[Bug] 用的官方的一键包，wan2.1模型都下完了，运行也没报错就总自动断开 #870
Description
4060ti 16g 32g内存，电脑其他软件啥都没开。模型全部从魔搭上自动下载的。运行不了，加载完模型就断开了。试过多种模型和注意力机制，也重启过软件和电脑，一直是一样的问题，求帮助！
最近一次日志如下：
2026-02-04 10:25:37.817 | INFO | lightx2v_platform.base.base:init_ai_device:15 - Initialized AI_DEVICE: cuda
2026-02-04 10:25:37.834 | INFO | lightx2v_platform.base.base:check_ai_device:32 - AI device for platform 'cuda' is available.
INFO 02-04 10:25:41 [init.py:244] Automatically detected platform cuda.
2026-02-04 10:25:43.891 | INFO | lightx2v.common.ops.attn.flash_attn::13 - flash_attn_varlen_func_v3 not found, please install flash_attn3 first
2026-02-04 10:25:43.928 | INFO | lightx2v.common.ops.attn.sage_attn::27 - sageattn3 not found, please install sageattention first
INFO: Could not find files for the given pattern(s).
2026-02-04 10:25:44.279 | WARNING | lightx2v.utils.quant_utils::7 - qtorch not found, please install qtorch.Please install qtorch (pip install qtorch).
2026-02-04 10:25:44.332 | INFO | lightx2v.models.networks.hunyuan_video.infer.attn_no_pad::21 - flash_attn_varlen_func_v3 not available
2026-02-04 10:25:44.332 | INFO | lightx2v.models.networks.hunyuan_video.infer.attn_no_pad::39 - sageattn3 not found, please install sageattention first
2026-02-04 10:25:44.875 | INFO | lightx2v.models.runners.wan.wan_animate_runner::14 - If you want to run animate model, please install decord.
2026-02-04 10:25:45.093 | INFO | main:main:1881 - 正在加载 Hugging Face 模型列表缓存...
2026-02-04 10:25:45.094 | INFO | main:load_hf_models_cache:152 - Loading models from Hugging Face lightx2v/Wan2.1-Distill-Models...
2026-02-04 10:25:46.337 | INFO | main:load_hf_models_cache:160 - Successfully loaded models from Hugging Face lightx2v/Wan2.1-Distill-Models
2026-02-04 10:25:46.338 | INFO | main:load_hf_models_cache:176 - Loaded 10 models from Hugging Face lightx2v/Wan2.1-Distill-Models
2026-02-04 10:25:46.339 | INFO | main:load_hf_models_cache:152 - Loading models from Hugging Face lightx2v/Wan2.1-Official-Models...
2026-02-04 10:25:47.530 | INFO | main:load_hf_models_cache:160 - Successfully loaded models from Hugging Face lightx2v/Wan2.1-Official-Models
2026-02-04 10:25:47.532 | INFO | main:load_hf_models_cache:176 - Loaded 0 models from Hugging Face lightx2v/Wan2.1-Official-Models
2026-02-04 10:25:47.532 | INFO | main:load_hf_models_cache:152 - Loading models from Hugging Face lightx2v/Wan2.2-Distill-Models...
2026-02-04 10:25:48.804 | INFO | main:load_hf_models_cache:160 - Successfully loaded models from Hugging Face lightx2v/Wan2.2-Distill-Models
2026-02-04 10:25:48.805 | INFO | main:load_hf_models_cache:176 - Loaded 13 models from Hugging Face lightx2v/Wan2.2-Distill-Models
2026-02-04 10:25:48.806 | INFO | main:load_hf_models_cache:152 - Loading models from Hugging Face lightx2v/Wan2.2-Official-Models...
2026-02-04 10:25:50.002 | INFO | main:load_hf_models_cache:160 - Successfully loaded models from Hugging Face lightx2v/Wan2.2-Official-Models
2026-02-04 10:25:50.003 | INFO | main:load_hf_models_cache:176 - Loaded 3 models from Hugging Face lightx2v/Wan2.2-Official-Models
2026-02-04 10:25:50.005 | INFO | main:load_hf_models_cache:152 - Loading models from Hugging Face lightx2v/Encoders...
2026-02-04 10:25:51.187 | INFO | main:load_hf_models_cache:160 - Successfully loaded models from Hugging Face lightx2v/Encoders
2026-02-04 10:25:51.188 | INFO | main:load_hf_models_cache:176 - Loaded 7 models from Hugging Face lightx2v/Encoders
2026-02-04 10:25:51.189 | INFO | main:load_hf_models_cache:152 - Loading models from Hugging Face lightx2v/Autoencoders...
2026-02-04 10:25:52.361 | INFO | main:load_hf_models_cache:160 - Successfully loaded models from Hugging Face lightx2v/Autoencoders
2026-02-04 10:25:52.362 | INFO | main:load_hf_models_cache:176 - Loaded 8 models from Hugging Face lightx2v/Autoencoders
2026-02-04 10:25:52.362 | INFO | main:main:1883 - 模型列表缓存加载完成
- Running on local URL: http://127.0.0.1:8032
- Running on public URL: https://f2801391f9db5cce70.gradio.live
This share link expires in 1 week. For free permanent hosting and GPU upgrades, run gradio deploy from the terminal in the working directory to deploy to Hugging Face Spaces (https://huggingface.co/spaces)
2026-02-04 10:26:40.767 | INFO | main:download_model_from_ms:766 - 开始从 ModelScope lightx2v/Wan2.1-Distill-Models 下载 wan2.1_i2v_480p_int8_lightx2v_4step.safetensors 到 D:\AI\gradio_app_v2\models/
2026-02-04 10:26:40.768 | INFO | main:download_model_from_ms:821 - 检测到 wan2.1_i2v_480p_int8_lightx2v_4step.safetensors 是文件，使用 snapshot_download
Downloading Model from https://www.modelscope.cn to directory: D:\AI\gradio_app_v2\models/.temp_wan2.1_i2v_480p_int8_lightx2v_4step.safetensors\lightx2v\Wan2.1-Distill-Models
2026-02-04 10:26:42,608 - modelscope - INFO - Got 1 files, start to download ...
Downloading [wan2.1_i2v_480p_int8_lightx2v_4step.safetensors]: 100%|██████████████| 16.0G/16.0G [13:32<00:00, 21.1MB/s]
Processing 1 items: 100%|████████████████████████████████████████████████████████████| 1.00/1.00 [13:32<00:00, 812s/it]
2026-02-04 10:40:14,893 - modelscope - INFO - Download model 'lightx2v/Wan2.1-Distill-Models' successfully.
2026-02-04 10:40:14,893 - modelscope - INFO - Creating symbolic link [D:\AI\gradio_app_v2\models/.temp_wan2.1_i2v_480p_int8_lightx2v_4step.safetensors\lightx2v\Wan2.1-Distill-Models].
2026-02-04 10:40:14,895 - modelscope - WARNING - Failed to create symbolic link D:\AI\gradio_app_v2\models/.temp_wan2.1_i2v_480p_int8_lightx2v_4step.safetensors\lightx2v\Wan2.1-Distill-Models for D:\AI\gradio_app_v2\models.temp_wan2.1_i2v_480p_int8_lightx2v_4step.safetensors\lightx2v\Wan2___1-Distill-Models.
2026-02-04 10:40:14.903 | INFO | main:download_model_from_ms:852 - 文件 wan2.1_i2v_480p_int8_lightx2v_4step.safetensors 下载完成，保存到 D:\AI\gradio_app_v2\models/wan2.1_i2v_480p_int8_lightx2v_4step.safetensors
2026-02-04 10:48:57.643 | INFO | main:run_inference:1260 - 自动确定 model_cls: wan2.1_distill (模型类型: wan2.1)
2026-02-04 10:48:57.644 | INFO | main:run_inference:1268 - 自动检测量化精度 - DIT: int8, T5: fp8, CLIP: fp8
2026-02-04 10:48:57.645 | INFO | main:run_inference:1372 - VAE 配置 - use_tae: False, use_lightvae: False, need_scaled: False (VAE编码器: Wan2.1_VAE.safetensors, VAE解码器: Wan2.1_VAE.safetensors, vae_path: D:\AI\gradio_app_v2\models/Wan2.1_VAE.safetensors, tae_path: None)
2026-02-04 10:48:57.646 | INFO | main:run_inference:1458 - 使用模型: D:\AI\gradio_app_v2\models/
2026-02-04 10:48:57.647 | INFO | main:run_inference:1459 - 推理配置:
{
"do_mm_calib": false,
"cpu_offload": true,
"max_area": false,
"vae_stride": [
4,
8,
8
],
"patch_size": [
1,
2,
2
],
"feature_caching": "NoCaching",
"teacache_thresh": 0.26,
"use_ret_steps": false,
"use_bfloat16": true,
"lora_configs": null,
"use_prompt_enhancer": false,
"parallel": false,
"seq_parallel": false,
"cfg_parallel": false,
"enable_cfg": false,
"use_image_encoder": true,
"model_cls": "wan2.1_distill",
"seed": 1620317221,
"task": "i2v",
"model_path": "D:\AI\gradio_app_v2\models/",
"prompt_enhancer": null,
"prompt": "两个男人在对话",
"negative_prompt": "镜头晃动，色调艳丽，过曝，静态，细节模糊不清，字幕，风格，作品，画作，画面，静止，整体发灰，最差质量，低质量，JPEG压缩残留，丑陋的，残缺的，多余的手指，画得不好的手部，画得不好的脸部，畸形的，毁容的，形态畸形的肢体，手指融合，静止不动的画面，杂乱的背景，三条腿，背景人很多，倒着走",
"image_path": "C:\Users\Administrator\AppData\Local\Temp\gradio\dd8e7e27c6479cc540801c77e5b6caeb9569887403da64dc505744e327a08fb7\test.jpg",
"save_result_path": "D:\AI\gradio_app_v2\outputs/20260204_104857.mp4",
"return_result_tensor": false,
"_class_name": "WanModel",
"_diffusers_version": "0.33.0",
"dim": 5120,
"eps": 1e-06,
"ffn_dim": 13824,
"freq_dim": 256,
"in_dim": 36,
"num_heads": 40,
"num_layers": 40,
"out_dim": 16,
"text_len": 512,
"infer_steps": 4,
"target_video_length": 81,
"resolution": "480p",
"resize_mode": "adaptive",
"self_attn_1_type": "sage_attn2",
"cross_attn_1_type": "sage_attn2",
"cross_attn_2_type": "sage_attn2",
"sample_guide_scale": 1,
"sample_shift": 5,
"fps": 16,
"parallel_attn_type": null,
"parallel_vae": false,
"lora_path": null,
"strength_model": 1.0,
"denoising_step_list": [
1000,
750,
500,
250
],
"offload_granularity": "phase",
"t5_cpu_offload": true,
"clip_cpu_offload": true,
"vae_cpu_offload": true,
"dit_quantized": true,
"dit_quant_scheme": "int8-q8f",
"dit_quantized_ckpt": "D:\AI\gradio_app_v2\models/wan2.1_i2v_480p_int8_lightx2v_4step.safetensors",
"dit_original_ckpt": null,
"high_noise_quantized_ckpt": null,
"low_noise_quantized_ckpt": null,
"high_noise_original_ckpt": null,
"low_noise_original_ckpt": null,
"t5_original_ckpt": null,
"t5_quantized": true,
"t5_quantized_ckpt": "D:\AI\gradio_app_v2\models/models_t5_umt5-xxl-enc-fp8.safetensors",
"t5_quant_scheme": "fp8-q8f",
"clip_original_ckpt": null,
"clip_quantized": true,
"clip_quantized_ckpt": "D:\AI\gradio_app_v2\models/models_clip_open-clip-xlm-roberta-large-vit-huge-14-fp8.safetensors",
"clip_quant_scheme": "fp8-q8f",
"vae_path": "D:\AI\gradio_app_v2\models/Wan2.1_VAE.safetensors",
"tae_path": null,
"use_tiling_vae": true,
"use_tae": false,
"use_lightvae": false,
"need_scaled": false,
"lazy_load": true,
"rope_chunk": false,
"rope_chunk_size": 100,
"clean_cuda_cache": false,
"unload_modules": true,
"warm_up_cpu_buffers": false,
"boundary_step_index": 2,
"boundary": 0.9,
"rope_type": "torch"
}
2026-02-04 10:48:57.667 | INFO | lightx2v.models.runners.default_runner:init_modules:65 - Initializing runner modules...
2026-02-04 10:48:57.693 | INFO | lightx2v.models.runners.default_runner:resize_image:45 - resize_image: torch.Size([1, 3, 1280, 720]) -> torch.Size([1, 3, 832, 480]), target_h: 832, target_w: 480
2026-02-04 10:48:57.694 | INFO | lightx2v.models.runners.default_runner:read_image_input:248 - resize_image target_h: 832, target_w: 480
2026-02-04 10:48:57.694 | INFO | lightx2v.models.runners.default_runner:read_image_input:260 - target_h: 832, target_w: 480, latent_h: 104, latent_w: 60
2026-02-04 10:48:57.823 | INFO | lightx2v.utils.utils:load_weights:388 - Loading weights from D:\AI\gradio_app_v2\models/models_clip_open-clip-xlm-roberta-large-vit-huge-14-fp8.safetensors
2026-02-04 10:48:59.815 | INFO | lightx2v.utils.profiler:exit:57 - [Profile] Single GPU - Level1_Log Run Image Encoder cost 2.112862 seconds
2026-02-04 10:48:59.869 | INFO | lightx2v.utils.utils:load_weights:388 - Loading weights from D:\AI\gradio_app_v2\models/Wan2.1_VAE.safetensors
2026-02-04 10:49:18.341 | INFO | lightx2v.utils.profiler:exit:57 - [Profile] Single GPU - Level1_Log Run VAE Encoder cost 18.525159 seconds
2026-02-04 10:50:40.629 | INFO | lightx2v.utils.utils:load_weights:388 - Loading weights from D:\AI\gradio_app_v2\models/models_t5_umt5-xxl-enc-fp8.safetensors
2026-02-04 10:50:52.393 | INFO | lightx2v.utils.profiler:exit:57 - [Profile] Single GPU - Level1_Log Run Text Encoder cost 94.050481 seconds
2026-02-04 10:50:52.644 | INFO | lightx2v.utils.profiler:exit:57 - [Profile] Single GPU - Level2_Log Run Encoders cost 114.975879 seconds
2026-02-04 10:50:52.645 | INFO | lightx2v.utils.custom_compiler:_discover_compiled_methods:120 - [Compile] Discovering compiled methods for WanDistillModel...
2026-02-04 10:50:52.645 | INFO | lightx2v.utils.custom_compiler:_discover_compiled_methods:125 - [Compile] Found compiled method: _infer_cond_uncond
2026-02-04 10:50:52.654 | INFO | lightx2v.models.networks.wan.model:_load_quant_ckpt:240 - Loading weights from D:\AI\gradio_app_v2\models/wan2.1_i2v_480p_int8_lightx2v_4step.safetensors
2026-02-04 10:50:55.701 | INFO | lightx2v.models.runners.default_runner:run_main:357 - 🔄 start segment 1/1
2026-02-04 10:50:55.702 | INFO | lightx2v.models.runners.default_runner:run_segment:177 - ==> step_index: 1 / 4
2026-02-04 10:50:55.703 | INFO | lightx2v.utils.profiler:exit:57 - [Profile] Single GPU - Level1_Log step_pre cost 0.001329 seconds