asynchronous inference with teacache

[ywlq]

Question 1: When using asynchronous inference with TeaCache, the output video quality is very poor. May I ask if the two features are incompatible? model_id="SkyReels-V2/SkyReels-V2-DF-1.3B-540P" python3 generate_video_df.py \ --model_id ${model_id} \ --resolution 540P \ --ar_step 5 \ --causal_block_size 5 \ --base_num_frames 97 \ --num_frames 737 \ --overlap_history 17 \ --prompt "A graceful white swan with a curved neck and delicate feathers swimming in a serene lake at dawn, its reflection perfectly mirrored in the still water as mist rises from the surface, with the swan occasionally dipping its head into the water to feed." \ --addnoise_condition 20 \ --offload \ --seed 1024 \ --teacache \ --teacache_thresh 0.1

Question 2: When I use synchronous inference, why is the video content not exactly the same when using TeaCache compared to when not using it? Furthermore, when use_ret_steps is not enabled, the difference in the video content is particularly significant.

[RoseRollZhu]

Thanks for yout question. DF model is not compatiable with teacache. You are supposed to disable teacache for all settings with DF version model.