[diffusion] feat: support torch compile for diffusers backend (#19673)

docs/diffusion/api/cli.md

@@ -286,6 +286,8 @@ SGLang diffusion supports a **diffusers backend** that allows you to run any dif
 | `--vae-slicing` | flag | Enable VAE slicing for lower memory usage (decodes slice-by-slice). |
 | `--dit-precision` | `fp16`, `bf16`, `fp32` | Precision for the diffusion transformer. |
 | `--vae-precision` | `fp16`, `bf16`, `fp32` | Precision for the VAE. |
+| `--enable-torch-compile` | flag | Enable `torch.compile` for diffusers pipelines. |
+| `--cache-dit-config` | `{PATH}` | Path to a Cache-DiT YAML/JSON config file for accelerating diffusers pipelines with Cache-DiT. |
 
 ### Example: Running Ovis-Image-7B
 

python/sglang/multimodal_gen/runtime/managers/scheduler.py

@@ -5,7 +5,6 @@ import asyncio
 import os
 import pickle
 from collections import deque
-from copy import deepcopy
 from typing import Any, List
 
 import zmq
@@ -233,7 +232,7 @@ class Scheduler:
                         height=height,
                         prompt="",
                     )
-                req.set_as_warmup()
+                req.set_as_warmup(self.server_args.warmup_steps)
                 self.waiting_queue.append((None, req))
             # if server is warmed-up, set this flag to avoid req-based warmup
             self.warmed_up = True
@@ -253,8 +252,7 @@ class Scheduler:
         # only the very first req through server's lifetime will be warmed up
         identity, req = recv_reqs[0]
         if isinstance(req, Req):
-            warmup_req = deepcopy(req)
+            warmup_req = req.copy_as_warmup(self.server_args.warmup_steps)
-            warmup_req.set_as_warmup()
             recv_reqs.insert(0, (identity, warmup_req))
             self._warmup_total = 1
             self._warmup_processed = 0

python/sglang/multimodal_gen/runtime/pipelines/diffusers_pipeline.py

@@ -378,6 +378,7 @@ class DiffusersPipeline(ComposedPipelineBase):
         self.memory_usages: dict[str, float] = {}
         self.post_init_called = False
         self.executor = executor or SyncExecutor(server_args=server_args)
+        self._cache_dit_enabled = False
 
         logger.info("Loading diffusers pipeline from %s", model_path)
         self.diffusers_pipe = self._load_diffusers_pipeline(model_path, server_args)
@@ -462,6 +463,8 @@ class DiffusersPipeline(ComposedPipelineBase):
         self._apply_attention_backend(pipe, server_args)
         # Apply cache-dit acceleration if configured
         pipe = self._apply_cache_dit(pipe, server_args)
+        # Apply torch.compile if enabled and supported
+        pipe = self._apply_torch_compile(pipe, server_args)
         logger.info("Loaded diffusers pipeline: %s", pipe.__class__.__name__)
         return pipe
 
@@ -562,6 +565,58 @@ class DiffusersPipeline(ComposedPipelineBase):
             raise
 
         logger.info("Enabled cache-dit for diffusers pipeline")
+        self._cache_dit_enabled = True
+        return pipe
+
+    def _apply_torch_compile(self, pipe: Any, server_args: ServerArgs) -> Any:
+        """Apply torch.compile to the pipeline if configured and supported."""
+        if not server_args.enable_torch_compile:
+            return pipe
+
+        # check if the pipeline has 'transformer' or 'unet' components which are
+        # typically the most expensive parts to compile. 'transformer_2' for some
+        # video pipelines, e.g, Wan 2.2 series, also check for that.
+        compilable_components = ["transformer", "transformer_2", "unet"]
+        if not any(hasattr(pipe, comp) for comp in compilable_components):
+            logger.warning(
+                "Pipeline does not have 'transformer' or 'unet' components. "
+                "torch.compile may not provide significant benefits and could increase latency."
+            )
+            return pipe
+
+        if self._cache_dit_enabled:
+            try:
+                import cache_dit
+
+                if hasattr(cache_dit, "set_compile_configs"):
+                    cache_dit.set_compile_configs()
+            except Exception as e:
+                logger.warning(
+                    f"Failed to set torch_compile configs for cache-dit: {e}"
+                )
+
+        for comp in compilable_components:
+            if hasattr(pipe, comp):
+                try:
+                    component = getattr(pipe, comp)
+                    # TODO(DefTruth): Add support for 'compile_repeated_blocks' for 'transformer'
+                    # modules which can significantly reduce compilation time for large models
+                    # with repeated blocks.
+                    if isinstance(component, torch.nn.Module) and hasattr(
+                        component, "compile"
+                    ):
+                        # Prefer in-place compilation if supported. According to PyTorch documentation:
+                        # https://docs.pytorch.org/docs/stable/generated/torch.compile.html
+                        component.compile()
+                    else:
+                        compiled_component = torch.compile(component)
+                        setattr(pipe, comp, compiled_component)
+                    logger.info(
+                        f"Applied torch.compile to {comp} component of the pipeline"
+                    )
+                except Exception as e:
+                    logger.warning(f"Failed to apply torch.compile to {comp}: {e}")
+
         return pipe
 
     def _get_dtype(self, server_args: ServerArgs) -> torch.dtype:

python/sglang/multimodal_gen/runtime/pipelines_core/schedule_batch.py

@@ -13,6 +13,7 @@ from __future__ import annotations
 
 import os
 import pprint
+from copy import deepcopy
 from dataclasses import MISSING, asdict, dataclass, field, fields
 from typing import Any, Optional
 
@@ -251,12 +252,17 @@ class Req:
             return None
         return os.path.join(self.output_path, output_file_name)
 
-    def set_as_warmup(self):
+    def set_as_warmup(self, warmup_steps: int = 1):
         self.is_warmup = True
         self.save_output = False
         self.suppress_logs = True
         self.extra["cache_dit_num_inference_steps"] = self.num_inference_steps
-        self.num_inference_steps = 1
+        self.num_inference_steps = warmup_steps
+
+    def copy_as_warmup(self, warmup_steps: int = 1) -> "Req":
+        req = deepcopy(self)
+        req.set_as_warmup(warmup_steps)
+        return req
 
     def validate(self):
         """Initialize dependent fields after dataclass initialization."""
@@ -270,9 +276,6 @@ class Req:
 
         self.metrics = RequestMetrics(request_id=self.request_id)
 
-        if self.is_warmup:
-            self.set_as_warmup()
-
     def adjust_size(self, server_args: ServerArgs):
         pass
 

python/sglang/multimodal_gen/runtime/server_args.py

@@ -274,6 +274,7 @@ class ServerArgs:
     # warmup
     warmup: bool = False
     warmup_resolutions: list[str] = None
+    warmup_steps: int = 1
 
     disable_autocast: bool | None = None
 
@@ -755,6 +756,12 @@ class ServerArgs:
             default=ServerArgs.warmup_resolutions,
             help="Specify resolutions for server to warmup. e.g., `--warmup-resolutions 256x256, 720x720`",
         )
+        parser.add_argument(
+            "--warmup-steps",
+            type=int,
+            default=ServerArgs.warmup_steps,
+            help="The number of warmup steps to perform for each resolution.",
+        )
 
         parser.add_argument(
             "--dit-cpu-offload",