexity: 43477.055 I0425 13:49:46.629949 135413858600768 checkpointing.py:794] Waiting for step 10 to finish before checkpoint... I0425 13:49:46.935194 135413858600768 checkpointing.py:798] Waited 0.30522727966308594 seconds for step 10 to finish before starting checkpointing. I0425 13:49:46.937583 135413858600768 checkpoint_manager.py:2020] [process=6][thread=MainThread][step=0][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0425 13:49:49.601649 135285918127872 async_checkpointer.py:160] [process=6][thread=async_save] Background save thread done. Time taken: 28.854780s. I0425 13:49:49.601920 135285421631232 async_checkpointer.py:288] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save. I0425 13:49:49.602047 135285421631232 async_checkpointer.py:298] [process=6][thread=save_finalize] No errors found in background save thread=async_save. I0425 13:49:49.602113 135285421631232 checkpoint_manager.py:2137] [process=6][thread=save_finalize][step=0] CheckpointManager Save Finalize is syncing with other hosts... I0425 13:49:49.604865 135285421631232 checkpoint_manager.py:2146] [process=6][thread=save_finalize][step=0] CheckpointManager Save Finalize is done on all hosts. I0425 13:49:49.605012 135413858600768 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. W0425 13:49:49.605082 135413858600768 checkpoint_manager.py:1452] Waiting for previous save to complete took 2.667503 seconds. If this number is high, consider checkpointing less frequently. I0425 13:49:49.607091 135413858600768 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 10 I0425 13:49:49.609012 135413858600768 event_tracking.py:70] [process=6] [async] Started save checkpoint @ gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260425_121355/nnx_xpk_feat_nnx_post_train_fixes_20260425_121355_05_fp8/checkpoints/10. I0425 13:49:49.902312 135413858600768 jax_array_handlers.py:360] Scheduling D2H of 153 prioritized jax.Array. I0425 13:49:49.902422 135413858600768 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 I0425 13:49:49.937509 135413858600768 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.037713s I0425 13:49:49.937658 135413858600768 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 29.148 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.05292844772338867 s) (per-host) I0425 13:49:49.937711 135413858600768 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.052994s (batch_requests_ready=0.005624s, total_serialization_initiated=0.047297s, others=0.000072s) I0425 13:49:49.937808 135413858600768 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.057083s (all_items=0.000015s, per_item={'items': '0.00001550'}, temp_paths=0.057068) I0425 13:49:49.938524 135413858600768 event_tracking.py:125] [process=6] [async] Finished blocking save in 0.33 seconds. Continuing save @ gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260425_121355/nnx_xpk_feat_nnx_post_train_fixes_20260425_121355_05_fp8/checkpoints/10. I0425 13:49:49.938862 135285421631232 async_checkpointer.py:76] [process=6][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-25 14:09:49.938825 I0425 13:49:49.947732 135413858600768 checkpoint_manager.py:1560] [process=6][thread=MainThread][step=10] Starting CheckpointManager Save Finalize thread=save_finalize I0425 13:49:49.948012 135282727335680 async_checkpointer.py:280] [process=6][thread=save_finalize] Waiting for background save thread=async_save. I0425 13:49:49.948157 135413858600768 standard_logger.py:34] {'step': 10, 'event_type': 'save', 'directory': 'gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260425_121355/nnx_xpk_feat_nnx_post_train_fixes_20260425_121355_05_fp8/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1777124986.9375541, 'wait_for_prev_duration_secs': 2.6675033569335938, 'time_between_consecutive_saves_sec': None, 'checkpointer_blocking_start_time': 1777124989.6071446, 'checkpointer_blocking_duration_secs': 0.33185863494873047, 'get_old_steps_start_time': 1777124989.9390266, 'get_old_steps_duration_secs': 2.9087066650390625e-05, 'checkpoint_manager_blocking_start_time': 1777124986.93566, 'checkpoint_manager_blocking_duration_secs': 3.0124616622924805} I0425 13:49:49.948328 135413858600768 checkpointing.py:409] Started an asynchronous checkpoint save for step 10 I0425 13:49:49.949003 135413858600768 metric_logger.py:196] completed step: 9, seconds: 0.008, TFLOP/s/device: 1703.929, Tokens/s/device: 256834.713, total_weights: 65536, loss: 10.656, lm_loss: 10.656, perplexity: 42467.320 I0425 13:49:49.960549 135413858600768 metric_logger.py:196] completed step: 10, seconds: 0.158, TFLOP/s/device: 85.768, Tokens/s/device: 12927.905, total_weights: 65536, loss: 10.636, lm_loss: 10.636, perplexity: 41601.941 I0425 13:49:50.112444 135413858600768 metric_logger.py:196] completed step: 11, seconds: 3.327, TFLOP/s/device: 4.083, Tokens/s/device: 615.480, total_weights: 65536, loss: 10.618, lm_loss: 10.618, perplexity: 40849.570 I0425 13:49:50.675924 135413858600768 metric_logger.py:196] completed step: 12, seconds: 0.010, TFLOP/s/device: 1336.132, Tokens/s/device: 201396.401, total_weights: 65536, loss: 10.602, lm_loss: 10.602, perplexity: 40203.926 I0425 13:49:50.832935 135413858600768 metric_logger.py:196] completed step: 13, seconds: 0.554, TFLOP/s/device: 24.535, Tokens/s/device: 3698.193, total_weights: 65536, loss: 10.588, lm_loss: 10.588, perplexity: 39652.145 I0425 13:49:50.989683 135413858600768 metric_logger.py:196] completed step: 14, seconds: 0.163, TFLOP/s/device: 83.429, Tokens/s/device: 12575.295, total_weights: 65536, loss: 10.577, lm_loss: 10.577, perplexity: 39212.879 I0425 13:49:51.146497 135413858600768 metric_logger.py:196] completed step: 15, seconds: 0.157, TFLOP/s/device: 86.653, Tokens/s/device: 13061.308, total_weights: 65536, loss: 10.568, lm_loss: 10.568, perplexity: 38879.223 I0425 13:49:51.303339 135413858600768 metric_logger.py:196] completed step: 16, seconds: 0.157, TFLOP/s/device: 86.582, Tokens/s/device: 13050.571, total_weights: 65536, loss: 10.562, lm_loss: 10.562, perplexity: 38619.105 I0425 13:49:51.460118 135413858600768 metric_logger.py:196] completed step: 17, seconds: 0.157, TFLOP/s/device: 86.600, Tokens/s/device: 13053.233, total_weights: 65536, loss: 10.556, lm_loss: 10.556, perplexity: 38395.816 I0425 13:49:51.616877 135413858600768 metric_logger.py:196] completed step: 18, seconds: 0.157, TFLOP/s/device: 86.786, Tokens/s/device: 13081.330, total_weights: 65536, loss: 10.552, lm_loss: 10.552, perplexity: 38236.652 I0425 13:49:51.775592 135413858600768 checkpointing.py:794] Waiting for step 19 to finish before checkpoint... I0425 13:49:51.776846 135413858600768 checkpointing.py:798] Waited 0.001264333724975586 seconds for step 19 to finish before starting checkpointing. I0425 13:49:51.778828 135413858600768 checkpoint_manager.py:2020] [process=6][thread=MainThread][step=10][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0425 13:49:55.596451 135283282540288 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_20260425_121355/nnx_xpk_feat_nnx_post_train_fixes_20260425_121355_05_fp8/checkpoints/10/items/array_metadatas/process_6 I0425 13:50:10.763076 135285421631232 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 75.665 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 20.878326892852783 s) (per-host) I0425 13:50:10.763196 135285421631232 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 20.824223s. I0425 13:50:18.877918 135285421631232 async_checkpointer.py:160] [process=6][thread=async_save] Background save thread done. Time taken: 28.938927s. I0425 13:50:18.878219 135282727335680 async_checkpointer.py:288] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save. I0425 13:50:18.878358 135282727335680 async_checkpointer.py:298] [process=6][thread=save_finalize] No errors found in background save thread=async_save. I0425 13:50:18.878412 135282727335680 checkpoint_manager.py:2137] [process=6][thread=save_finalize][step=10] CheckpointManager Save Finalize is syncing with other hosts... I0425 13:50:18.880866 135282727335680 checkpoint_manager.py:2146] [process=6][thread=save_finalize][step=10] CheckpointManager Save Finalize is done on all hosts. I0425 13:50:18.881032 135413858600768 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. W0425 13:50:18.881164 135413858600768 checkpoint_manager.py:1452] Waiting for previous save to complete took 27.102340 seconds. If this number is high, consider checkpointing less frequently. I0425 13:50:18.882887 135413858600768 checkpoint_manager.py:1512] [process=6] Saving checkpoint at step 19 I0425 13:50:18.884872 135413858600768 event_tracking.py:70] [process=6] [async] Started save checkpoint @ gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260425_121355/nnx_xpk_feat_nnx_post_train_fixes_20260425_121355_05_fp8/checkpoints/19. I0425 13:50:19.174561 135413858600768 jax_array_handlers.py:360] Scheduling D2H of 153 prioritized jax.Array. I0425 13:50:19.174715 135413858600768 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 I0425 13:50:19.213666 135413858600768 base_pytree_checkpoint_handler.py:154] [process=6][thread=MainThread] Initiated "orbax.checkpoint._src.serialization.jax_array_handlers.ArrayHandler".serialize. Time taken: 0.041581s I0425 13:50:19.213811 135413858600768 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/blocking_gbytes_per_sec: 27.414 GiB/s (total gbytes: 1.5 GiB) (time elapsed: 0.0562746524810791 s) (per-host) I0425 13:50:19.213859 135413858600768 base_pytree_checkpoint_handler.py:768] [process=6][thread=MainThread] Initiated Pytree async_save. Time taken: 0.056333s (batch_requests_ready=0.005669s, total_serialization_initiated=0.050601s, others=0.000063s) I0425 13:50:19.213947 135413858600768 composite_checkpoint_handler.py:715] [process=6][thread=MainThread] Initiated CompositeCheckpointHandler.async_save. Time taken: 0.060270s (all_items=0.000011s, per_item={'items': '0.00001073'}, temp_paths=0.060259) I0425 13:50:19.214668 135413858600768 event_tracking.py:125] [process=6] [async] Finished blocking save in 0.33 seconds. Continuing save @ gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260425_121355/nnx_xpk_feat_nnx_post_train_fixes_20260425_121355_05_fp8/checkpoints/19. I0425 13:50:19.215001 135282727335680 async_checkpointer.py:76] [process=6][thread=async_save] Background save thread started. Deadline for this save operation is 2026-04-25 14:10:19.214963 I0425 13:50:19.673412 135413858600768 checkpoint_manager.py:1560] [process=6][thread=MainThread][step=19] Starting CheckpointManager Save Finalize thread=save_finalize I0425 13:50:19.673781 135283265754880 async_checkpointer.py:280] [process=6][thread=save_finalize] Waiting for background save thread=async_save. I0425 13:50:19.673949 135413858600768 standard_logger.py:34] {'step': 19, 'event_type': 'save', 'directory': 'gs://lance-maxtext/nnx_ckpt_xpk_feat_nnx_post_train_fixes_20260425_121355/nnx_xpk_feat_nnx_post_train_fixes_20260425_121355_05_fp8/checkpoints', 'reached_preemption': False, 'preemption_received_at': None, 'synchronous': False, 'wait_for_prev_start_time': 1777124991.7788024, 'wait_for_prev_duration_secs': 27.10233974456787, 'time_between_consecutive_saves_sec': 2.1738975048065186, 'checkpointer_blocking_start_time': 1777125018.8829262, 'checkpointer_blocking_duration_secs': 0.33223772048950195, 'get_old_steps_start_time': 1777125019.2151825, 'get_old_steps_duration_secs': 2.574920654296875e-05, 'checkpoint_manager_blocking_start_time': 1777124991.777182, 'checkpoint_manager_blocking_duration_secs': 27.89673089981079} I0425 13:50:19.674075 135413858600768 checkpointing.py:409] Started an asynchronous checkpoint save for step 19 I0425 13:50:19.674137 135413858600768 checkpoint_manager.py:2020] [process=6][thread=MainThread][step=19][wait_until_finished] Waiting for Save Finalize thread (save_finalize) to complete. I0425 13:50:24.443926 135283274147584 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_20260425_121355/nnx_xpk_feat_nnx_post_train_fixes_20260425_121355_05_fp8/checkpoints/19/items/array_metadatas/process_6 I0425 13:50:40.123450 135282727335680 base_pytree_checkpoint_handler.py:130] [process=6] /jax/orbax/write/gbytes_per_sec: 75.349 MiB/s (total gbytes: 1.5 GiB) (time elapsed: 20.96586513519287 s) (per-host) I0425 13:50:40.123590 135282727335680 async_checkpointer.py:90] [process=6][thread=async_save] 3 Handler Commit operations completed. Time taken: 20.908462s. I0425 13:50:47.077143 135282727335680 async_checkpointer.py:160] [process=6][thread=async_save] Background save thread done. Time taken: 27.861997s. I0425 13:50:47.077413 135283265754880 async_checkpointer.py:288] [process=6][thread=save_finalize] Done with waiting for background save thread=async_save. I0425 13:50:47.077472 135283265754880 async_checkpointer.py:298] [process=6][thread=save_finalize] No errors found in background save thread=async_save. I0425 13:50:47.077520 135283265754880 checkpoint_manager.py:2137] [process=6][thread=save_finalize][step=19] CheckpointManager Save Finalize is syncing with other hosts... I0425 13:50:47.080069 135283265754880 checkpoint_manager.py:2146] [process=6][thread=save_finalize][step=19] CheckpointManager Save Finalize is done on all hosts. I0425 13:50:47.080240 135413858600768 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. I0425 13:50:47.080429 135413858600768 checkpoint_manager.py:2009] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0425 13:50:47.081475 135413858600768 metric_logger.py:196] completed step: 19, seconds: 0.157, TFLOP/s/device: 86.673, Tokens/s/device: 13064.224, 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: Sat Apr 25 13:50:57 UTC 2026 EXIT_CODE=0