Add the multiple model support for file type (#96)

Author: lingtonglu

app/api/schemas/openai.py

@@ -17,15 +17,21 @@ class OpenAIAudioModel(BaseModel):
     format: str
 
 
+class OpenAIContentFileModel(BaseModel):
+    filename: str | None = None
+    file_data: str | None = None
+    file_id: str | None = None
+
 class OpenAIContentModel(BaseModel):
-    type: str  # One of: "text", "image_url", "input_audio"
+    type: str  # One of: "text", "image_url", "input_audio", "file"
     text: str | None = None
     image_url: OpenAIContentImageUrlModel | None = None
     input_audio: OpenAIAudioModel | None = None
+    file: OpenAIContentFileModel | None = None
 
     def __init__(self, **data: Any):
         super().__init__(**data)
-        if self.type not in ["text", "image_url", "input_audio"]:
+        if self.type not in ["text", "image_url", "input_audio", "file"]:
             error_message = f"Invalid type: {self.type}. Must be one of: text, image_url, input_audio"
             logger.error(error_message)
             raise InvalidCompletionRequestException(
@@ -55,6 +61,13 @@ def __init__(self, **data: Any):
                 provider_name="openai",
                 error=ValueError(error_message)
             )
+        if self.type == "file" and self.file is None:
+            error_message = "file field must be set when type is 'file'"
+            logger.error(error_message)
+            raise InvalidCompletionRequestException(
+                provider_name="openai",
+                error=ValueError(error_message)
+            )
 
 
 # ---------------------------------------------------------------------------

app/exceptions/exceptions.py

@@ -76,15 +76,15 @@ class InvalidCompletionRequestException(BaseInvalidRequestException):
     def __init__(self, provider_name: str, error: Exception):
         self.provider_name = provider_name
         self.error = error
-        super().__init__(f"Provider {provider_name} completion request is invalid: {error}")
+        super().__init__(self.provider_name, self.error)
 
 class InvalidEmbeddingsRequestException(BaseInvalidRequestException):
     """Exception raised when a embeddings request is invalid."""
 
     def __init__(self, provider_name: str, error: Exception):
         self.provider_name = provider_name
         self.error = error
-        super().__init__(f"Provider {provider_name} embeddings request is invalid: {error}")
+        super().__init__(self.provider_name, self.error)
 
 class BaseInvalidForgeKeyException(BaseForgeException):
     """Exception raised when a Forge key is invalid."""

app/services/providers/anthropic_adapter.py

@@ -58,6 +58,48 @@ def format_anthropic_usage(usage_data: dict[str, Any], token_usage: dict[str, in
             },
         }
 
+    @staticmethod
+    async def convert_openai_file_content_to_anthropic(
+        msg: dict[str, Any], allow_url_download: bool = False
+    ) -> dict[str, Any]:
+        """Convert OpenAI file content to Anthropic file content"""
+        # Only support pdf & plain text files for now
+        file = msg["file"]
+        file_data = file.get("file_data")
+        if not file_data:
+            raise InvalidCompletionRequestException(
+                provider_name="anthropic", error=ValueError("file_data is required for file content in anthropic")
+            )
+        
+        if file_data.startswith("data:"):
+            # Extract media type and base64 data, assume it's a pdf file and return it as a base64 document
+            parts = file_data.split(",", 1)
+            media_type = parts[0].split(":")[1].split(";")[0]  # e.g., "application/pdf"
+            base64_data = parts[1]  # The actual base64 string without prefix
+            if not media_type == "application/pdf":
+                raise InvalidCompletionRequestException(
+                    provider_name="anthropic", error=ValueError("Only application/pdf files are supported for base64 file content in anthropic")
+                )
+            return {
+                "type": "document",
+                "source": {
+                    "data": base64_data,
+                    "media_type": media_type,
+                    "type": "base64",
+                }
+            }
+        else:
+            # Treat it as a plain text file
+            return {
+                "type": "document",
+                "source": {
+                    "data": file_data,
+                    "media_type": "text/plain",
+                    "type": "text",
+                }
+            }
+
+    
     @staticmethod
     async def convert_openai_image_content_to_anthropic(
         msg: dict[str, Any], allow_url_download: bool = False
@@ -154,6 +196,10 @@ async def convert_openai_content_to_anthropic(
                 result.append(
                     await AnthropicAdapter.convert_openai_image_content_to_anthropic(msg, allow_url_download=allow_url_download)
                 )
+            elif _type == "file":
+                result.append(
+                    await AnthropicAdapter.convert_openai_file_content_to_anthropic(msg, allow_url_download=allow_url_download)
+                )
             else:
                 error_message = f"{_type} is not supported"
                 logger.error(error_message)

app/utils/anthropic_converter.py

@@ -23,7 +23,6 @@
     ToolChoice,
     Usage,
 )
-from app.api.schemas.openai import ChatMessage, OpenAIContentModel
 from app.core.logger import get_logger
 
 logger = get_logger(name="anthropic_converter")