Fix OOM error in gr.load (#12928)

* Fix OOM error in gr.load

* add changeset

* Fix flaky test

* Serial

* Fix client tests

* Add code

* Back out js/css/ load from hub

---------

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

Author: freddyaboulton

.changeset/busy-heads-bet.md

@@ -0,0 +1,5 @@
+---
+"gradio": patch
+---
+
+fix:Fix OOM error in gr.load

client/python/test/test_client.py

@@ -282,34 +282,26 @@ def test_raises_exception(self, calculator_demo):
     def test_job_output_video(self, video_component):
         with connect(video_component) as client:
             job = client.submit(
-                {
-                    "video": handle_file(
-                        "https://huggingface.co/spaces/gradio/video_component/resolve/main/files/a.mp4"
-                    )
-                },
-                fn_index=0,
+                handle_file(
+                    "https://huggingface.co/datasets/freddyaboulton/bucket/resolve/main/ProgressNotifications.mp4"
+                ),
+                api_name="/predict",
             )
-            assert Path(job.result()["video"]).exists()
+            assert Path(job.result()).exists()
             assert (
-                Path(DEFAULT_TEMP_DIR).resolve()
-                in Path(job.result()["video"]).resolve().parents
+                Path(DEFAULT_TEMP_DIR).resolve() in Path(job.result()).resolve().parents
             )
 
         temp_dir = tempfile.mkdtemp()
         with connect(video_component, download_files=temp_dir) as client:
             job = client.submit(
-                {
-                    "video": handle_file(
-                        "https://huggingface.co/spaces/gradio/video_component/resolve/main/files/a.mp4"
-                    )
-                },
+                handle_file(
+                    "https://huggingface.co/spaces/gradio/video_component/resolve/main/files/a.mp4"
+                ),
                 fn_index=0,
             )
-            assert Path(job.result()["video"]).exists()
-            assert (
-                Path(temp_dir).resolve()
-                in Path(job.result()["video"]).resolve().parents
-            )
+            assert Path(job.result()).exists()
+            assert Path(temp_dir).resolve() in Path(job.result()).resolve().parents
 
     def test_progress_updates(self, progress_demo):
         with connect(progress_demo) as client:
@@ -1047,23 +1039,10 @@ def test_upload(self):
             with patch("builtins.open", MagicMock()):
                 with patch.object(pathlib.Path, "name") as mock_name:
                     mock_name.side_effect = lambda x: x
-                    results = client.endpoints[0]._upload(
-                        ["pre1", ["pre2", "pre3", "pre4"], ["pre5", "pre6"], "pre7"]
+                    results = client.endpoints[0]._upload_file(
+                        handle_file(__file__), data_index=0
                     )
-
-        res = []
-        for re in results:
-            if isinstance(re, list):
-                res.append([r["name"] for r in re])
-            else:
-                res.append(re["name"])
-
-        assert res == [
-            "file1",
-            ["file2", "file3", "file4"],
-            ["file5", "file6"],
-            "file7",
-        ]
+        assert results["path"] == "file1"
 
     @pytest.mark.flaky
     def test_download_private_file(self, gradio_temp_dir):

gradio/blocks.py

@@ -1261,7 +1261,7 @@ def iterate_over_children(children_list):
 
         derived_fields = ["types"]
 
-        with Blocks() as blocks:
+        with Blocks(theme=config.get("theme", None)) as blocks:
             # ID 0 should be the root Blocks component
             original_mapping[0] = root_block = Context.root_block or blocks
 

gradio/external.py

@@ -501,11 +501,13 @@ def query_huggingface_inference_endpoints(*data):
 
     kwargs = dict(interface_info, **kwargs)
 
-    fn = kwargs.pop("fn", None)
+    interface_fn = kwargs.pop("fn", None)
     inputs = kwargs.pop("inputs", None)
     outputs = kwargs.pop("outputs", None)
 
-    interface = gr.Interface(fn, inputs, outputs, **kwargs, api_name="predict")
+    interface = gr.Interface(
+        interface_fn, inputs, outputs, **kwargs, api_name="predict"
+    )
     return interface
 
 

gradio/external_utils.py

@@ -326,8 +326,13 @@ def handle_hf_error(e: Exception):
         raise TooManyRequestsError() from e
     elif "401" in str(e) or "You must provide an api_key" in str(e):
         raise Error("Unauthorized, please make sure you are signed in.") from e
+    elif isinstance(e, StopIteration):
+        raise Error(
+            "This model is not supported by any Hugging Face Inference Provider. "
+            "Please check the supported models at https://huggingface.co/docs/inference-providers."
+        ) from e
     else:
-        raise Error(str(e)) from e
+        raise Error(str(e) or f"An error occurred: {type(e).__name__}") from e
 
 
 def create_endpoint_fn(

test/test_external.py

@@ -22,8 +22,8 @@
 included in a separate file because of the above-mentioned dependency.
 """
 
-# Mark the whole module as flaky
-pytestmark = pytest.mark.flaky
+# Mark the whole module as flaky and serial
+pytestmark = [pytest.mark.flaky, pytest.mark.serial]
 
 os.environ["GRADIO_ANALYTICS_ENABLED"] = "False"
 
@@ -72,7 +72,7 @@ def test_text_generation(self):
     def test_summarization(self):
         model_type = "summarization"
         interface = gr.load(
-            "models/facebook/bart-large-cnn", hf_token=HF_TOKEN, alias=model_type
+            "models/facebook/bart-large-cnn", token=HF_TOKEN, alias=model_type
         )
         assert interface.__name__ == model_type
         assert interface.input_components and interface.output_components
@@ -82,7 +82,7 @@ def test_summarization(self):
     def test_translation(self):
         model_type = "translation"
         interface = gr.load(
-            "models/facebook/bart-large-cnn", hf_token=HF_TOKEN, alias=model_type
+            "models/facebook/bart-large-cnn", token=HF_TOKEN, alias=model_type
         )
         assert interface.__name__ == model_type
         assert interface.input_components and interface.output_components
@@ -93,7 +93,7 @@ def test_text_classification(self):
         model_type = "text-classification"
         interface = gr.load(
             "models/distilbert-base-uncased-finetuned-sst-2-english",
-            hf_token=HF_TOKEN,
+            token=HF_TOKEN,
             alias=model_type,
         )
         assert interface.__name__ == model_type
@@ -104,7 +104,7 @@ def test_text_classification(self):
     def test_fill_mask(self):
         model_type = "fill-mask"
         interface = gr.load(
-            "models/bert-base-uncased", hf_token=HF_TOKEN, alias=model_type
+            "models/bert-base-uncased", token=HF_TOKEN, alias=model_type
         )
         assert interface.__name__ == model_type
         assert interface.input_components and interface.output_components
@@ -114,7 +114,7 @@ def test_fill_mask(self):
     def test_zero_shot_classification(self):
         model_type = "zero-shot-classification"
         interface = gr.load(
-            "models/facebook/bart-large-mnli", hf_token=HF_TOKEN, alias=model_type
+            "models/facebook/bart-large-mnli", token=HF_TOKEN, alias=model_type
         )
         assert interface.__name__ == model_type
         assert interface.input_components and interface.output_components
@@ -126,7 +126,7 @@ def test_zero_shot_classification(self):
     def test_automatic_speech_recognition(self):
         model_type = "automatic-speech-recognition"
         interface = gr.load(
-            "models/facebook/wav2vec2-base-960h", hf_token=HF_TOKEN, alias=model_type
+            "models/facebook/wav2vec2-base-960h", token=HF_TOKEN, alias=model_type
         )
         assert interface.__name__ == model_type
         assert interface.input_components and interface.output_components
@@ -136,7 +136,7 @@ def test_automatic_speech_recognition(self):
     def test_image_classification(self):
         model_type = "image-classification"
         interface = gr.load(
-            "models/google/vit-base-patch16-224", hf_token=HF_TOKEN, alias=model_type
+            "models/google/vit-base-patch16-224", token=HF_TOKEN, alias=model_type
         )
         assert interface.__name__ == model_type
         assert interface.input_components and interface.output_components
@@ -147,7 +147,7 @@ def test_feature_extraction(self):
         model_type = "feature-extraction"
         interface = gr.load(
             "models/sentence-transformers/distilbert-base-nli-mean-tokens",
-            hf_token=HF_TOKEN,
+            token=HF_TOKEN,
             alias=model_type,
         )
         assert interface.__name__ == model_type
@@ -159,7 +159,7 @@ def test_sentence_similarity(self):
         model_type = "text-to-speech"
         interface = gr.load(
             "models/julien-c/ljspeech_tts_train_tacotron2_raw_phn_tacotron_g2p_en_no_space_train",
-            hf_token=HF_TOKEN,
+            token=HF_TOKEN,
             alias=model_type,
         )
         assert interface.__name__ == model_type
@@ -171,7 +171,7 @@ def test_text_to_speech(self):
         model_type = "text-to-speech"
         interface = gr.load(
             "models/julien-c/ljspeech_tts_train_tacotron2_raw_phn_tacotron_g2p_en_no_space_train",
-            hf_token=HF_TOKEN,
+            token=HF_TOKEN,
             alias=model_type,
         )
         assert interface.__name__ == model_type
@@ -187,7 +187,7 @@ def test_multiple_spaces_one_private(self):
         with gr.Blocks():
             gr.load(
                 "spaces/gradio-tests/not-actually-private-spacev4-sse",
-                hf_token=HF_TOKEN,
+                token=HF_TOKEN,
             )
             gr.load(
                 "spaces/gradio/test-loading-examplesv4-sse",
@@ -197,12 +197,12 @@ def test_multiple_spaces_one_private(self):
     def test_private_space_v4_sse_v1(self):
         io = gr.load(
             "spaces/gradio-tests/not-actually-private-spacev4-sse-v1",
-            hf_token=HF_TOKEN,
+            token=HF_TOKEN,
         )
         try:
             output = io("abc")
             assert output == "abc"
-            assert io.theme.name == "gradio/monochrome"
+            assert io._deprecated_theme == "gradio/monochrome"
         except TooManyRequestsError:
             pass
 
@@ -227,7 +227,7 @@ def test_interface_load_cache_examples(self, tmp_path):
                     name="models/google/vit-base-patch16-224",
                     examples=[Path(test_file_dir, "cheetah1.jpg")],
                     cache_examples=True,
-                    hf_token=HF_TOKEN,
+                    token=HF_TOKEN,
                 )
             except TooManyRequestsError:
                 pass
@@ -302,8 +302,8 @@ def check_dataframe(config):
         c for c in config["components"] if c["props"].get("label", "") == "Input Rows"
     )
     assert input_df["props"]["headers"] == ["a", "b"]
-    assert input_df["props"]["row_count"] == (1, "dynamic")
-    assert input_df["props"]["col_count"] == (2, "fixed")
+    assert input_df["props"]["row_count"] == [3, "dynamic"]
+    assert input_df["props"]["col_count"] == [2, "dynamic"]
 
 
 def check_dataset(config, readme_examples):
@@ -352,6 +352,8 @@ def test_use_api_name_in_call_method():
 
 
 def test_load_custom_component():
+    pytest.skip("Custom components not supported yet")
+
     from gradio_pdf import PDF  # noqa
 
     demo = gr.load("spaces/freddyaboulton/gradiopdf")
@@ -363,7 +365,7 @@ def test_load_custom_component():
 
 def test_load_inside_blocks():
     demo = gr.load("spaces/abidlabs/en2fr")
-    output = demo("Hello")
+    output = demo("Hello", api_name="predict")
     assert isinstance(output, str)