./scripts/hunyuan_video_15/run_hy15_i2v_720p.sh
LightX2V Base Environment Variables Summary:
lightx2v_path: /root/autodl-tmp/LightX2V
model_path: /root/autodl-tmp/HunyuanVideo-1.5/ckpts
Model Inference Data Type: BF16
Sensitive Layer Data Type: None
Performance Profiling Debug Level: 2
2025-12-02 00:20:35.624 | INFO | lightx2v.models.networks.hunyuan_video.infer.attn_no_pad::11 - flash_attn_varlen_func_v3 not available
2025-12-02 00:20:35.666 | INFO | lightx2v.models.networks.hunyuan_video.infer.attn_no_pad::29 - sageattn3 not found, please install sageattention first
2025-12-02 00:20:36.313 | INFO | lightx2v.common.ops.attn.flash_attn::15 - flash_attn_varlen_func_v3 not found, please install flash_attn3 first
2025-12-02 00:20:36.313 | INFO | lightx2v.common.ops.attn.flash_attn::21 - torch_mlu_ops not found.
2025-12-02 00:20:36.323 | INFO | lightx2v.common.ops.attn.sage_attn::26 - sageattn3 not found, please install sageattention first
2025-12-02 00:20:36.323 | INFO | lightx2v.common.ops.attn.sage_attn::33 - torch_mlu_ops not found.
2025-12-02 00:20:36.632 | WARNING | lightx2v.utils.quant_utils::7 - qtorch not found, please install qtorch.Please install qtorch (pip install qtorch).
2025-12-02 00:20:36.712 | INFO | lightx2v.utils.set_config:set_config:41 - Loading some config from /root/autodl-tmp/LightX2V/configs/hunyuan_video_15/hunyuan_video_i2v_720p.json
2025-12-02 00:20:36.712 | INFO | lightx2v.utils.set_config:print_config:115 - config:
{
"do_mm_calib": false,
"cpu_offload": false,
"max_area": false,
"vae_stride": [
4,
16,
16
],
"patch_size": [
1,
1,
1
],
"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": true,
"use_image_encoder": true,
"model_cls": "hunyuan_video_1.5",
"task": "i2v",
"model_path": "/root/autodl-tmp/HunyuanVideo-1.5/ckpts",
"sf_model_path": null,
"config_json": "/root/autodl-tmp/LightX2V/configs/hunyuan_video_15/hunyuan_video_i2v_720p.json",
"infer_steps": 50,
"transformer_model_name": "720p_i2v",
"fps": 24,
"target_video_length": 121,
"sample_shift": 7.0,
"sample_guide_scale": 6.0,
"attn_type": "sage_attn2",
"transformer_model_path": "/root/autodl-tmp/HunyuanVideo-1.5/ckpts/transformer/720p_i2v",
"_class_name": "HunyuanVideo_1_5_DiffusionTransformer",
"_diffusers_version": "0.35.0",
"attn_mode": "flash",
"attn_param": null,
"concat_condition": true,
"glyph_byT5_v2": true,
"guidance_embed": false,
"heads_num": 16,
"hidden_size": 2048,
"ideal_resolution": "720p",
"ideal_task": "i2v",
"in_channels": 32,
"is_reshape_temporal_channels": false,
"mlp_act_type": "gelu_tanh",
"mlp_width_ratio": 4,
"mm_double_blocks_depth": 54,
"mm_single_blocks_depth": 0,
"out_channels": 32,
"qk_norm": true,
"qk_norm_type": "rms",
"qkv_bias": true,
"rope_dim_list": [
16,
56,
56
],
"rope_theta": 256,
"text_pool_type": null,
"text_projection": "single_refiner",
"text_states_dim": 3584,
"text_states_dim_2": null,
"use_attention_mask": true,
"use_cond_type_embedding": true,
"use_meanflow": false,
"vision_projection": "linear",
"vision_states_dim": 1152
}
2025-12-02 00:20:36.835 | INFO | lightx2v.models.runners.default_runner:init_modules:38 - Initializing runner modules...
2025-12-02 00:20:36.835 | INFO | lightx2v.utils.custom_compiler:_discover_compiled_methods:120 - [Compile] Discovering compiled methods for HunyuanVideo15Model...
2025-12-02 00:20:36.836 | INFO | lightx2v.models.networks.hunyuan_video.model:_load_ckpt:170 - Loading weights from /root/autodl-tmp/HunyuanVideo-1.5/ckpts/transformer/720p_i2v/diffusion_pytorch_model.safetensors
2025-12-02 00:20:42.585 | INFO | lightx2v.models.runners.hunyuan_video.hunyuan_video_15_runner:load_text_encoder:81 - Loading text encoder from /root/autodl-tmp/HunyuanVideo-1.5/ckpts/text_encoder/llm
Loading checkpoint shards: 100%|████████████████████████████████████████████████████████████████████████████████████| 5/5 [00:01<00:00, 2.89it/s]
2025-12-02 00:20:51.140 | INFO | lightx2v.utils.profiler:exit:43 - [Profile] Single GPU - Level2_Log Load models cost 14.304974 seconds
2025-12-02 00:20:51.412 | INFO | lightx2v.utils.profiler:exit:43 - [Profile] Single GPU - Level1_Log Run Image Encoder cost 0.258388 seconds
2025-12-02 00:20:51.617 | INFO | lightx2v.utils.profiler:exit:43 - [Profile] Single GPU - Level1_Log Run VAE Encoder cost 0.204709 seconds
2025-12-02 00:20:51.971 | INFO | lightx2v.utils.profiler:exit:43 - [Profile] Single GPU - Level2_Log Run Encoders cost 0.829852 seconds
2025-12-02 00:20:51.978 | INFO | lightx2v.models.runners.default_runner:run_main:300 - 🔄 start segment 1/1
2025-12-02 00:20:51.979 | INFO | lightx2v.models.runners.default_runner:run_segment:150 - ==> step_index: 1 / 50
2025-12-02 00:20:51.979 | INFO | lightx2v.utils.profiler:exit:43 - [Profile] Single GPU - Level1_Log step_pre cost 0.000009 seconds
2025-12-02 00:20:52.319 | INFO | lightx2v.utils.profiler:exit:43 - [Profile] Single GPU - Level1_Log 🚀 infer_main cost 0.340484 seconds
2025-12-02 00:20:52.319 | INFO | lightx2v.utils.profiler:exit:43 - [Profile] Single GPU - Level1_Log Run Dit every step cost 0.340732 seconds
2025-12-02 00:20:52.319 | INFO | lightx2v.utils.profiler:exit:43 - [Profile] Single GPU - Level1_Log segment end2end 1/1 cost 0.340804 seconds
2025-12-02 00:20:52.319 | INFO | lightx2v.utils.profiler:exit:43 - [Profile] Single GPU - Level2_Log Run DiT cost 0.348629 seconds
2025-12-02 00:20:52.319 | INFO | lightx2v.utils.profiler:exit:43 - [Profile] Single GPU - Level1_Log RUN pipeline cost 1.178721 seconds
2025-12-02 00:20:52.319 | INFO | lightx2v.utils.profiler:exit:43 - [Profile] Single GPU - Level1_Log Total Cost cost 15.484251 seconds
Traceback (most recent call last):
File "", line 198, in _run_module_as_main
File "", line 88, in _run_code
File "/root/autodl-tmp/LightX2V/lightx2v/infer.py", line 133, in
main()
File "/root/autodl-tmp/LightX2V/lightx2v/infer.py", line 124, in main
runner.run_pipeline(input_info)
File "/root/autodl-tmp/LightX2V/lightx2v/utils/profiler.py", line 77, in sync_wrapper
return func(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^
File "/root/autodl-tmp/LightX2V/lightx2v/models/runners/default_runner.py", line 386, in run_pipeline
gen_video_final = self.run_main()
^^^^^^^^^^^^^^^
File "/root/autodl-tmp/LightX2V/lightx2v/utils/profiler.py", line 77, in sync_wrapper
return func(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^
File "/root/autodl-tmp/LightX2V/lightx2v/models/runners/default_runner.py", line 311, in run_main
latents = self.run_segment(segment_idx)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/root/autodl-tmp/LightX2V/lightx2v/utils/memory_profiler.py", line 18, in wrapper
result = func(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^
File "/root/autodl-tmp/LightX2V/lightx2v/models/runners/default_runner.py", line 156, in run_segment
self.model.infer(self.inputs)
File "/root/miniconda3/envs/lightx2v/lib/python3.12/site-packages/torch/utils/_contextlib.py", line 120, in decorate_context
return func(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^
File "/root/autodl-tmp/LightX2V/lightx2v/models/networks/hunyuan_video/model.py", line 226, in infer
noise_pred_cond = self._infer_cond_uncond(inputs, infer_condition=True)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/root/miniconda3/envs/lightx2v/lib/python3.12/site-packages/torch/utils/_contextlib.py", line 120, in decorate_context
return func(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^
File "/root/autodl-tmp/LightX2V/lightx2v/models/networks/hunyuan_video/model.py", line 250, in _infer_cond_uncond
x = self.transformer_infer.infer(self.transformer_weights, pre_infer_out)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/root/miniconda3/envs/lightx2v/lib/python3.12/site-packages/torch/utils/_contextlib.py", line 120, in decorate_context
return func(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^
File "/root/autodl-tmp/LightX2V/lightx2v/models/networks/hunyuan_video/infer/transformer_infer.py", line 126, in infer
self.infer_func(weights, infer_module_out)
File "/root/miniconda3/envs/lightx2v/lib/python3.12/site-packages/torch/utils/_contextlib.py", line 120, in decorate_context
return func(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^
File "/root/autodl-tmp/LightX2V/lightx2v/models/networks/hunyuan_video/infer/transformer_infer.py", line 133, in infer_without_offload
infer_module_out.img, infer_module_out.txt = self.infer_double_block(weights.double_blocks[i], infer_module_out)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/root/miniconda3/envs/lightx2v/lib/python3.12/site-packages/torch/utils/_contextlib.py", line 120, in decorate_context
return func(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^
File "/root/autodl-tmp/LightX2V/lightx2v/models/networks/hunyuan_video/infer/transformer_infer.py", line 146, in infer_double_block
img_q, img_k, img_v, img_branch_out = self._infer_img_branch_before_attn(weights, infer_module_out)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/root/miniconda3/envs/lightx2v/lib/python3.12/site-packages/torch/utils/_contextlib.py", line 120, in decorate_context
return func(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^
File "/root/autodl-tmp/LightX2V/lightx2v/models/networks/hunyuan_video/infer/transformer_infer.py", line 173, in _infer_img_branch_before_attn
img_q, img_k = self.apply_rope_func(img_q.unsqueeze(0), img_k.unsqueeze(0), cos_sin_cache=self.scheduler.cos_sin)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/root/autodl-tmp/LightX2V/lightx2v/models/networks/hunyuan_video/infer/transformer_infer.py", line 53, in apply_hunyuan_rope_with_flashinfer
apply_rope_with_cos_sin_cache_inplace(
TypeError: 'NoneType' object is not callable
ModelTC
