daniel/h3-blackwell-runtime
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Compute upstream VAE temporal overlap constants

Browse source
This commit is contained in:
Daniel Maddern 2026-08-14 13:04:58 +07:00
parent 81714ed50f
commit b73d08b231
1 changed files with 4 additions and 2 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

6
src/h3_blackwell_runtime/vae_decoder.py
View file

@ -162,8 +162,10 @@ class MiniMaxH3VideoVAE(nn.Module):
super().__init__()
self.vae_ratio, self.vae_ratio_t = 16, 4
self.clip_length, self.token_drop = 17, 3
self.tokens_chunk_size, self.token_overlap = 5, 3
self.frame_pre_padding, self.frame_overlap = 3, 9
self.frame_pre_padding = (-self.clip_length) % self.vae_ratio_t
self.tokens_chunk_size = math.ceil(self.clip_length / self.vae_ratio_t)
self.token_overlap = (-self.token_drop) % self.tokens_chunk_size
self.frame_overlap = max(self.token_overlap * self.vae_ratio_t - self.frame_pre_padding, 0)
self.tiling, self.tile_size, self.tile_overlap_min = tiling, 256, 64
self.post_quant_conv = nn.Conv3d(24, 24, 1, device=device)
self.decoder = ViT3DDecoder(device=device)