Update app.py

app.py

@@ -25,6 +25,7 @@ import spaces
 import gradio as gr
 
 import pk_workflow as pk
+from h3_dpmpp_2s_ancestral import use_dpmpp_2s_ancestral
 
 MODEL_REPO = os.environ.get("H3_MODEL_REPO", "MiniMaxAI/MiniMax-H3")
 CONDITIONER_SPACE = os.environ.get("H3_CONDITIONER", "dagloop5/qwen3vl-conditioner")
@@ -111,6 +112,7 @@ SAMPLERS = {
     "er_sde": "er_sde",
     "dpmpp_2m_sde_gpu": "dpmpp_2m_sde_gpu",
     "dpmpp_3m_sde_gpu": "dpmpp_3m_sde_gpu",
+    "dpmpp_2s_ancestral": "dpmpp_2s_ancestral",
 }
 DEFAULT_SAMPLER = "euler"
 
@@ -559,18 +561,20 @@ def _generate(
 
     started = time.time()
     with pk.use_schedule(PIPE, steps, schedule, video_shift, audio_shift, sampler_name=sampler, seed=int(seed)):
-        state = PIPE(
-            prompt_embeds=prompt_embeds.to("cuda"),
-            text_token_tags=text_token_tags,
-            image=first_frame,
-            last_image=last_frame,
-            height=height,
-            width=width,
-            num_frames=num_frames,
-            num_inference_steps=requested_steps,
-            output_type="pt",
-            generator=torch.Generator("cpu").manual_seed(int(seed)),
-        )
+        with use_dpmpp_2s_ancestral(PIPE, int(seed), enabled=(sampler == "dpmpp_2s_ancestral")):
+            state = PIPE(
+                prompt_embeds=prompt_embeds.to("cuda"),
+                text_token_tags=text_token_tags,
+                image=first_frame,
+                last_image=last_frame,
+                height=height,
+                width=width,
+                num_frames=num_frames,
+                num_inference_steps=requested_steps,
+                output_type="pt",
+                generator=torch.Generator("cpu").manual_seed(int(seed)),
+            )
+            
     denoised = time.time() - started
 
     video = state.get("videos")[0]  # (frames, 3, H, W), float in [0, 1], on the card