abstractcore 2.4.5__py3-none-any.whl → 2.4.7__py3-none-any.whl

abstractcore/providers/openai_provider.py

@@ -50,8 +50,7 @@ class OpenAIProvider(BaseProvider):
         # Preflight check: validate model exists
         self._validate_model_exists()
 
-        # Store configuration (remove duplicate max_tokens)
-        self.temperature = kwargs.get("temperature", 0.7)
+        # Store provider-specific configuration
         self.top_p = kwargs.get("top_p", 1.0)
         self.frequency_penalty = kwargs.get("frequency_penalty", 0.0)
         self.presence_penalty = kwargs.get("presence_penalty", 0.0)
@@ -125,6 +124,11 @@ class OpenAIProvider(BaseProvider):
             call_params["top_p"] = kwargs.get("top_p", self.top_p)
             call_params["frequency_penalty"] = kwargs.get("frequency_penalty", self.frequency_penalty)
             call_params["presence_penalty"] = kwargs.get("presence_penalty", self.presence_penalty)
+            
+            # Add seed if provided (OpenAI supports seed for deterministic outputs)
+            seed_value = kwargs.get("seed", self.seed)
+            if seed_value is not None:
+                call_params["seed"] = seed_value
 
         # Handle different token parameter names for different model families
         if self._uses_max_completion_tokens():
@@ -165,8 +169,14 @@ class OpenAIProvider(BaseProvider):
             if stream:
                 return self._stream_response(call_params, tools)
             else:
+                # Track generation time
+                start_time = time.time()
                 response = self.client.chat.completions.create(**call_params)
+                gen_time = round((time.time() - start_time) * 1000, 1)
+                
                 formatted = self._format_response(response)
+                # Add generation time to response
+                formatted.gen_time = gen_time
 
                 # Handle tool execution for OpenAI native responses
                 if tools and formatted.has_tool_calls():
@@ -212,13 +222,16 @@ class OpenAIProvider(BaseProvider):
                     "arguments": tc.function.arguments
                 })
 
-        # Build usage dict with detailed breakdown
+        # Build usage dict with consistent terminology
         usage = None
         if hasattr(response, 'usage'):
             usage = {
+                "input_tokens": response.usage.prompt_tokens,
+                "output_tokens": response.usage.completion_tokens,
+                "total_tokens": response.usage.total_tokens,
+                # Keep legacy keys for backward compatibility
                 "prompt_tokens": response.usage.prompt_tokens,
-                "completion_tokens": response.usage.completion_tokens,
-                "total_tokens": response.usage.total_tokens
+                "completion_tokens": response.usage.completion_tokens
             }
 
             # Add detailed token breakdown for reasoning models