Skip to content

Open-Sora: Democratizing Efficient Video Production for All

Python

add cache dir

Zangwei Zheng4d338419a727
diff --git a/configs/opensora-v1-1/inference/sample.py b/configs/opensora-v1-1/inference/sample.py
index f929f56..a052b30 100644
--- a/configs/opensora-v1-1/inference/sample.py
+++ b/configs/opensora-v1-1/inference/sample.py
@@ -6,7 +6,7 @@ multi_resolution = "STDiT2"
# Define model
model = dict(
type="STDiT2-XL/2",
- from_pretrained="PixArt-XL-2-1024-MS.pth",
+ from_pretrained=None,
input_sq_size=512,
enable_flashattn=True,
enable_layernorm_kernel=True,
@@ -14,11 +14,13 @@ model = dict(
vae = dict(
type="VideoAutoencoderKL",
from_pretrained="stabilityai/sd-vae-ft-ema",
+ cache_dir=None, # "/mnt/hdd/cached_models",
micro_batch_size=4,
)
text_encoder = dict(
type="t5",
from_pretrained="DeepFloyd/t5-v1_1-xxl",
+ cache_dir=None, # "/mnt/hdd/cached_models",
model_max_length=300,
)
scheduler = dict(
diff --git a/opensora/models/vae/vae.py b/opensora/models/vae/vae.py
index 79f78b1..9c79c5c 100644
--- a/opensora/models/vae/vae.py
+++ b/opensora/models/vae/vae.py
@@ -8,9 +8,9 @@ from opensora.registry import MODELS
@MODELS.register_module()
class VideoAutoencoderKL(nn.Module):
- def __init__(self, from_pretrained=None, micro_batch_size=None):
+ def __init__(self, from_pretrained=None, micro_batch_size=None, cache_dir=None):
super().__init__()
- self.module = AutoencoderKL.from_pretrained(from_pretrained)
+ self.module = AutoencoderKL.from_pretrained(from_pretrained, cache_dir=cache_dir)
self.out_channels = self.module.config.latent_channels
self.patch_size = (1, 8, 8)
self.micro_batch_size = micro_batch_size
@@ -70,9 +70,9 @@ class VideoAutoencoderKL(nn.Module):
@MODELS.register_module()
class VideoAutoencoderKLTemporalDecoder(nn.Module):
- def __init__(self, from_pretrained=None):
+ def __init__(self, from_pretrained=None, cache_dir=None):
super().__init__()
- self.module = AutoencoderKLTemporalDecoder.from_pretrained(from_pretrained)
+ self.module = AutoencoderKLTemporalDecoder.from_pretrained(from_pretrained, cache_dir=cache_dir)
self.out_channels = self.module.config.latent_channels
self.patch_size = (1, 8, 8)