Skip to content

Open-Sora: Democratizing Efficient Video Production for All

Python

fixed checkpoint (#92)

Frank Lee82f15b30f905
diff --git a/opensora/models/stdit/stdit3.py b/opensora/models/stdit/stdit3.py
index b7de22f..4bec074 100644
--- a/opensora/models/stdit/stdit3.py
+++ b/opensora/models/stdit/stdit3.py
@@ -164,7 +164,6 @@ class STDiT3Config(PretrainedConfig):
drop_path=0.0,
caption_channels=4096,
model_max_length=300,
- dtype=torch.float32,
qk_norm=False,
enable_flash_attn=False,
enable_layernorm_kernel=False,
@@ -185,7 +184,6 @@ class STDiT3Config(PretrainedConfig):
self.drop_path = drop_path
self.caption_channels = caption_channels
self.model_max_length = model_max_length
- self.dtype = dtype
self.qk_norm = qk_norm
self.enable_flash_attn = enable_flash_attn
self.enable_layernorm_kernel = enable_layernorm_kernel