Training in progress, step 20

config.json

@@ -55,7 +55,7 @@
   "topk_method": "greedy",
   "torch_dtype": "bfloat16",
   "transformers_version": "4.49.0",
-  "use_cache": false,
+  "use_cache": true,
   "v_head_dim": 128,
   "vocab_size": 102400
 }

expert_selection.log

@@ -0,0 +1,555 @@
+2025-04-24 23:33:57 - INFO - __main__ - Model parameters ModelConfig(model_name_or_path='deepseek-ai/DeepSeek-Coder-V2-Lite-Instruct', model_revision='main', torch_dtype='bfloat16', trust_remote_code=True, attn_implementation='flash_attention_2', use_peft=False, lora_r=16, lora_alpha=32, lora_dropout=0.05, lora_target_modules=None, lora_modules_to_save=None, lora_task_type='CAUSAL_LM', use_rslora=False, load_in_8bit=False, load_in_4bit=False, bnb_4bit_quant_type='nf4', use_bnb_nested_quant=False)
+2025-04-24 23:33:57 - INFO - __main__ - Script parameters ScriptArguments(dataset_name='open-r1/OpenR1-Math-220k', dataset_config=None, dataset_train_split='train', dataset_test_split='test', gradient_checkpointing_use_reentrant=False, ignore_bias_buffers=False)
+2025-04-24 23:33:57 - INFO - __main__ - Training parameters EfficientDistillationConfig(
+_n_gpu=1,
+accelerator_config={'split_batches': False, 'dispatch_batches': None, 'even_batches': True, 'use_seedable_sampler': True, 'non_blocking': False, 'gradient_accumulation_kwargs': None, 'use_configured_state': False},
+adafactor=False,
+adam_beta1=0.9,
+adam_beta2=0.999,
+adam_epsilon=1e-08,
+auto_find_batch_size=False,
+average_tokens_across_devices=False,
+batch_eval_metrics=False,
+benchmarks=[],
+bf16=True,
+bf16_full_eval=False,
+callbacks=[],
+chars_per_token=<CHARS_PER_TOKEN>,
+chat_template=None,
+data_seed=None,
+dataloader_drop_last=False,
+dataloader_num_workers=0,
+dataloader_persistent_workers=False,
+dataloader_pin_memory=True,
+dataloader_prefetch_factor=None,
+dataset_batch_size=None,
+dataset_kwargs=None,
+dataset_num_proc=None,
+dataset_text_field=text,
+ddp_backend=None,
+ddp_broadcast_buffers=None,
+ddp_bucket_cap_mb=None,
+ddp_find_unused_parameters=None,
+ddp_timeout=180000000,
+debug=[],
+deepspeed=None,
+disable_dropout=True,
+disable_tqdm=False,
+dispatch_batches=None,
+do_eval=True,
+do_predict=False,
+do_train=False,
+eval_accumulation_steps=None,
+eval_delay=0,
+eval_do_concat_batches=True,
+eval_on_start=False,
+eval_packing=None,
+eval_steps=None,
+eval_strategy=IntervalStrategy.NO,
+eval_use_gather_object=False,
+evaluation_strategy=None,
+fp16=False,
+fp16_backend=auto,
+fp16_full_eval=False,
+fp16_opt_level=O1,
+fsdp=[],
+fsdp_config={'min_num_params': 0, 'xla': False, 'xla_fsdp_v2': False, 'xla_fsdp_grad_ckpt': False},
+fsdp_min_num_params=0,
+fsdp_transformer_layer_cls_to_wrap=None,
+full_determinism=False,
+gradient_accumulation_steps=4,
+gradient_checkpointing=True,
+gradient_checkpointing_kwargs={'use_reentrant': False},
+greater_is_better=None,
+group_by_length=False,
+half_precision_backend=auto,
+hub_always_push=False,
+hub_model_id=Deepseek-Coder-V2-Lite-13B-Instruct-Open-R1-Distill,
+hub_model_revision=main,
+hub_private_repo=None,
+hub_strategy=HubStrategy.EVERY_SAVE,
+hub_token=<HUB_TOKEN>,
+ignore_data_skip=False,
+include_for_metrics=[],
+include_inputs_for_metrics=False,
+include_num_input_tokens_seen=False,
+include_tokens_per_second=False,
+jit_mode_eval=False,
+label_names=None,
+label_smoothing_factor=0.0,
+learning_rate=5e-05,
+length_column_name=length,
+lmbda=0.0,
+load_best_model_at_end=False,
+local_rank=0,
+log_level=info,
+log_level_replica=warning,
+log_on_each_node=True,
+logging_dir=data/DeepSeek-Coder-V2-Lite-Instruct/distill/runs/Apr24_23-33-57_q-h100,
+logging_first_step=False,
+logging_nan_inf_filter=True,
+logging_steps=1,
+logging_strategy=IntervalStrategy.STEPS,
+loss_type=forward_kl,
+lr_scheduler_kwargs={'min_lr_rate': 0.1},
+lr_scheduler_type=SchedulerType.COSINE_WITH_MIN_LR,
+max_grad_norm=1.0,
+max_length=2048,
+max_new_tokens=128,
+max_seq_length=None,
+max_steps=-1,
+metric_for_best_model=None,
+model_init_kwargs=None,
+mp_parameters=,
+neftune_noise_alpha=None,
+no_cuda=False,
+num_of_sequences=None,
+num_train_epochs=3,
+optim=OptimizerNames.ADAMW_TORCH,
+optim_args=None,
+optim_target_modules=None,
+output_dir=data/DeepSeek-Coder-V2-Lite-Instruct/distill,
+overwrite_hub_revision=False,
+overwrite_output_dir=True,
+packing=False,
+past_index=-1,
+per_device_eval_batch_size=16,
+per_device_train_batch_size=4,
+prediction_loss_only=False,
+push_to_hub=True,
+push_to_hub_model_id=None,
+push_to_hub_organization=None,
+push_to_hub_revision=False,
+push_to_hub_token=<PUSH_TO_HUB_TOKEN>,
+ray_scope=last,
+reduction=sum,
+remove_unused_columns=True,
+report_to=['wandb'],
+restore_callback_states_from_checkpoint=False,
+resume_from_checkpoint=None,
+run_name=data/DeepSeek-Coder-V2-Lite-Instruct/distill,
+save_on_each_node=False,
+save_only_model=False,
+save_safetensors=True,
+save_steps=200,
+save_strategy=SaveStrategy.STEPS,
+save_total_limit=1,
+seed=42,
+skip_memory_metrics=True,
+split_batches=None,
+system_prompt=None,
+teacher_model_init_kwargs=None,
+teacher_model_name_or_path=None,
+temperature=0.9,
+tf32=None,
+torch_compile=False,
+torch_compile_backend=None,
+torch_compile_mode=None,
+torch_empty_cache_steps=None,
+torchdynamo=None,
+tpu_metrics_debug=False,
+tpu_num_cores=None,
+use_cpu=False,
+use_ipex=False,
+use_legacy_prediction_loop=False,
+use_liger=False,
+use_liger_kernel=False,
+use_mps_device=False,
+wandb_entity=None,
+wandb_project=None,
+warmup_ratio=0.1,
+warmup_steps=0,
+weight_decay=0.0,
+)
+2025-04-24 23:34:00 - INFO - __main__ - *** Initializing model kwargs ***
+2025-04-24 23:34:00 - INFO - __main__ - Model memory in step 1, before model initialization (0):Memory allocated: 0.0
+Memory reserved: 0.0
+2025-04-24 23:34:19 - INFO - __main__ - Model memory in step 1, after model initialization:Memory allocated: 8743.642578125
+Memory reserved: 9596.0
+2025-04-24 23:34:19 - INFO - __main__ - Model memory in step 2, before data collator initialization:Memory allocated: 8743.642578125
+Memory reserved: 8782.0
+2025-04-24 23:34:19 - INFO - __main__ - Model memory in step 2, after data collator initialization:Memory allocated: 8743.642578125
+Memory reserved: 8782.0
+2025-04-24 23:34:19 - INFO - __main__ - Model memory in step 4, before trainer initialization:Memory allocated: 8743.642578125
+Memory reserved: 8782.0
+2025-04-24 23:34:19 - INFO - __main__ - Model memory in step 4, after trainer initialization:Memory allocated: 8743.64306640625
+Memory reserved: 8782.0
+2025-04-24 23:34:19 - INFO - __main__ - Model memory in step 5, before prediction:Memory allocated: 8743.64306640625
+Memory reserved: 8782.0
+2025-04-24 23:34:19 - INFO - __main__ - Running prediction on test subset to record expert activations...
+2025-04-24 23:34:27 - INFO - __main__ - Model memory in step 5, after prediction:Memory allocated: 7696.69775390625
+Memory reserved: 15308.0
+2025-04-24 23:34:27 - INFO - __main__ - Top k experts selected: {'model.layers.1.mlp': [45, 51, 44], 'model.layers.2.mlp': [25, 18, 27], 'model.layers.3.mlp': [54, 28, 25], 'model.layers.4.mlp': [11, 21, 49], 'model.layers.5.mlp': [35, 54, 20], 'model.layers.6.mlp': [45, 22, 1], 'model.layers.7.mlp': [58, 24, 43], 'model.layers.8.mlp': [47, 39, 54], 'model.layers.9.mlp': [31, 22, 32], 'model.layers.10.mlp': [22, 47, 42], 'model.layers.11.mlp': [11, 17, 29], 'model.layers.12.mlp': [4, 3, 59], 'model.layers.13.mlp': [17, 10, 47], 'model.layers.14.mlp': [51, 7, 27], 'model.layers.15.mlp': [24, 14, 17], 'model.layers.16.mlp': [61, 33, 19], 'model.layers.17.mlp': [32, 29, 26], 'model.layers.18.mlp': [56, 5, 2], 'model.layers.19.mlp': [24, 36, 40], 'model.layers.20.mlp': [1, 56, 38], 'model.layers.21.mlp': [19, 5, 28], 'model.layers.22.mlp': [32, 14, 58], 'model.layers.23.mlp': [20, 58, 0], 'model.layers.24.mlp': [7, 63, 47], 'model.layers.25.mlp': [45, 39, 46], 'model.layers.26.mlp': [6, 46, 49]}
+2025-04-24 23:34:27 - INFO - __main__ - Top k experts saved to: data/DeepSeek-Coder-V2-Lite-Instruct/distill/top_k_experts.json
+2025-04-24 23:34:27 - INFO - __main__ - Model memory before cleanup:Memory allocated: 7696.69775390625
+Memory reserved: 15308.0
+2025-04-24 23:34:30 - INFO - __main__ - Model memory after cleanup:Memory allocated: 7695.6298828125
+Memory reserved: 7814.0
+2025-04-24 23:34:30 - INFO - __main__ - Expert selection completed successfully. Run part 2 for training.
+2025-04-28 20:35:37 - INFO - __main__ - Model parameters ModelConfig(model_name_or_path='deepseek-ai/DeepSeek-Coder-V2-Lite-Instruct', model_revision='main', torch_dtype='bfloat16', trust_remote_code=True, attn_implementation='flash_attention_2', use_peft=False, lora_r=16, lora_alpha=32, lora_dropout=0.05, lora_target_modules=None, lora_modules_to_save=None, lora_task_type='CAUSAL_LM', use_rslora=False, load_in_8bit=False, load_in_4bit=False, bnb_4bit_quant_type='nf4', use_bnb_nested_quant=False)
+2025-04-28 20:35:37 - INFO - __main__ - Script parameters ScriptArguments(dataset_name='open-r1/OpenR1-Math-220k', dataset_config=None, dataset_train_split='train', dataset_test_split='test', gradient_checkpointing_use_reentrant=False, ignore_bias_buffers=False)
+2025-04-28 20:35:37 - INFO - __main__ - Training parameters EfficientDistillationConfig(
+_n_gpu=1,
+accelerator_config={'split_batches': False, 'dispatch_batches': None, 'even_batches': True, 'use_seedable_sampler': True, 'non_blocking': False, 'gradient_accumulation_kwargs': None, 'use_configured_state': False},
+adafactor=False,
+adam_beta1=0.9,
+adam_beta2=0.999,
+adam_epsilon=1e-08,
+auto_find_batch_size=False,
+average_tokens_across_devices=False,
+batch_eval_metrics=False,
+benchmarks=[],
+bf16=True,
+bf16_full_eval=False,
+callbacks=[],
+chars_per_token=<CHARS_PER_TOKEN>,
+chat_template=None,
+data_seed=None,
+dataloader_drop_last=False,
+dataloader_num_workers=0,
+dataloader_persistent_workers=False,
+dataloader_pin_memory=True,
+dataloader_prefetch_factor=None,
+dataset_batch_size=None,
+dataset_kwargs=None,
+dataset_num_proc=None,
+dataset_text_field=text,
+ddp_backend=None,
+ddp_broadcast_buffers=None,
+ddp_bucket_cap_mb=None,
+ddp_find_unused_parameters=None,
+ddp_timeout=1800000000,
+debug=[],
+deepspeed=None,
+disable_dropout=True,
+disable_tqdm=False,
+dispatch_batches=None,
+do_eval=True,
+do_predict=False,
+do_train=False,
+eval_accumulation_steps=None,
+eval_delay=0,
+eval_do_concat_batches=True,
+eval_on_start=False,
+eval_packing=None,
+eval_steps=None,
+eval_strategy=IntervalStrategy.NO,
+eval_use_gather_object=False,
+evaluation_strategy=None,
+fp16=False,
+fp16_backend=auto,
+fp16_full_eval=False,
+fp16_opt_level=O1,
+fsdp=[],
+fsdp_config={'min_num_params': 0, 'xla': False, 'xla_fsdp_v2': False, 'xla_fsdp_grad_ckpt': False},
+fsdp_min_num_params=0,
+fsdp_transformer_layer_cls_to_wrap=None,
+full_determinism=False,
+gradient_accumulation_steps=4,
+gradient_checkpointing=False,
+gradient_checkpointing_kwargs={'use_reentrant': False},
+greater_is_better=None,
+group_by_length=False,
+half_precision_backend=auto,
+hub_always_push=False,
+hub_model_id=Deepseek-Coder-V2-Lite-13B-Instruct-Open-R1-Distill,
+hub_model_revision=main,
+hub_private_repo=None,
+hub_strategy=HubStrategy.EVERY_SAVE,
+hub_token=<HUB_TOKEN>,
+ignore_data_skip=False,
+include_for_metrics=[],
+include_inputs_for_metrics=False,
+include_num_input_tokens_seen=False,
+include_tokens_per_second=False,
+jit_mode_eval=False,
+label_names=None,
+label_smoothing_factor=0.0,
+learning_rate=5e-05,
+length_column_name=length,
+lmbda=0.0,
+load_best_model_at_end=False,
+local_rank=0,
+log_level=info,
+log_level_replica=warning,
+log_on_each_node=True,
+logging_dir=data/DeepSeek-Coder-V2-Lite-Instruct/distill/runs/Apr28_20-35-36_q-h100,
+logging_first_step=False,
+logging_nan_inf_filter=True,
+logging_steps=1,
+logging_strategy=IntervalStrategy.STEPS,
+loss_type=forward_kl,
+lr_scheduler_kwargs={'min_lr_rate': 0.1},
+lr_scheduler_type=SchedulerType.COSINE_WITH_MIN_LR,
+max_grad_norm=1.0,
+max_length=4096,
+max_new_tokens=1024,
+max_seq_length=None,
+max_steps=-1,
+metric_for_best_model=None,
+model_init_kwargs=None,
+mp_parameters=,
+neftune_noise_alpha=None,
+no_cuda=False,
+num_of_sequences=None,
+num_train_epochs=3,
+optim=OptimizerNames.ADAMW_TORCH,
+optim_args=None,
+optim_target_modules=None,
+output_dir=data/DeepSeek-Coder-V2-Lite-Instruct/distill,
+overwrite_hub_revision=False,
+overwrite_output_dir=True,
+packing=False,
+past_index=-1,
+per_device_eval_batch_size=16,
+per_device_train_batch_size=4,
+prediction_loss_only=False,
+push_to_hub=True,
+push_to_hub_model_id=None,
+push_to_hub_organization=None,
+push_to_hub_revision=False,
+push_to_hub_token=<PUSH_TO_HUB_TOKEN>,
+ray_scope=last,
+reduction=sum,
+remove_unused_columns=True,
+report_to=['wandb'],
+restore_callback_states_from_checkpoint=False,
+resume_from_checkpoint=None,
+run_name=data/DeepSeek-Coder-V2-Lite-Instruct/distill,
+save_on_each_node=False,
+save_only_model=False,
+save_safetensors=True,
+save_steps=200,
+save_strategy=SaveStrategy.STEPS,
+save_total_limit=1,
+seed=42,
+skip_memory_metrics=True,
+split_batches=None,
+system_prompt=None,
+teacher_model_init_kwargs=None,
+teacher_model_name_or_path=None,
+temperature=0.9,
+tf32=None,
+torch_compile=False,
+torch_compile_backend=None,
+torch_compile_mode=None,
+torch_empty_cache_steps=None,
+torchdynamo=None,
+tpu_metrics_debug=False,
+tpu_num_cores=None,
+use_cpu=False,
+use_ipex=False,
+use_legacy_prediction_loop=False,
+use_liger=False,
+use_liger_kernel=False,
+use_mps_device=False,
+wandb_entity=None,
+wandb_project=None,
+warmup_ratio=0.1,
+warmup_steps=0,
+weight_decay=0.0,
+)
+2025-04-28 20:35:39 - INFO - __main__ - *** Initializing model kwargs ***
+2025-04-28 20:35:39 - INFO - __main__ - Model memory in step 1, before model initialization (0):Memory allocated: 0.0
+Memory reserved: 0.0
+2025-04-28 20:35:49 - INFO - __main__ - Model memory in step 1, after model initialization:Memory allocated: 0.0
+Memory reserved: 0.0
+2025-04-28 20:35:49 - INFO - __main__ - Model memory in step 2, before data collator initialization:Memory allocated: 0.0
+Memory reserved: 0.0
+2025-04-28 20:35:49 - INFO - __main__ - Model memory in step 2, after data collator initialization:Memory allocated: 0.0
+Memory reserved: 0.0
+2025-04-28 20:35:49 - INFO - __main__ - Model memory in step 4, before trainer initialization:Memory allocated: 0.0
+Memory reserved: 0.0
+2025-04-28 20:35:52 - INFO - __main__ - Model memory in step 4, after trainer initialization:Memory allocated: 0.00048828125
+Memory reserved: 2.0
+2025-04-28 20:35:52 - INFO - __main__ - Model memory in step 5, before prediction:Memory allocated: 0.00048828125
+Memory reserved: 2.0
+2025-04-28 20:35:52 - INFO - __main__ - Running prediction on test subset to record expert activations...
+2025-04-28 20:38:13 - INFO - __main__ - Model parameters ModelConfig(model_name_or_path='deepseek-ai/DeepSeek-Coder-V2-Lite-Instruct', model_revision='main', torch_dtype='bfloat16', trust_remote_code=True, attn_implementation='flash_attention_2', use_peft=False, lora_r=16, lora_alpha=32, lora_dropout=0.05, lora_target_modules=None, lora_modules_to_save=None, lora_task_type='CAUSAL_LM', use_rslora=False, load_in_8bit=False, load_in_4bit=False, bnb_4bit_quant_type='nf4', use_bnb_nested_quant=False)
+2025-04-28 20:38:13 - INFO - __main__ - Script parameters ScriptArguments(dataset_name='open-r1/OpenR1-Math-220k', dataset_config=None, dataset_train_split='train', dataset_test_split='test', gradient_checkpointing_use_reentrant=False, ignore_bias_buffers=False)
+2025-04-28 20:38:13 - INFO - __main__ - Training parameters EfficientDistillationConfig(
+_n_gpu=1,
+accelerator_config={'split_batches': False, 'dispatch_batches': None, 'even_batches': True, 'use_seedable_sampler': True, 'non_blocking': False, 'gradient_accumulation_kwargs': None, 'use_configured_state': False},
+adafactor=False,
+adam_beta1=0.9,
+adam_beta2=0.999,
+adam_epsilon=1e-08,
+auto_find_batch_size=False,
+average_tokens_across_devices=False,
+batch_eval_metrics=False,
+benchmarks=[],
+bf16=True,
+bf16_full_eval=False,
+callbacks=[],
+chars_per_token=<CHARS_PER_TOKEN>,
+chat_template=None,
+data_seed=None,
+dataloader_drop_last=False,
+dataloader_num_workers=0,
+dataloader_persistent_workers=False,
+dataloader_pin_memory=True,
+dataloader_prefetch_factor=None,
+dataset_batch_size=None,
+dataset_kwargs=None,
+dataset_num_proc=None,
+dataset_text_field=text,
+ddp_backend=None,
+ddp_broadcast_buffers=None,
+ddp_bucket_cap_mb=None,
+ddp_find_unused_parameters=None,
+ddp_timeout=1800000000,
+debug=[],
+deepspeed=None,
+disable_dropout=True,
+disable_tqdm=False,
+dispatch_batches=None,
+do_eval=True,
+do_predict=False,
+do_train=False,
+eval_accumulation_steps=None,
+eval_delay=0,
+eval_do_concat_batches=True,
+eval_on_start=False,
+eval_packing=None,
+eval_steps=None,
+eval_strategy=IntervalStrategy.NO,
+eval_use_gather_object=False,
+evaluation_strategy=None,
+fp16=False,
+fp16_backend=auto,
+fp16_full_eval=False,
+fp16_opt_level=O1,
+fsdp=[],
+fsdp_config={'min_num_params': 0, 'xla': False, 'xla_fsdp_v2': False, 'xla_fsdp_grad_ckpt': False},
+fsdp_min_num_params=0,
+fsdp_transformer_layer_cls_to_wrap=None,
+full_determinism=False,
+gradient_accumulation_steps=4,
+gradient_checkpointing=False,
+gradient_checkpointing_kwargs={'use_reentrant': False},
+greater_is_better=None,
+group_by_length=False,
+half_precision_backend=auto,
+hub_always_push=False,
+hub_model_id=Deepseek-Coder-V2-Lite-13B-Instruct-Open-R1-Distill,
+hub_model_revision=main,
+hub_private_repo=None,
+hub_strategy=HubStrategy.EVERY_SAVE,
+hub_token=<HUB_TOKEN>,
+ignore_data_skip=False,
+include_for_metrics=[],
+include_inputs_for_metrics=False,
+include_num_input_tokens_seen=False,
+include_tokens_per_second=False,
+jit_mode_eval=False,
+label_names=None,
+label_smoothing_factor=0.0,
+learning_rate=5e-05,
+length_column_name=length,
+lmbda=0.0,
+load_best_model_at_end=False,
+local_rank=0,
+log_level=info,
+log_level_replica=warning,
+log_on_each_node=True,
+logging_dir=data/DeepSeek-Coder-V2-Lite-Instruct/distill/runs/Apr28_20-38-13_q-h100,
+logging_first_step=False,
+logging_nan_inf_filter=True,
+logging_steps=1,
+logging_strategy=IntervalStrategy.STEPS,
+loss_type=forward_kl,
+lr_scheduler_kwargs={'min_lr_rate': 0.1},
+lr_scheduler_type=SchedulerType.COSINE_WITH_MIN_LR,
+max_grad_norm=1.0,
+max_length=4096,
+max_new_tokens=1024,
+max_seq_length=None,
+max_steps=-1,
+metric_for_best_model=None,
+model_init_kwargs=None,
+mp_parameters=,
+neftune_noise_alpha=None,
+no_cuda=False,
+num_of_sequences=None,
+num_train_epochs=3,
+optim=OptimizerNames.ADAMW_TORCH,
+optim_args=None,
+optim_target_modules=None,
+output_dir=data/DeepSeek-Coder-V2-Lite-Instruct/distill,
+overwrite_hub_revision=False,
+overwrite_output_dir=True,
+packing=False,
+past_index=-1,
+per_device_eval_batch_size=16,
+per_device_train_batch_size=4,
+prediction_loss_only=False,
+push_to_hub=True,
+push_to_hub_model_id=None,
+push_to_hub_organization=None,
+push_to_hub_revision=False,
+push_to_hub_token=<PUSH_TO_HUB_TOKEN>,
+ray_scope=last,
+reduction=sum,
+remove_unused_columns=True,
+report_to=['wandb'],
+restore_callback_states_from_checkpoint=False,
+resume_from_checkpoint=None,
+run_name=data/DeepSeek-Coder-V2-Lite-Instruct/distill,
+save_on_each_node=False,
+save_only_model=False,
+save_safetensors=True,
+save_steps=200,
+save_strategy=SaveStrategy.STEPS,
+save_total_limit=1,
+seed=42,
+skip_memory_metrics=True,
+split_batches=None,
+system_prompt=None,
+teacher_model_init_kwargs=None,
+teacher_model_name_or_path=None,
+temperature=0.9,
+tf32=None,
+torch_compile=False,
+torch_compile_backend=None,
+torch_compile_mode=None,
+torch_empty_cache_steps=None,
+torchdynamo=None,
+tpu_metrics_debug=False,
+tpu_num_cores=None,
+use_cpu=False,
+use_ipex=False,
+use_legacy_prediction_loop=False,
+use_liger=False,
+use_liger_kernel=False,
+use_mps_device=False,
+wandb_entity=None,
+wandb_project=None,
+warmup_ratio=0.1,
+warmup_steps=0,
+weight_decay=0.0,
+)
+2025-04-28 20:38:15 - INFO - __main__ - *** Initializing model kwargs ***
+2025-04-28 20:38:15 - INFO - __main__ - Model memory in step 1, before model initialization (0):Memory allocated: 0.0
+Memory reserved: 0.0
+2025-04-28 20:38:39 - INFO - __main__ - Model memory in step 1, after model initialization:Memory allocated: 6091.439453125
+Memory reserved: 7478.0
+2025-04-28 20:38:39 - INFO - __main__ - Model memory in step 2, before data collator initialization:Memory allocated: 6091.439453125
+Memory reserved: 6556.0
+2025-04-28 20:38:39 - INFO - __main__ - Model memory in step 2, after data collator initialization:Memory allocated: 6091.439453125
+Memory reserved: 6556.0
+2025-04-28 20:38:39 - INFO - __main__ - Model memory in step 4, before trainer initialization:Memory allocated: 6091.439453125
+Memory reserved: 6556.0
+2025-04-28 20:38:39 - INFO - __main__ - Model memory in step 4, after trainer initialization:Memory allocated: 6091.43994140625
+Memory reserved: 6556.0
+2025-04-28 20:38:39 - INFO - __main__ - Model memory in step 5, before prediction:Memory allocated: 6091.43994140625
+Memory reserved: 6556.0
+2025-04-28 20:38:39 - INFO - __main__ - Running prediction on test subset to record expert activations...
+2025-04-28 20:38:47 - INFO - __main__ - Model memory in step 5, after prediction:Memory allocated: 5044.49462890625
+Memory reserved: 15056.0
+2025-04-28 20:38:47 - INFO - __main__ - Top k experts selected: {'model.layers.1.mlp': [45, 51, 44, 61, 22, 14], 'model.layers.2.mlp': [25, 18, 27, 13, 23, 3], 'model.layers.3.mlp': [54, 28, 25, 41, 23, 57], 'model.layers.4.mlp': [11, 21, 49, 33, 14, 37], 'model.layers.5.mlp': [35, 54, 20, 9, 47, 52], 'model.layers.6.mlp': [45, 22, 1, 42, 47, 13], 'model.layers.7.mlp': [58, 24, 43, 62, 18, 44], 'model.layers.8.mlp': [47, 39, 54, 58, 30, 56], 'model.layers.9.mlp': [31, 22, 32, 13, 12, 24], 'model.layers.10.mlp': [22, 47, 42, 19, 2, 13], 'model.layers.11.mlp': [11, 17, 29, 10, 22, 59], 'model.layers.12.mlp': [4, 3, 59, 56, 5, 26], 'model.layers.13.mlp': [17, 10, 47, 14, 42, 58], 'model.layers.14.mlp': [51, 7, 27, 31, 61, 18], 'model.layers.15.mlp': [24, 14, 17, 55, 41, 5], 'model.layers.16.mlp': [61, 33, 19, 49, 9, 63], 'model.layers.17.mlp': [32, 29, 26, 43, 0, 27], 'model.layers.18.mlp': [56, 5, 2, 36, 1, 42], 'model.layers.19.mlp': [24, 36, 40, 0, 23, 2], 'model.layers.20.mlp': [1, 56, 38, 48, 58, 20], 'model.layers.21.mlp': [19, 5, 28, 15, 13, 10], 'model.layers.22.mlp': [32, 14, 58, 31, 3, 45], 'model.layers.23.mlp': [20, 58, 0, 42, 33, 45], 'model.layers.24.mlp': [7, 63, 47, 42, 10, 62], 'model.layers.25.mlp': [45, 39, 46, 11, 38, 48], 'model.layers.26.mlp': [6, 46, 49, 13, 57, 11]}
+2025-04-28 20:38:47 - INFO - __main__ - Top k experts saved to: data/DeepSeek-Coder-V2-Lite-Instruct/distill/top_6_experts.json
+2025-04-28 20:38:47 - INFO - __main__ - Model memory before cleanup:Memory allocated: 5044.49462890625
+Memory reserved: 15056.0
+2025-04-28 20:38:50 - INFO - __main__ - Model memory after cleanup:Memory allocated: 5043.4267578125
+Memory reserved: 5948.0
+2025-04-28 20:38:50 - INFO - __main__ - Expert selection completed successfully. Run part 2 for training.

model-00001-of-00002.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6708ea601efb4f587b288726e2d5a65eb4a7a4c404bf4e0e06af1f7d213b42aa
+size 4902968072

model-00002-of-00002.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2824f10339f6804c027eb8c08240cd840beb37ac66ebf73cd53f8907acbaf324
+size 419430528

tokenizer.json

@@ -1,6 +1,11 @@
 {
   "version": "1.0",
-  "truncation": null,
+  "truncation": {
+    "direction": "Right",
+    "max_length": 7000,
+    "strategy": "LongestFirst",
+    "stride": 0
+  },
   "padding": null,
   "added_tokens": [
     {

tokenizer_config.json

@@ -153,6 +153,7 @@
   "clean_up_tokenization_spaces": false,
   "eos_token": "<｜end▁of▁sentence｜>",
   "extra_special_tokens": {},
+  "fast_tokenizer": true,
   "legacy": true,
   "model_max_length": 16384,
   "pad_token": "<｜end▁of▁sentence｜>",

top_6_experts.json

@@ -0,0 +1 @@
+{"model.layers.1.mlp": [45, 51, 44, 61, 22, 14], "model.layers.2.mlp": [25, 18, 27, 13, 23, 3], "model.layers.3.mlp": [54, 28, 25, 41, 23, 57], "model.layers.4.mlp": [11, 21, 49, 33, 14, 37], "model.layers.5.mlp": [35, 54, 20, 9, 47, 52], "model.layers.6.mlp": [45, 22, 1, 42, 47, 13], "model.layers.7.mlp": [58, 24, 43, 62, 18, 44], "model.layers.8.mlp": [47, 39, 54, 58, 30, 56], "model.layers.9.mlp": [31, 22, 32, 13, 12, 24], "model.layers.10.mlp": [22, 47, 42, 19, 2, 13], "model.layers.11.mlp": [11, 17, 29, 10, 22, 59], "model.layers.12.mlp": [4, 3, 59, 56, 5, 26], "model.layers.13.mlp": [17, 10, 47, 14, 42, 58], "model.layers.14.mlp": [51, 7, 27, 31, 61, 18], "model.layers.15.mlp": [24, 14, 17, 55, 41, 5], "model.layers.16.mlp": [61, 33, 19, 49, 9, 63], "model.layers.17.mlp": [32, 29, 26, 43, 0, 27], "model.layers.18.mlp": [56, 5, 2, 36, 1, 42], "model.layers.19.mlp": [24, 36, 40, 0, 23, 2], "model.layers.20.mlp": [1, 56, 38, 48, 58, 20], "model.layers.21.mlp": [19, 5, 28, 15, 13, 10], "model.layers.22.mlp": [32, 14, 58, 31, 3, 45], "model.layers.23.mlp": [20, 58, 0, 42, 33, 45], "model.layers.24.mlp": [7, 63, 47, 42, 10, 62], "model.layers.25.mlp": [45, 39, 46, 11, 38, 48], "model.layers.26.mlp": [6, 46, 49, 13, 57, 11]}

top_k_experts.json

@@ -0,0 +1 @@
+{"model.layers.1.mlp": [45, 51, 44], "model.layers.2.mlp": [25, 18, 27], "model.layers.3.mlp": [54, 28, 25], "model.layers.4.mlp": [11, 21, 49], "model.layers.5.mlp": [35, 54, 20], "model.layers.6.mlp": [45, 22, 1], "model.layers.7.mlp": [58, 24, 43], "model.layers.8.mlp": [47, 39, 54], "model.layers.9.mlp": [31, 22, 32], "model.layers.10.mlp": [22, 47, 42], "model.layers.11.mlp": [11, 17, 29], "model.layers.12.mlp": [4, 3, 59], "model.layers.13.mlp": [17, 10, 47], "model.layers.14.mlp": [51, 7, 27], "model.layers.15.mlp": [24, 14, 17], "model.layers.16.mlp": [61, 33, 19], "model.layers.17.mlp": [32, 29, 26], "model.layers.18.mlp": [56, 5, 2], "model.layers.19.mlp": [24, 36, 40], "model.layers.20.mlp": [1, 56, 38], "model.layers.21.mlp": [19, 5, 28], "model.layers.22.mlp": [32, 14, 58], "model.layers.23.mlp": [20, 58, 0], "model.layers.24.mlp": [7, 63, 47], "model.layers.25.mlp": [45, 39, 46], "model.layers.26.mlp": [6, 46, 49]}

training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:784c8ddede97b768ac7df72d26cc44295ab6c84ab394fde376d273efc9af5e4c
-size 7544
+oid sha256:3f1ee949d3c02c1e467580bf3863307c1318d6d8f14e61d099b4d6190e4e75be
+size 7864