Merge branch 'dev' of https://github.com/Tauffer-Consulting/openai_do…

…mino_pieces into dev

.github/workflows/tests-dev.yml

@@ -38,7 +38,7 @@ jobs:
 
       - name: Run organize and build images
         run: |
-          domino piece organize --build-images --source-url=https://github.com/${{github.repository}} --tag-overwrite=dev
+          domino piece organize --build-images --source-url=https://github.com/${{github.repository}} --tag-overwrite=development
 
       - name: Install Tests Dependencies
         run: pip install -r requirements-tests.txt

dependencies/Dockerfile_whisper

@@ -0,0 +1,20 @@
+FROM ghcr.io/tauffer-consulting/domino-base-piece:gpu
+
+
+# Install specific requirements
+ENV TZ=UTC
+ARG DEBIAN_FRONTEND=noninteractive
+RUN apt update
+RUN apt install ffmpeg -y
+RUN apt install git -y
+
+# Need to copy operators source code
+COPY config.toml domino/pieces_repository/
+COPY pieces domino/pieces_repository/pieces
+COPY .domino domino/pieces_repository/.domino
+
+# Install specific python dependencies
+RUN pip install -U openai-whisper==20231106
+
+# Dowload Whisper model
+RUN python3 -c "import whisper; whisper.load_model('tiny')"

pieces/AudioTranscriptionLocalPiece/metadata.json

@@ -0,0 +1,22 @@
+{
+    "name": "AudioTranscriptionLocalPiece",
+    "description": "Runs transcription locally using Whisper, a general-purpose speech recognition model. Ref: https://github.com/openai/whisper",
+    "dependency": {
+        "dockerfile": "Dockerfile_whisper"
+    },
+    "container_resources": {
+        "use_gpu": true,
+        "requests": {
+            "cpu": "1000m",
+            "memory": "3Gi"
+        },
+        "limits": {
+            "cpu": "5000m",
+            "memory": "15Gi"
+        }
+    },
+    "style": {
+        "node_label": "Audio Transcription Local",
+        "icon_class_name": "fa-solid:comment-dots"
+    }
+}

pieces/AudioTranscriptionLocalPiece/models.py

@@ -0,0 +1,45 @@
+from pydantic import BaseModel, Field, FilePath
+from typing import Union
+from enum import Enum
+
+
+class ModelSizeType(str, Enum):
+    tiny = "tiny"
+    base = "base"
+    small = "small"
+    medium = "medium"
+    large = "large"
+
+
+class OutputTypeType(str, Enum):
+    string = "string"
+    file = "file"
+    both = "both"
+
+
+class InputModel(BaseModel):
+    audio_file_path: str = Field(
+        description='The path to the audio file to process.',
+        json_schema_extra={
+            "from_upstream": "always"
+        }
+    )
+    output_type: OutputTypeType = Field(
+        default=OutputTypeType.string,
+        description='The type of output for the result text. Options are `string`, `file` or `both`. Default is `string`.',
+    )
+    model_size: ModelSizeType = Field(
+        description='The size of the model to use. Default is tiny.',
+        default=ModelSizeType.tiny
+    )
+
+
+class OutputModel(BaseModel):
+    transcription_result: str = Field(
+        default="",
+        description="The result transcription text as a string."
+    )
+    file_path_transcription_result: Union[FilePath, str] = Field(
+        default="",
+        description="The path to the text file with the transcription result."
+    )

pieces/AudioTranscriptionLocalPiece/piece.py

@@ -0,0 +1,54 @@
+from domino.base_piece import BasePiece
+from .models import InputModel, OutputModel
+import whisper
+
+
+class AudioTranscriptionLocalPiece(BasePiece):
+
+    def piece_function(self, input_data: InputModel):
+
+        self.logger.info("Loading model...")
+        model = whisper.load_model(input_data.model_size)
+
+        self.logger.info("Transcribing audio file...")
+        result = model.transcribe(str(input_data.audio_file_path))["text"]
+
+        if input_data.output_type == "string":
+            self.logger.info("Transcription complete successfully. Result returned as string.")
+            transcription_result = result
+            output_file_path = ""
+        elif input_data.output_type == "file":
+            self.logger.info("Transcription complete successfully. Result returned as file.")
+            transcription_result = ""
+            output_file_path = "transcription_result.txt"
+            with open(output_file_path, "w") as f:
+                f.write(result)
+        else:
+            self.logger.info("Transcription complete successfully. Result returned as string and file.")
+            transcription_result = result
+            output_file_path = "transcription_result.txt"
+            with open(output_file_path, "w") as f:
+                f.write(result)
+
+        # Display result in the Domino GUI
+        self.format_display_result(input_data=input_data, string_transcription_result=result)
+
+        return OutputModel(
+            transcription_result=transcription_result,
+            file_path_transcription_result=output_file_path
+        )
+
+    def format_display_result(self, input_data: InputModel, string_transcription_result: str):
+        md_text = f"""## Audio Transcription Result  \n
+{string_transcription_result}  \n
+
+## Args
+**model_size**: {input_data.model_size.value}\n
+"""
+        file_path = f"{self.results_path}/display_result.md"
+        with open(file_path, "w") as f:
+            f.write(md_text)
+        self.display_result = {
+            "file_type": "md",
+            "file_path": file_path
+        }

pieces/AudioTranscriptionLocalPiece/test_localtranscription_piece.py

@@ -0,0 +1,20 @@
+from domino.testing import piece_dry_run
+from pathlib import Path
+
+
+test_file = str(Path(__file__).parent / "test-audio-to-transcribe.mp3")
+
+
+def test_whisper_piece():
+    input_data = {
+        "audio_file_path": test_file,
+        "model_size": "tiny",
+        "output_type": "both"
+    }
+    piece_output = piece_dry_run(
+        piece_name="AudioTranscriptionLocalPiece",
+        input_data=input_data,
+    )
+    assert piece_output["transcription_result"]
+    assert piece_output["file_path_transcription_result"]
+    assert "audio" in piece_output.get("transcription_result", "").lower()

pieces/AudioTranscriptionPiece/metadata.json

@@ -19,6 +19,6 @@
     ],
     "style": {
         "node_label": "OpenAI Audio Transcript",
-        "icon_class_name": "fa-solid:headset"
+        "icon_class_name": "fa-solid:comment-dots"
     }
 }

pieces/AudioTranscriptionPiece/models.py

@@ -2,60 +2,44 @@
 from enum import Enum
 
 
-
 class OutputTypeType(str, Enum):
-    """
-    Output type for the result text
-    """
     file = "file"
     string = "string"
-    file_and_string = "file_and_string"
+    both = "both"
 
 
 class InputModel(BaseModel):
-    """
-    AudioTranscriptPiece input model
-    """
     audio_file_path: str = Field(
-        ...,
         description='The path to the audio file to process.',
-
+        json_schema_extra={
+            "from_upstream": "always"
+        }
     )
     output_type: OutputTypeType = Field(
         default=OutputTypeType.string,
-        description='The type of output for the result text',
-
+        description='The type of output for the result text. Options are `string`, `file` or `both`. Default is `string`.',
     )
     temperature: float = Field(
         description="What sampling temperature to use, between 0 and 1",
         default=0.1,
         gt=0.,
         le=1,
-        
+
     )
 
 
 class OutputModel(BaseModel):
-    """
-    AudioTranscriptPiece output model
-    """
-    message: str = Field(
-        description="Output message to log"
-    )
-    string_transcription_result: str = Field(
-        default=None,
+    transcription_result: str = Field(
+        default="",
         description="The result transcription text as a string."
     )
     file_path_transcription_result: str = Field(
-        default=None,
-        description="The result transcription text as a file path."
+        default="",
+        description="The path to the text file with the transcription result."
     )
 
 
 class SecretsModel(BaseModel):
-    """
-    AudioTranscriptPiece secret model
-    """
     OPENAI_API_KEY: str = Field(
         description="OpenAI API key"
-    )
+    )

pieces/AudioTranscriptionPiece/piece.py

@@ -16,7 +16,7 @@ def piece_function(self, input_data: InputModel, secrets_data: SecretsModel):
 
         # Input arguments are retrieved from the Input model object
         file_path = input_data.audio_file_path
-        
+
         print("Making OpenAI audio transcription request...")
         try:
             full_audio = AudioSegment.from_mp3(file_path)
@@ -31,9 +31,9 @@ def piece_function(self, input_data: InputModel, secrets_data: SecretsModel):
                 endpoint = min((i+1)*ten_minutes, total_time-1)
                 minutes = full_audio[i*ten_minutes:endpoint]
                 minutes.export(f"audio_piece_{i}.mp3", format="mp3")
-                audio_file= open(f"audio_piece_{i}.mp3", "rb")
+                audio_file = open(f"audio_piece_{i}.mp3", "rb")
                 transcript = client.audio.transcriptions.create(
-                    model="whisper-1", 
+                    model="whisper-1",
                     file=audio_file,
                     temperature=input_data.temperature
                 )
@@ -48,35 +48,31 @@ def piece_function(self, input_data: InputModel, secrets_data: SecretsModel):
 
         # Display result in the Domino GUI
         self.format_display_result(input_data=input_data, string_transcription_result=full_transcript)
-        
+
         if input_data.output_type == "string":
             self.logger.info("Transcription complete successfully. Result returned as string.")
-            msg = f"Transcription complete successfully. Result returned as string."
             return OutputModel(
-                message=msg,
-                string_transcription_result=full_transcript
+                transcription_result=full_transcript,
+                file_path_transcription_result=""
             )
-                
+
         output_file_path = f"{self.results_path}/audio_transcription_result.txt"
         with open(output_file_path, "w") as f:
             f.write(full_transcript)
 
         if input_data.output_type == "file":
             self.logger.info(f"Transcription complete successfully. Result returned as file in {output_file_path}")
-            msg = f"Transcription complete successfully. Result returned as file."
             return OutputModel(
-                message=msg,
+                transcription_result="",
                 file_path_transcription_result=output_file_path
             )
 
         self.logger.info(f"Transcription complete successfully. Result returned as string and file in {output_file_path}")
-        msg = f"Transcription complete successfully. Result returned as string and file."
         return OutputModel(
-            message=msg,
-            string_transcription_result=full_transcript,
+            transcription_result=full_transcript,
             file_path_transcription_result=output_file_path
         )
-    
+
     def format_display_result(self, input_data: InputModel, string_transcription_result: str):
         md_text = f"""
 ## Generated transcription:  \n
@@ -91,4 +87,4 @@ def format_display_result(self, input_data: InputModel, string_transcription_res
         self.display_result = {
             "file_type": "md",
             "file_path": file_path
-        }
+        }

pieces/ImageGeneratorPiece/models.py

@@ -34,7 +34,6 @@ class InputModel(BaseModel):
     ImageGeneratorPiece input model
     """
     prompt: str = Field(
-        ...,
         description="A text description of the desired image",
     )
     size: ImageSize = Field(
@@ -43,7 +42,7 @@ class InputModel(BaseModel):
     )
     image_format: ImageFormat = Field(
         default=ImageFormat.url,
-        description="The format in which the generated image is returned",   
+        description="The format in which the generated image is returned",
     )
     output_type: OutputTypeType = Field(
         default=OutputTypeType.string,