perf: pre-warm ImageToVideo SFAST pipeline

This commit ensures that the model is pretraced when SFAST is enabled
for the ImageToVideo pipeline. Without this pre-tracing the first
request will be slower than a non SFAST call.

runner/app/pipelines/image_to_image.py

@@ -92,15 +92,25 @@ def __init__(self, model_id: str):
                 model_id, **kwargs
             ).to(torch_device)
 
-        if os.environ.get("SFAST"):
+        if os.getenv("SFAST", "").strip().lower() == "true":
             logger.info(
-                "ImageToImagePipeline will be dynamicallly compiled with stable-fast for %s",
+                "ImageToImagePipeline will be dynamically compiled with stable-fast "
+                "for %s",
                 model_id,
             )
             from app.pipelines.sfast import compile_model
 
             self.ldm = compile_model(self.ldm)
 
+            # Warm-up the pipeline.
+            # TODO: Not yet supported for ImageToImagePipeline.
+            if os.getenv("SFAST_WARMUP", "true").lower() == "true":
+                logger.warning(
+                    "The 'SFAST_WARMUP' flag is not yet supported for the "
+                    "ImageToImagePipeline and will be ignored. As a result the first "
+                    "call may be slow if 'SFAST' is enabled."
+                )
+
     def __call__(self, prompt: str, image: PIL.Image, **kwargs) -> List[PIL.Image]:
         seed = kwargs.pop("seed", None)
         if seed is not None:

runner/app/pipelines/image_to_video.py

@@ -8,13 +8,16 @@
 from typing import List
 import logging
 import os
+import time
 
 from PIL import ImageFile
 
 ImageFile.LOAD_TRUNCATED_IMAGES = True
 
 logger = logging.getLogger(__name__)
 
+SFAST_WARMUP_ITERATIONS = 2  # Model warm-up iterations when SFAST is enabled.
+
 
 class ImageToVideoPipeline(Pipeline):
     def __init__(self, model_id: str):
@@ -40,17 +43,52 @@ def __init__(self, model_id: str):
         self.ldm = StableVideoDiffusionPipeline.from_pretrained(model_id, **kwargs)
         self.ldm.to(get_torch_device())
 
-        if os.environ.get("SFAST"):
+        if os.getenv("SFAST", "").strip().lower() == "true":
             logger.info(
-                "ImageToVideoPipeline will be dynamicallly compiled with stable-fast for %s",
+                "ImageToVideoPipeline will be dynamically compiled with stable-fast "
+                "for %s",
                 model_id,
             )
             from app.pipelines.sfast import compile_model
 
             self.ldm = compile_model(self.ldm)
 
+            # Warm-up the pipeline.
+            # NOTE: Initial calls may be slow due to compilation. Subsequent calls will
+            # be faster.
+            if os.getenv("SFAST_WARMUP", "true").lower() == "true":
+                # Retrieve default model params.
+                # TODO: Retrieve defaults from Pydantic class in route.
+                warmup_kwargs = {
+                    "image": PIL.Image.new("RGB", (576, 1024)),
+                    "height": 576,
+                    "width": 1024,
+                    "fps": 6,
+                    "motion_bucket_id": 127,
+                    "noise_aug_strength": 0.02,
+                    "decode_chunk_size": 25,
+                }
+
+                logger.info("Warming up ImageToVideoPipeline pipeline...")
+                total_time = 0
+                for ii in range(SFAST_WARMUP_ITERATIONS):
+                    t = time.time()
+                    try:
+                        self.ldm(**warmup_kwargs).frames
+                    except Exception as e:
+                        # FIXME: When out of memory, pipeline is corrupted.
+                        logger.error(f"ImageToVideoPipeline warmup error: {e}")
+                        raise e
+                    iteration_time = time.time() - t
+                    total_time += iteration_time
+                    logger.info(
+                        "Warmup iteration %s took %s seconds", ii + 1, iteration_time
+                    )
+                logger.info("Total warmup time: %s seconds", total_time)
+
     def __call__(self, image: PIL.Image, **kwargs) -> List[List[PIL.Image]]:
         if "decode_chunk_size" not in kwargs:
+            # Decrease decode_chunk_size to reduce memory usage.
             kwargs["decode_chunk_size"] = 4
 
         seed = kwargs.pop("seed", None)

runner/app/pipelines/text_to_image.py

@@ -14,11 +14,13 @@
 from typing import List
 import logging
 import os
+import time
 
 logger = logging.getLogger(__name__)
 
 SDXL_LIGHTNING_MODEL_ID = "ByteDance/SDXL-Lightning"
-
+WARMUP_ITERATIONS = 3  # Warm-up calls count when SFAST is enabled.
+WARMUP_BATCH_SIZE = 3  # Max batch size for warm-up calls when SFAST is enabled.
 
 class TextToImagePipeline(Pipeline):
     def __init__(self, model_id: str):
@@ -111,15 +113,25 @@ def __init__(self, model_id: str):
                 self.ldm.vae.decode, mode="max-autotune", fullgraph=True
             )
 
-        if os.environ.get("SFAST"):
+        if os.getenv("SFAST", "").strip().lower() == "true":
             logger.info(
-                "TextToImagePipeline will be dynamicallly compiled with stable-fast for %s",
+                "TextToImagePipeline will be dynamically compiled with stable-fast for "
+                "%s",
                 model_id,
             )
             from app.pipelines.sfast import compile_model
 
             self.ldm = compile_model(self.ldm)
 
+            # Warm-up the pipeline.
+            # TODO: Not yet supported for ImageToImagePipeline.
+            if os.getenv("SFAST_WARMUP", "true").lower() == "true":
+                logger.warning(
+                    "The 'SFAST_WARMUP' flag is not yet supported for the "
+                    "TextToImagePipeline and will be ignored. As a result the first "
+                    "call may be slow if 'SFAST' is enabled."
+                )
+
     def __call__(self, prompt: str, **kwargs) -> List[PIL.Image]:
         seed = kwargs.pop("seed", None)
         if seed is not None: