Repository navigation
[tests] refactor remaining s series pipelines - #14654
Conversation
|
❌ |
|
/diffusers-bot pytest tests/pipelines/skyreels_v2/ |
|
❌ |
|
/diffusers-bot pytest tests/pipelines/stable_audio_3/ |
|
|
|
/diffusers-bot pytest tests/pipelines/stable_audio/ |
|
✅ |
|
/diffusers-bot pytest tests/pipelines/stable_audio_3/ |
|
✅ |
|
/diffusers-bot pytest tests/pipelines/skyreels_v2/ |
|
|
|
The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update. |
|
|
||
| if long_video_iter == 0: | ||
| # `video` is preprocessed in float32; the VAE may be running in another dtype (fp16/bf16). | ||
| video = video.to(self.vae.dtype) |
There was a problem hiding this comment.
Mimicking what the I2V pipeline does in prepare_latents()
| if last_image is not None: | ||
| latents = latents[:, :, :-prefix_video_latents_frames, :, :].to(self.vae.dtype) | ||
| latents = latents[:, :, :-prefix_video_latents_frames, :, :] | ||
| latents = latents.to(self.vae.dtype) |
There was a problem hiding this comment.
The latents needs to go to the vae.dtype regardless of the condition.
|
/diffusers-bot pytest tests/pipelines/skyreels_v2/ |

No description provided.