| { |
| "use_ray": false, |
| "ray_exp_name": null, |
| "device_groups": null, |
| "model": "/dnn_training_sys/ningpu/megatron_sft/models/Tongyi-DeepResearch-30B-A3B", |
| "model_type": "qwen3_moe", |
| "model_revision": null, |
| "task_type": "causal_lm", |
| "torch_dtype": "bfloat16", |
| "attn_impl": null, |
| "experts_impl": null, |
| "new_special_tokens": [], |
| "num_labels": null, |
| "problem_type": null, |
| "rope_scaling": null, |
| "device_map": null, |
| "max_memory": {}, |
| "max_model_len": null, |
| "local_repo_path": null, |
| "init_strategy": null, |
| "template": "qwen3", |
| "system": null, |
| "max_length": 131072, |
| "truncation_strategy": "delete", |
| "max_pixels": null, |
| "agent_template": null, |
| "norm_bbox": null, |
| "use_chat_template": true, |
| "padding_side": "right", |
| "padding_free": true, |
| "loss_scale": "default", |
| "sequence_parallel_size": 2, |
| "template_backend": "swift", |
| "response_prefix": null, |
| "enable_thinking": null, |
| "add_non_thinking_prefix": true, |
| "dataset": [], |
| "val_dataset": [], |
| "cached_dataset": [ |
| "/dnn_training_sys/ningpu/megatron_sft/datasets/v7/setting5/train_cached/train" |
| ], |
| "cached_val_dataset": [], |
| "split_dataset_ratio": 0.0, |
| "data_seed": 42, |
| "dataset_num_proc": 8, |
| "load_from_cache_file": false, |
| "dataset_shuffle": true, |
| "val_dataset_shuffle": false, |
| "streaming": false, |
| "interleave_prob": null, |
| "stopping_strategy": "first_exhausted", |
| "shuffle_buffer_size": 1000, |
| "download_mode": "reuse_dataset_if_exists", |
| "columns": {}, |
| "strict": false, |
| "remove_unused_columns": true, |
| "disable_auto_column_mapping": false, |
| "model_name": null, |
| "model_author": null, |
| "custom_dataset_info": [], |
| "quant_method": null, |
| "quant_bits": null, |
| "hqq_axis": null, |
| "bnb_4bit_compute_dtype": "bfloat16", |
| "bnb_4bit_quant_type": "nf4", |
| "bnb_4bit_use_double_quant": true, |
| "bnb_4bit_quant_storage": null, |
| "max_new_tokens": null, |
| "temperature": 0.9, |
| "top_k": -1, |
| "top_p": 1.0, |
| "repetition_penalty": 1.0, |
| "num_beams": 1, |
| "stream": false, |
| "stop_words": [], |
| "logprobs": false, |
| "top_logprobs": null, |
| "structured_outputs_regex": null, |
| "tuner_backend": "peft", |
| "tuner_type": "full", |
| "adapters": [], |
| "external_plugins": [], |
| "custom_register_path": [], |
| "seed": 42, |
| "model_kwargs": {}, |
| "enable_npu_model_patch": true, |
| "load_args": false, |
| "load_data_args": false, |
| "packing": true, |
| "packing_length": 131072, |
| "packing_num_proc": 1, |
| "lazy_tokenize": false, |
| "use_hf": true, |
| "hub_token": null, |
| "ddp_timeout": 18000000, |
| "ddp_backend": "nccl", |
| "ignore_args_error": false, |
| "use_swift_lora": false, |
| "freeze_llm": false, |
| "freeze_vit": true, |
| "freeze_aligner": true, |
| "freeze_parameters": [], |
| "freeze_parameters_regex": null, |
| "freeze_parameters_ratio": 0.0, |
| "trainable_parameters": [], |
| "trainable_parameters_regex": null, |
| "target_modules": [ |
| "all-linear" |
| ], |
| "target_regex": null, |
| "modules_to_save": [], |
| "lora_rank": 8, |
| "lora_alpha": 32, |
| "lora_dropout": 0.05, |
| "lora_bias": "none", |
| "lora_dtype": null, |
| "use_rslora": false, |
| "rlhf_type": null, |
| "loss_type": null, |
| "mcore_ref_model": null, |
| "mcore_ref_adapter": null, |
| "beta": null, |
| "rpo_alpha": null, |
| "reference_free": false, |
| "label_smoothing": 0.0, |
| "f_divergence_type": "reverse_kl", |
| "desirable_weight": 1.0, |
| "undesirable_weight": 1.0, |
| "calculate_KL": null, |
| "center_rewards_coefficient": null, |
| "teacher_model": null, |
| "teacher_model_type": null, |
| "teacher_model_revision": null, |
| "teacher_model_server": null, |
| "gkd_logits_topk": null, |
| "lmbda": 0.5, |
| "seq_kd": false, |
| "offload_teacher_model": false, |
| "sft_alpha": 0.0, |
| "generation_batch_size": null, |
| "steps_per_generation": null, |
| "num_generations": 8, |
| "num_generations_eval": null, |
| "max_completion_length": 512, |
| "importance_sampling_level": "token", |
| "tau_pos": 1.0, |
| "tau_neg": 1.05, |
| "epsilon": 0.2, |
| "epsilon_high": null, |
| "delta": null, |
| "use_vllm": true, |
| "vllm_mode": null, |
| "vllm_enable_prefix_caching": true, |
| "vllm_gpu_memory_utilization": 0.9, |
| "vllm_tensor_parallel_size": 1, |
| "vllm_max_model_len": null, |
| "vllm_enforce_eager": false, |
| "vllm_limit_mm_per_prompt": null, |
| "vllm_disable_cascade_attn": false, |
| "vllm_max_num_seqs": null, |
| "vllm_mm_processor_cache_gb": null, |
| "vllm_engine_kwargs": null, |
| "vllm_enable_lora": false, |
| "sleep_level": 0, |
| "offload_optimizer": false, |
| "offload_model": false, |
| "offload_bridge": false, |
| "vllm_server_base_url": null, |
| "vllm_server_host": null, |
| "vllm_server_port": [ |
| 8000 |
| ], |
| "vllm_server_timeout": 240.0, |
| "vllm_server_group_port": null, |
| "reward_funcs": [], |
| "reward_weights": null, |
| "cosine_min_len_value_wrong": -0.5, |
| "cosine_max_len_value_wrong": 0.0, |
| "cosine_min_len_value_correct": 1.0, |
| "cosine_max_len_value_correct": 0.5, |
| "cosine_max_len": null, |
| "repetition_n_grams": 3, |
| "repetition_max_penalty": -1.0, |
| "soft_max_length": null, |
| "soft_cache_length": null, |
| "dynamic_sample": false, |
| "max_resample_times": 3, |
| "overlong_filter": false, |
| "scale_rewards": "group", |
| "advantage_estimator": "grpo", |
| "kl_in_reward": false, |
| "wandb_log_unique_prompts": null, |
| "log_completions": false, |
| "rollout_importance_sampling_mode": null, |
| "rollout_importance_sampling_threshold": 2.0, |
| "log_rollout_offpolicy_metrics": false, |
| "off_policy_sequence_mask_delta": null, |
| "log_entropy": false, |
| "top_entropy_quantile": 1.0, |
| "router_replay_mode": "disabled", |
| "reward_model": null, |
| "reward_model_plugin": null, |
| "sync_ref_model": false, |
| "ref_model_sync_steps": 512, |
| "ref_model_mixup_alpha": 0.6, |
| "async_generate": false, |
| "move_model_batches": null, |
| "multi_turn_scheduler": null, |
| "max_turns": null, |
| "completion_length_limit_scope": "per_round", |
| "vllm_server_pass_dataset": false, |
| "num_iterations": 1, |
| "micro_batch_size": 1, |
| "global_batch_size": 128, |
| "recompute_granularity": "selective", |
| "recompute_method": null, |
| "recompute_num_layers": null, |
| "recompute_modules": [ |
| "core_attn" |
| ], |
| "train_iters": null, |
| "num_train_epochs": 3, |
| "masked_softmax_fusion": true, |
| "bias_dropout_fusion": true, |
| "bias_activation_fusion": true, |
| "apply_rope_fusion": true, |
| "gradient_accumulation_fusion": false, |
| "cross_entropy_loss_fusion": true, |
| "cross_entropy_fusion_impl": "te", |
| "calculate_per_token_loss": true, |
| "attention_backend": "<AttnBackend.flash: 1>", |
| "optimizer": "adam", |
| "optimizer_cpu_offload": false, |
| "optimizer_offload_fraction": 1.0, |
| "use_precision_aware_optimizer": false, |
| "main_grads_dtype": "float32", |
| "main_params_dtype": "float32", |
| "exp_avg_dtype": "float32", |
| "exp_avg_sq_dtype": "float32", |
| "manual_gc": false, |
| "manual_gc_steps": 0, |
| "manual_gc_eval": true, |
| "train_dataloader_shuffle": true, |
| "dataloader_num_workers": 8, |
| "dataloader_pin_memory": true, |
| "dataloader_persistent_workers": true, |
| "dataloader_prefetch_factor": 2, |
| "data_sharding": false, |
| "group_by_length": false, |
| "te_rng_tracker": false, |
| "data_parallel_random_init": false, |
| "mlp_padding_free": false, |
| "lr_warmup_init": 0.0, |
| "lr": 5e-05, |
| "lr_decay_style": "cosine", |
| "lr_decay_iters": null, |
| "lr_warmup_iters": 0, |
| "lr_warmup_fraction": 0.03, |
| "min_lr": 1e-06, |
| "lr_wsd_decay_style": "exponential", |
| "lr_wsd_decay_iters": null, |
| "weight_decay": 0.01, |
| "weight_decay_incr_style": "constant", |
| "start_weight_decay": 0.01, |
| "end_weight_decay": 0.01, |
| "clip_grad": 1.0, |
| "adam_beta1": 0.9, |
| "adam_beta2": 0.95, |
| "adam_eps": 1e-08, |
| "sgd_momentum": 0.9, |
| "muon_momentum": 0.9, |
| "muon_split_qkv": true, |
| "muon_use_nesterov": false, |
| "muon_scale_mode": "spectral", |
| "muon_fp32_matmul_prec": "medium", |
| "muon_num_ns_steps": 5, |
| "muon_tp_mode": "blockwise", |
| "muon_extra_scale_factor": 1.0, |
| "output_dir": "/dnn_training_sys/ningpu/megatron_sft/outputs/v7/setting5/v1-20260512-131221", |
| "save_steps": 100, |
| "no_save_optim": true, |
| "no_save_rng": true, |
| "mcore_model": null, |
| "mcore_adapter": null, |
| "no_load_optim": false, |
| "no_load_rng": false, |
| "finetune": true, |
| "perform_initialization": false, |
| "use_cpu_initialization": false, |
| "async_save": false, |
| "save_total_limit": null, |
| "metric_for_best_model": "loss", |
| "greater_is_better": false, |
| "use_persistent_ckpt_worker": false, |
| "dist_ckpt_save_pre_mcore_014": false, |
| "dist_ckpt_optim_fully_reshardable": false, |
| "distrib_optim_fully_reshardable_mem_efficient": false, |
| "local_rank": 0, |
| "use_distributed_optimizer": true, |
| "tensor_model_parallel_size": 4, |
| "pipeline_model_parallel_size": 2, |
| "decoder_first_pipeline_num_layers": null, |
| "decoder_last_pipeline_num_layers": null, |
| "account_for_embedding_in_pipeline_split": false, |
| "account_for_loss_in_pipeline_split": false, |
| "overlap_p2p_comm": false, |
| "align_param_gather": false, |
| "sequence_parallel": true, |
| "context_parallel_size": 2, |
| "tp_comm_overlap": false, |
| "overlap_grad_reduce": true, |
| "overlap_param_gather": true, |
| "overlap_param_gather_with_optimizer_step": false, |
| "align_grad_reduce": true, |
| "virtual_pipeline_model_parallel_size": null, |
| "microbatch_group_size_per_vp_stage": null, |
| "pipeline_model_parallel_layout": null, |
| "expert_model_parallel_size": 4, |
| "expert_tensor_parallel_size": 1, |
| "report_to": [ |
| "tensorboard", |
| "wandb" |
| ], |
| "logging_steps": 1, |
| "tensorboard_dir": "/dnn_training_sys/ningpu/megatron_sft/outputs/v7/setting5/v1-20260512-131221/runs", |
| "tensorboard_queue_size": 50, |
| "wandb_project": "ma-explore", |
| "wandb_exp_name": "v7_chinese_must_visit_short0_5_sub038", |
| "swanlab_project": "megatron-swift", |
| "swanlab_exp_name": null, |
| "eval_iters": -1, |
| "eval_steps": 100, |
| "fp8_format": null, |
| "fp8_recipe": "delayed", |
| "fp8_amax_history_len": 1024, |
| "fp8_amax_compute_algo": "max", |
| "fp8_param_gather": false, |
| "fp16": false, |
| "bf16": true, |
| "apply_query_key_layer_scaling": false, |
| "attention_softmax_in_fp32": true, |
| "accumulate_allreduce_grads_in_fp32": true, |
| "moe_router_load_balancing_type": null, |
| "moe_router_dtype": "fp32", |
| "moe_token_dispatcher_type": null, |
| "moe_enable_deepep": false, |
| "moe_grouped_gemm": true, |
| "moe_permute_fusion": true, |
| "moe_aux_loss_coeff": 0.001, |
| "moe_z_loss_coeff": null, |
| "moe_shared_expert_overlap": true, |
| "moe_layer_recompute": false, |
| "moe_expert_capacity_factor": null, |
| "moe_pad_expert_input_to_capacity": false, |
| "moe_token_drop_policy": null, |
| "mtp_num_layers": null, |
| "mtp_loss_scaling_factor": 0.1, |
| "mtp_decoder_input_detach": false, |
| "mtp_shared_weights": false, |
| "save_safetensors": true, |
| "ref_model": null, |
| "ref_adapters": [], |
| "merge_lora": true, |
| "max_shard_size": "5GB", |
| "vit_gradient_checkpointing": false, |
| "vit_gradient_checkpointing_kwargs": null, |
| "vit_attn_impl": null, |
| "vit_lr": null, |
| "aligner_lr": null, |
| "dsa_indexer_loss_coeff": null, |
| "dsa_indexer_use_sparse_loss": false, |
| "check_model": true, |
| "apply_wd_to_qk_layernorm": false, |
| "linear_decoupled_in_proj": false, |
| "enable_dft_loss": false, |
| "enable_channel_loss": false, |
| "save_strategy": "steps", |
| "callbacks": [ |
| "print", |
| "default_flow", |
| "tensorboard", |
| "wandb" |
| ], |
| "add_version": true, |
| "create_checkpoint_symlink": false, |
| "swift_version": "4.2.0", |
| "ckpt_dir": null, |
| "rank": 0, |
| "global_world_size": 64, |
| "local_world_size": 8, |
| "model_suffix": "Tongyi-DeepResearch-30B-A3B", |
| "model_info": "ModelInfo(model_type='qwen3_moe', model_dir='/dnn_training_sys/ningpu/megatron_sft/models/Tongyi-DeepResearch-30B-A3B', torch_dtype=torch.bfloat16, max_model_len=131072, quant_method=None, quant_bits=None, rope_scaling=None, is_moe_model=True, is_multimodal=False, config=None, task_type='causal_lm', num_labels=None)", |
| "model_meta": "ModelMeta(model_type='qwen3_moe', model_groups=[ModelGroup(models=[Model(ms_model_id='Qwen/Qwen3-30B-A3B-Base', hf_model_id='Qwen/Qwen3-30B-A3B-Base', model_path=None, ms_revision=None, hf_revision=None), Model(ms_model_id='Qwen/Qwen3-30B-A3B', hf_model_id='Qwen/Qwen3-30B-A3B', model_path=None, ms_revision=None, hf_revision=None), Model(ms_model_id='Qwen/Qwen3-235B-A22B', hf_model_id='Qwen/Qwen3-235B-A22B', model_path=None, ms_revision=None, hf_revision=None), Model(ms_model_id='Qwen/Qwen3-30B-A3B-FP8', hf_model_id='Qwen/Qwen3-30B-A3B-FP8', model_path=None, ms_revision=None, hf_revision=None), Model(ms_model_id='Qwen/Qwen3-235B-A22B-FP8', hf_model_id='Qwen/Qwen3-235B-A22B-FP8', model_path=None, ms_revision=None, hf_revision=None), Model(ms_model_id='swift/Qwen3-30B-A3B-AWQ', hf_model_id='cognitivecomputations/Qwen3-30B-A3B-AWQ', model_path=None, ms_revision=None, hf_revision=None), Model(ms_model_id='swift/Qwen3-235B-A22B-AWQ', hf_model_id='cognitivecomputations/Qwen3-235B-A22B-AWQ', model_path=None, ms_revision=None, hf_revision=None)], template='qwen3', ignore_patterns=None, requires=None, tags=[]), ModelGroup(models=[Model(ms_model_id='iic/Tongyi-DeepResearch-30B-A3B', hf_model_id='Alibaba-NLP/Tongyi-DeepResearch-30B-A3B', model_path=None, ms_revision=None, hf_revision=None)], template='qwen3', ignore_patterns=None, requires=None, tags=[]), ModelGroup(models=[Model(ms_model_id='Qwen/Qwen3-30B-A3B-Instruct-2507', hf_model_id='Qwen/Qwen3-30B-A3B-Instruct-2507', model_path=None, ms_revision=None, hf_revision=None), Model(ms_model_id='Qwen/Qwen3-30B-A3B-Instruct-2507-FP8', hf_model_id='Qwen/Qwen3-30B-A3B-Instruct-2507-FP8', model_path=None, ms_revision=None, hf_revision=None), Model(ms_model_id='Qwen/Qwen3-235B-A22B-Instruct-2507', hf_model_id='Qwen/Qwen3-235B-A22B-Instruct-2507', model_path=None, ms_revision=None, hf_revision=None), Model(ms_model_id='Qwen/Qwen3-235B-A22B-Instruct-2507-FP8', hf_model_id='Qwen/Qwen3-235B-A22B-Instruct-2507-FP8', model_path=None, ms_revision=None, hf_revision=None), Model(ms_model_id='swift/Qwen3-235B-A22B-Instruct-2507-AWQ', hf_model_id=None, model_path=None, ms_revision=None, hf_revision=None)], template='qwen3_nothinking', ignore_patterns=None, requires=None, tags=[]), ModelGroup(models=[Model(ms_model_id='Qwen/Qwen3-Coder-30B-A3B-Instruct', hf_model_id='Qwen/Qwen3-Coder-30B-A3B-Instruct', model_path=None, ms_revision=None, hf_revision=None), Model(ms_model_id='Qwen/Qwen3-Coder-30B-A3B-Instruct-FP8', hf_model_id='Qwen/Qwen3-Coder-30B-A3B-Instruct-FP8', model_path=None, ms_revision=None, hf_revision=None), Model(ms_model_id='Qwen/Qwen3-Coder-480B-A35B-Instruct', hf_model_id='Qwen/Qwen3-Coder-480B-A35B-Instruct', model_path=None, ms_revision=None, hf_revision=None), Model(ms_model_id='Qwen/Qwen3-Coder-480B-A35B-Instruct-FP8', hf_model_id='Qwen/Qwen3-Coder-480B-A35B-Instruct-FP8', model_path=None, ms_revision=None, hf_revision=None), Model(ms_model_id='swift/Qwen3-Coder-480B-A35B-Instruct-AWQ', hf_model_id=None, model_path=None, ms_revision=None, hf_revision=None)], template='qwen3_coder', ignore_patterns=None, requires=None, tags=['coding']), ModelGroup(models=[Model(ms_model_id='iic/QwenLong-L1.5-30B-A3B', hf_model_id='Tongyi-Zhiwen/QwenLong-L1.5-30B-A3B', model_path=None, ms_revision=None, hf_revision=None)], template='qwen3_thinking', ignore_patterns=None, requires=None, tags=[]), ModelGroup(models=[Model(ms_model_id='Qwen/Qwen3-30B-A3B-Thinking-2507', hf_model_id='Qwen/Qwen3-30B-A3B-Thinking-2507', model_path=None, ms_revision=None, hf_revision=None), Model(ms_model_id='Qwen/Qwen3-30B-A3B-Thinking-2507-FP8', hf_model_id='Qwen/Qwen3-30B-A3B-Thinking-2507-FP8', model_path=None, ms_revision=None, hf_revision=None), Model(ms_model_id='Qwen/Qwen3-235B-A22B-Thinking-2507', hf_model_id='Qwen/Qwen3-235B-A22B-Thinking-2507', model_path=None, ms_revision=None, hf_revision=None), Model(ms_model_id='Qwen/Qwen3-235B-A22B-Thinking-2507-FP8', hf_model_id='Qwen/Qwen3-235B-A22B-Thinking-2507-FP8', model_path=None, ms_revision=None, hf_revision=None), Model(ms_model_id='swift/Qwen3-235B-A22B-Thinking-2507-AWQ', hf_model_id=None, model_path=None, ms_revision=None, hf_revision=None)], template='qwen3_thinking', ignore_patterns=None, requires=None, tags=[]), ModelGroup(models=[Model(ms_model_id='AIDC-AI/Marco-Nano-Base', hf_model_id='AIDC-AI/Marco-Nano-Base', model_path=None, ms_revision=None, hf_revision=None), Model(ms_model_id='AIDC-AI/Marco-Nano-Instruct', hf_model_id='AIDC-AI/Marco-Nano-Instruct', model_path=None, ms_revision=None, hf_revision=None), Model(ms_model_id='AIDC-AI/Marco-Mini-Base', hf_model_id='AIDC-AI/Marco-Mini-Base', model_path=None, ms_revision=None, hf_revision=None), Model(ms_model_id='AIDC-AI/Marco-Mini-Instruct', hf_model_id='AIDC-AI/Marco-Mini-Instruct', model_path=None, ms_revision=None, hf_revision=None), Model(ms_model_id='AIDC-AI/Marco-Mini-Global-Base', hf_model_id='AIDC-AI/Marco-Mini-Global-Base', model_path=None, ms_revision=None, hf_revision=None)], template='qwen3_nothinking', ignore_patterns=None, requires=None, tags=[])], loader=<class 'swift.model.register.ModelLoader'>, template='qwen3', model_arch=None, mcore_model_type=None, architectures=['Qwen3MoeForCausalLM'], additional_saved_files=[], torch_dtype=None, is_multimodal=False, is_reward=False, task_type=None, ignore_patterns=None, requires=['transformers>=4.51'], tags=[])", |
| "model_dir": "/dnn_training_sys/ningpu/megatron_sft/models/Tongyi-DeepResearch-30B-A3B", |
| "template_meta": "Qwen3MixedTemplateMeta(template_type='qwen3', prefix=[], prompt=['<|im_start|>user\\n{{QUERY}}<|im_end|>\\n<|im_start|>assistant\\n'], chat_sep=['<|im_end|>\\n'], suffix=['<|im_end|>\\n'], template_cls=<class 'swift.template.base.Template'>, system_prefix=['<|im_start|>system\\n{{SYSTEM}}<|im_end|>\\n'], default_system=None, auto_add_bos=False, stop_words=['<|endoftext|>'], agent_template='hermes', is_thinking=True, thinking_prefix='', non_thinking_prefix='<think>\\n\\n</think>\\n\\n', history_thinking_prefix='')", |
| "_val_dataset_exists": false, |
| "hub": "<class 'swift.hub.hub.HFHub'>", |
| "seq_length": 131072, |
| "is_multimodal": false, |
| "megatron_model_meta": "ModelMeta(model_type='gpt', model_types=['qwen2', 'llama', 'qwen3', 'qwen2_moe', 'qwen3_moe', 'internlm3', 'mimo', 'deepseek', 'deepseek_v2', 'deepseek_v3', 'deepseek_v32', 'kimi_k2', 'dots1', 'ernie4_5', 'ernie4_5_moe', 'glm4_moe', 'glm4_moe_lite', 'glm_moe_dsa', 'gpt_oss'], bridge_cls=<class 'mcore_bridge.bridge.gpt_bridge.GPTBridge'>, visual_cls=None, is_multimodal=False, loader=<class 'mcore_bridge.model.register.ModelLoader'>)", |
| "fp8": null, |
| "world_size": 64, |
| "data_parallel_size": 4, |
| "num_microbatches": 32 |
| } |
