JLNTV_IPD/CogVideo
4
0
Fork 0
mirror of https://github.com/THUDM/CogVideo.git synced 2025-09-21 22:00:00 +08:00
Code Issues Packages Projects Releases Wiki Activity

perf: cast VAE and text encoder to target dtype before precomputing cache

Browse Source 
Before precomputing the latent cache and text embeddings, cast the VAE and
text encoder to the target training dtype (fp16/bf16) instead of keeping them
in fp32. This reduces memory usage during the precomputation phase.

The change occurs in prepare_dataset() where the models are moved to device
and cast to weight_dtype before being used to generate the cache.
This commit is contained in:
OleehyO 2025-01-08 01:38:13 +00:00
parent 0e21d41b12
commit 10de04fc08
1 changed files with 4 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

6
finetune/trainer.py
View File

@ -185,10 +185,12 @@ class Trainer:
raise ValueError(f"Invalid model type: {self.args.model_type}")
# Prepare VAE and text encoder for encoding
self.components.vae = self.components.vae.to(self.accelerator.device)
self.components.vae.requires_grad_(False)
self.components.text_encoder = self.components.text_encoder.to(self.accelerator.device)
self.components.text_encoder.requires_grad_(False)
self.components.vae = self.components.vae.to(self.accelerator.device, dtype=self.state.weight_dtype)
self.components.text_encoder = self.components.text_encoder.to(
self.accelerator.device, dtype=self.state.weight_dtype
)
# Precompute latent for video and prompt embedding
logger.info("Precomputing latent for video and prompt embedding ...")