Skip to content

Open-Sora: Democratizing Efficient Video Production for All

Python

update config

pxy2af0f79f6dd7
diff --git a/configs/dit/inference/16x256x256.py b/configs/dit/inference/16x256x256.py
index 50c3fa5..9ed4d84 100644
--- a/configs/dit/inference/16x256x256.py
+++ b/configs/dit/inference/16x256x256.py
@@ -7,7 +7,7 @@ image_size = (256, 256)
model = dict(
type="DiT-XL/2",
condition="text",
- from_pretrained="YOUR_MODEL_PATH",
+ from_pretrained="PRETRAINED_MODEL",
)
vae = dict(
type="VideoAutoencoderKL",
diff --git a/configs/dit/inference/1x256x256-class.py b/configs/dit/inference/1x256x256-class.py
index 7e81ad4..24d1c8a 100644
--- a/configs/dit/inference/1x256x256-class.py
+++ b/configs/dit/inference/1x256x256-class.py
@@ -1,9 +1,8 @@
-# sample size
num_frames = 1
fps = 1
image_size = (256, 256)
-# model config
+# Define model
model = dict(
type="DiT-XL/2",
no_temporal_pos_emb=True,
@@ -25,7 +24,7 @@ scheduler = dict(
)
dtype = "fp16"
-# prompts
+# Others
batch_size = 2
seed = 42
prompt_path = "./assets/texts/imagenet_id.txt"
diff --git a/configs/dit/inference/1x256x256.py b/configs/dit/inference/1x256x256.py
index d99d7a7..31a5b9f 100644
--- a/configs/dit/inference/1x256x256.py
+++ b/configs/dit/inference/1x256x256.py
@@ -1,14 +1,13 @@
-# sample size
num_frames = 1
fps = 1
image_size = (256, 256)
-# model config
+# Define model
model = dict(
type="DiT-XL/2",
no_temporal_pos_emb=True,
condition="text",
- from_pretrained="YOUR_MODEL_PATH",
+ from_pretrained="PRETRAINED_MODEL",
)
vae = dict(
type="VideoAutoencoderKL",
@@ -26,7 +25,7 @@ scheduler = dict(
)
dtype = "fp16"
-# prompts
+# Others
batch_size = 2
seed = 42
prompt_path = "./assets/texts/imagenet_labels.txt"
diff --git a/configs/dit/train/16x256x256.py b/configs/dit/train/16x256x256.py
index df809e5..67d3736 100644
--- a/configs/dit/train/16x256x256.py
+++ b/configs/dit/train/16x256x256.py
@@ -1,21 +1,20 @@
-# sample size
num_frames = 16
frame_interval = 3
image_size = (256, 256)
-# dataset
+# Define dataset
root = None
-data_path = "/mnt/hdd/data/csv/ucf101_videos.csv"
+data_path = "CSV_PATH"
use_image_transform = False
num_workers = 4
-# acceleration
-dtype = "fp16"
+# Define acceleration
+dtype = "bf16"
grad_checkpoint = True
plugin = "zero2"
sp_size = 1
-# model config
+# Define model
model = dict(
type="DiT-XL/2",
from_pretrained="DiT-XL-2-256x256.pt",
@@ -36,7 +35,7 @@ scheduler = dict(
timestep_respacing="",
)
-# runtime
+# Others
seed = 42
outputs = "outputs"
wandb = False
diff --git a/configs/dit/train/1x256x256.py b/configs/dit/train/1x256x256.py
index 8cb8f9a..f8bd8d3 100644
--- a/configs/dit/train/1x256x256.py
+++ b/configs/dit/train/1x256x256.py
@@ -1,21 +1,20 @@
-# sample size
num_frames = 1
frame_interval = 1
image_size = (256, 256)
-# dataset
+# Define dataset
root = None
-data_path = "/mnt/hdd/data/csv/imagenet_train.csv"
+data_path = "CSV_PATH"
use_image_transform = True
num_workers = 4
-# acceleration
-dtype = "fp16"
+# Define acceleration
+dtype = "bf16"
grad_checkpoint = True
plugin = "zero2"
sp_size = 1
-# model config
+# Define model
model = dict(
type="DiT-XL/2",
no_temporal_pos_emb=True,
@@ -36,7 +35,7 @@ scheduler = dict(
timestep_respacing="",
)
-# runtime
+# Others
seed = 42
outputs = "outputs"
wandb = False
diff --git a/configs/latte/inference/16x256x256-class.py b/configs/latte/inference/16x256x256-class.py
index e08acb8..c46f4bc 100644
--- a/configs/latte/inference/16x256x256-class.py
+++ b/configs/latte/inference/16x256x256-class.py
@@ -1,9 +1,8 @@
-# sample size
num_frames = 16
fps = 8
image_size = (256, 256)
-# model config
+# Define model
model = dict(
type="Latte-XL/2",
condition="label_101",
@@ -24,7 +23,7 @@ scheduler = dict(
)
dtype = "fp16"
-# prompts
+# Others
batch_size = 2
seed = 42
prompt_path = "./assets/texts/ucf101_id.txt"
diff --git a/configs/latte/inference/16x256x256.py b/configs/latte/inference/16x256x256.py
index d74c995..cb50237 100644
--- a/configs/latte/inference/16x256x256.py
+++ b/configs/latte/inference/16x256x256.py
@@ -1,13 +1,12 @@
-# sample size
num_frames = 16
fps = 8
image_size = (256, 256)
-# model config
+# Define model
model = dict(
type="Latte-XL/2",
condition="text",
- from_pretrained="YOUR_MODEL_PATH",
+ from_pretrained="PRETRAINED_MODEL",
)
vae = dict(
type="VideoAutoencoderKL",
@@ -25,7 +24,7 @@ scheduler = dict(
)
dtype = "fp16"
-# prompts
+# Others
batch_size = 2
seed = 42
prompt_path = "./assets/texts/ucf101_labels.txt"
diff --git a/configs/latte/train/16x256x256.py b/configs/latte/train/16x256x256.py
index 685996d..0bf6bd4 100644
--- a/configs/latte/train/16x256x256.py
+++ b/configs/latte/train/16x256x256.py
@@ -1,21 +1,20 @@
-# sample size
num_frames = 16
frame_interval = 3
image_size = (256, 256)
-# dataset
+# Define dataset
root = None
-data_path = "/mnt/hdd/data/csv/ucf101_videos.csv"
+data_path = "CSV_PATH"
use_image_transform = False
num_workers = 4
-# acceleration
-dtype = "fp16"
+# Define acceleration
+dtype = "bf16"
grad_checkpoint = True
plugin = "zero2"
sp_size = 1
-# model config
+# Define model
model = dict(
type="Latte-XL/2",
enable_flashattn=True,
@@ -35,7 +34,7 @@ scheduler = dict(
timestep_respacing="",
)
-# runtime
+# Others
seed = 42
outputs = "outputs"
wandb = False
diff --git a/configs/pixart/inference/16x256x256.py b/configs/pixart/inference/16x256x256.py
index 11ce918..6fc8ee6 100644
--- a/configs/pixart/inference/16x256x256.py
+++ b/configs/pixart/inference/16x256x256.py
@@ -1,9 +1,8 @@
-# sample size
num_frames = 16
fps = 8
image_size = (256, 256)
-# model config
+# Define model
model = dict(
type="PixArt-XL/2",
space_scale=0.5,
@@ -26,7 +25,7 @@ scheduler = dict(
)
dtype = "fp16"
-# prompts
+# Others
batch_size = 2
seed = 42
prompt_path = "./assets/texts/t2v_samples.txt"
diff --git a/configs/pixart/inference/1x1024MS.py b/configs/pixart/inference/1x1024MS.py
index 5e864b8..41cc97a 100644
--- a/configs/pixart/inference/1x1024MS.py
+++ b/configs/pixart/inference/1x1024MS.py
@@ -1,10 +1,9 @@
-# sample size
num_frames = 1
fps = 1
image_size = (1920, 512)
multi_resolution = True
-# model config
+# Define model
model = dict(
type="PixArtMS-XL/2",
space_scale=2.0,
@@ -28,7 +27,7 @@ scheduler = dict(
)
dtype = "fp16"
-# prompts
+# Others
batch_size = 2
seed = 42
prompt_path = "./assets/texts/t2i_samples.txt"
diff --git a/configs/pixart/inference/1x256x256.py b/configs/pixart/inference/1x256x256.py
index 6dc614d..6dcbbba 100644
--- a/configs/pixart/inference/1x256x256.py
+++ b/configs/pixart/inference/1x256x256.py
@@ -1,9 +1,8 @@
-# sample size
num_frames = 1
fps = 1
image_size = (256, 256)
-# model config
+# Define model
model = dict(
type="PixArt-XL/2",
space_scale=1.0,
@@ -29,7 +28,7 @@ scheduler = dict(
)
dtype = "fp16"
-# prompts
+# Others
batch_size = 2
seed = 42
prompt_path = "./assets/texts/t2i_samples.txt"
diff --git a/configs/pixart/inference/1x512x512.py b/configs/pixart/inference/1x512x512.py
index 8c8c3fe..5674259 100644
--- a/configs/pixart/inference/1x512x512.py
+++ b/configs/pixart/inference/1x512x512.py
@@ -1,9 +1,8 @@
-# sample size
num_frames = 1
fps = 1
image_size = (512, 512)
-# model config
+# Define model
model = dict(
type="PixArt-XL/2",
space_scale=1.0,
@@ -27,7 +26,7 @@ scheduler = dict(
)
dtype = "fp16"
-# prompts
+# Others
batch_size = 2
seed = 42
prompt_path = "./assets/texts/t2i_samples.txt"
diff --git a/configs/pixart/train/16x256x256.py b/configs/pixart/train/16x256x256.py
index 4db54ff..6819573 100644
--- a/configs/pixart/train/16x256x256.py
+++ b/configs/pixart/train/16x256x256.py
@@ -1,22 +1,20 @@
-# sample size
num_frames = 16
frame_interval = 3
image_size = (256, 256)
-# dataset
+# Define dataset
root = None
-# data_path = "/mnt/hdd/data/csv/bak_00/pexels_inter4k_fmin_48_rp.csv"
-data_path = "/mnt/hdd/data/csv/ucf101_videos.csv"
+data_path = "CSV_PATH"
use_image_transform = False
num_workers = 4
-# acceleration
-dtype = "fp16"
+# Define acceleration
+dtype = "bf16"
grad_checkpoint = True
plugin = "zero2"
sp_size = 1
-# model config
+# Define model
model = dict(
type="PixArt-XL/2",
space_scale=0.5,
@@ -40,7 +38,7 @@ scheduler = dict(
timestep_respacing="",
)
-# runtime
+# Others
seed = 42
outputs = "outputs"
wandb = False
diff --git a/configs/pixart/train/1x512x512.py b/configs/pixart/train/1x512x512.py
index 56dc00a..619c9aa 100644
--- a/configs/pixart/train/1x512x512.py
+++ b/configs/pixart/train/1x512x512.py
@@ -1,21 +1,20 @@
-# sample size
num_frames = 1
frame_interval = 1
image_size = (512, 512)
-# dataset
+# Define dataset
root = None
-data_path = "/mnt/hdd/data/csv/imagenet_train.csv"
+data_path = "CSV_PATH"
use_image_transform = True
num_workers = 4
-# acceleration
-dtype = "fp16"
+# Define acceleration
+dtype = "bf16"
grad_checkpoint = True
plugin = "zero2"
sp_size = 1
-# model config
+# Define model
model = dict(
type="PixArt-XL/2",
space_scale=1.0,
@@ -40,7 +39,7 @@ scheduler = dict(
timestep_respacing="",
)
-# runtime
+# Others
seed = 42
outputs = "outputs"
wandb = False