Updated example to fix deprication issues. (triton-inference-server#5117

)
shixiaoyu321 · Nov 29, 2022 · e78ea71 · e78ea71
1 parent 48b7063
commit e78ea71
Show file tree

Hide file tree

Showing 2 changed files with 7 additions and 7 deletions.
diff --git a/docs/examples/stable_diffusion/README.md b/docs/examples/stable_diffusion/README.md
@@ -51,7 +51,8 @@ docker run --gpus=all -it --shm-size=256m --rm -p8000:8000 -p8001:8001 -p8002:80
 ```
 # PyTorch & Transformers Lib
 pip install torch torchvision torchaudio
-pip install transformers ftfy scipy accelerate diffusers
+pip install transformers ftfy scipy accelerate
+pip install diffusers==0.9.0
 pip install transformers[onnxruntime]
 huggingface-cli login
 ```
@@ -64,7 +65,7 @@ docker run -it --gpus all -p 8888:8888 -v ${PWD}:/mount nvcr.io/nvidia/pytorch:y
 
 pip install transformers ftfy scipy
 pip install transformers[onnxruntime]
-pip install diffusers
+pip install diffusers==0.9.0
 huggingface-cli login
 cd /mount
 python export.py

diff --git a/docs/examples/stable_diffusion/model_repository/pipeline/1/model.py b/docs/examples/stable_diffusion/model_repository/pipeline/1/model.py
@@ -113,21 +113,20 @@ def execute(self, requests):
 
             for i, t in tqdm(enumerate(self.scheduler.timesteps)):
                 latent_model_input = torch.cat([latents] * 2)
-                sigma = self.scheduler.sigmas[i]
-                latent_model_input = latent_model_input / ((sigma**2 + 1)**0.5)
+                latent_model_input = self.scheduler.scale_model_input(latent_model_input, t)
 
                 with torch.no_grad(), torch.autocast("cuda"):
                     noise_pred = self.unet(
                         latent_model_input,
                         t,
-                        encoder_hidden_states=text_embeddings)["sample"]
+                        encoder_hidden_states=text_embeddings).sample
 
                 noise_pred_uncond, noise_pred_text = noise_pred.chunk(2)
                 noise_pred = noise_pred_uncond + guidance_scale * (
                     noise_pred_text - noise_pred_uncond)
 
-                latents = self.scheduler.step(noise_pred, i,
-                                              latents)["prev_sample"]
+                latents = self.scheduler.step(noise_pred, self.scheduler.timesteps[i],
+                                              latents).prev_sample
 
             # VAE decoding
             latents = 1 / 0.18215 * latents