MCPcopy Create free account
hub / github.com/Netflix/void-model / _batch_encode_vae

Function _batch_encode_vae

scripts/cogvideox_fun/train.py:1608–1617  ·  view source on GitHub ↗
(pixel_values)

Source from the content-addressed store, hash-verified

1606 with torch.no_grad():
1607 # This way is quicker when batch grows up
1608 def _batch_encode_vae(pixel_values):
1609 pixel_values = rearrange(pixel_values, "b f c h w -> b c f h w")
1610 bs = args.vae_mini_batch
1611 new_pixel_values = []
1612 for i in range(0, pixel_values.shape[0], bs):
1613 pixel_values_bs = pixel_values[i : i + bs]
1614 pixel_values_bs = vae.encode(pixel_values_bs)[0]
1615 pixel_values_bs = pixel_values_bs.sample()
1616 new_pixel_values.append(pixel_values_bs)
1617 return torch.cat(new_pixel_values, dim = 0)
1618 if vae_stream_1 is not None:
1619 vae_stream_1.wait_stream(torch.cuda.current_stream())
1620 with torch.cuda.stream(vae_stream_1):

Callers 1

mainFunction · 0.70

Calls 1

encodeMethod · 0.80

Tested by

no test coverage detected