commit

PaddlePaddle · Sep 20, 2024 · 2aba4ec · 2aba4ec
1 parent e2e2879
commit 2aba4ec
Show file tree

Hide file tree

Showing 2 changed files with 6 additions and 9 deletions.
diff --git a/ppdiffusers/deploy/sd3/README.md b/ppdiffusers/deploy/sd3/README.md
@@ -8,14 +8,11 @@ python -m pip install triton
 python -m pip install git+https://github.com/zhoutianzi666/UseTritonInPaddle.git
 python -c "import use_triton_in_paddle; use_triton_in_paddle.make_triton_compatible_with_paddle()"
 
-# 安装develop版本的paddle
+# 安装develop版本的paddle，请根据自己的cuda版本选择对应的paddle版本，这里选择12.3的cuda版本
 python -m pip install --pre paddlepaddle-gpu -i https://www.paddlepaddle.org.cn/packages/nightly/cu123/
 
-# 指定Tensor-RT的lib路径
-export LD_LIBRARY_PATH=/your_TensorRT_dir//lib:$LD_LIBRARY_PATH
-
-# 指定cutlass包路径
-export LD_LIBRARY_PATH=/your_dir/Paddle/paddle/phi/kernels/fusion/cutlass/conv2d/build:$LD_LIBRARY_PATH
+# 指定 libCutlassGemmEpilogue.so 的路径
+# 详情请参考 https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/phi/kernels/fusion/cutlass/gemm_epilogue/README.md
 export LD_LIBRARY_PATH=/your_dir/Paddle/paddle/phi/kernels/fusion/cutlass/gemm_epilogue/build:$LD_LIBRARY_PATH
 ```
 

diff --git a/ppdiffusers/ppdiffusers/pipelines/stable_diffusion_3/pipeline_stable_diffusion_3.py b/ppdiffusers/ppdiffusers/pipelines/stable_diffusion_3/pipeline_stable_diffusion_3.py
@@ -802,7 +802,7 @@ def __call__(
                 # broadcast to batch dimension in a way that's compatible with ONNX/Core ML
                 timestep = t.expand(latent_model_input.shape[0])
 
-                noise_pred_out = self.transformer(
+                model_output = self.transformer(
                     hidden_states=latent_model_input,
                     timestep=timestep,
                     encoder_hidden_states=prompt_embeds,
@@ -812,9 +812,9 @@ def __call__(
                 )
 
                 if is_inference_mode(self.transformer):
-                    noise_pred = noise_pred_out
+                    noise_pred = model_output
                 else:
-                    noise_pred = noise_pred_out[0]
+                    noise_pred = model_output[0]
 
                 # perform guidance
                 if self.do_classifier_free_guidance: