Run pre/post processing in threadpool (#7327)

* Add code

* Add code

* Add code

* add changeset

---------

Co-authored-by: gradio-pr-bot <gradio-pr-bot@users.noreply.github.com>

.changeset/hot-taxis-jump.md

@@ -0,0 +1,5 @@
+---
+"gradio": patch
+---
+
+fix:Run pre/post processing in threadpool

gradio/blocks.py

@@ -1521,6 +1521,9 @@ def handle_streaming_diffs(
 
         return data
 
+    def run_fn_batch(self, fn, batch, fn_index, state):
+        return [fn(fn_index, list(i), state) for i in zip(*batch)]
+
     async def process_api(
         self,
         fn_index: int,
@@ -1565,10 +1568,14 @@ async def process_api(
                 raise ValueError(
                     f"Batch size ({batch_size}) exceeds the max_batch_size for this function ({max_batch_size})"
                 )
-
-            inputs = [
-                self.preprocess_data(fn_index, list(i), state) for i in zip(*inputs)
-            ]
+            inputs = await anyio.to_thread.run_sync(
+                self.run_fn_batch,
+                self.preprocess_data,
+                inputs,
+                fn_index,
+                state,
+                limiter=self.limiter,
+            )
             result = await self.call_function(
                 fn_index,
                 list(zip(*inputs)),
@@ -1579,17 +1586,24 @@ async def process_api(
                 in_event_listener,
             )
             preds = result["prediction"]
-            data = [
-                self.postprocess_data(fn_index, list(o), state) for o in zip(*preds)
-            ]
+            data = await anyio.to_thread.run_sync(
+                self.run_fn_batch,
+                self.postprocess_data,
+                preds,
+                fn_index,
+                state,
+                limiter=self.limiter,
+            )
             data = list(zip(*data))
             is_generating, iterator = None, None
         else:
             old_iterator = iterator
             if old_iterator:
                 inputs = []
             else:
-                inputs = self.preprocess_data(fn_index, inputs, state)
+                inputs = await anyio.to_thread.run_sync(
+                    self.preprocess_data, fn_index, inputs, state, limiter=self.limiter
+                )
             was_generating = old_iterator is not None
             result = await self.call_function(
                 fn_index,
@@ -1600,7 +1614,13 @@ async def process_api(
                 event_data,
                 in_event_listener,
             )
-            data = self.postprocess_data(fn_index, result["prediction"], state)
+            data = await anyio.to_thread.run_sync(
+                self.postprocess_data,
+                fn_index,  # type: ignore
+                result["prediction"],
+                state,
+                limiter=self.limiter,
+            )
             is_generating, iterator = result["is_generating"], result["iterator"]
             if is_generating or was_generating:
                 run = id(old_iterator) if was_generating else id(iterator)

gradio/components/gallery.py

@@ -2,6 +2,7 @@
 
 from __future__ import annotations
 
+from concurrent.futures import ThreadPoolExecutor
 from pathlib import Path
 from typing import Any, Callable, List, Literal, Optional, Tuple, Union
 from urllib.parse import urlparse
@@ -165,7 +166,8 @@ def postprocess(
         if value is None:
             return GalleryData(root=[])
         output = []
-        for img in value:
+
+        def _save(img):
             url = None
             caption = None
             orig_name = None
@@ -194,11 +196,14 @@ def postprocess(
                 orig_name = img.name
             else:
                 raise ValueError(f"Cannot process type as image: {type(img)}")
-            entry = GalleryImage(
+            return GalleryImage(
                 image=FileData(path=file_path, url=url, orig_name=orig_name),
                 caption=caption,
             )
-            output.append(entry)
+
+        with ThreadPoolExecutor() as executor:
+            for o in executor.map(_save, value):
+                output.append(o)
         return GalleryData(root=output)
 
     @staticmethod

gradio/processing_utils.py

@@ -135,7 +135,7 @@ def save_pil_to_cache(
     temp_dir = Path(cache_dir) / hash_bytes(bytes_data)
     temp_dir.mkdir(exist_ok=True, parents=True)
     filename = str((temp_dir / f"{name}.{format}").resolve())
-    img.save(filename, pnginfo=get_pil_metadata(img))
+    (temp_dir / f"{name}.{format}").resolve().write_bytes(bytes_data)
     return filename