Testing streaming outputs

Various version of this PR are for testing streaming implementations of the HuggingFace model parsers

cookbooks/Gradio/hf_model_parsers.py

@@ -0,0 +1,33 @@
+from aiconfig_extension_hugging_face import (
+    HuggingFaceAutomaticSpeechRecognitionTransformer,
+    HuggingFaceImage2TextTransformer,
+    HuggingFaceTextSummarizationTransformer,
+    HuggingFaceText2ImageDiffusor,
+    HuggingFaceText2SpeechTransformer,
+    HuggingFaceTextGenerationTransformer,
+    HuggingFaceTextTranslationTransformer,
+)
+from aiconfig import (AIConfigRuntime, ModelParserRegistry)
+
+def register_model_parsers() -> None:
+    """Register model parsers for HuggingFace models.
+    """
+    # Audio --> Text
+    # AIConfigRuntime.register_model_parser(HuggingFaceAutomaticSpeechRecognitionTransformer(), "AutomaticSpeechRecognition")
+
+    # # Image --> Text
+    # AIConfigRuntime.register_model_parser(HuggingFaceImage2TextTransformer(), "Image2Text")
+
+    # # Text --> Image
+    # AIConfigRuntime.register_model_parser(HuggingFaceText2ImageDiffusor(), "Text2Image")
+
+    # # Text --> Audio
+    # AIConfigRuntime.register_model_parser(HuggingFaceText2SpeechTransformer(), "Text2Speech")
+
+    # # Text --> Text
+    # AIConfigRuntime.register_model_parser(HuggingFaceTextGenerationTransformer(), "TextGeneration")
+    # AIConfigRuntime.register_model_parser(HuggingFaceTextSummarizationTransformer(), "TextSummarization")
+    ModelParserRegistry.register_model_parser(HuggingFaceText2SpeechTransformer())
+    ModelParserRegistry.register_model_parser(HuggingFaceTextGenerationTransformer())
+    ModelParserRegistry.register_model_parser(HuggingFaceImage2TextTransformer())
+    # AIConfigRuntime.register_model_parser(HuggingFaceTextTranslationTransformer(), "TextTranslation")

cookbooks/Gradio/huggingface.aiconfig.json

@@ -1,19 +1,73 @@
 {
-  "name": "",
+  "name": "The Tale of the Quick Brown Fox",
   "schema_version": "latest",
   "metadata": {
     "parameters": {},
-    "models": {}
+    "models": {
+      "Salesforce/blip-image-captioning-base": {}
+    },
+    "default_model": "Salesforce/blip-image-captioning-base",
+    "model_parsers": {
+      "Salesforce/blip-image-captioning-base": "HuggingFaceImage2TextTransformer"
+    }
   },
-  "description": "",
+  "description": "The Tale of the Quick Brown Fox",
   "prompts": [
     {
-      "name": "prompt_1",
-      "input": "",
+      "name": "prompt_2",
+      "input": {
+        "attachments": [
+          {
+            "data": "/Users/rossdancraig/Downloads/fox_in_forest.png",
+            "mime_type": "image/png"
+          },
+          {
+            "data": "/Users/rossdancraig/Downloads/trex.png",
+            "mime_type": "image/png"
+          }
+        ]
+      },
       "metadata": {
-        "model": "gpt-4",
+        "model": {
+          "name": "Salesforce/blip-image-captioning-base",
+          "settings": {
+            "max_new_tokens": 4
+          }
+        },
         "parameters": {}
       },
+      "outputs": [
+        {
+          "output_type": "execute_result",
+          "execution_count": 0,
+          "data": "a red fox in",
+          "metadata": {}
+        },
+        {
+          "output_type": "execute_result",
+          "execution_count": 1,
+          "data": "a dinosaur in the",
+          "metadata": {}
+        }
+      ]
+    },
+    {
+      "name": "Generate a caption based on this image",
+      "input": "Once upon a time, in the booming metropolis of Santa Cruz",
+      "metadata": {
+        "model": {
+          "name": "Salesforce/blip-image-captioning-base",
+          "settings": {
+            "max_length": 100,
+            "min_length": 50,
+            "num_beams": 1,
+            "stream": false
+          }
+        },
+        "parameters": {
+          "city": "New York"
+        }
+      },
       "outputs": []
     }
   ],

python/src/aiconfig/editor/server/server.py

@@ -293,9 +293,9 @@ def kill_thread(thread_id: int | None):
 
             # Yea I know time.sleep() isn't super accurate, but it's fine,
             # we can fix later
-            time.sleep(0.1)
+            # time.sleep(0.1)
             wait_time_in_seconds += SLEEP_DELAY_SECONDS
-            print(f"Output queue is currently empty. Waiting for {wait_time_in_seconds:.1f}s...")
+            # print(f"Output queue is currently empty. Waiting for {wait_time_in_seconds:.1f}s...")
 
         # Yield in flask is weird and you either need to send responses as a
         # string, or artificially wrap them around "[" and "]"