Skip to content

Open-Sora: Democratizing Efficient Video Production for All

Python

fix typo

gxyesbbc64bcb7256
diff --git a/configs/opensora-v1-3/inference/t2v.py b/configs/opensora-v1-3/inference/t2v.py
index 8d3ee7c..1ddf535 100644
--- a/configs/opensora-v1-3/inference/t2v.py
+++ b/configs/opensora-v1-3/inference/t2v.py
@@ -12,7 +12,7 @@ dtype = "bf16"
model = dict(
type="STDiT3-XL/2",
- from_pretrained="/home/guoxinying/open_source_video_ocean_V1/OpenSora-STDiT-v4",
+ from_pretrained="hpcaitech/OpenSora-STDiT-v4",
qk_norm=True,
enable_flash_attn=True,
enable_layernorm_kernel=True,
@@ -22,7 +22,7 @@ model = dict(
)
vae = dict(
type="OpenSoraVAE_V1_3",
- from_pretrained="/home/guoxinying/open_source_video_ocean_V1/OpenSora-VAE-v1.3",
+ from_pretrained="hpcaitech/OpenSora-VAE-v1.3",
z_channels=16,
micro_batch_size=1,
micro_batch_size_2d=4,
@@ -35,7 +35,7 @@ vae = dict(
)
text_encoder = dict(
type="t5",
- from_pretrained="/mnt/jfs-hdd/sora/checkpoints/pretrained_models/t5-v1_1-xxl",
+ from_pretrained="DeepFloyd/t5-v1_1-xxl",
model_max_length=300,
)
scheduler = dict(
diff --git a/configs/opensora-v1-3/inference/v2v.py b/configs/opensora-v1-3/inference/v2v.py
index 949d91f..4fb91b8 100644
--- a/configs/opensora-v1-3/inference/v2v.py
+++ b/configs/opensora-v1-3/inference/v2v.py
@@ -18,7 +18,7 @@ use_oscillation_guidance_for_image = True
model = dict(
type="STDiT3-XL/2",
- from_pretrained="/home/guoxinying/open_source_video_ocean_V1/OpenSora-STDiT-v4",
+ from_pretrained="hpcaitech/OpenSora-STDiT-v4",
qk_norm=True,
enable_flash_attn=True,
enable_layernorm_kernel=True,
@@ -29,7 +29,7 @@ model = dict(
)
vae = dict(
type="OpenSoraVAE_V1_3",
- from_pretrained="/home/guoxinying/open_source_video_ocean_V1/OpenSora-VAE-v1.3",
+ from_pretrained="hpcaitech/OpenSora-VAE-v1.3",
z_channels=16,
micro_batch_size=1,
micro_batch_size_2d=4,
@@ -42,7 +42,7 @@ vae = dict(
)
text_encoder = dict(
type="t5",
- from_pretrained="/mnt/jfs-hdd/sora/checkpoints/pretrained_models/t5-v1_1-xxl",
+ from_pretrained="DeepFloyd/t5-v1_1-xxl",
model_max_length=300,
)
scheduler = dict(
diff --git a/configs/opensora-v1-3/misc/eval_loss.py b/configs/opensora-v1-3/misc/eval_loss.py
index 0da015e..0c2efb9 100644
--- a/configs/opensora-v1-3/misc/eval_loss.py
+++ b/configs/opensora-v1-3/misc/eval_loss.py
@@ -25,7 +25,7 @@ model = dict(
)
vae = dict(
type="OpenSoraVAE_V1_3",
- from_pretrained="/home/guoxinying/open_source_video_ocean_V1/OpenSora-VAE-v1.3",
+ from_pretrained="hpcaitech/OpenSora-VAE-v1.3",
z_channels=16,
micro_batch_size=1,
micro_batch_size_2d=4,
diff --git a/configs/opensora-v1-3/train/adapt.py b/configs/opensora-v1-3/train/adapt.py
index d8ff572..34174fc 100644
--- a/configs/opensora-v1-3/train/adapt.py
+++ b/configs/opensora-v1-3/train/adapt.py
@@ -26,22 +26,16 @@ plugin = "zero2"
# Model settings
model = dict(
type="STDiT3-XL/2",
- from_pretrained="outputs/0373-STDiT3-XL-2/epoch3-global_step36000/ema.pt",
+ from_pretrained=None,
qk_norm=True,
enable_flash_attn=True,
enable_layernorm_kernel=True,
adapt_16ch=True,
skip_temporal=True,
)
-# vae = dict(
-# type="VideoAutoencoderKL",
-# from_pretrained="PixArt-alpha/pixart_sigma_sdxlvae_T5_diffusers",
-# subfolder="vae",
-# scaling_factor=0.13025,
-# )
vae = dict(
type="OpenSoraVAE_V1_3",
- from_pretrained="/home/guoxinying/open_source_video_ocean_V1/OpenSora-VAE-v1.3",
+ from_pretrained="hpcaitech/OpenSora-VAE-v1.3",
z_channels=16,
micro_batch_size=1,
micro_batch_size_2d=4,
diff --git a/configs/opensora-v1-3/train/adapt_i2v.py b/configs/opensora-v1-3/train/adapt_i2v.py
index 4723536..80a0d8c 100644
--- a/configs/opensora-v1-3/train/adapt_i2v.py
+++ b/configs/opensora-v1-3/train/adapt_i2v.py
@@ -63,7 +63,7 @@ model = dict(
)
vae = dict(
type="OpenSoraVAE_V1_3",
- from_pretrained="/home/guoxinying/open_source_video_ocean_V1/OpenSora-VAE-v1.3",
+ from_pretrained="hpcaitech/OpenSora-VAE-v1.3",
z_channels=16,
micro_batch_size=1,
micro_batch_size_2d=4,
@@ -76,7 +76,7 @@ vae = dict(
)
text_encoder = dict(
type="t5",
- from_pretrained="pretrained_models/t5-v1_1-xxl",
+ from_pretrained="DeepFloyd/t5-v1_1-xxl",
model_max_length=300,
shardformer=True,
)
diff --git a/configs/opensora-v1-3/train/stage1.py b/configs/opensora-v1-3/train/stage1.py
index 56f3195..77674a0 100644
--- a/configs/opensora-v1-3/train/stage1.py
+++ b/configs/opensora-v1-3/train/stage1.py
@@ -33,7 +33,7 @@ model = dict(
)
vae = dict(
type="OpenSoraVAE_V1_3",
- from_pretrained="/home/guoxinying/open_source_video_ocean_V1/OpenSora-VAE-v1.3",
+ from_pretrained="hpcaitech/OpenSora-VAE-v1.3",
z_channels=16,
micro_batch_size=1,
micro_batch_size_2d=4,
diff --git a/configs/opensora-v1-3/train/stage1_hq.py b/configs/opensora-v1-3/train/stage1_hq.py
index bb06888..af56d6f 100644
--- a/configs/opensora-v1-3/train/stage1_hq.py
+++ b/configs/opensora-v1-3/train/stage1_hq.py
@@ -53,7 +53,7 @@ model = dict(
)
vae = dict(
type="OpenSoraVAE_V1_3",
- from_pretrained="/home/guoxinying/open_source_video_ocean_V1/OpenSora-VAE-v1.3",
+ from_pretrained="hpcaitech/OpenSora-VAE-v1.3",
z_channels=16,
micro_batch_size=1,
micro_batch_size_2d=4,
diff --git a/configs/opensora-v1-3/train/stage1_nk.py b/configs/opensora-v1-3/train/stage1_nk.py
index 3fed68e..600ec4b 100644
--- a/configs/opensora-v1-3/train/stage1_nk.py
+++ b/configs/opensora-v1-3/train/stage1_nk.py
@@ -24,14 +24,14 @@ plugin = "zero2"
# Model settings
model = dict(
type="STDiT3-XL/2",
- from_pretrained="outputs/0461-STDiT3-XL-2/epoch3-global_step40000/ema.pt",
+ from_pretrained=None,
qk_norm=True,
enable_flash_attn=True,
enable_layernorm_kernel=True,
)
vae = dict(
type="OpenSoraVAE_V1_3",
- from_pretrained="/home/guoxinying/open_source_video_ocean_V1/OpenSora-VAE-v1.3",
+ from_pretrained="hpcaitech/OpenSora-VAE-v1.3",
z_channels=16,
micro_batch_size=1,
micro_batch_size_2d=4,
diff --git a/configs/opensora-v1-3/train/stage1_noise.py b/configs/opensora-v1-3/train/stage1_noise.py
index 100771d..d7dd214 100644
--- a/configs/opensora-v1-3/train/stage1_noise.py
+++ b/configs/opensora-v1-3/train/stage1_noise.py
@@ -33,7 +33,7 @@ model = dict(
)
vae = dict(
type="OpenSoraVAE_V1_3",
- from_pretrained="/home/guoxinying/open_source_video_ocean_V1/OpenSora-VAE-v1.3",
+ from_pretrained="hpcaitech/OpenSora-VAE-v1.3",
z_channels=16,
micro_batch_size=1,
micro_batch_size_2d=4,
diff --git a/configs/opensora-v1-3/train/stage2.py b/configs/opensora-v1-3/train/stage2.py
index 7269439..89ec442 100644
--- a/configs/opensora-v1-3/train/stage2.py
+++ b/configs/opensora-v1-3/train/stage2.py
@@ -33,7 +33,7 @@ model = dict(
)
vae = dict(
type="OpenSoraVAE_V1_3",
- from_pretrained="/home/guoxinying/open_source_video_ocean_V1/OpenSora-VAE-v1.3",
+ from_pretrained="hpcaitech/OpenSora-VAE-v1.3",
z_channels=16,
micro_batch_size=1,
micro_batch_size_2d=4,