Exporting mxfp8 quantization with `use_cpu_initialization` enabled seemingly causes a VRAM leak

Checklist / 检查清单

• I have searched existing issues, and this is a new bug report. / 我已经搜索过现有的 issues，确认这是一个新的 bug report。

Bug Description / Bug 描述

在我尝试使用megatron量化模型为mxfp8时, 显存被占满且产生驱动错误:

这台设备的显存足以完成同一个模型的微调, 而且由显存占用曲线看, 似乎是存在泄露.

How to Reproduce / 如何复现

以下是启动参数和运行日志:

(swift_megatron) root@hypnoshv:~/maica_llm/esft# bash quant_fp8.sh 
BMC is:
/root/maica_llm/esft/megatron_output/Qwen3.5-397B-A17B/v42-20260622-143501/checkpoint-15-merged
run sh: `/root/swift_megatron/bin/python -m torch.distributed.run --nproc_per_node 10 /root/swift/swift/cli/_megatron/export.py --model /root/maica_llm/esft/megatron_output/Qwen3.5-397B-A17B/v42-20260622-143501/checkpoint-15-merged --quant_method fp8 --output_dir /root/maica_llm/esft/megatron_output/Qwen3.5-397B-A17B/v42-20260622-143501/checkpoint-15-merged-FP8 --to_hf true --fp8_recipe mxfp8 --fp8_format e4m3 --fp8_param_gather true --linear_decoupled_in_proj true --pipeline_model_parallel_size 10 --pipeline_model_parallel_layout Et*6,(|t*6)*8,|t*6L --device_map cpu --use_cpu_initialization true`

*****************************************
Setting OMP_NUM_THREADS environment variable for each process to be 1 in default, to avoid your system being overloaded, please further tune the variable for optimal performance in your application as needed. 
*****************************************
/root/swift_megatron/lib/python3.11/site-packages/transformer_engine/jax/sharding.py:23: DeprecationWarning: jax.interpreters.pxla.thread_resources is deprecated as of JAX v0.8.2. Please switch to using `with jax.set_mesh(mesh)` instead of `with mesh:` and then use `jax.sharding.get_abstract_mesh()` to get the current mesh.
  _PXLA_THREAD_RESOURCES = pxla.thread_resources
/root/swift_megatron/lib/python3.11/site-packages/transformer_engine/jax/sharding.py:23: DeprecationWarning: jax.interpreters.pxla.thread_resources is deprecated as of JAX v0.8.2. Please switch to using `with jax.set_mesh(mesh)` instead of `with mesh:` and then use `jax.sharding.get_abstract_mesh()` to get the current mesh.
  _PXLA_THREAD_RESOURCES = pxla.thread_resources
/root/swift_megatron/lib/python3.11/site-packages/transformer_engine/jax/sharding.py:23: DeprecationWarning: jax.interpreters.pxla.thread_resources is deprecated as of JAX v0.8.2. Please switch to using `with jax.set_mesh(mesh)` instead of `with mesh:` and then use `jax.sharding.get_abstract_mesh()` to get the current mesh.
  _PXLA_THREAD_RESOURCES = pxla.thread_resources
/root/swift_megatron/lib/python3.11/site-packages/transformer_engine/jax/sharding.py:23: DeprecationWarning: jax.interpreters.pxla.thread_resources is deprecated as of JAX v0.8.2. Please switch to using `with jax.set_mesh(mesh)` instead of `with mesh:` and then use `jax.sharding.get_abstract_mesh()` to get the current mesh.
  _PXLA_THREAD_RESOURCES = pxla.thread_resources
/root/swift_megatron/lib/python3.11/site-packages/transformer_engine/jax/sharding.py:23: DeprecationWarning: jax.interpreters.pxla.thread_resources is deprecated as of JAX v0.8.2. Please switch to using `with jax.set_mesh(mesh)` instead of `with mesh:` and then use `jax.sharding.get_abstract_mesh()` to get the current mesh.
  _PXLA_THREAD_RESOURCES = pxla.thread_resources
/root/swift_megatron/lib/python3.11/site-packages/transformer_engine/jax/sharding.py:23: DeprecationWarning: jax.interpreters.pxla.thread_resources is deprecated as of JAX v0.8.2. Please switch to using `with jax.set_mesh(mesh)` instead of `with mesh:` and then use `jax.sharding.get_abstract_mesh()` to get the current mesh.
  _PXLA_THREAD_RESOURCES = pxla.thread_resources
/root/swift_megatron/lib/python3.11/site-packages/transformer_engine/jax/sharding.py:23: DeprecationWarning: jax.interpreters.pxla.thread_resources is deprecated as of JAX v0.8.2. Please switch to using `with jax.set_mesh(mesh)` instead of `with mesh:` and then use `jax.sharding.get_abstract_mesh()` to get the current mesh.
  _PXLA_THREAD_RESOURCES = pxla.thread_resources
/root/swift_megatron/lib/python3.11/site-packages/transformer_engine/jax/sharding.py:23: DeprecationWarning: jax.interpreters.pxla.thread_resources is deprecated as of JAX v0.8.2. Please switch to using `with jax.set_mesh(mesh)` instead of `with mesh:` and then use `jax.sharding.get_abstract_mesh()` to get the current mesh.
  _PXLA_THREAD_RESOURCES = pxla.thread_resources
/root/swift_megatron/lib/python3.11/site-packages/transformer_engine/jax/sharding.py:23: DeprecationWarning: jax.interpreters.pxla.thread_resources is deprecated as of JAX v0.8.2. Please switch to using `with jax.set_mesh(mesh)` instead of `with mesh:` and then use `jax.sharding.get_abstract_mesh()` to get the current mesh.
  _PXLA_THREAD_RESOURCES = pxla.thread_resources
/root/swift_megatron/lib/python3.11/site-packages/transformer_engine/jax/sharding.py:23: DeprecationWarning: jax.interpreters.pxla.thread_resources is deprecated as of JAX v0.8.2. Please switch to using `with jax.set_mesh(mesh)` instead of `with mesh:` and then use `jax.sharding.get_abstract_mesh()` to get the current mesh.
  _PXLA_THREAD_RESOURCES = pxla.thread_resources
[INFO:swift] mcore_bridge.__version__: 1.4.3
[INFO:swift] megatron.core.__version__: 0.16.1
[INFO:mcore_bridge] Setting USE_MCORE_GDN: True. You can adjust this hyperparameter through the environment variable: `USE_MCORE_GDN`.
[INFO:swift] Successfully registered `/root/swift/swift/dataset/data/dataset_info.json`.
[INFO:swift] Successfully loaded /root/maica_llm/esft/megatron_output/Qwen3.5-397B-A17B/v42-20260622-143501/checkpoint-15-merged/args.json.
[INFO:swift] rank: 0, local_rank: 0, world_size: 10, local_world_size: 10
[transformers] `torch_dtype` is deprecated! Use `dtype` instead!
[transformers] `torch_dtype` is deprecated! Use `dtype` instead!
[transformers] `torch_dtype` is deprecated! Use `dtype` instead!
[transformers] `torch_dtype` is deprecated! Use `dtype` instead!
[transformers] `torch_dtype` is deprecated! Use `dtype` instead!
[transformers] `torch_dtype` is deprecated! Use `dtype` instead!
[transformers] `torch_dtype` is deprecated! Use `dtype` instead!
[transformers] `torch_dtype` is deprecated! Use `dtype` instead!
[INFO:swift] Setting args.lazy_tokenize: True
[INFO:swift] args.output_dir: `/root/maica_llm/esft/megatron_output/Qwen3.5-397B-A17B/v42-20260622-143501/checkpoint-15-merged-FP8`
[transformers] `torch_dtype` is deprecated! Use `dtype` instead!
[transformers] `torch_dtype` is deprecated! Use `dtype` instead!
[INFO:swift] Setting torch_dtype: torch.bfloat16
[INFO:swift] TP: 1, PP: 10, VPP: None, CP: 1, EP: 1, ETP: 1
[INFO:swift] Setting random seeds to 42.
[INFO:swift] args: MegatronExportArguments(use_ray=False, ray_exp_name=None, device_groups=None, model='/root/maica_llm/esft/megatron_output/Qwen3.5-397B-A17B/v42-20260622-143501/checkpoint-15-merged', model_type='qwen3_5_moe', model_revision=None, task_type='causal_lm', torch_dtype=torch.bfloat16, attn_impl=None, experts_impl=None, new_special_tokens=[], num_labels=None, problem_type=None, rope_scaling=None, device_map='cpu', max_memory={}, max_model_len=None, local_repo_path=None, init_strategy=None, template='qwen3_5', system=None, max_length=262144, truncation_strategy='delete', max_pixels=None, agent_template=None, norm_bbox=None, use_chat_template=True, padding_side='right', padding_free=False, loss_scale='default+ignore_empty_think', sequence_parallel_size=1, is_binary_loss_scale=None, template_backend='swift', response_prefix=None, enable_thinking=None, preserve_thinking=None, add_non_thinking_prefix=True, disable_ignore_empty_think=False, dataset=[], val_dataset=[], cached_dataset=[], cached_val_dataset=[], split_dataset_ratio=0.0, data_seed=42, dataset_num_proc=1, load_from_cache_file=False, dataset_shuffle=True, val_dataset_shuffle=False, streaming=False, interleave_prob=None, 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=None, model_author=None, custom_dataset_info=[], quant_method='fp8', quant_bits=None, hqq_axis=None, bnb_4bit_compute_dtype=torch.bfloat16, bnb_4bit_quant_type='nf4', bnb_4bit_use_double_quant=True, bnb_4bit_quant_storage=None, max_new_tokens=None, 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=None, structured_outputs_regex=None, tuner_backend='peft', tuner_type='lora', adapters=[], external_plugins=[], custom_register_path=[], seed=42, model_kwargs={}, enable_npu_model_patch=True, load_args=True, load_data_args=False, packing=False, packing_length=None, packing_num_proc=1, lazy_tokenize=True, use_hf=False, hub_token=None, 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=None, freeze_parameters_ratio=0.0, trainable_parameters=[], trainable_parameters_regex=None, target_modules=['all-linear'], target_regex=None, modules_to_save=[], lora_rank=64, lora_alpha=128, lora_dropout=0.05, lora_bias='none', lora_dtype=None, use_rslora=False, rlhf_type=None, loss_type=None, mcore_ref_model=None, mcore_ref_adapter=None, beta=None, rpo_alpha=None, reference_free=False, label_smoothing=0.0, f_divergence_type='reverse_kl', desirable_weight=1.0, undesirable_weight=1.0, calculate_KL=None, center_rewards_coefficient=None, teacher_model=None, teacher_model_type=None, teacher_model_revision=None, teacher_model_server=None, gkd_logits_topk=None, lmbda=0.5, seq_kd=False, offload_teacher_model=False, sft_alpha=0.0, generation_batch_size=None, steps_per_generation=None, num_generations=8, num_generations_eval=None, max_completion_length=512, importance_sampling_level='token', tau_pos=1.0, tau_neg=1.05, real_tau=0.5, fipo_decay_rate=32.0, fipo_clip_range=0.2, fipo_clip_high_only=True, fipo_safety_threshold=4.0, epsilon=0.2, epsilon_high=None, delta=None, use_vllm=True, vllm_mode=None, vllm_enable_prefix_caching=True, vllm_gpu_memory_utilization=0.9, vllm_tensor_parallel_size=1, vllm_max_model_len=None, vllm_enforce_eager=False, vllm_limit_mm_per_prompt=None, vllm_disable_cascade_attn=False, vllm_max_num_seqs=None, vllm_mm_processor_cache_gb=None, vllm_engine_kwargs=None, vllm_enable_lora=False, sleep_level=0, offload_optimizer=False, offload_model=False, offload_bridge=False, vllm_server_base_url=None, vllm_server_host=None, vllm_server_port=[8000], vllm_server_timeout=240.0, vllm_server_group_port=None, reward_funcs=[], reward_weights=None, 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=None, repetition_n_grams=3, repetition_max_penalty=-1.0, soft_max_length=None, soft_cache_length=None, dynamic_sample=False, max_resample_times=3, overlong_filter=False, scale_rewards='group', advantage_estimator='grpo', kl_in_reward=False, wandb_log_unique_prompts=None, log_completions=False, rollout_importance_sampling_mode=None, rollout_importance_sampling_threshold=2.0, log_rollout_offpolicy_metrics=False, off_policy_sequence_mask_delta=None, log_entropy=False, top_entropy_quantile=1.0, router_replay_mode='disabled', reward_model=None, reward_model_plugin=None, sync_ref_model=False, ref_model_sync_steps=512, ref_model_mixup_alpha=0.6, async_generate=False, move_model_batches=None, multi_turn_scheduler=None, max_turns=None, completion_length_limit_scope='per_round', vllm_server_pass_dataset=False, use_gym_env=None, gym_env=None, num_iterations=1, micro_batch_size=1, global_batch_size=16, recompute_granularity=None, recompute_method=None, recompute_num_layers=None, recompute_modules=['core_attn'], train_iters=None, num_train_epochs=None, masked_softmax_fusion=True, bias_dropout_fusion=True, bias_activation_fusion=True, apply_rope_fusion=False, gradient_accumulation_fusion=True, cross_entropy_loss_fusion=True, cross_entropy_fusion_impl='native', calculate_per_token_loss=True, attention_backend=<AttnBackend.unfused: 3>, optimizer='adam', optimizer_cpu_offload=False, optimizer_offload_fraction=1.0, use_precision_aware_optimizer=False, main_grads_dtype=torch.float32, main_params_dtype=torch.float32, exp_avg_dtype=torch.float32, exp_avg_sq_dtype=torch.float32, manual_gc=False, manual_gc_steps=0, manual_gc_eval=True, train_dataloader_shuffle=True, dataloader_num_workers=4, dataloader_pin_memory=True, dataloader_persistent_workers=False, 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=0.0001, lr_decay_style='cosine', lr_decay_iters=None, lr_warmup_iters=0, lr_warmup_fraction=None, min_lr=0, lr_wsd_decay_style='exponential', lr_wsd_decay_iters=None, weight_decay=0.1, weight_decay_incr_style='constant', start_weight_decay=0.1, end_weight_decay=0.1, 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_coefficient_type='quintic', muon_num_ns_steps=5, muon_tp_mode='blockwise', muon_extra_scale_factor=1.0, muon_scalar_optimizer='adam', output_dir='/root/maica_llm/esft/megatron_output/Qwen3.5-397B-A17B/v42-20260622-143501/checkpoint-15-merged-FP8', save_steps=500, no_save_optim=True, no_save_rng=True, mcore_model='/root/maica_llm/esft/megatron_weights/Qwen/Qwen3.5-397B-A17B-mcore', mcore_adapter=None, no_load_optim=True, no_load_rng=True, finetune=True, perform_initialization=False, use_cpu_initialization=True, async_save=False, save_total_limit=None, 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=1, pipeline_model_parallel_size=10, decoder_first_pipeline_num_layers=None, decoder_last_pipeline_num_layers=None, account_for_embedding_in_pipeline_split=False, account_for_loss_in_pipeline_split=False, overlap_p2p_comm=False, batch_p2p_comm=True, align_param_gather=False, sequence_parallel=False, context_parallel_size=1, tp_comm_overlap=False, overlap_grad_reduce=False, overlap_param_gather=False, overlap_param_gather_with_optimizer_step=False, align_grad_reduce=True, virtual_pipeline_model_parallel_size=None, microbatch_group_size_per_vp_stage=None, pipeline_model_parallel_layout='Et*6,(|t*6)*8,|t*6L', expert_model_parallel_size=1, expert_tensor_parallel_size=1, report_to=['tensorboard'], logging_steps=5, tensorboard_dir=None, tensorboard_queue_size=50, wandb_project='megatron-swift', wandb_exp_name=None, swanlab_project='megatron-swift', swanlab_exp_name=None, eval_iters=-1, eval_steps=500, fp8_format='e4m3', fp8_recipe='mxfp8', fp8_param_gather=True, fp8_amax_history_len=1024, fp8_amax_compute_algo='max', fp4_format=None, fp4_recipe='nvfp4', fp4_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=None, moe_router_dtype='fp32', moe_token_dispatcher_type='alltoall', moe_enable_deepep=False, moe_grouped_gemm=True, moe_permute_fusion=False, moe_aux_loss_coeff=None, moe_z_loss_coeff=None, moe_shared_expert_overlap=False, moe_layer_recompute=False, moe_expert_capacity_factor=None, moe_pad_expert_input_to_capacity=False, moe_token_drop_policy='probs', mtp_num_layers=None, mtp_loss_scaling_factor=0.1, mtp_decoder_input_detach=False, mtp_shared_weights=False, save_safetensors=True, ref_model=None, ref_adapters=[], merge_lora=True, max_shard_size='5GB', vit_gradient_checkpointing=False, vit_gradient_checkpointing_kwargs=None, vit_attn_impl=None, vit_lr=None, aligner_lr=None, dsa_indexer_loss_coeff=0.0, dsa_indexer_use_sparse_loss=False, csa_dense_mode=False, use_fused_mhc=False, mhc_recompute_layer_num=None, megatron_extra_kwargs=None, language_model_only=False, check_model=True, apply_wd_to_qk_layernorm=False, linear_decoupled_in_proj=True, enable_dft_loss=False, enable_channel_loss=False, mrl_dims=None, save_strategy='steps', callbacks=['print', 'default_flow', 'tensorboard'], skip_megatron_init=False, to_mcore=False, to_hf=True, test_convert_precision=False, test_convert_dtype=torch.float32, exist_ok=False)
[INFO:swift] Global seed set to 42
[INFO:swift] Start time of running main: 2026-06-23 11:44:45.830301
[INFO:swift] swift.__version__: 4.4.0.dev0
[INFO:swift] Setting torch_dtype: torch.bfloat16
[transformers] The fast path is not available because one of the required library is not installed. Falling back to torch implementation. To install follow https://github.com/fla-org/flash-linear-attention#installation and https://github.com/Dao-AILab/causal-conv1d
[transformers] The fast path is not available because one of the required library is not installed. Falling back to torch implementation. To install follow https://github.com/fla-org/flash-linear-attention#installation and https://github.com/Dao-AILab/causal-conv1d
[INFO:swift] Setting max_ratio: 200. You can adjust this hyperparameter through the environment variable: `MAX_RATIO`.
[INFO:swift] Setting frame_factor: 2. You can adjust this hyperparameter through the environment variable: `FRAME_FACTOR`.
[INFO:swift] Setting fps: 2.0. You can adjust this hyperparameter through the environment variable: `FPS`.
[INFO:swift] Setting fps_min_frames: 4. You can adjust this hyperparameter through the environment variable: `FPS_MIN_FRAMES`.
[INFO:swift] Setting fps_max_frames: 768. You can adjust this hyperparameter through the environment variable: `FPS_MAX_FRAMES`.
[INFO:swift] Setting image_max_token_num: 16384. You can adjust this hyperparameter through the environment variable: `IMAGE_MAX_TOKEN_NUM`.
[INFO:swift] Setting image_min_token_num: 4. You can adjust this hyperparameter through the environment variable: `IMAGE_MIN_TOKEN_NUM`.
[INFO:swift] Setting spatial_merge_size: 2. You can adjust this hyperparameter through the environment variable: `SPATIAL_MERGE_SIZE`.
[INFO:swift] Setting video_max_token_num: 768. You can adjust this hyperparameter through the environment variable: `VIDEO_MAX_TOKEN_NUM`.
[INFO:swift] Setting video_min_token_num: 128. You can adjust this hyperparameter through the environment variable: `VIDEO_MIN_TOKEN_NUM`.
[transformers] The fast path is not available because one of the required library is not installed. Falling back to torch implementation. To install follow https://github.com/fla-org/flash-linear-attention#installation and https://github.com/Dao-AILab/causal-conv1d
[transformers] The fast path is not available because one of the required library is not installed. Falling back to torch implementation. To install follow https://github.com/fla-org/flash-linear-attention#installation and https://github.com/Dao-AILab/causal-conv1d
[INFO:swift] model_kwargs: {'device_map': 'cuda:0', 'dtype': torch.bfloat16}
[transformers] The fast path is not available because one of the required library is not installed. Falling back to torch implementation. To install follow https://github.com/fla-org/flash-linear-attention#installation and https://github.com/Dao-AILab/causal-conv1d
[transformers] The fast path is not available because one of the required library is not installed. Falling back to torch implementation. To install follow https://github.com/fla-org/flash-linear-attention#installation and https://github.com/Dao-AILab/causal-conv1d
[transformers] The fast path is not available because one of the required library is not installed. Falling back to torch implementation. To install follow https://github.com/fla-org/flash-linear-attention#installation and https://github.com/Dao-AILab/causal-conv1d
[transformers] The fast path is not available because one of the required library is not installed. Falling back to torch implementation. To install follow https://github.com/fla-org/flash-linear-attention#installation and https://github.com/Dao-AILab/causal-conv1d
[transformers] The fast path is not available because one of the required library is not installed. Falling back to torch implementation. To install follow https://github.com/fla-org/flash-linear-attention#installation and https://github.com/Dao-AILab/causal-conv1d
[transformers] The fast path is not available because one of the required library is not installed. Falling back to torch implementation. To install follow https://github.com/fla-org/flash-linear-attention#installation and https://github.com/Dao-AILab/causal-conv1d
/root/swift_megatron/lib/python3.11/site-packages/megatron/core/transformer/transformer_config.py:1705: UserWarning: full scope is deprecated. Use empty cuda_graph_scope to capture the whole layer.
  warnings.warn(
/root/swift_megatron/lib/python3.11/site-packages/megatron/core/transformer/transformer_config.py:1705: UserWarning: full scope is deprecated. Use empty cuda_graph_scope to capture the whole layer.
  warnings.warn(
/root/swift_megatron/lib/python3.11/site-packages/megatron/core/transformer/transformer_config.py:1705: UserWarning: full scope is deprecated. Use empty cuda_graph_scope to capture the whole layer.
  warnings.warn(
/root/swift_megatron/lib/python3.11/site-packages/megatron/core/transformer/transformer_config.py:1705: UserWarning: full scope is deprecated. Use empty cuda_graph_scope to capture the whole layer.
  warnings.warn(
/root/swift_megatron/lib/python3.11/site-packages/megatron/core/transformer/transformer_config.py:1705: UserWarning: full scope is deprecated. Use empty cuda_graph_scope to capture the whole layer.
  warnings.warn(
/root/swift_megatron/lib/python3.11/site-packages/megatron/core/transformer/transformer_config.py:1705: UserWarning: full scope is deprecated. Use empty cuda_graph_scope to capture the whole layer.
  warnings.warn(
[INFO:swift] default_system: None
[INFO:swift] max_length: 262144
[INFO:swift] response_prefix: None
[INFO:swift] agent_template: qwen3_5
[INFO:swift] norm_bbox: norm1000
[INFO:swift] Setting ROOT_IMAGE_DIR: None. You can adjust this hyperparameter through the environment variable: `ROOT_IMAGE_DIR`.
[INFO:swift] Setting QWENVL_BBOX_FORMAT: legacy. You can adjust this hyperparameter through the environment variable: `QWENVL_BBOX_FORMAT`.
/root/swift_megatron/lib/python3.11/site-packages/megatron/core/transformer/transformer_config.py:1705: UserWarning: full scope is deprecated. Use empty cuda_graph_scope to capture the whole layer.
  warnings.warn(
/root/swift_megatron/lib/python3.11/site-packages/megatron/core/transformer/transformer_config.py:1705: UserWarning: full scope is deprecated. Use empty cuda_graph_scope to capture the whole layer.
  warnings.warn(
/root/swift_megatron/lib/python3.11/site-packages/megatron/core/transformer/transformer_config.py:1705: UserWarning: full scope is deprecated. Use empty cuda_graph_scope to capture the whole layer.
  warnings.warn(
/root/swift_megatron/lib/python3.11/site-packages/megatron/core/transformer/transformer_config.py:1705: UserWarning: full scope is deprecated. Use empty cuda_graph_scope to capture the whole layer.
  warnings.warn(

此时显存被占满, 且执行卡住. 尝试禁用/启用device_map没有效果. 尝试添加/去除mtp层没有效果.

Additional Information / 补充信息

No response

Start with swift/cli/_megatron/export.py and reproduce the command using --quant_method fp8, --fp8_recipe mxfp8, --device_map cpu, and --use_cpu_initialization true. Compare memory behavior with CPU initialization disabled and trace the export path until the VRAM growth is explained; done means the export completes without exhausting VRAM or triggering a driver error.