Merge pull request #39 from cloudera/dev

Refactor and Path management

Author: Khauneesh-AI

app/core/config.py

@@ -1,6 +1,10 @@
 from enum import Enum
 from typing import Dict, List, Optional
 from pydantic import BaseModel
+from fastapi import FastAPI, HTTPException, Request, status
+import requests
+import json
+from fastapi.responses import JSONResponse
 
 class UseCase(str, Enum):
     CODE_GENERATION = "code_generation"
@@ -58,7 +62,11 @@ class UseCaseMetadata(BaseModel):
     budget DECIMAL(15,2)
 );
 """
-
+bedrock_list = ['us.anthropic.claude-3-5-haiku-20241022-v1:0', 'us.anthropic.claude-3-5-sonnet-20241022-v2:0',
+                              'us.anthropic.claude-3-opus-20240229-v1:0','anthropic.claude-instant-v1', 
+                              'us.meta.llama3-2-11b-instruct-v1:0','us.meta.llama3-2-90b-instruct-v1:0', 'us.meta.llama3-1-70b-instruct-v1:0', 
+                                'mistral.mixtral-8x7b-instruct-v0:1', 'mistral.mistral-large-2402-v1:0',
+                                   'mistral.mistral-small-2402-v1:0'  ]
 
 # Detailed use case configurations with examples
 USE_CASE_CONFIGS = {
@@ -198,4 +206,92 @@ def get_examples_for_topic(use_case: UseCase, topic: str) -> List[Dict[str, str]
     if not topic_metadata:
         return use_case_config.default_examples
 
-    return topic_metadata.example_questions
+    return topic_metadata.example_questions
+
+
+responses = {
+    # 4XX Client Errors
+    status.HTTP_400_BAD_REQUEST: {
+        "description": "Bad Request - Invalid input parameters",
+        "content": {
+            "application/json": {
+                "example": {
+                    "status": "failed",
+                    "error": "Invalid input: No topics provided"
+                }
+            }
+        }
+    },
+    status.HTTP_404_NOT_FOUND: {
+        "description": "Resource not found",
+        "content": {
+            "application/json": {
+                "example": {
+                    "status": "failed",
+                    "error": "Requested resource not found"
+                }
+            }
+        }
+    },
+    status.HTTP_422_UNPROCESSABLE_ENTITY: {
+        "description": "Validation Error",
+        "content": {
+            "application/json": {
+                "example": {
+                    "status": "failed",
+                    "error": "Invalid request parameters"
+                }
+            }
+        }
+    },
+
+    # 5XX Server Errors
+    status.HTTP_500_INTERNAL_SERVER_ERROR: {
+        "description": "Internal server error",
+        "content": {
+            "application/json": {
+                "example": {
+                    "status": "failed",
+                    "error": "Internal server error occurred"
+                }
+            }
+        }
+    },
+    status.HTTP_503_SERVICE_UNAVAILABLE: {
+        "description": "Service temporarily unavailable",
+        "content": {
+            "application/json": {
+                "example": {
+                    "status": "failed",
+                    "error": "The CAII endpoint is downscaled, please try after >15 minutes"
+                }
+            }
+        }
+    },
+    status.HTTP_504_GATEWAY_TIMEOUT: {
+        "description": "Request timed out",
+        "content": {
+            "application/json": {
+                "example": {
+                    "status": "failed",
+                    "error": "Operation timed out after specified seconds"
+                }
+            }
+        }
+    }
+}
+
+
+def caii_check(caii_endpoint):
+    API_KEY = json.load(open("/tmp/jwt"))["access_token"]
+    headers = {
+        "Authorization": f"Bearer {API_KEY}"
+    }
+    
+    
+    if caii_endpoint:
+        caii_endpoint = caii_endpoint.removesuffix('/chat/completions') 
+        caii_endpoint = caii_endpoint + "/models"
+        response = requests.get(caii_endpoint, headers=headers, timeout=3)  # Will raise RequestException if fails
+        
+    return response

app/core/exceptions.py

@@ -1,15 +1,64 @@
+from typing import Optional, Dict, Any
+
 class APIError(Exception):
-    def __init__(self, message: str, status_code: int = 400):
+    """Base exception for API errors"""
+    def __init__(self, message: str, status_code: int = 400, details: Optional[Dict[str, Any]] = None):
         self.message = message
         self.status_code = status_code
+        self.details = details or {}
         super().__init__(message)
-        
+
 class ModelHandlerError(APIError):
+    """Errors related to model handling"""
     pass
 
 class InvalidModelError(ModelHandlerError):
-    def __init__(self, model_id: str):
+    """Invalid model configuration or access errors
+    
+    Status code 404 (Not Found) is used because:
+    - The requested model resource cannot be found or accessed
+    - This includes cases where the model exists but cannot be accessed due to configuration
+    - Follows REST principles for resource not found scenarios
+    """
+    def __init__(self, model_id: str, error: str):
+        # Extract specific error conditions from AWS error messages
+        details = {
+            "model_id": model_id,
+            "error_type": "invalid_model_error"
+        }
+        
+        if "on-demand throughput isn't supported" in error:
+            details["error_subtype"] = "throughput_configuration_error"
+            message = (f"Model {model_id} requires an inference profile configuration. "
+                      "Please use a model with on-demand throughput support or configure an inference profile.")
+        elif "model identifier is invalid" in error:
+            details["error_subtype"] = "invalid_identifier_error"
+            message = f"Model {model_id} is not a valid model identifier."
+        else:
+            details["error_subtype"] = "general_model_error"
+            message = f"Invalid model configuration for {model_id}: {error}"
+            
+        super().__init__(
+            message=message,
+            status_code=404,
+            details=details
+        )
+
+class JSONParsingError(APIError):
+    """Specific exception for JSON parsing failures
+    
+    Status code 422 (Unprocessable Entity) is used because:
+    - The request syntax is valid (unlike 400 Bad Request)
+    - The server understands the content type (unlike 415 Unsupported Media Type)
+    - But the server cannot process the contained instructions/content
+    """
+    def __init__(self, message: str, raw_text: str, parsing_context: str = ""):
         super().__init__(
-            message=f"Invalid model identifier: {model_id}",
-            status_code=400
+            message=f"JSON parsing error: {message}",
+            status_code=422,  # Unprocessable Entity
+            details={
+                "parsing_context": parsing_context,
+                "raw_text_sample": raw_text[:500] if raw_text else "No text provided",
+                "error_type": "json_parsing_error"
+            }
         )