XPK Start: Thu Apr 23 21:37:17 UTC 2026 2026-04-23 21:37:34.926518: 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) Unrecognized keys in `rope_scaling` for 'rope_type'='yarn': {'rope_theta'} `rope_scaling`'s factor field must be a float >= 1, got 40 `rope_scaling`'s beta_fast field must be a float, got 32 `rope_scaling`'s beta_slow field must be a float, got 1 Unrecognized keys in `rope_scaling` for 'rope_type'='yarn': {'rope_theta'} Unrecognized keys in `rope_scaling` for 'rope_type'='yarn': {'rope_theta'} Unrecognized keys in `rope_scaling` for 'rope_type'='yarn': {'rope_theta'} I0423 21:37:38.946377 133240883222336 max_utils.py:273] Attempting to initialize the jax distributed system... INFO:2026-04-23 21:37:47,985:jax._src.distributed:149: Starting JAX distributed service on [::]:8482 I0423 21:37:47.985188 133240883222336 distributed.py:149] Starting JAX distributed service on [::]:8482 INFO:2026-04-23 21:37:47,987:jax._src.distributed:166: Connecting to JAX distributed service on mt-07-distill-smoke-cj7qb-slice-job-0-0.mt-07-distill-smoke-cj7qb:8482 I0423 21:37:47.987637 133240883222336 distributed.py:166] Connecting to JAX distributed service on mt-07-distill-smoke-cj7qb-slice-job-0-0.mt-07-distill-smoke-cj7qb:8482 I0423 21:37:48.979138 133240883222336 max_utils.py:284] Jax distributed system initialized! I0423 21:37:55.639220 133240883222336 max_utils.py:244] Jax distributed system is already initialized. W0423 21:37:55.770690 133240883222336 pyconfig.py:111] base_output_directory is not provided; Using local directory called maxtext_output I0423 21:37:55.830936 133240883222336 max_utils.py:244] Jax distributed system is already initialized. I0423 21:37:55.832119 133240883222336 pyconfig.py:471] Config param abort_on_inf_loss: True I0423 21:37:55.832166 133240883222336 pyconfig.py:471] Config param abort_on_nan_loss: True I0423 21:37:55.832193 133240883222336 pyconfig.py:471] Config param act_quantization_calibration_method: absmax I0423 21:37:55.832214 133240883222336 pyconfig.py:471] Config param activation_dropout_for_audio: 0.0 I0423 21:37:55.832236 133240883222336 pyconfig.py:471] Config param activation_function_for_audio: gelu I0423 21:37:55.832255 133240883222336 pyconfig.py:471] Config param activations_in_float32: False I0423 21:37:55.832273 133240883222336 pyconfig.py:471] Config param adam_b1: 0.9 I0423 21:37:55.832293 133240883222336 pyconfig.py:471] Config param adam_b2: 0.95 I0423 21:37:55.832311 133240883222336 pyconfig.py:471] Config param adam_eps: 1e-08 I0423 21:37:55.832334 133240883222336 pyconfig.py:471] Config param adam_eps_root: 0.0 I0423 21:37:55.832351 133240883222336 pyconfig.py:471] Config param adam_weight_decay: 0.1 I0423 21:37:55.832368 133240883222336 pyconfig.py:471] Config param adamw_mask: [] I0423 21:37:55.832384 133240883222336 pyconfig.py:471] Config param add_bos: True I0423 21:37:55.832400 133240883222336 pyconfig.py:471] Config param add_eos: True I0423 21:37:55.832417 133240883222336 pyconfig.py:471] Config param allow_split_physical_axes: False I0423 21:37:55.832432 133240883222336 pyconfig.py:471] Config param ar_cache_axis_order: 1,2,0,3 I0423 21:37:55.832449 133240883222336 pyconfig.py:471] Config param async_checkpointing: True I0423 21:37:55.832466 133240883222336 pyconfig.py:471] Config param async_scheduling: False I0423 21:37:55.832483 133240883222336 pyconfig.py:471] Config param attention: dot_product I0423 21:37:55.832499 133240883222336 pyconfig.py:471] Config param attention_bias: False I0423 21:37:55.832515 133240883222336 pyconfig.py:471] Config param attention_dropout_for_audio: 0.0 I0423 21:37:55.832531 133240883222336 pyconfig.py:471] Config param attention_out: RematLocation.REMAT I0423 21:37:55.832557 133240883222336 pyconfig.py:471] Config param attention_output_dim: -1 I0423 21:37:55.832575 133240883222336 pyconfig.py:471] Config param attention_sink: False I0423 21:37:55.832590 133240883222336 pyconfig.py:471] Config param attention_type: global I0423 21:37:55.832606 133240883222336 pyconfig.py:471] Config param attn_logits_soft_cap: None I0423 21:37:55.832621 133240883222336 pyconfig.py:471] Config param audio_path: I0423 21:37:55.832638 133240883222336 pyconfig.py:471] Config param audio_placeholder: <|audio|> I0423 21:37:55.832654 133240883222336 pyconfig.py:471] Config param autoregressive_decode_assert: I0423 21:37:55.832670 133240883222336 pyconfig.py:471] Config param base_config: base.yml I0423 21:37:55.832692 133240883222336 pyconfig.py:471] Config param base_emb_dim: 16 I0423 21:37:55.832709 133240883222336 pyconfig.py:471] Config param base_mlp_dim: 64 I0423 21:37:55.832725 133240883222336 pyconfig.py:471] Config param base_moe_mlp_dim: -1 I0423 21:37:55.832740 133240883222336 pyconfig.py:471] Config param base_num_decoder_layers: 1 I0423 21:37:55.832756 133240883222336 pyconfig.py:471] Config param base_num_kv_heads: 2 I0423 21:37:55.832771 133240883222336 pyconfig.py:471] Config param base_num_query_heads: 2 I0423 21:37:55.832787 133240883222336 pyconfig.py:471] Config param base_output_directory: /deps/maxtext_output I0423 21:37:55.832802 133240883222336 pyconfig.py:471] Config param batch_size: 1 I0423 21:37:55.832817 133240883222336 pyconfig.py:471] Config param batch_split_factor: 1 I0423 21:37:55.832832 133240883222336 pyconfig.py:471] Config param beta_fast: 32 I0423 21:37:55.832847 133240883222336 pyconfig.py:471] Config param beta_slow: 1 I0423 21:37:55.832864 133240883222336 pyconfig.py:471] Config param bwd_quantization_calibration_method: absmax I0423 21:37:55.832880 133240883222336 pyconfig.py:471] Config param capacity_factor: -1.0 I0423 21:37:55.832897 133240883222336 pyconfig.py:471] Config param cast_logits_to_fp32: True I0423 21:37:55.832912 133240883222336 pyconfig.py:471] Config param chat_template: I0423 21:37:55.832927 133240883222336 pyconfig.py:471] Config param chat_template_path: I0423 21:37:55.832944 133240883222336 pyconfig.py:471] Config param checkpoint_conversion_fn: None I0423 21:37:55.832960 133240883222336 pyconfig.py:471] Config param checkpoint_dir: /deps/maxtext_output/gpt3-52k_2026-04-23-21-37/checkpoints/ I0423 21:37:55.832977 133240883222336 pyconfig.py:471] Config param checkpoint_is_quantized: False I0423 21:37:55.832993 133240883222336 pyconfig.py:471] Config param checkpoint_period: 2000 I0423 21:37:55.833008 133240883222336 pyconfig.py:471] Config param checkpoint_storage_concurrent_gb: 96 I0423 21:37:55.833024 133240883222336 pyconfig.py:471] Config param checkpoint_storage_target_data_file_size_bytes: 2147483648 I0423 21:37:55.833039 133240883222336 pyconfig.py:471] Config param checkpoint_storage_use_ocdbt: True I0423 21:37:55.833055 133240883222336 pyconfig.py:471] Config param checkpoint_storage_use_zarr3: True I0423 21:37:55.833069 133240883222336 pyconfig.py:471] Config param checkpoint_todelete_full_path: None I0423 21:37:55.833085 133240883222336 pyconfig.py:471] Config param checkpoint_todelete_subdir: None I0423 21:37:55.833122 133240883222336 pyconfig.py:471] Config param chips_per_vm: 4 I0423 21:37:55.833138 133240883222336 pyconfig.py:471] Config param chunk_attn_window_size: 0 I0423 21:37:55.833153 133240883222336 pyconfig.py:471] Config param collect_stack_trace: False I0423 21:37:55.833170 133240883222336 pyconfig.py:471] Config param colocated_python_checkpointing: False I0423 21:37:55.833185 133240883222336 pyconfig.py:471] Config param colocated_python_data_input: False I0423 21:37:55.833201 133240883222336 pyconfig.py:471] Config param compile_topology: I0423 21:37:55.833216 133240883222336 pyconfig.py:471] Config param compile_topology_num_slices: -1 I0423 21:37:55.833231 133240883222336 pyconfig.py:471] Config param compile_xla_flags: I0423 21:37:55.833246 133240883222336 pyconfig.py:471] Config param compiled_trainstep_file: I0423 21:37:55.833261 133240883222336 pyconfig.py:471] Config param compute_axis_order: 0,1,2,3 I0423 21:37:55.833276 133240883222336 pyconfig.py:471] Config param constant_bound_config: [] I0423 21:37:55.833291 133240883222336 pyconfig.py:471] Config param context: RematLocation.REMAT I0423 21:37:55.833307 133240883222336 pyconfig.py:471] Config param context_parallel_load_balance: True I0423 21:37:55.833321 133240883222336 pyconfig.py:471] Config param context_parallel_reorder_strategy: ReorderStrategy.AUTO I0423 21:37:55.833339 133240883222336 pyconfig.py:471] Config param context_parallel_size: 1 I0423 21:37:55.833353 133240883222336 pyconfig.py:471] Config param context_parallel_strategy: all_gather I0423 21:37:55.833370 133240883222336 pyconfig.py:471] Config param context_sharding: context I0423 21:37:55.833385 133240883222336 pyconfig.py:471] Config param conv_chunksize_for_audio: 500 I0423 21:37:55.833401 133240883222336 pyconfig.py:471] Config param conv_stride_for_vit: 14 I0423 21:37:55.833415 133240883222336 pyconfig.py:471] Config param convert_checkpoint_if_possible: False I0423 21:37:55.833431 133240883222336 pyconfig.py:471] Config param cost_estimate_flops_bwd: -1 I0423 21:37:55.833447 133240883222336 pyconfig.py:471] Config param cost_estimate_flops_fwd: -1 I0423 21:37:55.833463 133240883222336 pyconfig.py:471] Config param custom_mesh: I0423 21:37:55.833477 133240883222336 pyconfig.py:471] Config param custom_mesh_and_rule: I0423 21:37:55.833492 133240883222336 pyconfig.py:471] Config param d_model_for_audio: 256 I0423 21:37:55.833507 133240883222336 pyconfig.py:471] Config param data_sharding: (('data', 'stage', 'fsdp', 'fsdp_transpose', 'sequence', 'context', 'context_autoregressive', 'tensor', 'tensor_transpose', 'tensor_sequence', 'expert', 'autoregressive'),) I0423 21:37:55.833526 133240883222336 pyconfig.py:471] Config param data_shuffle_seed: 0 I0423 21:37:55.833540 133240883222336 pyconfig.py:471] Config param dataset_name: c4/en:3.0.1 I0423 21:37:55.833560 133240883222336 pyconfig.py:471] Config param dataset_path: I0423 21:37:55.833575 133240883222336 pyconfig.py:471] Config param dataset_type: DatasetType.HF I0423 21:37:55.833592 133240883222336 pyconfig.py:471] Config param dcn_autoregressive_parallelism: 1 I0423 21:37:55.833606 133240883222336 pyconfig.py:471] Config param dcn_context_autoregressive_parallelism: 1 I0423 21:37:55.833622 133240883222336 pyconfig.py:471] Config param dcn_context_parallelism: 1 I0423 21:37:55.833636 133240883222336 pyconfig.py:471] Config param dcn_data_parallelism: -1 I0423 21:37:55.833652 133240883222336 pyconfig.py:471] Config param dcn_diloco_parallelism: 1 I0423 21:37:55.833666 133240883222336 pyconfig.py:471] Config param dcn_expert_parallelism: 1 I0423 21:37:55.833681 133240883222336 pyconfig.py:471] Config param dcn_fsdp_parallelism: 1 I0423 21:37:55.833696 133240883222336 pyconfig.py:471] Config param dcn_fsdp_transpose_parallelism: 1 I0423 21:37:55.833711 133240883222336 pyconfig.py:471] Config param dcn_parallelism: [1, -1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1] I0423 21:37:55.833728 133240883222336 pyconfig.py:471] Config param dcn_pipeline_parallelism: 1 I0423 21:37:55.833743 133240883222336 pyconfig.py:471] Config param dcn_sequence_parallelism: 1 I0423 21:37:55.833758 133240883222336 pyconfig.py:471] Config param dcn_tensor_parallelism: 1 I0423 21:37:55.833773 133240883222336 pyconfig.py:471] Config param dcn_tensor_sequence_parallelism: 1 I0423 21:37:55.833789 133240883222336 pyconfig.py:471] Config param dcn_tensor_transpose_parallelism: 1 I0423 21:37:55.833803 133240883222336 pyconfig.py:471] Config param debug: {'rl': False} I0423 21:37:55.833819 133240883222336 pyconfig.py:471] Config param debug_sharding: False I0423 21:37:55.833835 133240883222336 pyconfig.py:471] Config param decode_sampling_nucleus_p: -1 I0423 21:37:55.833849 133240883222336 pyconfig.py:471] Config param decode_sampling_strategy: SamplingStrategy.GREEDY I0423 21:37:55.833867 133240883222336 pyconfig.py:471] Config param decode_sampling_temperature: 1.0 I0423 21:37:55.833882 133240883222336 pyconfig.py:471] Config param decode_sampling_top_k: 0 I0423 21:37:55.833898 133240883222336 pyconfig.py:471] Config param decoder_block: DecoderBlockType.GPT3 I0423 21:37:55.833913 133240883222336 pyconfig.py:471] Config param decoder_layer_input: RematLocation.DEVICE I0423 21:37:55.833930 133240883222336 pyconfig.py:471] Config param deepstack_visual_indexes_for_vit: [] I0423 21:37:55.833944 133240883222336 pyconfig.py:471] Config param degenerate_group_masking: True I0423 21:37:55.833959 133240883222336 pyconfig.py:471] Config param dense_init_scale: 1.0 I0423 21:37:55.833974 133240883222336 pyconfig.py:471] Config param diloco_outer_lr: 0.3 I0423 21:37:55.833990 133240883222336 pyconfig.py:471] Config param diloco_outer_momentum: 0.9 I0423 21:37:55.834005 133240883222336 pyconfig.py:471] Config param diloco_sync_period: 36 I0423 21:37:55.834021 133240883222336 pyconfig.py:471] Config param distill_alpha: 0.5 I0423 21:37:55.834035 133240883222336 pyconfig.py:471] Config param distill_alpha_end: None I0423 21:37:55.834051 133240883222336 pyconfig.py:471] Config param distill_alpha_schedule: constant I0423 21:37:55.834066 133240883222336 pyconfig.py:471] Config param distill_beta: 0.0 I0423 21:37:55.834081 133240883222336 pyconfig.py:471] Config param distill_beta_end: None I0423 21:37:55.834104 133240883222336 pyconfig.py:471] Config param distill_beta_schedule: constant I0423 21:37:55.834120 133240883222336 pyconfig.py:471] Config param distill_feature_loss_type: cosine I0423 21:37:55.834134 133240883222336 pyconfig.py:471] Config param distill_layer_indices: None I0423 21:37:55.834150 133240883222336 pyconfig.py:471] Config param distill_temperature: 1.0 I0423 21:37:55.834166 133240883222336 pyconfig.py:471] Config param distill_temperature_end: None I0423 21:37:55.834180 133240883222336 pyconfig.py:471] Config param distill_temperature_schedule: constant I0423 21:37:55.834195 133240883222336 pyconfig.py:471] Config param downsample_hidden_size_for_audio: 256 I0423 21:37:55.834210 133240883222336 pyconfig.py:471] Config param dpo_beta: 0.1 I0423 21:37:55.834225 133240883222336 pyconfig.py:471] Config param dpo_label_smoothing: 0.0 I0423 21:37:55.834240 133240883222336 pyconfig.py:471] Config param dq_reduction_steps: 0 I0423 21:37:55.834256 133240883222336 pyconfig.py:471] Config param dropout_rate: 0.0 I0423 21:37:55.834270 133240883222336 pyconfig.py:471] Config param dtype: bfloat16 I0423 21:37:55.834299 133240883222336 pyconfig.py:471] Config param dtype_mm: float32 I0423 21:37:55.834314 133240883222336 pyconfig.py:471] Config param dump_hlo: False I0423 21:37:55.834330 133240883222336 pyconfig.py:471] Config param dump_hlo_delete_local_after: True I0423 21:37:55.834345 133240883222336 pyconfig.py:471] Config param dump_hlo_gcs_dir: /deps/maxtext_output/gpt3-52k_2026-04-23-21-37/xla_dump I0423 21:37:55.834359 133240883222336 pyconfig.py:471] Config param dump_hlo_local_dir: /tmp/xla_dump/ I0423 21:37:55.834375 133240883222336 pyconfig.py:471] Config param dump_hlo_local_module_name: jit_train_step I0423 21:37:55.834391 133240883222336 pyconfig.py:471] Config param dump_hlo_module_name: jit_train_step I0423 21:37:55.834405 133240883222336 pyconfig.py:471] Config param dump_hlo_upload_all: False I0423 21:37:55.834420 133240883222336 pyconfig.py:471] Config param dump_hlo_xla_flags: I0423 21:37:55.834436 133240883222336 pyconfig.py:471] Config param dump_jaxpr: False I0423 21:37:55.834452 133240883222336 pyconfig.py:471] Config param dump_jaxpr_delete_local_after: True I0423 21:37:55.834467 133240883222336 pyconfig.py:471] Config param dump_jaxpr_gcs_dir: /deps/maxtext_output/gpt3-52k_2026-04-23-21-37/jaxpr_dump I0423 21:37:55.834483 133240883222336 pyconfig.py:471] Config param dump_jaxpr_local_dir: /tmp/jaxpr_dump/ I0423 21:37:55.834497 133240883222336 pyconfig.py:471] Config param dump_step: -1 I0423 21:37:55.834513 133240883222336 pyconfig.py:471] Config param elastic_enabled: False I0423 21:37:55.834527 133240883222336 pyconfig.py:471] Config param elastic_max_retries: 10 I0423 21:37:55.834547 133240883222336 pyconfig.py:471] Config param elastic_timeout_seconds: 300 I0423 21:37:55.834562 133240883222336 pyconfig.py:471] Config param emb_dim: 16 I0423 21:37:55.834578 133240883222336 pyconfig.py:471] Config param enable_autocheckpoint: False I0423 21:37:55.834594 133240883222336 pyconfig.py:471] Config param enable_checkpoint_cloud_logger: False I0423 21:37:55.834609 133240883222336 pyconfig.py:471] Config param enable_checkpointing: True I0423 21:37:55.834625 133240883222336 pyconfig.py:471] Config param enable_continuous_checkpointing: False I0423 21:37:55.834640 133240883222336 pyconfig.py:471] Config param enable_data_shuffling: True I0423 21:37:55.834655 133240883222336 pyconfig.py:471] Config param enable_diloco: False I0423 21:37:55.834670 133240883222336 pyconfig.py:471] Config param enable_dp_attention: False I0423 21:37:55.834686 133240883222336 pyconfig.py:471] Config param enable_dropout: False I0423 21:37:55.834700 133240883222336 pyconfig.py:471] Config param enable_emergency_checkpoint: False I0423 21:37:55.834716 133240883222336 pyconfig.py:471] Config param enable_expert_parallel: False I0423 21:37:55.834731 133240883222336 pyconfig.py:471] Config param enable_gcp_goodput_metrics: True I0423 21:37:55.834746 133240883222336 pyconfig.py:471] Config param enable_gcp_step_deviation_metrics: True I0423 21:37:55.834761 133240883222336 pyconfig.py:471] Config param enable_goodput_recording: False I0423 21:37:55.834775 133240883222336 pyconfig.py:471] Config param enable_jax_profiler: False I0423 21:37:55.834791 133240883222336 pyconfig.py:471] Config param enable_llm_inference_pool: False I0423 21:37:55.834805 133240883222336 pyconfig.py:471] Config param enable_model_warmup: False I0423 21:37:55.834821 133240883222336 pyconfig.py:471] Config param enable_multi_tier_checkpointing: False I0423 21:37:55.834835 133240883222336 pyconfig.py:471] Config param enable_nnx: False I0423 21:37:55.834850 133240883222336 pyconfig.py:471] Config param enable_orbax_v1: False I0423 21:37:55.834866 133240883222336 pyconfig.py:471] Config param enable_padding_causal_mask: True I0423 21:37:55.834880 133240883222336 pyconfig.py:471] Config param enable_pathways_goodput: False I0423 21:37:55.834896 133240883222336 pyconfig.py:471] Config param enable_prefix_caching: False I0423 21:37:55.834910 133240883222336 pyconfig.py:471] Config param enable_rampup_batch_size: False I0423 21:37:55.834926 133240883222336 pyconfig.py:471] Config param enable_single_controller: False I0423 21:37:55.834940 133240883222336 pyconfig.py:471] Config param enable_single_replica_ckpt_restoring: False I0423 21:37:55.834956 133240883222336 pyconfig.py:471] Config param enable_tensorboard: True I0423 21:37:55.834971 133240883222336 pyconfig.py:471] Config param enable_tunix_perf_metrics: False I0423 21:37:55.834986 133240883222336 pyconfig.py:471] Config param encoder_attention_heads_for_audio: 4 I0423 21:37:55.835000 133240883222336 pyconfig.py:471] Config param encoder_ffn_dim_for_audio: 512 I0423 21:37:55.835016 133240883222336 pyconfig.py:471] Config param encoder_layers_for_audio: 2 I0423 21:37:55.835031 133240883222336 pyconfig.py:471] Config param engram: RematLocation.REMAT I0423 21:37:55.835047 133240883222336 pyconfig.py:471] Config param engram_head_dim: 1280 I0423 21:37:55.835062 133240883222336 pyconfig.py:471] Config param engram_kernel_size: 4 I0423 21:37:55.835077 133240883222336 pyconfig.py:471] Config param engram_layers: [] I0423 21:37:55.835101 133240883222336 pyconfig.py:471] Config param engram_max_ngram_size: 3 I0423 21:37:55.835118 133240883222336 pyconfig.py:471] Config param engram_num_heads: 8 I0423 21:37:55.835132 133240883222336 pyconfig.py:471] Config param engram_seed: 0 I0423 21:37:55.835146 133240883222336 pyconfig.py:471] Config param engram_vocab_bases: [] I0423 21:37:55.835161 133240883222336 pyconfig.py:471] Config param epsilon_high: None I0423 21:37:55.835177 133240883222336 pyconfig.py:471] Config param eval_corr_lst: False I0423 21:37:55.835192 133240883222336 pyconfig.py:471] Config param eval_data_columns: ['text'] I0423 21:37:55.835207 133240883222336 pyconfig.py:471] Config param eval_dataset_name: c4/en:3.0.1 I0423 21:37:55.835222 133240883222336 pyconfig.py:471] Config param eval_image_column: image I0423 21:37:55.835236 133240883222336 pyconfig.py:471] Config param eval_interval: -1 I0423 21:37:55.835252 133240883222336 pyconfig.py:471] Config param eval_make_lst: False I0423 21:37:55.835268 133240883222336 pyconfig.py:471] Config param eval_per_device_batch_size: 2 I0423 21:37:55.835283 133240883222336 pyconfig.py:471] Config param eval_sampling_strategy: greedy I0423 21:37:55.835298 133240883222336 pyconfig.py:471] Config param eval_split: validation I0423 21:37:55.835314 133240883222336 pyconfig.py:471] Config param eval_steps: -1 I0423 21:37:55.835328 133240883222336 pyconfig.py:471] Config param expansion_factor_real_data: -1.0 I0423 21:37:55.835343 133240883222336 pyconfig.py:471] Config param final_logits_soft_cap: None I0423 21:37:55.835359 133240883222336 pyconfig.py:471] Config param first_num_dense_layers: 0 I0423 21:37:55.835375 133240883222336 pyconfig.py:471] Config param float32_gate_logits: False I0423 21:37:55.835389 133240883222336 pyconfig.py:471] Config param float32_logits: False I0423 21:37:55.835405 133240883222336 pyconfig.py:471] Config param float32_qk_product: False I0423 21:37:55.835419 133240883222336 pyconfig.py:471] Config param float32_weight_sum: True I0423 21:37:55.835434 133240883222336 pyconfig.py:471] Config param force_q_layout: False I0423 21:37:55.835450 133240883222336 pyconfig.py:471] Config param force_unroll: False I0423 21:37:55.835464 133240883222336 pyconfig.py:471] Config param freeze_audio_encoder_params: True I0423 21:37:55.835479 133240883222336 pyconfig.py:471] Config param freeze_vision_encoder_params: True I0423 21:37:55.835495 133240883222336 pyconfig.py:471] Config param fused_mlp: False I0423 21:37:55.835509 133240883222336 pyconfig.py:471] Config param fused_qkv: True I0423 21:37:55.835524 133240883222336 pyconfig.py:471] Config param gcs_metrics: False I0423 21:37:55.835539 133240883222336 pyconfig.py:471] Config param gdn_chunk_size: 64 I0423 21:37:55.835557 133240883222336 pyconfig.py:471] Config param gdn_conv_kernel_dim: 4 I0423 21:37:55.835572 133240883222336 pyconfig.py:471] Config param gdn_key_head_dim: 128 I0423 21:37:55.835587 133240883222336 pyconfig.py:471] Config param gdn_num_key_heads: 16 I0423 21:37:55.835601 133240883222336 pyconfig.py:471] Config param gdn_num_value_heads: 32 I0423 21:37:55.835617 133240883222336 pyconfig.py:471] Config param gdn_value_head_dim: 128 I0423 21:37:55.835631 133240883222336 pyconfig.py:471] Config param generate_padding_batch_eval: False I0423 21:37:55.835646 133240883222336 pyconfig.py:471] Config param generate_padding_batch_train: False I0423 21:37:55.835661 133240883222336 pyconfig.py:471] Config param generate_slice: v5e-16 I0423 21:37:55.835676 133240883222336 pyconfig.py:471] Config param generation_configs: {} I0423 21:37:55.835690 133240883222336 pyconfig.py:471] Config param global_batch_size_to_eval_on: 64 I0423 21:37:55.835706 133240883222336 pyconfig.py:471] Config param global_batch_size_to_load: 512 I0423 21:37:55.835721 133240883222336 pyconfig.py:471] Config param global_batch_size_to_load_eval: 64 I0423 21:37:55.835736 133240883222336 pyconfig.py:471] Config param global_batch_size_to_load_increment: None I0423 21:37:55.835752 133240883222336 pyconfig.py:471] Config param global_batch_size_to_load_start: None I0423 21:37:55.835767 133240883222336 pyconfig.py:471] Config param global_batch_size_to_train_on: 512 I0423 21:37:55.835782 133240883222336 pyconfig.py:471] Config param global_head_dim: 0 I0423 21:37:55.835797 133240883222336 pyconfig.py:471] Config param global_num_kv_heads: 0 I0423 21:37:55.835812 133240883222336 pyconfig.py:471] Config param global_parameter_scale: 1 I0423 21:37:55.835827 133240883222336 pyconfig.py:471] Config param global_rampup_samples: 500 I0423 21:37:55.835843 133240883222336 pyconfig.py:471] Config param global_rope_max_timescale: -1 I0423 21:37:55.835858 133240883222336 pyconfig.py:471] Config param global_rope_proportion: 0.25 I0423 21:37:55.835874 133240883222336 pyconfig.py:471] Config param goodput_upload_interval_seconds: 30 I0423 21:37:55.835889 133240883222336 pyconfig.py:471] Config param grad_dtype: float32 I0423 21:37:55.835923 133240883222336 pyconfig.py:471] Config param gradient_accumulation_steps: 8 I0423 21:37:55.835940 133240883222336 pyconfig.py:471] Config param gradient_clipping_threshold: 1.0 I0423 21:37:55.835957 133240883222336 pyconfig.py:471] Config param grain_data_source_max_workers: 16 I0423 21:37:55.835972 133240883222336 pyconfig.py:471] Config param grain_eval_files: I0423 21:37:55.835987 133240883222336 pyconfig.py:471] Config param grain_file_type: arrayrecord I0423 21:37:55.836003 133240883222336 pyconfig.py:471] Config param grain_num_threads: 16 I0423 21:37:55.836017 133240883222336 pyconfig.py:471] Config param grain_num_threads_eval: 16 I0423 21:37:55.836033 133240883222336 pyconfig.py:471] Config param grain_packing_type: first_fit I0423 21:37:55.836047 133240883222336 pyconfig.py:471] Config param grain_per_worker_buffer_size: 1 I0423 21:37:55.836062 133240883222336 pyconfig.py:471] Config param grain_per_worker_buffer_size_eval: 1 I0423 21:37:55.836078 133240883222336 pyconfig.py:471] Config param grain_prefetch_buffer_size: 500 I0423 21:37:55.836107 133240883222336 pyconfig.py:471] Config param grain_prefetch_buffer_size_eval: 500 I0423 21:37:55.836123 133240883222336 pyconfig.py:471] Config param grain_ram_budget_mb: 1024 I0423 21:37:55.836139 133240883222336 pyconfig.py:471] Config param grain_shuffle_buffer_size: 100 I0423 21:37:55.836153 133240883222336 pyconfig.py:471] Config param grain_train_files: I0423 21:37:55.836169 133240883222336 pyconfig.py:471] Config param grain_train_mixture_config_path: I0423 21:37:55.836184 133240883222336 pyconfig.py:471] Config param grain_worker_count: 1 I0423 21:37:55.836200 133240883222336 pyconfig.py:471] Config param grain_worker_count_eval: 1 I0423 21:37:55.836216 133240883222336 pyconfig.py:471] Config param grpo_beta: 0.08 I0423 21:37:55.836231 133240883222336 pyconfig.py:471] Config param grpo_epsilon: 0.2 I0423 21:37:55.836246 133240883222336 pyconfig.py:471] Config param hardware: tpu I0423 21:37:55.836262 133240883222336 pyconfig.py:471] Config param hbm_utilization_vllm: 0.72 I0423 21:37:55.836277 133240883222336 pyconfig.py:471] Config param head_dim: 8 I0423 21:37:55.836293 133240883222336 pyconfig.py:471] Config param heartbeat_reporting_interval_in_seconds: 5 I0423 21:37:55.836309 133240883222336 pyconfig.py:471] Config param hf_data_dir: None I0423 21:37:55.836323 133240883222336 pyconfig.py:471] Config param hf_eval_files: None I0423 21:37:55.836339 133240883222336 pyconfig.py:471] Config param hf_eval_split: None I0423 21:37:55.836353 133240883222336 pyconfig.py:471] Config param hf_name: None I0423 21:37:55.836369 133240883222336 pyconfig.py:471] Config param hf_path: OptimalScale/ClimbMix I0423 21:37:55.836383 133240883222336 pyconfig.py:471] Config param hf_train_files: None I0423 21:37:55.836399 133240883222336 pyconfig.py:471] Config param hidden_size_for_vit: 1408 I0423 21:37:55.836413 133240883222336 pyconfig.py:471] Config param hide_profiler_step_metric: False I0423 21:37:55.836429 133240883222336 pyconfig.py:471] Config param ici_autoregressive_parallelism: 1 I0423 21:37:55.836444 133240883222336 pyconfig.py:471] Config param ici_context_autoregressive_parallelism: 1 I0423 21:37:55.836458 133240883222336 pyconfig.py:471] Config param ici_context_parallelism: 1 I0423 21:37:55.836474 133240883222336 pyconfig.py:471] Config param ici_data_parallelism: 1 I0423 21:37:55.836489 133240883222336 pyconfig.py:471] Config param ici_diloco_parallelism: 1 I0423 21:37:55.836505 133240883222336 pyconfig.py:471] Config param ici_expert_parallelism: 1 I0423 21:37:55.836519 133240883222336 pyconfig.py:471] Config param ici_fsdp_parallelism: -1 I0423 21:37:55.836534 133240883222336 pyconfig.py:471] Config param ici_fsdp_transpose_parallelism: 1 I0423 21:37:55.836552 133240883222336 pyconfig.py:471] Config param ici_parallelism: [1, 1, 1, -1, 1, 1, 1, 1, 1, 1, 1, 1, 1] I0423 21:37:55.836568 133240883222336 pyconfig.py:471] Config param ici_pipeline_parallelism: 1 I0423 21:37:55.836584 133240883222336 pyconfig.py:471] Config param ici_sequence_parallelism: 1 I0423 21:37:55.836599 133240883222336 pyconfig.py:471] Config param ici_tensor_parallelism: 1 I0423 21:37:55.836613 133240883222336 pyconfig.py:471] Config param ici_tensor_sequence_parallelism: 1 I0423 21:37:55.836628 133240883222336 pyconfig.py:471] Config param ici_tensor_transpose_parallelism: 1 I0423 21:37:55.836644 133240883222336 pyconfig.py:471] Config param image_path: I0423 21:37:55.836659 133240883222336 pyconfig.py:471] Config param image_placeholder: <|image|> I0423 21:37:55.836673 133240883222336 pyconfig.py:471] Config param image_size_for_vit: 896 I0423 21:37:55.836689 133240883222336 pyconfig.py:471] Config param indexer_head_dim: 128 I0423 21:37:55.836704 133240883222336 pyconfig.py:471] Config param indexer_loss_scaling_factor: 0.0 I0423 21:37:55.836720 133240883222336 pyconfig.py:471] Config param indexer_n_heads: 64 I0423 21:37:55.836735 133240883222336 pyconfig.py:471] Config param indexer_sparse_training: False I0423 21:37:55.836750 133240883222336 pyconfig.py:471] Config param indexer_topk: 2048 I0423 21:37:55.836764 133240883222336 pyconfig.py:471] Config param inference_benchmark_test: False I0423 21:37:55.836780 133240883222336 pyconfig.py:471] Config param inference_metadata_file: I0423 21:37:55.836797 133240883222336 pyconfig.py:471] Config param inference_microbenchmark_log_file_path: I0423 21:37:55.836814 133240883222336 pyconfig.py:471] Config param inference_microbenchmark_loop_iters: 10 I0423 21:37:55.836830 133240883222336 pyconfig.py:471] Config param inference_microbenchmark_num_samples: [1, 2, 3, 4, 5] I0423 21:37:55.836846 133240883222336 pyconfig.py:471] Config param inference_microbenchmark_prefill_lengths: 64,128,256,512,1024 I0423 21:37:55.836863 133240883222336 pyconfig.py:471] Config param inference_microbenchmark_stages: prefill,generate I0423 21:37:55.836880 133240883222336 pyconfig.py:471] Config param inference_server: MaxtextInterleavedServer I0423 21:37:55.836894 133240883222336 pyconfig.py:471] Config param inhomogeneous_layer_cycle_interval: 1 I0423 21:37:55.836910 133240883222336 pyconfig.py:471] Config param init_weights_seed: 0 I0423 21:37:55.836925 133240883222336 pyconfig.py:471] Config param input_data_sharding_logical_axes: ['activation_embed_and_logits_batch', 'activation_norm_length'] I0423 21:37:55.836941 133240883222336 pyconfig.py:471] Config param interleave_moe_layer_step: 1 I0423 21:37:55.836956 133240883222336 pyconfig.py:471] Config param intermediate_size_for_vit: 5632 I0423 21:37:55.836971 133240883222336 pyconfig.py:471] Config param internal_compile: False I0423 21:37:55.836986 133240883222336 pyconfig.py:471] Config param internal_compile_num_devices: -1 I0423 21:37:55.837002 133240883222336 pyconfig.py:471] Config param jax_cache_dir: ~/jax_cache I0423 21:37:55.837018 133240883222336 pyconfig.py:471] Config param jax_debug_log_modules: I0423 21:37:55.837032 133240883222336 pyconfig.py:471] Config param jax_distributed_initialization_timeout: 300 I0423 21:37:55.837048 133240883222336 pyconfig.py:471] Config param jax_profiler_port: 9999 I0423 21:37:55.837062 133240883222336 pyconfig.py:471] Config param key_proj: RematLocation.REMAT I0423 21:37:55.837078 133240883222336 pyconfig.py:471] Config param kv_cache_buffer: 256 I0423 21:37:55.837110 133240883222336 pyconfig.py:471] Config param kv_lora_rank: 512 I0423 21:37:55.837126 133240883222336 pyconfig.py:471] Config param kv_quant_axis: KvQuantAxis.HEADS_AND_DKV I0423 21:37:55.837144 133240883222336 pyconfig.py:471] Config param kv_quant_dtype: int8 I0423 21:37:55.837161 133240883222336 pyconfig.py:471] Config param kv_wa_proj: RematLocation.REMAT I0423 21:37:55.837177 133240883222336 pyconfig.py:471] Config param learning_rate: 0.0002 I0423 21:37:55.837194 133240883222336 pyconfig.py:471] Config param learning_rate_final_fraction: 0.1 I0423 21:37:55.837208 133240883222336 pyconfig.py:471] Config param learning_rate_schedule_steps: 200000 I0423 21:37:55.837224 133240883222336 pyconfig.py:471] Config param load_balance_loss_weight: 0.0 I0423 21:37:55.837238 133240883222336 pyconfig.py:471] Config param load_checkpoint_only_once: False I0423 21:37:55.837254 133240883222336 pyconfig.py:471] Config param load_from_prefill_dir: False I0423 21:37:55.837270 133240883222336 pyconfig.py:471] Config param load_full_state_path: I0423 21:37:55.837287 133240883222336 pyconfig.py:471] Config param load_parameters_path: gs://lance-maxtext/pt_seed_ckpts/pt_seed_ckpts/pt_seed_ckpt_gpt352k_v32k_linen/checkpoints/4/items I0423 21:37:55.837304 133240883222336 pyconfig.py:471] Config param local_checkpoint_directory: I0423 21:37:55.837320 133240883222336 pyconfig.py:471] Config param local_checkpoint_period: 0 I0423 21:37:55.837336 133240883222336 pyconfig.py:471] Config param local_rope_max_timescale: -1 I0423 21:37:55.837350 133240883222336 pyconfig.py:471] Config param local_rope_proportion: 1.0 I0423 21:37:55.837366 133240883222336 pyconfig.py:471] Config param log_config: True I0423 21:37:55.837380 133240883222336 pyconfig.py:471] Config param log_period: 10 I0423 21:37:55.837395 133240883222336 pyconfig.py:471] Config param logical_axis_rules: (('activation_embed_and_logits_batch', ('data', 'stage', 'fsdp', 'fsdp_transpose', 'expert')), ('activation_embed_and_logits_batch_sequence', ('data', 'stage', 'fsdp', 'fsdp_transpose', 'sequence', 'context', 'expert')), ('activation_vocab', ('tensor', 'tensor_transpose', 'tensor_sequence')), ('activation_vocab', ('tensor', 'tensor_transpose')), ('activation_vocab', 'tensor_sequence'), ('activation_vocab', ('sequence', 'context')), ('vocab', ('tensor', 'tensor_transpose', 'tensor_sequence', 'autoregressive')), ('embed_vocab', ('fsdp', 'fsdp_transpose', 'sequence', 'context', 'expert')), ('activation_heads', ('tensor', 'tensor_transpose', 'sequence', 'tensor_sequence', 'autoregressive')), ('activation_kv_heads', ('tensor', 'tensor_transpose', 'sequence', 'tensor_sequence')), ('activation_attn_length', ('sequence', 'context')), ('activation_attn_length', ('context',)), ('activation_q_length', ('context',)), ('activation_kv_length', ()), ('activation_attn_embed', ('tensor', 'tensor_transpose')), ('activation_kv', ('tensor', 'tensor_transpose', 'tensor_sequence')), ('activation_kv_batch', ('data', 'fsdp', 'fsdp_transpose', 'expert')), ('activation_kv_head_dim', ('tensor', 'tensor_transpose', 'tensor_sequence')), ('heads', ('tensor', 'tensor_transpose', 'tensor_sequence', 'autoregressive')), ('q_heads', ('tensor', 'tensor_transpose', 'tensor_sequence', 'autoregressive')), ('kv_heads', ('tensor', 'tensor_transpose', 'tensor_sequence', 'autoregressive')), ('qkv', ()), ('kv', ()), ('kv_head_dim', ()), ('q_lora', ('fsdp', 'fsdp_transpose', 'sequence', 'context', 'tensor_transpose', 'expert')), ('q_lora', ('fsdp', 'sequence', 'context', 'tensor_transpose', 'expert')), ('q_lora', ('fsdp', 'fsdp_transpose', 'sequence', 'context', 'expert')), ('q_lora', ('fsdp', 'sequence', 'context', 'expert')), ('q_lora_up_proj', ()), ('kv_lora', ('fsdp', 'fsdp_transpose', 'sequence', 'context', 'tensor_transpose', 'expert')), ('kv_lora', ('fsdp', 'sequence', 'context', 'tensor_transpose', 'expert')), ('kv_lora', ('fsdp', 'fsdp_transpose', 'sequence', 'context', 'expert')), ('kv_lora', ('fsdp', 'sequence', 'context', 'expert')), ('kv_lora_up_proj', ()), ('activation_batch_moe', ('data', 'fsdp', 'fsdp_transpose')), ('activation_length_moe', ('sequence', 'context')), ('activation_length_moe', ('context',)), ('activation_norm_length_moe', ('tensor_sequence', 'context', 'sequence')), ('activation_embed_moe', ('tensor', 'tensor_transpose')), ('activation_mlp_moe', ('tensor', 'tensor_transpose', 'tensor_sequence')), ('activation_exp', ('expert',)), ('exp', 'expert'), ('mlp_moe', ('fsdp_transpose', 'tensor', 'tensor_sequence', 'autoregressive')), ('embed_moe', ('fsdp', 'fsdp_transpose', 'sequence', 'tensor_transpose', 'context')), ('embed_moe', ('fsdp', 'sequence', 'tensor_transpose', 'context')), ('embed_moe', ('fsdp', 'fsdp_transpose', 'sequence', 'context')), ('embed_moe', ('fsdp', 'sequence', 'context')), ('activation_mlp', ('tensor', 'tensor_transpose', 'tensor_sequence')), ('activation_batch', ('data', 'fsdp', 'fsdp_transpose', 'expert')), ('activation_length', ('sequence', 'context')), ('activation_length', ('context',)), ('activation_norm_length', ('tensor_sequence', 'context', 'sequence')), ('activation_embed', ('tensor', 'tensor_transpose')), ('activation_stage', 'stage'), ('mlp', ('fsdp_transpose', 'tensor', 'tensor_sequence', 'autoregressive')), ('embed', ('fsdp', 'fsdp_transpose', 'sequence', 'tensor_transpose', 'context', 'expert')), ('embed', ('fsdp', 'sequence', 'tensor_transpose', 'context', 'expert')), ('embed', ('fsdp', 'fsdp_transpose', 'sequence', 'context', 'expert')), ('embed', ('fsdp', 'sequence', 'context', 'expert')), ('norm', ('tensor', 'tensor_transpose')), ('layers', 'stage'), ('diloco', 'diloco'), ('engram_dim', ('tensor',)), ('dense_layers', ()), ('moe_layers', ()), ('mhc', ()), ('prefill_activation_length', ('sequence', 'context')), ('prefill_activation_norm_length', ('tensor_sequence', 'context', 'sequence')), ('activation_prefill_kv_batch', ('data', 'fsdp', 'fsdp_transpose', 'expert')), ('decode_batch', ('data', 'fsdp', 'fsdp_transpose', 'expert')), ('decode_length', ('sequence',)), ('cache_heads', ('autoregressive', 'tensor', 'tensor_transpose', 'tensor_sequence')), ('cache_heads', ('autoregressive', 'tensor', 'tensor_sequence')), ('paged_kv_heads', ('tensor',)), ('cache_batch_prefill', ()), ('cache_batch', ()), ('cache_heads_none', ()), ('cache_kv', ()), ('cache_sequence', ()), ('num_pages', ()), ('tokens_per_page', ()), ('paged_kv_head_dim_size', ()), ('mlp_no_fsdp', ('tensor', 'tensor_sequence', 'autoregressive')), ('embed_tensor_transpose', ('tensor_transpose',)), ('exp_with_fsdp', 'fsdp')) I0423 21:37:55.837466 133240883222336 pyconfig.py:471] Config param logits_dot_in_fp32: False I0423 21:37:55.837483 133240883222336 pyconfig.py:471] Config param logits_via_embedding: True I0423 21:37:55.837499 133240883222336 pyconfig.py:471] Config param lora_input_adapters_path: I0423 21:37:55.837515 133240883222336 pyconfig.py:471] Config param loss_algo: grpo I0423 21:37:55.837530 133240883222336 pyconfig.py:471] Config param lr_schedule_type: LearningRateScheduleType.COSINE I0423 21:37:55.837553 133240883222336 pyconfig.py:471] Config param managed_mldiagnostics: False I0423 21:37:55.837568 133240883222336 pyconfig.py:471] Config param managed_mldiagnostics_dir: /deps/maxtext_output/gpt3-52k_2026-04-23-21-37/managed-mldiagnostics I0423 21:37:55.837584 133240883222336 pyconfig.py:471] Config param managed_mldiagnostics_run_group: I0423 21:37:55.837599 133240883222336 pyconfig.py:471] Config param matmul_precision: MatmulPrecision.DEFAULT I0423 21:37:55.837616 133240883222336 pyconfig.py:471] Config param max_checkify: False I0423 21:37:55.837632 133240883222336 pyconfig.py:471] Config param max_concurrency: 256 I0423 21:37:55.837647 133240883222336 pyconfig.py:471] Config param max_corpus_chars: 10000000 I0423 21:37:55.837663 133240883222336 pyconfig.py:471] Config param max_num_batched_tokens: None I0423 21:37:55.837677 133240883222336 pyconfig.py:471] Config param max_num_checkpoints_to_keep: None I0423 21:37:55.837693 133240883222336 pyconfig.py:471] Config param max_num_images_per_example: -1 I0423 21:37:55.837709 133240883222336 pyconfig.py:471] Config param max_num_seqs: None I0423 21:37:55.837723 133240883222336 pyconfig.py:471] Config param max_position_embeddings: 163840 I0423 21:37:55.837738 133240883222336 pyconfig.py:471] Config param max_prefill_predict_length: 64 I0423 21:37:55.837754 133240883222336 pyconfig.py:471] Config param max_sample_len_for_audio: 10000 I0423 21:37:55.837768 133240883222336 pyconfig.py:471] Config param max_segments_per_seq: -1 I0423 21:37:55.837784 133240883222336 pyconfig.py:471] Config param max_source_positions_for_audio: 1500 I0423 21:37:55.837798 133240883222336 pyconfig.py:471] Config param max_target_length: 2048 I0423 21:37:55.837814 133240883222336 pyconfig.py:471] Config param max_timescale_for_audio: 10000.0 I0423 21:37:55.837829 133240883222336 pyconfig.py:471] Config param megablox: True I0423 21:37:55.837844 133240883222336 pyconfig.py:471] Config param merge_gating_gmm: False I0423 21:37:55.837858 133240883222336 pyconfig.py:471] Config param mesh_axes: ['diloco', 'data', 'stage', 'fsdp', 'fsdp_transpose', 'sequence', 'context', 'context_autoregressive', 'tensor', 'tensor_transpose', 'tensor_sequence', 'expert', 'autoregressive'] I0423 21:37:55.837876 133240883222336 pyconfig.py:471] Config param metrics_dir: /deps/maxtext_output/gpt3-52k_2026-04-23-21-37/metrics/ I0423 21:37:55.837890 133240883222336 pyconfig.py:471] Config param metrics_file: I0423 21:37:55.837905 133240883222336 pyconfig.py:471] Config param mhc_expansion_rate: 1 I0423 21:37:55.837920 133240883222336 pyconfig.py:471] Config param micro_batch_size_to_eval_on: 64 I0423 21:37:55.837936 133240883222336 pyconfig.py:471] Config param micro_batch_size_to_train_on: 64 I0423 21:37:55.837950 133240883222336 pyconfig.py:471] Config param mla_kv: RematLocation.REMAT I0423 21:37:55.837966 133240883222336 pyconfig.py:471] Config param mla_naive_kvcache: True I0423 21:37:55.837980 133240883222336 pyconfig.py:471] Config param mla_q: RematLocation.REMAT I0423 21:37:55.837996 133240883222336 pyconfig.py:471] Config param mlp_activations: ['gelu'] I0423 21:37:55.838012 133240883222336 pyconfig.py:471] Config param mlp_activations_limit: -1.0 I0423 21:37:55.838028 133240883222336 pyconfig.py:471] Config param mlp_bias: False I0423 21:37:55.838042 133240883222336 pyconfig.py:471] Config param mlp_dim: 64 I0423 21:37:55.838057 133240883222336 pyconfig.py:471] Config param mlpwi: RematLocation.REMAT I0423 21:37:55.838074 133240883222336 pyconfig.py:471] Config param mlpwi_0: RematLocation.REMAT I0423 21:37:55.838088 133240883222336 pyconfig.py:471] Config param mlpwi_1: RematLocation.REMAT I0423 21:37:55.838112 133240883222336 pyconfig.py:471] Config param mlpwo: RematLocation.REMAT I0423 21:37:55.838128 133240883222336 pyconfig.py:471] Config param moba: False I0423 21:37:55.838142 133240883222336 pyconfig.py:471] Config param moba_chunk_size: 1024 I0423 21:37:55.838156 133240883222336 pyconfig.py:471] Config param moba_topk: 8 I0423 21:37:55.838170 133240883222336 pyconfig.py:471] Config param model_call_mode: I0423 21:37:55.838186 133240883222336 pyconfig.py:471] Config param model_name: gpt3-52k I0423 21:37:55.838200 133240883222336 pyconfig.py:471] Config param moe_expert_input_dim: -1 I0423 21:37:55.838215 133240883222336 pyconfig.py:471] Config param moe_fsdp_use_two_stage_all_gather: False I0423 21:37:55.838230 133240883222336 pyconfig.py:471] Config param moe_mlp_dim: -1 I0423 21:37:55.838245 133240883222336 pyconfig.py:471] Config param moe_mlpwi_0: RematLocation.REMAT I0423 21:37:55.838260 133240883222336 pyconfig.py:471] Config param moe_mlpwi_1: RematLocation.REMAT I0423 21:37:55.838275 133240883222336 pyconfig.py:471] Config param moe_mlpwo: RematLocation.REMAT I0423 21:37:55.838290 133240883222336 pyconfig.py:471] Config param monitor_goodput: False I0423 21:37:55.838305 133240883222336 pyconfig.py:471] Config param monitor_step_time_deviation: True I0423 21:37:55.838320 133240883222336 pyconfig.py:471] Config param mrope_section: [24, 20, 20] I0423 21:37:55.838336 133240883222336 pyconfig.py:471] Config param mscale: 1.0 I0423 21:37:55.838350 133240883222336 pyconfig.py:471] Config param mtc_data_parallelism: 0 I0423 21:37:55.838366 133240883222336 pyconfig.py:471] Config param mtp_eval_target_module: 0 I0423 21:37:55.838380 133240883222336 pyconfig.py:471] Config param mtp_loss_scaling_factor: 0.1 I0423 21:37:55.838396 133240883222336 pyconfig.py:471] Config param mtp_num_layers: 0 I0423 21:37:55.838410 133240883222336 pyconfig.py:471] Config param mu_dtype: float32 I0423 21:37:55.838434 133240883222336 pyconfig.py:471] Config param multi_sampling: False I0423 21:37:55.838450 133240883222336 pyconfig.py:471] Config param multi_tier_checkpointing_backup_interval_minutes: 0 I0423 21:37:55.838465 133240883222336 pyconfig.py:471] Config param muon_beta: 0.95 I0423 21:37:55.838480 133240883222336 pyconfig.py:471] Config param muon_consistent_rms: None I0423 21:37:55.838496 133240883222336 pyconfig.py:471] Config param muon_weight_decay: 0.0 I0423 21:37:55.838512 133240883222336 pyconfig.py:471] Config param n_routing_groups: -1 I0423 21:37:55.838527 133240883222336 pyconfig.py:471] Config param n_window_for_audio: 50 I0423 21:37:55.838543 133240883222336 pyconfig.py:471] Config param n_window_infer_for_audio: 800 I0423 21:37:55.838563 133240883222336 pyconfig.py:471] Config param nope_layer_interval: -1 I0423 21:37:55.838580 133240883222336 pyconfig.py:471] Config param norm_topk_prob: False I0423 21:37:55.838594 133240883222336 pyconfig.py:471] Config param normalization_layer_epsilon: 1e-05 I0423 21:37:55.838612 133240883222336 pyconfig.py:471] Config param normalize_embedding_logits: False I0423 21:37:55.838626 133240883222336 pyconfig.py:471] Config param num_attention_heads_for_vit: 16 I0423 21:37:55.838641 133240883222336 pyconfig.py:471] Config param num_batches: 4 I0423 21:37:55.838656 133240883222336 pyconfig.py:471] Config param num_channels_for_vit: 3 I0423 21:37:55.838670 133240883222336 pyconfig.py:471] Config param num_conv_layers_for_audio: 3 I0423 21:37:55.838685 133240883222336 pyconfig.py:471] Config param num_decoder_layers: 1 I0423 21:37:55.838700 133240883222336 pyconfig.py:471] Config param num_diloco_replicas: 1 I0423 21:37:55.838716 133240883222336 pyconfig.py:471] Config param num_epoch: 1 I0423 21:37:55.838731 133240883222336 pyconfig.py:471] Config param num_eval_passes: 1 I0423 21:37:55.838746 133240883222336 pyconfig.py:471] Config param num_experts: 1 I0423 21:37:55.838761 133240883222336 pyconfig.py:471] Config param num_experts_per_tok: 1 I0423 21:37:55.838777 133240883222336 pyconfig.py:471] Config param num_generations: 2 I0423 21:37:55.838792 133240883222336 pyconfig.py:471] Config param num_hidden_layers_for_vit: 34 I0423 21:37:55.838808 133240883222336 pyconfig.py:471] Config param num_iterations: 1 I0423 21:37:55.838822 133240883222336 pyconfig.py:471] Config param num_kv_heads: 2 I0423 21:37:55.838838 133240883222336 pyconfig.py:471] Config param num_layers_per_pipeline_stage: 1 I0423 21:37:55.838854 133240883222336 pyconfig.py:471] Config param num_mel_bins_for_audio: 128 I0423 21:37:55.838869 133240883222336 pyconfig.py:471] Config param num_pipeline_microbatches: -1 I0423 21:37:55.838884 133240883222336 pyconfig.py:471] Config param num_pipeline_repeats: -1 I0423 21:37:55.838899 133240883222336 pyconfig.py:471] Config param num_position_embeddings_for_vit: 1024 I0423 21:37:55.838914 133240883222336 pyconfig.py:471] Config param num_query_heads: 2 I0423 21:37:55.838930 133240883222336 pyconfig.py:471] Config param num_samplers_slices: -1 I0423 21:37:55.838944 133240883222336 pyconfig.py:471] Config param num_slices: 1 I0423 21:37:55.838959 133240883222336 pyconfig.py:471] Config param num_target_devices: 32 I0423 21:37:55.838974 133240883222336 pyconfig.py:471] Config param num_test_batches: 5 I0423 21:37:55.838989 133240883222336 pyconfig.py:471] Config param num_trainer_slices: -1 I0423 21:37:55.839003 133240883222336 pyconfig.py:471] Config param num_vocab_tiling: 1 I0423 21:37:55.839018 133240883222336 pyconfig.py:471] Config param off_policy_steps: 0 I0423 21:37:55.839032 133240883222336 pyconfig.py:471] Config param offline_data_dir: None I0423 21:37:55.839048 133240883222336 pyconfig.py:471] Config param opt_type: OptimizerType.ADAM_PAX I0423 21:37:55.839066 133240883222336 pyconfig.py:471] Config param optimize_mesh_for_tpu_v6e: False I0423 21:37:55.839081 133240883222336 pyconfig.py:471] Config param optimizer_memory_host_offload: False I0423 21:37:55.839107 133240883222336 pyconfig.py:471] Config param original_max_position_embeddings: 4096 I0423 21:37:55.839123 133240883222336 pyconfig.py:471] Config param out_hidden_size_for_vit: 512 I0423 21:37:55.839138 133240883222336 pyconfig.py:471] Config param out_proj: RematLocation.REMAT I0423 21:37:55.839155 133240883222336 pyconfig.py:471] Config param output_dim_for_audio: 512 I0423 21:37:55.839170 133240883222336 pyconfig.py:471] Config param override_logical_axis_rules: False I0423 21:37:55.839186 133240883222336 pyconfig.py:471] Config param override_model_config: True I0423 21:37:55.839201 133240883222336 pyconfig.py:471] Config param packing: True I0423 21:37:55.839217 133240883222336 pyconfig.py:471] Config param pagedattn_head_dim_alignment: 128 I0423 21:37:55.839231 133240883222336 pyconfig.py:471] Config param pagedattn_max_pages_per_group: -1 I0423 21:37:55.839246 133240883222336 pyconfig.py:471] Config param pagedattn_num_pages: 64 I0423 21:37:55.839261 133240883222336 pyconfig.py:471] Config param pagedattn_pages_per_compute_block: 4 I0423 21:37:55.839276 133240883222336 pyconfig.py:471] Config param pagedattn_tokens_per_page: 32 I0423 21:37:55.839291 133240883222336 pyconfig.py:471] Config param param_scan_axis: 1 I0423 21:37:55.839306 133240883222336 pyconfig.py:471] Config param parameter_memory_host_offload: False I0423 21:37:55.839321 133240883222336 pyconfig.py:471] Config param partial_rotary_factor: 1.0 I0423 21:37:55.839337 133240883222336 pyconfig.py:471] Config param patch_size_for_vit: 14 I0423 21:37:55.839352 133240883222336 pyconfig.py:471] Config param penalty_incorrect_answer: -1.0 I0423 21:37:55.839368 133240883222336 pyconfig.py:471] Config param penalty_incorrect_format: -0.5 I0423 21:37:55.839384 133240883222336 pyconfig.py:471] Config param per_device_batch_size: 2 I0423 21:37:55.839400 133240883222336 pyconfig.py:471] Config param per_device_batch_size_increment: 2.0 I0423 21:37:55.839415 133240883222336 pyconfig.py:471] Config param per_device_batch_size_start: 4.0 I0423 21:37:55.839431 133240883222336 pyconfig.py:471] Config param pipeline_delay_activation_forwarding: False I0423 21:37:55.839445 133240883222336 pyconfig.py:471] Config param pipeline_fsdp_ag_once: False I0423 21:37:55.839461 133240883222336 pyconfig.py:471] Config param pipeline_fsdp_ag_per_repeat: False I0423 21:37:55.839475 133240883222336 pyconfig.py:471] Config param pipeline_parallel_layers: 1 I0423 21:37:55.839490 133240883222336 pyconfig.py:471] Config param pixel_shuffle_ratio_for_vit: 0.5 I0423 21:37:55.839506 133240883222336 pyconfig.py:471] Config param posemb_type_for_vit: learn I0423 21:37:55.839520 133240883222336 pyconfig.py:471] Config param position_id_per_seconds: 25 I0423 21:37:55.839536 133240883222336 pyconfig.py:471] Config param prefill_cache_axis_order: 1,2,0,3 I0423 21:37:55.839554 133240883222336 pyconfig.py:471] Config param prefill_cache_dir: I0423 21:37:55.839569 133240883222336 pyconfig.py:471] Config param prefill_chunk_size: 256 I0423 21:37:55.839584 133240883222336 pyconfig.py:471] Config param prefill_slice: v5e-16 I0423 21:37:55.839599 133240883222336 pyconfig.py:471] Config param prefix_caching_dram_byte: 100000000000 I0423 21:37:55.839614 133240883222336 pyconfig.py:471] Config param prefix_caching_hbm_byte: 10000000000 I0423 21:37:55.839629 133240883222336 pyconfig.py:471] Config param profile_cleanly: True I0423 21:37:55.839644 133240883222336 pyconfig.py:471] Config param profile_periodically_period: -1 I0423 21:37:55.839658 133240883222336 pyconfig.py:471] Config param profile_power_events: False I0423 21:37:55.839672 133240883222336 pyconfig.py:471] Config param profiler: ProfilerType.NONE I0423 21:37:55.839689 133240883222336 pyconfig.py:471] Config param profiler_steps: 5 I0423 21:37:55.839704 133240883222336 pyconfig.py:471] Config param projector_dropout_for_vit: 0.0 I0423 21:37:55.839718 133240883222336 pyconfig.py:471] Config param projector_input_dim_for_vit: 4096 I0423 21:37:55.839734 133240883222336 pyconfig.py:471] Config param projector_output_dim_for_vit: 4096 I0423 21:37:55.839751 133240883222336 pyconfig.py:471] Config param prometheus_port: 0 I0423 21:37:55.839765 133240883222336 pyconfig.py:471] Config param prompt: I love to I0423 21:37:55.839780 133240883222336 pyconfig.py:471] Config param pure_nnx: False I0423 21:37:55.839796 133240883222336 pyconfig.py:471] Config param pure_nnx_decoder: False I0423 21:37:55.839810 133240883222336 pyconfig.py:471] Config param q_lora_rank: 0 I0423 21:37:55.839826 133240883222336 pyconfig.py:471] Config param qk_clip_threshold: 100.0 I0423 21:37:55.839843 133240883222336 pyconfig.py:471] Config param qk_nope_head_dim: 128 I0423 21:37:55.839858 133240883222336 pyconfig.py:471] Config param qk_norm_with_scale: True I0423 21:37:55.839873 133240883222336 pyconfig.py:471] Config param qk_rope_head_dim: 64 I0423 21:37:55.839889 133240883222336 pyconfig.py:471] Config param qkv_proj: RematLocation.REMAT I0423 21:37:55.839905 133240883222336 pyconfig.py:471] Config param quant_cfg_path: I0423 21:37:55.839919 133240883222336 pyconfig.py:471] Config param quantization: QuantizationType.NONE I0423 21:37:55.839936 133240883222336 pyconfig.py:471] Config param quantization_local_shard_count: 4 I0423 21:37:55.839952 133240883222336 pyconfig.py:471] Config param quantize_kvcache: False I0423 21:37:55.839966 133240883222336 pyconfig.py:471] Config param query_proj: RematLocation.REMAT I0423 21:37:55.839982 133240883222336 pyconfig.py:471] Config param query_wa_proj: RematLocation.REMAT I0423 21:37:55.839998 133240883222336 pyconfig.py:471] Config param ragged_block_size: 256 I0423 21:37:55.840013 133240883222336 pyconfig.py:471] Config param ragged_buffer_factor: -1.0 I0423 21:37:55.840027 133240883222336 pyconfig.py:471] Config param rampup_end_step: 0 I0423 21:37:55.840043 133240883222336 pyconfig.py:471] Config param rampup_samples_per_increment_to_load: None I0423 21:37:55.840059 133240883222336 pyconfig.py:471] Config param reasoning_end_token: </reasoning> I0423 21:37:55.840074 133240883222336 pyconfig.py:471] Config param reasoning_start_token: <reasoning> I0423 21:37:55.840089 133240883222336 pyconfig.py:471] Config param record_internal_nn_metrics: 0 I0423 21:37:55.840111 133240883222336 pyconfig.py:471] Config param remat_policy: full I0423 21:37:55.840126 133240883222336 pyconfig.py:471] Config param remat_policy_for_vit: minimal I0423 21:37:55.840142 133240883222336 pyconfig.py:471] Config param remove_size_one_mesh_axis_from_type: True I0423 21:37:55.840157 133240883222336 pyconfig.py:471] Config param replicate_quant_scale: False I0423 21:37:55.840172 133240883222336 pyconfig.py:471] Config param replicator_backup_interval_minutes: 0 I0423 21:37:55.840186 133240883222336 pyconfig.py:471] Config param report_heartbeat_metric_for_gcp_monitoring: False I0423 21:37:55.840201 133240883222336 pyconfig.py:471] Config param report_performance_metric_for_gcp_monitoring: False I0423 21:37:55.840216 133240883222336 pyconfig.py:471] Config param reshape_q: False I0423 21:37:55.840231 133240883222336 pyconfig.py:471] Config param return_log_prob: False I0423 21:37:55.840247 133240883222336 pyconfig.py:471] Config param reuse_example_batch: 0 I0423 21:37:55.840263 133240883222336 pyconfig.py:471] Config param reward_exact_answer: 5.0 I0423 21:37:55.840280 133240883222336 pyconfig.py:471] Config param reward_exact_format_match: 3.0 I0423 21:37:55.840294 133240883222336 pyconfig.py:471] Config param reward_partial_format_match: 0.5 I0423 21:37:55.840310 133240883222336 pyconfig.py:471] Config param reward_ratio_guess_to_answer_high: 0.5 I0423 21:37:55.840324 133240883222336 pyconfig.py:471] Config param reward_ratio_guess_to_answer_low: 0.25 I0423 21:37:55.840340 133240883222336 pyconfig.py:471] Config param reward_white_space_format_match: 1.5 I0423 21:37:55.840357 133240883222336 pyconfig.py:471] Config param rl: {'num_generations': 2, 'num_iterations': 1, 'grpo_beta': 0.08, 'grpo_epsilon': 0.2, 'loss_algo': 'grpo', 'use_agentic_rollout': False, 'max_concurrency': 256, 'off_policy_steps': 0, 'system_prompt': '', 'degenerate_group_masking': True, 'epsilon_high': None} I0423 21:37:55.840378 133240883222336 pyconfig.py:471] Config param rollout_data_parallelism: -1 I0423 21:37:55.840393 133240883222336 pyconfig.py:471] Config param rollout_expert_parallelism: 1 I0423 21:37:55.840408 133240883222336 pyconfig.py:471] Config param rollout_micro_batch_size: -1 I0423 21:37:55.840423 133240883222336 pyconfig.py:471] Config param rollout_tensor_parallelism: -1 I0423 21:37:55.840438 133240883222336 pyconfig.py:471] Config param rope_attention_scaling: False I0423 21:37:55.840454 133240883222336 pyconfig.py:471] Config param rope_factor: 40 I0423 21:37:55.840470 133240883222336 pyconfig.py:471] Config param rope_interleave: True I0423 21:37:55.840485 133240883222336 pyconfig.py:471] Config param rope_linear_scaling_factor: 1.0 I0423 21:37:55.840499 133240883222336 pyconfig.py:471] Config param rope_max_timescale: 10000 I0423 21:37:55.840516 133240883222336 pyconfig.py:471] Config param rope_min_timescale: 1 I0423 21:37:55.840532 133240883222336 pyconfig.py:471] Config param rope_theta_for_vit: 10000 I0423 21:37:55.840549 133240883222336 pyconfig.py:471] Config param rope_truncate: True I0423 21:37:55.840565 133240883222336 pyconfig.py:471] Config param rope_type: RopeType.DEFAULT I0423 21:37:55.840581 133240883222336 pyconfig.py:471] Config param rope_use_scale: True I0423 21:37:55.840596 133240883222336 pyconfig.py:471] Config param routed_bias: False I0423 21:37:55.840610 133240883222336 pyconfig.py:471] Config param routed_bias_update_rate: 0.0 I0423 21:37:55.840626 133240883222336 pyconfig.py:471] Config param routed_scaling_factor: 1.0 I0423 21:37:55.840642 133240883222336 pyconfig.py:471] Config param routed_score_func: I0423 21:37:55.840656 133240883222336 pyconfig.py:471] Config param run_name: gpt3-52k_2026-04-23-21-37 I0423 21:37:55.840672 133240883222336 pyconfig.py:471] Config param sa_block_kv: 512 I0423 21:37:55.840688 133240883222336 pyconfig.py:471] Config param sa_block_kv_compute: 512 I0423 21:37:55.840704 133240883222336 pyconfig.py:471] Config param sa_block_kv_dkv: 512 I0423 21:37:55.840720 133240883222336 pyconfig.py:471] Config param sa_block_kv_dkv_compute: 512 I0423 21:37:55.840736 133240883222336 pyconfig.py:471] Config param sa_block_kv_dq: 512 I0423 21:37:55.840751 133240883222336 pyconfig.py:471] Config param sa_block_q: 512 I0423 21:37:55.840766 133240883222336 pyconfig.py:471] Config param sa_block_q_dkv: 512 I0423 21:37:55.840780 133240883222336 pyconfig.py:471] Config param sa_block_q_dq: 512 I0423 21:37:55.840795 133240883222336 pyconfig.py:471] Config param sa_k_layout: HEAD_DIM_MINOR I0423 21:37:55.840810 133240883222336 pyconfig.py:471] Config param sa_q_layout: HEAD_DIM_MINOR I0423 21:37:55.840825 133240883222336 pyconfig.py:471] Config param sa_use_fused_bwd_kernel: False I0423 21:37:55.840839 133240883222336 pyconfig.py:471] Config param sa_v_layout: HEAD_DIM_MINOR I0423 21:37:55.840855 133240883222336 pyconfig.py:471] Config param sampler_devices_fraction: 0.5 I0423 21:37:55.840871 133240883222336 pyconfig.py:471] Config param save_checkpoint_on_completion: True I0423 21:37:55.840887 133240883222336 pyconfig.py:471] Config param save_config_to_gcs: False I0423 21:37:55.840903 133240883222336 pyconfig.py:471] Config param save_quantized_params_path: I0423 21:37:55.840918 133240883222336 pyconfig.py:471] Config param scale_embedding_for_audio: True I0423 21:37:55.840933 133240883222336 pyconfig.py:471] Config param scan_layers: True I0423 21:37:55.840948 133240883222336 pyconfig.py:471] Config param scan_layers_per_stage: False I0423 21:37:55.840963 133240883222336 pyconfig.py:471] Config param scan_pipeline_iterations: True I0423 21:37:55.840978 133240883222336 pyconfig.py:471] Config param scan_pipeline_repeats: False I0423 21:37:55.840993 133240883222336 pyconfig.py:471] Config param set_remat_policy_on_layers_per_stage: False I0423 21:37:55.841007 133240883222336 pyconfig.py:471] Config param set_remat_policy_on_pipeline_iterations: True I0423 21:37:55.841023 133240883222336 pyconfig.py:471] Config param sft_train_on_completion_only: False I0423 21:37:55.841037 133240883222336 pyconfig.py:471] Config param shard_exp_on_fsdp: False I0423 21:37:55.841053 133240883222336 pyconfig.py:471] Config param shard_mode: ShardMode.AUTO I0423 21:37:55.841070 133240883222336 pyconfig.py:471] Config param shard_optimizer_over_data: False I0423 21:37:55.841084 133240883222336 pyconfig.py:471] Config param sharding_strategy: None I0423 21:37:55.841113 133240883222336 pyconfig.py:471] Config param sharding_tolerance: 0.02 I0423 21:37:55.841128 133240883222336 pyconfig.py:471] Config param shardy: True I0423 21:37:55.841144 133240883222336 pyconfig.py:471] Config param share_kv_projections: False I0423 21:37:55.841158 133240883222336 pyconfig.py:471] Config param shared_experts: 0 I0423 21:37:55.841174 133240883222336 pyconfig.py:471] Config param sinkhorn_iterations: 20 I0423 21:37:55.841189 133240883222336 pyconfig.py:471] Config param skip_first_n_steps_for_profiler: 1 I0423 21:37:55.841204 133240883222336 pyconfig.py:471] Config param skip_jax_distributed_system: False I0423 21:37:55.841219 133240883222336 pyconfig.py:471] Config param skip_step_interval: 128 I0423 21:37:55.841234 133240883222336 pyconfig.py:471] Config param skip_step_on_spikes: False I0423 21:37:55.841250 133240883222336 pyconfig.py:471] Config param skip_step_scaling_factor: 6.0 I0423 21:37:55.841265 133240883222336 pyconfig.py:471] Config param sliding_window_size: 0 I0423 21:37:55.841280 133240883222336 pyconfig.py:471] Config param solution_end_token: </answer> I0423 21:37:55.841294 133240883222336 pyconfig.py:471] Config param solution_start_token: <answer> I0423 21:37:55.841310 133240883222336 pyconfig.py:471] Config param source_checkpoint_layout: orbax I0423 21:37:55.841326 133240883222336 pyconfig.py:471] Config param sparse_matmul: True I0423 21:37:55.841340 133240883222336 pyconfig.py:471] Config param spatial_merge_size_for_vit: 2 I0423 21:37:55.841356 133240883222336 pyconfig.py:471] Config param stack_prefill_result_cache: False I0423 21:37:55.841370 133240883222336 pyconfig.py:471] Config param stack_trace_interval_seconds: 600 I0423 21:37:55.841386 133240883222336 pyconfig.py:471] Config param stack_trace_to_cloud: False I0423 21:37:55.841402 133240883222336 pyconfig.py:471] Config param step_deviation_interval_seconds: 30 I0423 21:37:55.841418 133240883222336 pyconfig.py:471] Config param steps: 200000 I0423 21:37:55.841433 133240883222336 pyconfig.py:471] Config param stop_strings: None I0423 21:37:55.841447 133240883222336 pyconfig.py:471] Config param student_overrides: {'model_name': 'llama3.1-8b'} I0423 21:37:55.841464 133240883222336 pyconfig.py:471] Config param student_params_to_update: None I0423 21:37:55.841480 133240883222336 pyconfig.py:471] Config param subslice_shape: I0423 21:37:55.841495 133240883222336 pyconfig.py:471] Config param swap_space_vllm_gb: 2 I0423 21:37:55.841511 133240883222336 pyconfig.py:471] Config param system_prompt: I0423 21:37:55.841525 133240883222336 pyconfig.py:471] Config param target_eval_loss: 0.0 I0423 21:37:55.841541 133240883222336 pyconfig.py:471] Config param teacher_overrides: {'model_name': 'llama3.1-8b'} I0423 21:37:55.841559 133240883222336 pyconfig.py:471] Config param temperature_tuning: False I0423 21:37:55.841574 133240883222336 pyconfig.py:471] Config param temporal_patch_size_for_vit: 2 I0423 21:37:55.841588 133240883222336 pyconfig.py:471] Config param tensorboard_dir: /deps/maxtext_output/gpt3-52k_2026-04-23-21-37/tensorboard/ I0423 21:37:55.841604 133240883222336 pyconfig.py:471] Config param tensors_on_device: None I0423 21:37:55.841618 133240883222336 pyconfig.py:471] Config param tensors_to_offload: None I0423 21:37:55.841633 133240883222336 pyconfig.py:471] Config param test_batch_start_index: 0 I0423 21:37:55.841647 133240883222336 pyconfig.py:471] Config param tile_size_for_vit: 336 I0423 21:37:55.841663 133240883222336 pyconfig.py:471] Config param tokenize_eval_data: True I0423 21:37:55.841679 133240883222336 pyconfig.py:471] Config param tokenize_train_data: True I0423 21:37:55.841693 133240883222336 pyconfig.py:471] Config param tokenizer_path: meta-llama/Llama-3.1-8B I0423 21:37:55.841709 133240883222336 pyconfig.py:471] Config param tokenizer_type: TokenizerType.HUGGINGFACE I0423 21:37:55.841727 133240883222336 pyconfig.py:471] Config param topk_routing_group: -1 I0423 21:37:55.841741 133240883222336 pyconfig.py:471] Config param train_data_columns: ['text'] I0423 21:37:55.841756 133240883222336 pyconfig.py:471] Config param train_fraction: 1.0 I0423 21:37:55.841773 133240883222336 pyconfig.py:471] Config param train_image_column: image I0423 21:37:55.841789 133240883222336 pyconfig.py:471] Config param train_micro_batch_size: -1 I0423 21:37:55.841804 133240883222336 pyconfig.py:471] Config param train_split: train I0423 21:37:55.841820 133240883222336 pyconfig.py:471] Config param trainable_parameters_mask: [] I0423 21:37:55.841834 133240883222336 pyconfig.py:471] Config param trainable_position_size: 2048 I0423 21:37:55.841848 133240883222336 pyconfig.py:471] Config param trainer_devices_fraction: 0.5 I0423 21:37:55.841863 133240883222336 pyconfig.py:471] Config param upload_all_profiler_results: False I0423 21:37:55.841878 133240883222336 pyconfig.py:471] Config param use_2d_fsdp_sharding: False I0423 21:37:55.841892 133240883222336 pyconfig.py:471] Config param use_agentic_rollout: False I0423 21:37:55.841908 133240883222336 pyconfig.py:471] Config param use_audio: False I0423 21:37:55.841922 133240883222336 pyconfig.py:471] Config param use_audio_in_video: False I0423 21:37:55.841938 133240883222336 pyconfig.py:471] Config param use_batch_split_schedule: False I0423 21:37:55.841953 133240883222336 pyconfig.py:471] Config param use_chat_template: False I0423 21:37:55.841968 133240883222336 pyconfig.py:471] Config param use_chunked_prefill: False I0423 21:37:55.841983 133240883222336 pyconfig.py:471] Config param use_custom_sort_vjp: True I0423 21:37:55.841997 133240883222336 pyconfig.py:471] Config param use_dpo: False I0423 21:37:55.842012 133240883222336 pyconfig.py:471] Config param use_gather_mosaic_kernel: False I0423 21:37:55.842028 133240883222336 pyconfig.py:471] Config param use_grpo: True I0423 21:37:55.842044 133240883222336 pyconfig.py:471] Config param use_indexer: False I0423 21:37:55.842058 133240883222336 pyconfig.py:471] Config param use_iota_embed: True I0423 21:37:55.842073 133240883222336 pyconfig.py:471] Config param use_jax_splash: False I0423 21:37:55.842088 133240883222336 pyconfig.py:471] Config param use_max_logit_estimate: -1 I0423 21:37:55.842112 133240883222336 pyconfig.py:471] Config param use_mrope: False I0423 21:37:55.842127 133240883222336 pyconfig.py:471] Config param use_multimodal: False I0423 21:37:55.842141 133240883222336 pyconfig.py:471] Config param use_nnx_pipeline: False I0423 21:37:55.842159 133240883222336 pyconfig.py:471] Config param use_pathways: True I0423 21:37:55.842173 133240883222336 pyconfig.py:471] Config param use_post_attn_norm: False I0423 21:37:55.842188 133240883222336 pyconfig.py:471] Config param use_post_ffw_norm: False I0423 21:37:55.842203 133240883222336 pyconfig.py:471] Config param use_qk_clip: False I0423 21:37:55.842218 133240883222336 pyconfig.py:471] Config param use_qk_norm: False I0423 21:37:55.842233 133240883222336 pyconfig.py:471] Config param use_qk_norm_in_gdn: True I0423 21:37:55.842247 133240883222336 pyconfig.py:471] Config param use_qwix_quantization: False I0423 21:37:55.842263 133240883222336 pyconfig.py:471] Config param use_ragged_attention: False I0423 21:37:55.842277 133240883222336 pyconfig.py:471] Config param use_random_routing: False I0423 21:37:55.842293 133240883222336 pyconfig.py:471] Config param use_replicator_service: False I0423 21:37:55.842307 133240883222336 pyconfig.py:471] Config param use_ring_of_experts: False I0423 21:37:55.842322 133240883222336 pyconfig.py:471] Config param use_sft: False I0423 21:37:55.842338 133240883222336 pyconfig.py:471] Config param use_splash_scheduler: False I0423 21:37:55.842354 133240883222336 pyconfig.py:471] Config param use_tokamax_gmm: False I0423 21:37:55.842368 133240883222336 pyconfig.py:471] Config param use_tokamax_splash: False I0423 21:37:55.842384 133240883222336 pyconfig.py:471] Config param use_truncation: True I0423 21:37:55.842398 133240883222336 pyconfig.py:471] Config param use_tunix_gradient_accumulation: False I0423 21:37:55.842414 133240883222336 pyconfig.py:471] Config param use_untrainable_positional_embedding: False I0423 21:37:55.842428 133240883222336 pyconfig.py:471] Config param use_vertex_tensorboard: False I0423 21:37:55.842443 133240883222336 pyconfig.py:471] Config param using_pipeline_parallelism: False I0423 21:37:55.842459 133240883222336 pyconfig.py:471] Config param v_head_dim: 128 I0423 21:37:55.842474 133240883222336 pyconfig.py:471] Config param v_norm_with_scale: True I0423 21:37:55.842490 133240883222336 pyconfig.py:471] Config param value_proj: RematLocation.REMAT I0423 21:37:55.842504 133240883222336 pyconfig.py:471] Config param vertex_tensorboard_project: I0423 21:37:55.842520 133240883222336 pyconfig.py:471] Config param vertex_tensorboard_region: I0423 21:37:55.842534 133240883222336 pyconfig.py:471] Config param video_path: I0423 21:37:55.842553 133240883222336 pyconfig.py:471] Config param video_placeholder: <|video|> I0423 21:37:55.842568 133240883222336 pyconfig.py:471] Config param vision_output_dim_for_vit: 4096 I0423 21:37:55.842582 133240883222336 pyconfig.py:471] Config param vision_output_length: -1 I0423 21:37:55.842598 133240883222336 pyconfig.py:471] Config param vllm_additional_config: {} I0423 21:37:55.842612 133240883222336 pyconfig.py:471] Config param vllm_hf_config_path: I0423 21:37:55.842628 133240883222336 pyconfig.py:471] Config param vllm_hf_overrides: {} I0423 21:37:55.842642 133240883222336 pyconfig.py:471] Config param vocab_size: 32000 I0423 21:37:55.842657 133240883222336 pyconfig.py:471] Config param warmup_steps_fraction: 0.1 I0423 21:37:55.842672 133240883222336 pyconfig.py:471] Config param weight_dtype: float32 I0423 21:37:55.842697 133240883222336 pyconfig.py:471] Config param weight_quantization_calibration_method: absmax I0423 21:37:55.842711 133240883222336 pyconfig.py:471] Config param wi_tile_dlhs_batch_seq: 512 I0423 21:37:55.842727 133240883222336 pyconfig.py:471] Config param wi_tile_dlhs_embed_dim: 1024 I0423 21:37:55.842741 133240883222336 pyconfig.py:471] Config param wi_tile_dlhs_mlp_dim: 1024 I0423 21:37:55.842757 133240883222336 pyconfig.py:471] Config param wi_tile_drhs_batch_seq: 512 I0423 21:37:55.842772 133240883222336 pyconfig.py:471] Config param wi_tile_drhs_embed_dim: 1024 I0423 21:37:55.842787 133240883222336 pyconfig.py:471] Config param wi_tile_drhs_mlp_dim: 1024 I0423 21:37:55.842802 133240883222336 pyconfig.py:471] Config param wi_tile_fwd_batch_seq: 512 I0423 21:37:55.842816 133240883222336 pyconfig.py:471] Config param wi_tile_fwd_embed_dim: 1024 I0423 21:37:55.842831 133240883222336 pyconfig.py:471] Config param wi_tile_fwd_mlp_dim: 1024 I0423 21:37:55.842845 133240883222336 pyconfig.py:471] Config param wo_tile_dlhs_batch_seq: 512 I0423 21:37:55.842861 133240883222336 pyconfig.py:471] Config param wo_tile_dlhs_embed_dim: 1024 I0423 21:37:55.842875 133240883222336 pyconfig.py:471] Config param wo_tile_dlhs_mlp_dim: 1024 I0423 21:37:55.842890 133240883222336 pyconfig.py:471] Config param wo_tile_drhs_batch_seq: 512 I0423 21:37:55.842905 133240883222336 pyconfig.py:471] Config param wo_tile_drhs_embed_dim: 1024 I0423 21:37:55.842920 133240883222336 pyconfig.py:471] Config param wo_tile_drhs_mlp_dim: 1024 I0423 21:37:55.842936 133240883222336 pyconfig.py:471] Config param wo_tile_fwd_batch_seq: 512 I0423 21:37:55.842950 133240883222336 pyconfig.py:471] Config param wo_tile_fwd_embed_dim: 1024 I0423 21:37:55.842966 133240883222336 pyconfig.py:471] Config param wo_tile_fwd_mlp_dim: 1024 I0423 21:37:55.842980 133240883222336 pyconfig.py:471] Config param wsd_decay_steps_fraction: 0.1 I0423 21:37:55.842995 133240883222336 pyconfig.py:471] Config param wsd_decay_style: WsdDecayStyle.LINEAR I0423 21:37:55.843012 133240883222336 pyconfig.py:471] Config param xprof_e2e_enable_fw_power_level_event: False I0423 21:37:55.843028 133240883222336 pyconfig.py:471] Config param xprof_e2e_enable_fw_thermal_event: False I0423 21:37:55.843043 133240883222336 pyconfig.py:471] Config param xprof_e2e_enable_fw_throttle_event: False I0423 21:37:55.843057 133240883222336 pyconfig.py:471] Config param xprof_tpu_power_trace_level: 0 I0423 21:37:55.843075 133240883222336 pyconfig.py:471] Config param z_loss_multiplier: 0.0 I0423 21:37:55.843393 133240883222336 tokenizer.py:245] Tokenizer path: meta-llama/Llama-2-7b-chat-hf I0423 21:37:55.843428 133240883222336 tokenizer.py:224] Loading HF tokenizer: meta-llama/Llama-2-7b-chat-hf I0423 21:37:59.491345 133240883222336 _schedule.py:129] A polynomial schedule was set with a non-positive `transition_steps` value; this results in a constant schedule with value `init_value`. I0423 21:37:59.496378 133240883222336 maxtext_utils.py:1565] Num_devices: 32, shape (1, 4, 1, 8, 1, 1, 1, 1, 1, 1, 1, 1, 1) I0423 21:37:59.496497 133240883222336 train_distill.py:596] Applying logical axis rules for model initialization and training... I0423 21:37:59.496572 133240883222336 train_distill.py:600] Loading Student from ... I0423 21:37:59.496601 133240883222336 train_distill.py:169] --- Student Configuration --- I0423 21:37:59.496623 133240883222336 train_distill.py:170] Model Name: gpt3-52k I0423 21:37:59.496645 133240883222336 train_distill.py:171] Dimensions: 1 Layers, 16 Emb Dim, 8 Head Dim I0423 21:37:59.496663 133240883222336 train_distill.py:174] Attention Heads: 2 Query, 2 KV I0423 21:37:59.496681 133240883222336 train_distill.py:175] Vocab Size: 32000 I0423 21:37:59.496698 133240883222336 train_distill.py:176] Checkpoint: I0423 21:37:59.496716 133240883222336 train_distill.py:465] Initializing model: gpt3-52k... I0423 21:38:00.769405 133240883222336 train_distill.py:614] Loading Teacher from gs://lance-maxtext/pt_seed_ckpts/pt_seed_ckpts/pt_seed_ckpt_gpt352k_v32k_linen/checkpoints/4/items... I0423 21:38:00.769520 133240883222336 train_distill.py:169] --- Teacher Configuration --- I0423 21:38:00.769550 133240883222336 train_distill.py:170] Model Name: gpt3-52k I0423 21:38:00.769574 133240883222336 train_distill.py:171] Dimensions: 1 Layers, 16 Emb Dim, 8 Head Dim I0423 21:38:00.769596 133240883222336 train_distill.py:174] Attention Heads: 2 Query, 2 KV I0423 21:38:00.769616 133240883222336 train_distill.py:175] Vocab Size: 32000 I0423 21:38:00.769633 133240883222336 train_distill.py:176] Checkpoint: gs://lance-maxtext/pt_seed_ckpts/pt_seed_ckpts/pt_seed_ckpt_gpt352k_v32k_linen/checkpoints/4/items I0423 21:38:00.769651 133240883222336 train_distill.py:465] Initializing model: gpt3-52k... I0423 21:38:01.907245 133240883222336 pytree_checkpoint_handler.py:577] save_device_host_concurrent_bytes=None I0423 21:38:01.907696 133240883222336 base_pytree_checkpoint_handler.py:411] Created BasePyTreeCheckpointHandler: use_ocdbt=True, use_zarr3=True, pytree_metadata_options=PyTreeMetadataOptions(support_rich_types=False), array_metadata_store=<orbax.checkpoint._src.metadata.array_metadata_store.Store object at 0x792dd805dd90>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0423 21:38:01.907769 133240883222336 abstract_checkpointer.py:35] orbax-checkpoint version: 0.11.28 W0423 21:38:02.412318 133240883222336 checkpoint.py:202] Metadata file does not exist: gs://lance-maxtext/pt_seed_ckpts/pt_seed_ckpts/pt_seed_ckpt_gpt352k_v32k_linen/checkpoints/4/items/_CHECKPOINT_METADATA I0423 21:38:03.385963 2145 google_auth_provider.cc:181] Running on GCE, using service account 562977990677-compute@developer.gserviceaccount.com I0423 21:38:04.533379 133240883222336 checkpointer.py:304] Restoring checkpoint from gs://lance-maxtext/pt_seed_ckpts/pt_seed_ckpts/pt_seed_ckpt_gpt352k_v32k_linen/checkpoints/4/items. W0423 21:38:06.690189 133240883222336 transform_utils.py:230] The transformations API will eventually be replaced by an upgraded design. The current API will not be removed until this point, but it will no longer be actively worked on. I0423 21:38:06.690559 133240883222336 transform_utils.py:288] The following keys are not loaded from the original tree after applying specified transforms: params/params/decoder/to_nnx__rngs/aqt/count, params/params/decoder/to_nnx__rngs/aqt/key, params/params/decoder/to_nnx__rngs/dropout/count, params/params/decoder/to_nnx__rngs/dropout/key, params/params/decoder/to_nnx__rngs/params/count, params/params/decoder/to_nnx__rngs/params/key I0423 21:38:07.174389 133240883222336 checkpointer.py:318] Finished restoring checkpoint in 3.02 seconds from gs://lance-maxtext/pt_seed_ckpts/pt_seed_ckpts/pt_seed_ckpt_gpt352k_v32k_linen/checkpoints/4/items. I0423 21:38:07.876790 133240883222336 train_distill.py:640] Initializing Data Iterators via MaxText pipeline... I0423 21:38:07.943328 133240883222336 config.py:112] TensorFlow version 2.20.0 available. I0423 21:38:07.943817 133240883222336 config.py:125] JAX version 0.8.3 available. E0423 21:38:09.942083 133240883222336 packing.py:209] PackAndBatchOperation is deprecated. Please use lazy_dataset.FirstFitPackIterDataset instead. I0423 21:38:09.942322 133240883222336 data_loader.py:408] Adding CopyNumPyArrayToSharedMemory MapTransform. I0423 21:38:09.945394 133240883222336 train_distill.py:410] Input Pipeline Checkpointing: DISABLED I0423 21:38:09.945457 133240883222336 train_distill.py:414] Reason: Iterator 'MultiHostDataLoadIterator' is not recognized as Grain (dataset_type='DatasetType.HF', has_save=False) I0423 21:38:09.945524 133240883222336 pytree_checkpoint_handler.py:577] save_device_host_concurrent_bytes=None I0423 21:38:09.945600 133240883222336 base_pytree_checkpoint_handler.py:411] Created BasePyTreeCheckpointHandler: use_ocdbt=True, use_zarr3=False, pytree_metadata_options=PyTreeMetadataOptions(support_rich_types=False), array_metadata_store=<orbax.checkpoint._src.metadata.array_metadata_store.Store object at 0x792dd805dd90>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0423 21:38:09.945641 133240883222336 pytree_checkpoint_handler.py:577] save_device_host_concurrent_bytes=None I0423 21:38:09.945672 133240883222336 base_pytree_checkpoint_handler.py:411] Created BasePyTreeCheckpointHandler: use_ocdbt=True, use_zarr3=False, pytree_metadata_options=PyTreeMetadataOptions(support_rich_types=False), array_metadata_store=<orbax.checkpoint._src.metadata.array_metadata_store.Store object at 0x792dd805dd90>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0423 21:38:09.945715 133240883222336 checkpoint_manager.py:702] [process=6][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=None, item_handlers={'model_params': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x792c85b33140>, 'optimizer_state': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7923d13e7410>, 'custom_metadata': <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7916f96ff590>}, handler_registry=None I0423 21:38:09.945908 133240883222336 composite_checkpoint_handler.py:237] Deferred registration for item: "model_params". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x792c85b33140>` for item "model_params" and save args `<class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeSaveArgs'>` and restore args `<class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>` to `_handler_registry`. I0423 21:38:09.945947 133240883222336 composite_checkpoint_handler.py:237] Deferred registration for item: "optimizer_state". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7923d13e7410>` for item "optimizer_state" and save args `<class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeSaveArgs'>` and restore args `<class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>` to `_handler_registry`. I0423 21:38:09.945973 133240883222336 composite_checkpoint_handler.py:237] Deferred registration for item: "custom_metadata". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7916f96ff590>` for item "custom_metadata" and save args `<class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>` and restore args `<class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>` to `_handler_registry`. I0423 21:38:09.945996 133240883222336 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7916f96ff200>` for item "metrics" and save args `<class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>` and restore args `<class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>` to `_handler_registry`. I0423 21:38:09.946023 133240883222336 composite_checkpoint_handler.py:505] Initialized registry DefaultCheckpointHandlerRegistry({('model_params', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeSaveArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x792c85b33140>, ('model_params', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x792c85b33140>, ('optimizer_state', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeSaveArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7923d13e7410>, ('optimizer_state', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7923d13e7410>, ('custom_metadata', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7916f96ff590>, ('custom_metadata', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7916f96ff590>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7916f96ff200>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7916f96ff200>}). I0423 21:38:09.946497 133240883222336 async_checkpointer.py:177] [process=6][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7916f9386480> timeout: 600 secs and primary_host=0 for async checkpoint writes I0423 21:38:12.813174 133240883222336 checkpoint_manager.py:1788] Found 0 checkpoint steps in gs://lance-maxtext/pt_ckpt_xpk_test_pipeline_scan_nnx_20260423_211626/pt_distill_nnx_xpk_test_pipeline_scan_nnx_20260423_211626_07_distill_smoke/checkpoints I0423 21:38:12.928549 133240883222336 checkpoint_manager.py:921] [process=6][thread=MainThread] CheckpointManager created, primary_host=0, CheckpointManagerOptions=CheckpointManagerOptions(save_interval_steps=2000, max_to_keep=None, keep_time_interval=None, keep_period=None, should_keep_fn=None, best_fn=None, best_mode='max', keep_checkpoints_without_metrics=True, step_prefix=None, step_format_fixed_length=None, step_name_format=None, create=True, cleanup_tmp_directories=False, save_on_steps=frozenset(), single_host_load_and_broadcast=False, todelete_subdir=None, todelete_full_path=None, enable_hns=False, enable_background_delete=False, read_only=False, enable_async_checkpointing=True, async_options=None, multiprocessing_options=MultiprocessingOptions(primary_host=0, active_processes=None, barrier_sync_key_prefix=None), should_save_fn=None, file_options=FileOptions(path_permission_mode=None), save_root_metadata=True, temporary_path_class=None, save_decision_policy=None, preservation_policy=None, prevent_write_metrics=False, enable_should_save_is_saving_in_progress_check=True, enable_per_process_directory_creation=False), root_directory=gs://lance-maxtext/pt_ckpt_xpk_test_pipeline_scan_nnx_20260423_211626/pt_distill_nnx_xpk_test_pipeline_scan_nnx_20260423_211626_07_distill_smoke/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7916f96ff560> I0423 21:38:12.928720 133240883222336 pytree_checkpoint_handler.py:577] save_device_host_concurrent_bytes=None I0423 21:38:12.928786 133240883222336 base_pytree_checkpoint_handler.py:411] Created BasePyTreeCheckpointHandler: use_ocdbt=True, use_zarr3=False, pytree_metadata_options=PyTreeMetadataOptions(support_rich_types=False), array_metadata_store=<orbax.checkpoint._src.metadata.array_metadata_store.Store object at 0x792dd805dd90>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0423 21:38:12.928822 133240883222336 pytree_checkpoint_handler.py:577] save_device_host_concurrent_bytes=None I0423 21:38:12.928853 133240883222336 base_pytree_checkpoint_handler.py:411] Created BasePyTreeCheckpointHandler: use_ocdbt=True, use_zarr3=False, pytree_metadata_options=PyTreeMetadataOptions(support_rich_types=False), array_metadata_store=<orbax.checkpoint._src.metadata.array_metadata_store.Store object at 0x792dd805dd90>, enable_pinned_host_transfer=False, save_concurrent_bytes: 96000000000 (89.4 GiB), restore_concurrent_bytes: 96000000000 (89.4 GiB) I0423 21:38:12.928891 133240883222336 checkpoint_manager.py:1983] [process=6][thread=MainThread][wait_until_finished] No Save Finalize thread to wait for. Returning. I0423 21:38:12.928946 133240883222336 checkpoint.py:459] Closing _NonBlockingMetadataStore(enable_write=True, _write_lock=<locked _thread.RLock object owner=133240883222336 count=1 at 0x7923d1c9a340>, _store_impl=<orbax.checkpoint._src.metadata.checkpoint._MetadataStoreImpl object at 0x7916f96ff350>, _single_thread_executor=<concurrent.futures.thread.ThreadPoolExecutor object at 0x7916f96ff320>, _write_futures=[]) I0423 21:38:12.929362 133240883222336 checkpoint.py:459] Closing _NonBlockingMetadataStore(enable_write=True, _write_lock=<locked _thread.RLock object owner=133240883222336 count=1 at 0x7923d1c9a340>, _store_impl=<orbax.checkpoint._src.metadata.checkpoint._MetadataStoreImpl object at 0x7916f96ff350>, _single_thread_executor=<concurrent.futures.thread.ThreadPoolExecutor object at 0x7916f96ff320>, _write_futures=[]) I0423 21:38:12.929390 133240883222336 checkpoint.py:459] Closing _NonBlockingMetadataStore(enable_write=True, _write_lock=<locked _thread.RLock object owner=133240883222336 count=1 at 0x7923d1c9a340>, _store_impl=<orbax.checkpoint._src.metadata.checkpoint._MetadataStoreImpl object at 0x7916f96ff350>, _single_thread_executor=<concurrent.futures.thread.ThreadPoolExecutor object at 0x7916f96ff320>, _write_futures=[]) I0423 21:38:12.929423 133240883222336 checkpoint_manager.py:702] [process=6][thread=MainThread] CheckpointManager init: checkpointers=None, item_names=None, item_handlers={'model_params': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7916f96ff530>, 'optimizer_state': <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7916f93ce720>, 'custom_metadata': <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7916f93ceae0>, 'iter': <maxtext.common.checkpointing.GrainCheckpointHandler object at 0x7916f93cfcb0>}, handler_registry=None I0423 21:38:12.929536 133240883222336 composite_checkpoint_handler.py:237] Deferred registration for item: "model_params". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7916f96ff530>` for item "model_params" and save args `<class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeSaveArgs'>` and restore args `<class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>` to `_handler_registry`. I0423 21:38:12.929571 133240883222336 composite_checkpoint_handler.py:237] Deferred registration for item: "optimizer_state". Adding handler `<orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7916f93ce720>` for item "optimizer_state" and save args `<class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeSaveArgs'>` and restore args `<class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>` to `_handler_registry`. I0423 21:38:12.929595 133240883222336 composite_checkpoint_handler.py:237] Deferred registration for item: "custom_metadata". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7916f93ceae0>` for item "custom_metadata" and save args `<class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>` and restore args `<class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>` to `_handler_registry`. I0423 21:38:12.929624 133240883222336 composite_checkpoint_handler.py:237] Deferred registration for item: "iter". Adding handler `<maxtext.common.checkpointing.GrainCheckpointHandler object at 0x7916f93cfcb0>` for item "iter" and save args `<class 'maxtext.common.checkpointing.GrainCheckpointSave'>` and restore args `<class 'maxtext.common.checkpointing.GrainCheckpointRestore'>` to `_handler_registry`. I0423 21:38:12.929647 133240883222336 composite_checkpoint_handler.py:237] Deferred registration for item: "metrics". Adding handler `<orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7916f93cde20>` for item "metrics" and save args `<class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>` and restore args `<class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>` to `_handler_registry`. I0423 21:38:12.929672 133240883222336 composite_checkpoint_handler.py:505] Initialized registry DefaultCheckpointHandlerRegistry({('model_params', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeSaveArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7916f96ff530>, ('model_params', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7916f96ff530>, ('optimizer_state', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeSaveArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7916f93ce720>, ('optimizer_state', <class 'orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeRestoreArgs'>): <orbax.checkpoint._src.handlers.pytree_checkpoint_handler.PyTreeCheckpointHandler object at 0x7916f93ce720>, ('custom_metadata', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7916f93ceae0>, ('custom_metadata', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7916f93ceae0>, ('iter', <class 'maxtext.common.checkpointing.GrainCheckpointSave'>): <maxtext.common.checkpointing.GrainCheckpointHandler object at 0x7916f93cfcb0>, ('iter', <class 'maxtext.common.checkpointing.GrainCheckpointRestore'>): <maxtext.common.checkpointing.GrainCheckpointHandler object at 0x7916f93cfcb0>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonSaveArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7916f93cde20>, ('metrics', <class 'orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonRestoreArgs'>): <orbax.checkpoint._src.handlers.json_checkpoint_handler.JsonCheckpointHandler object at 0x7916f93cde20>}). I0423 21:38:12.929751 133240883222336 async_checkpointer.py:177] [process=6][thread=MainThread] Using barrier_sync_fn: <function get_barrier_sync_fn.<locals>._fn at 0x7916f93865c0> timeout: 600 secs and primary_host=0 for async checkpoint writes I0423 21:38:13.303512 133240883222336 checkpoint_manager.py:1788] Found 0 checkpoint steps in gs://lance-maxtext/pt_ckpt_xpk_test_pipeline_scan_nnx_20260423_211626/pt_distill_nnx_xpk_test_pipeline_scan_nnx_20260423_211626_07_distill_smoke/checkpoints I0423 21:38:13.330581 133240883222336 checkpoint_manager.py:921] [process=6][thread=MainThread] CheckpointManager created, primary_host=0, CheckpointManagerOptions=CheckpointManagerOptions(save_interval_steps=2000, max_to_keep=None, keep_time_interval=None, keep_period=None, should_keep_fn=None, best_fn=None, best_mode='max', keep_checkpoints_without_metrics=True, step_prefix=None, step_format_fixed_length=None, step_name_format=None, create=True, cleanup_tmp_directories=False, save_on_steps=frozenset(), single_host_load_and_broadcast=False, todelete_subdir=None, todelete_full_path=None, enable_hns=False, enable_background_delete=False, read_only=False, enable_async_checkpointing=True, async_options=None, multiprocessing_options=MultiprocessingOptions(primary_host=0, active_processes=None, barrier_sync_key_prefix=None), should_save_fn=None, file_options=FileOptions(path_permission_mode=None), save_root_metadata=True, temporary_path_class=None, save_decision_policy=None, preservation_policy=None, prevent_write_metrics=False, enable_should_save_is_saving_in_progress_check=True, enable_per_process_directory_creation=False), root_directory=gs://lance-maxtext/pt_ckpt_xpk_test_pipeline_scan_nnx_20260423_211626/pt_distill_nnx_xpk_test_pipeline_scan_nnx_20260423_211626_07_distill_smoke/checkpoints: <orbax.checkpoint.checkpoint_manager.CheckpointManager object at 0x7916f96fec60> I0423 21:38:13.331042 133240883222336 train_distill.py:691] Starting Distillation Training... I0423 21:38:13.331162 133240883222336 peft_trainer.py:590] Training with mesh: Mesh('diloco': 1, 'data': 4, 'stage': 1, 'fsdp': 8, 'fsdp_transpose': 1, 'sequence': 1, 'context': 1, 'context_autoregressive': 1, 'tensor': 1, 'tensor_transpose': 1, 'tensor_sequence': 1, 'expert': 1, 'autoregressive': 1, axis_types=(Auto, Auto, Auto, Auto, Auto, Auto, Auto, Auto, Auto, Auto, Auto, Auto, Auto)) I0423 21:38:14.447522 133240883222336 peft_trainer.py:600] Compiled train_step cache size: 0 Training: 0%| | 0/5 [00:00<?, ?step/s]I0423 21:38:14.449539 133096582137600 grain_pool.py:367] Grain pool will use 1 processes. I0423 21:38:14.476680 133096582137600 grain_pool.py:440] Grain pool will start child processes. I0423 21:38:14.481830 133096582137600 grain_pool.py:448] Grain pool started all child processes. 2026-04-23 21:38:20.526801: 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) Unrecognized keys in `rope_scaling` for 'rope_type'='yarn': {'rope_theta'} `rope_scaling`'s factor field must be a float >= 1, got 40 `rope_scaling`'s beta_fast field must be a float, got 32 `rope_scaling`'s beta_slow field must be a float, got 1 Unrecognized keys in `rope_scaling` for 'rope_type'='yarn': {'rope_theta'} Unrecognized keys in `rope_scaling` for 'rope_type'='yarn': {'rope_theta'} Unrecognized keys in `rope_scaling` for 'rope_type'='yarn': {'rope_theta'} I0423 21:38:23.875899 133240883222336 utils.py:86] Train loop finished in: 9.4277 seconds Traceback (most recent call last): File "<frozen runpy>", line 198, in _run_module_as_main File "<frozen runpy>", line 88, in _run_code File "/deps/src/maxtext/trainers/post_train/distillation/train_distill.py", line 765, in <module> app.run(main) File "/usr/local/lib/python3.12/site-packages/absl/app.py", line 316, in run _run_main(main, args) File "/usr/local/lib/python3.12/site-packages/absl/app.py", line 261, in _run_main sys.exit(main(argv)) ^^^^^^^^^^ File "/deps/src/maxtext/trainers/post_train/distillation/train_distill.py", line 761, in main train_distill(student_config, teacher_config, is_offline, global_config.offline_data_dir) File "/deps/src/maxtext/trainers/post_train/distillation/train_distill.py", line 693, in train_distill trainer.train(train_iter, eval_iter) File "/usr/local/lib/python3.12/site-packages/tunix/sft/peft_trainer.py", line 659, in train train_example = sharding_utils.shard_input( ^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/usr/local/lib/python3.12/site-packages/tunix/sft/sharding_utils.py", line 58, in shard_input return jax.tree.map( ^^^^^^^^^^^^^ File "/usr/local/lib/python3.12/site-packages/jax/_src/tree.py", line 155, in map return tree_util.tree_map(f, tree, *rest, is_leaf=is_leaf) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/usr/local/lib/python3.12/site-packages/jax/_src/tree_util.py", line 362, in tree_map return treedef.unflatten(f(*xs) for xs in zip(*all_leaves)) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/usr/local/lib/python3.12/site-packages/jax/_src/tree_util.py", line 362, in <genexpr> return treedef.unflatten(f(*xs) for xs in zip(*all_leaves)) ^^^^^^ File "/usr/local/lib/python3.12/site-packages/tunix/sft/sharding_utils.py", line 59, in <lambda> lambda x: jax.make_array_from_process_local_data( ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/usr/local/lib/python3.12/site-packages/jax/_src/array.py", line 986, in make_array_from_process_local_data out = [_array_from_process_local_data(data, s, shape) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/usr/local/lib/python3.12/site-packages/jax/_src/array.py", line 1048, in _array_from_process_local_data return make_array_from_callback(global_shape, sharding, cb) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/usr/local/lib/python3.12/site-packages/jax/_src/array.py", line 845, in make_array_from_callback per_device_values = api.device_put(per_device_values, devices) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/usr/local/lib/python3.12/site-packages/jax/_src/api.py", line 2729, in device_put out_flat = dispatch._batched_device_put_impl( ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/usr/local/lib/python3.12/site-packages/jax/_src/dispatch.py", line 558, in _batched_device_put_impl y = _device_put_impl(x, device=device, src=src, copy=cp, aval=aval) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/usr/local/lib/python3.12/site-packages/jax/_src/dispatch.py", line 545, in _device_put_impl return _device_put_sharding_impl(x, aval, device, copy) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/usr/local/lib/python3.12/site-packages/jax/_src/dispatch.py", line 487, in _device_put_sharding_impl raise ValueError( ValueError: device_put's first argument must be a fully addressable array, but got value with devices {TpuDevice(id=0, process_index=0, coords=(0,0,0), core_on_chip=0), TpuDevice(id=15, process_index=3, coords=(3,3,0), core_on_chip=0), TpuDevice(id=4, process_index=0, coords=(0,1,0), core_on_chip=0), TpuDevice(id=16, process_index=4, coords=(0,4,0), core_on_chip=0), TpuDevice(id=25, process_index=6, coords=(1,6,0), core_on_chip=0), TpuDevice(id=31, process_index=7, coords=(3,7,0), core_on_chip=0), TpuDevice(id=18, process_index=5, coords=(2,4,0), core_on_chip=0), TpuDevice(id=17, process_index=4, coords=(1,4,0), core_on_chip=0), TpuDevice(id=22, process_index=5, coords=(2,5,0), core_on_chip=0), TpuDevice(id=7, process_index=1, coords=(3,1,0), core_on_chip=0), TpuDevice(id=8, process_index=2, coords=(0,2,0), core_on_chip=0), TpuDevice(id=28, process_index=6, coords=(0,7,0), core_on_chip=0), TpuDevice(id=19, process_index=5, coords=(3,4,0), core_on_chip=0), TpuDevice(id=9, process_index=2, coords=(1,2,0), core_on_chip=0), TpuDevice(id=5, process_index=0, coords=(1,1,0), core_on_chip=0), TpuDevice(id=29, process_index=6, coords=(1,7,0), core_on_chip=0), TpuDevice(id=14, process_index=3, coords=(2,3,0), core_on_chip=0), TpuDevice(id=11, process_index=3, coords=(3,2,0), core_on_chip=0), TpuDevice(id=27, process_index=7, coords=(3,6,0), core_on_chip=0), TpuDevice(id=12, process_index=2, coords=(0,3,0), core_on_chip=0), TpuDevice(id=2, process_index=1, coords=(2,0,0), core_on_chip=0), TpuDevice(id=20, process_index=4, coords=(0,5,0), core_on_chip=0), TpuDevice(id=3, process_index=1, coords=(3,0,0), core_on_chip=0), TpuDevice(id=1, process_index=0, coords=(1,0,0), core_on_chip=0), TpuDevice(id=6, process_index=1, coords=(2,1,0), core_on_chip=0), TpuDevice(id=26, process_index=7, coords=(2,6,0), core_on_chip=0), TpuDevice(id=23, process_index=5, coords=(3,5,0), core_on_chip=0), TpuDevice(id=13, process_index=2, coords=(1,3,0), core_on_chip=0), TpuDevice(id=21, process_index=4, coords=(1,5,0), core_on_chip=0), TpuDevice(id=10, process_index=3, coords=(2,2,0), core_on_chip=0), TpuDevice(id=30, process_index=7, coords=(2,7,0), core_on_chip=0), TpuDevice(id=24, process_index=6, coords=(0,6,0), core_on_chip=0)} I0423 21:38:24.217622 133096582137600 grain_pool.py:542] Grain pool is exiting. I0423 21:38:24.217721 133096582137600 grain_pool.py:547] Shutting down multiprocessing system. I0423 21:38:25.662690 133096582137600 grain_pool.py:547] Shutting down multiprocessing system. Training: 0%| | 0/5 [00:12<?, ?step/s] /usr/local/lib/python3.12/multiprocessing/resource_tracker.py:279: UserWarning: resource_tracker: There appear to be 15 leaked shared_memory objects to clean up at shutdown warnings.warn('resource_tracker: There appear to be %d ' XPK End: Thu Apr 23 21:38:32 UTC 2026 EXIT_CODE=1