Fix bug in lora training && Update zero3 cpu offload && Teacache thresholds (#209)

examples/wan2.1/post_infer_queue.py

@@ -103,8 +103,13 @@ if __name__ == '__main__':
         
     # Support TeaCache.
     enable_teacache     = True
-    # Recommended to be set between 0.05 and 0.20. A larger threshold can cache more steps, speeding up the inference process, 
+    # Recommended to be set between 0.05 and 0.30. A larger threshold can cache more steps, speeding up the inference process, 
     # but it may cause slight differences between the generated content and the original content.
+    # # --------------------------------------------------------------------------------------------------- #
+    # | Model Name          | threshold | Model Name          | threshold | Model Name          | threshold |
+    # | Wan2.1-T2V-1.3B     | 0.05~0.10 | Wan2.1-T2V-14B      | 0.10~0.15 | Wan2.1-I2V-14B-720P | 0.20~0.30 |
+    # | Wan2.1-I2V-14B-480P | 0.20~0.25 | Wan2.1-Fun-*-1.3B-* | 0.05~0.10 | Wan2.1-Fun-*-14B-*  | 0.20~0.30 |
+    # # --------------------------------------------------------------------------------------------------- #
     teacache_threshold  = 0.10
     # The number of steps to skip TeaCache at the beginning of the inference process, which can
     # reduce the impact of TeaCache on generated video quality.

examples/wan2.1/post_infer_queue_i2v.py

@@ -119,8 +119,13 @@ if __name__ == '__main__':
         
     # Support TeaCache.
     enable_teacache     = True
-    # Recommended to be set between 0.05 and 0.20. A larger threshold can cache more steps, speeding up the inference process, 
+    # Recommended to be set between 0.05 and 0.30. A larger threshold can cache more steps, speeding up the inference process, 
     # but it may cause slight differences between the generated content and the original content.
+    # # --------------------------------------------------------------------------------------------------- #
+    # | Model Name          | threshold | Model Name          | threshold | Model Name          | threshold |
+    # | Wan2.1-T2V-1.3B     | 0.05~0.10 | Wan2.1-T2V-14B      | 0.10~0.15 | Wan2.1-I2V-14B-720P | 0.20~0.30 |
+    # | Wan2.1-I2V-14B-480P | 0.20~0.25 | Wan2.1-Fun-*-1.3B-* | 0.05~0.10 | Wan2.1-Fun-*-14B-*  | 0.20~0.30 |
+    # # --------------------------------------------------------------------------------------------------- #
     teacache_threshold  = 0.10
     # The number of steps to skip TeaCache at the beginning of the inference process, which can
     # reduce the impact of TeaCache on generated video quality.

examples/wan2.1/predict_i2v.py

@@ -55,8 +55,13 @@ compile_dit         = False
 
 # Support TeaCache.
 enable_teacache     = True
-# Recommended to be set between 0.05 and 0.20. A larger threshold can cache more steps, speeding up the inference process, 
+# Recommended to be set between 0.05 and 0.30. A larger threshold can cache more steps, speeding up the inference process, 
 # but it may cause slight differences between the generated content and the original content.
+# # --------------------------------------------------------------------------------------------------- #
+# | Model Name          | threshold | Model Name          | threshold | Model Name          | threshold |
+# | Wan2.1-T2V-1.3B     | 0.05~0.10 | Wan2.1-T2V-14B      | 0.10~0.15 | Wan2.1-I2V-14B-720P | 0.20~0.30 |
+# | Wan2.1-I2V-14B-480P | 0.20~0.25 | Wan2.1-Fun-*-1.3B-* | 0.05~0.10 | Wan2.1-Fun-*-14B-*  | 0.20~0.30 |
+# # --------------------------------------------------------------------------------------------------- #
 teacache_threshold  = 0.10
 # The number of steps to skip TeaCache at the beginning of the inference process, which can
 # reduce the impact of TeaCache on generated video quality.

examples/wan2.1/predict_t2v.py

@@ -54,8 +54,13 @@ compile_dit         = False
 
 # TeaCache config
 enable_teacache     = True
-# Recommended to be set between 0.05 and 0.20. A larger threshold can cache more steps, speeding up the inference process, 
+# Recommended to be set between 0.05 and 0.30. A larger threshold can cache more steps, speeding up the inference process, 
 # but it may cause slight differences between the generated content and the original content.
+# # --------------------------------------------------------------------------------------------------- #
+# | Model Name          | threshold | Model Name          | threshold | Model Name          | threshold |
+# | Wan2.1-T2V-1.3B     | 0.05~0.10 | Wan2.1-T2V-14B      | 0.10~0.15 | Wan2.1-I2V-14B-720P | 0.20~0.30 |
+# | Wan2.1-I2V-14B-480P | 0.20~0.25 | Wan2.1-Fun-*-1.3B-* | 0.05~0.10 | Wan2.1-Fun-*-14B-*  | 0.20~0.30 |
+# # --------------------------------------------------------------------------------------------------- #
 teacache_threshold  = 0.10
 # The number of steps to skip TeaCache at the beginning of the inference process, which can
 # reduce the impact of TeaCache on generated video quality.

examples/wan2.1_fun/post_infer_queue.py

@@ -103,8 +103,13 @@ if __name__ == '__main__':
         
     # Support TeaCache.
     enable_teacache     = True
-    # Recommended to be set between 0.05 and 0.20. A larger threshold can cache more steps, speeding up the inference process, 
+    # Recommended to be set between 0.05 and 0.30. A larger threshold can cache more steps, speeding up the inference process, 
     # but it may cause slight differences between the generated content and the original content.
+    # # --------------------------------------------------------------------------------------------------- #
+    # | Model Name          | threshold | Model Name          | threshold | Model Name          | threshold |
+    # | Wan2.1-T2V-1.3B     | 0.05~0.10 | Wan2.1-T2V-14B      | 0.10~0.15 | Wan2.1-I2V-14B-720P | 0.20~0.30 |
+    # | Wan2.1-I2V-14B-480P | 0.20~0.25 | Wan2.1-Fun-*-1.3B-* | 0.05~0.10 | Wan2.1-Fun-*-14B-*  | 0.20~0.30 |
+    # # --------------------------------------------------------------------------------------------------- #
     teacache_threshold  = 0.10
     # The number of steps to skip TeaCache at the beginning of the inference process, which can
     # reduce the impact of TeaCache on generated video quality.

examples/wan2.1_fun/post_infer_queue_i2v.py

@@ -120,8 +120,13 @@ if __name__ == '__main__':
         
     # Support TeaCache.
     enable_teacache     = True
-    # Recommended to be set between 0.05 and 0.20. A larger threshold can cache more steps, speeding up the inference process, 
+    # Recommended to be set between 0.05 and 0.30. A larger threshold can cache more steps, speeding up the inference process, 
     # but it may cause slight differences between the generated content and the original content.
+    # # --------------------------------------------------------------------------------------------------- #
+    # | Model Name          | threshold | Model Name          | threshold | Model Name          | threshold |
+    # | Wan2.1-T2V-1.3B     | 0.05~0.10 | Wan2.1-T2V-14B      | 0.10~0.15 | Wan2.1-I2V-14B-720P | 0.20~0.30 |
+    # | Wan2.1-I2V-14B-480P | 0.20~0.25 | Wan2.1-Fun-*-1.3B-* | 0.05~0.10 | Wan2.1-Fun-*-14B-*  | 0.20~0.30 |
+    # # --------------------------------------------------------------------------------------------------- #
     teacache_threshold  = 0.10
     # The number of steps to skip TeaCache at the beginning of the inference process, which can
     # reduce the impact of TeaCache on generated video quality.

examples/wan2.1_fun/post_infer_queue_v2v_control.py

@@ -134,8 +134,13 @@ if __name__ == '__main__':
         
     # Support TeaCache.
     enable_teacache     = True
-    # Recommended to be set between 0.05 and 0.20. A larger threshold can cache more steps, speeding up the inference process, 
+    # Recommended to be set between 0.05 and 0.30. A larger threshold can cache more steps, speeding up the inference process, 
     # but it may cause slight differences between the generated content and the original content.
+    # # --------------------------------------------------------------------------------------------------- #
+    # | Model Name          | threshold | Model Name          | threshold | Model Name          | threshold |
+    # | Wan2.1-T2V-1.3B     | 0.05~0.10 | Wan2.1-T2V-14B      | 0.10~0.15 | Wan2.1-I2V-14B-720P | 0.20~0.30 |
+    # | Wan2.1-I2V-14B-480P | 0.20~0.25 | Wan2.1-Fun-*-1.3B-* | 0.05~0.10 | Wan2.1-Fun-*-14B-*  | 0.20~0.30 |
+    # # --------------------------------------------------------------------------------------------------- #
     teacache_threshold  = 0.10
     # The number of steps to skip TeaCache at the beginning of the inference process, which can
     # reduce the impact of TeaCache on generated video quality.

examples/wan2.1_fun/predict_i2v.py

@@ -55,8 +55,13 @@ compile_dit         = False
 
 # Support TeaCache.
 enable_teacache     = True
-# Recommended to be set between 0.05 and 0.20. A larger threshold can cache more steps, speeding up the inference process, 
+# Recommended to be set between 0.05 and 0.30. A larger threshold can cache more steps, speeding up the inference process, 
 # but it may cause slight differences between the generated content and the original content.
+# # --------------------------------------------------------------------------------------------------- #
+# | Model Name          | threshold | Model Name          | threshold | Model Name          | threshold |
+# | Wan2.1-T2V-1.3B     | 0.05~0.10 | Wan2.1-T2V-14B      | 0.10~0.15 | Wan2.1-I2V-14B-720P | 0.20~0.30 |
+# | Wan2.1-I2V-14B-480P | 0.20~0.25 | Wan2.1-Fun-*-1.3B-* | 0.05~0.10 | Wan2.1-Fun-*-14B-*  | 0.20~0.30 |
+# # --------------------------------------------------------------------------------------------------- #
 teacache_threshold  = 0.10
 # The number of steps to skip TeaCache at the beginning of the inference process, which can
 # reduce the impact of TeaCache on generated video quality.

examples/wan2.1_fun/predict_t2v.py

@@ -55,8 +55,13 @@ compile_dit         = False
 
 # Support TeaCache.
 enable_teacache     = True
-# Recommended to be set between 0.05 and 0.20. A larger threshold can cache more steps, speeding up the inference process, 
+# Recommended to be set between 0.05 and 0.30. A larger threshold can cache more steps, speeding up the inference process, 
 # but it may cause slight differences between the generated content and the original content.
+# # --------------------------------------------------------------------------------------------------- #
+# | Model Name          | threshold | Model Name          | threshold | Model Name          | threshold |
+# | Wan2.1-T2V-1.3B     | 0.05~0.10 | Wan2.1-T2V-14B      | 0.10~0.15 | Wan2.1-I2V-14B-720P | 0.20~0.30 |
+# | Wan2.1-I2V-14B-480P | 0.20~0.25 | Wan2.1-Fun-*-1.3B-* | 0.05~0.10 | Wan2.1-Fun-*-14B-*  | 0.20~0.30 |
+# # --------------------------------------------------------------------------------------------------- #
 teacache_threshold  = 0.10
 # The number of steps to skip TeaCache at the beginning of the inference process, which can
 # reduce the impact of TeaCache on generated video quality.

examples/wan2.1_fun/predict_v2v_control.py

@@ -58,8 +58,13 @@ compile_dit         = False
 
 # Support TeaCache.
 enable_teacache     = True
-# Recommended to be set between 0.05 and 0.20. A larger threshold can cache more steps, speeding up the inference process, 
+# Recommended to be set between 0.05 and 0.30. A larger threshold can cache more steps, speeding up the inference process, 
 # but it may cause slight differences between the generated content and the original content.
+# # --------------------------------------------------------------------------------------------------- #
+# | Model Name          | threshold | Model Name          | threshold | Model Name          | threshold |
+# | Wan2.1-T2V-1.3B     | 0.05~0.10 | Wan2.1-T2V-14B      | 0.10~0.15 | Wan2.1-I2V-14B-720P | 0.20~0.30 |
+# | Wan2.1-I2V-14B-480P | 0.20~0.25 | Wan2.1-Fun-*-1.3B-* | 0.05~0.10 | Wan2.1-Fun-*-14B-*  | 0.20~0.30 |
+# # --------------------------------------------------------------------------------------------------- #
 teacache_threshold  = 0.10
 # The number of steps to skip TeaCache at the beginning of the inference process, which can
 # reduce the impact of TeaCache on generated video quality.

examples/wan2.1_fun/predict_v2v_control_camera.py

@@ -58,8 +58,13 @@ compile_dit         = False
 
 # Support TeaCache.
 enable_teacache     = True
-# Recommended to be set between 0.05 and 0.20. A larger threshold can cache more steps, speeding up the inference process, 
+# Recommended to be set between 0.05 and 0.30. A larger threshold can cache more steps, speeding up the inference process, 
 # but it may cause slight differences between the generated content and the original content.
+# # --------------------------------------------------------------------------------------------------- #
+# | Model Name          | threshold | Model Name          | threshold | Model Name          | threshold |
+# | Wan2.1-T2V-1.3B     | 0.05~0.10 | Wan2.1-T2V-14B      | 0.10~0.15 | Wan2.1-I2V-14B-720P | 0.20~0.30 |
+# | Wan2.1-I2V-14B-480P | 0.20~0.25 | Wan2.1-Fun-*-1.3B-* | 0.05~0.10 | Wan2.1-Fun-*-14B-*  | 0.20~0.30 |
+# # --------------------------------------------------------------------------------------------------- #
 teacache_threshold  = 0.10
 # The number of steps to skip TeaCache at the beginning of the inference process, which can
 # reduce the impact of TeaCache on generated video quality.

examples/wan2.1_fun/predict_v2v_control_ref.py

@@ -58,8 +58,13 @@ compile_dit         = False
 
 # Support TeaCache.
 enable_teacache     = True
-# Recommended to be set between 0.05 and 0.20. A larger threshold can cache more steps, speeding up the inference process, 
+# Recommended to be set between 0.05 and 0.30. A larger threshold can cache more steps, speeding up the inference process, 
 # but it may cause slight differences between the generated content and the original content.
+# # --------------------------------------------------------------------------------------------------- #
+# | Model Name          | threshold | Model Name          | threshold | Model Name          | threshold |
+# | Wan2.1-T2V-1.3B     | 0.05~0.10 | Wan2.1-T2V-14B      | 0.10~0.15 | Wan2.1-I2V-14B-720P | 0.20~0.30 |
+# | Wan2.1-I2V-14B-480P | 0.20~0.25 | Wan2.1-Fun-*-1.3B-* | 0.05~0.10 | Wan2.1-Fun-*-14B-*  | 0.20~0.30 |
+# # --------------------------------------------------------------------------------------------------- #
 teacache_threshold  = 0.10
 # The number of steps to skip TeaCache at the beginning of the inference process, which can
 # reduce the impact of TeaCache on generated video quality.