Better LLM Explanations CF-597

codeflash/api/aiservice.py

@@ -278,6 +278,76 @@ def optimize_python_code_refinement(self, request: list[AIServiceRefinerRequest]
         console.rule()
         return []
 
+    def get_new_explanation(  # noqa: D417
+        self,
+        source_code: str,
+        optimized_code: str,
+        dependency_code: str,
+        trace_id: str,
+        original_line_profiler_results: str,
+        optimized_line_profiler_results: str,
+        original_code_runtime: str,
+        optimized_code_runtime: str,
+        speedup: str,
+        annotated_tests: str,
+        optimization_id: str,
+    ) -> str:
+        """Optimize the given python code for performance by making a request to the Django endpoint.
+
+        Parameters
+        ----------
+        - source_code (str): The python code to optimize.
+        - optimized_code (str): The python code generated by the AI service.
+        - dependency_code (str): The dependency code used as read-only context for the optimization
+        - original_line_profiler_results: str - line profiler results for the baseline code
+        - optimized_line_profiler_results: str - line profiler results for the optimized code
+        - original_code_runtime: str - runtime for the baseline code
+        - optimized_code_runtime: str - runtime for the optimized code
+        - speedup: str - speedup of the optimized code
+        - annotated_tests: str - test functions annotated with runtime
+        - optimization_id: str - unique id of opt candidate
+
+        Returns
+        -------
+        - List[OptimizationCandidate]: A list of Optimization Candidates.
+
+        """
+        payload = {
+            "trace_id": trace_id,
+            "source_code": source_code,
+            "optimized_code": optimized_code,
+            "original_line_profiler_results": original_line_profiler_results,
+            "optimized_line_profiler_results": optimized_line_profiler_results,
+            "original_code_runtime": original_code_runtime,
+            "optimized_code_runtime": optimized_code_runtime,
+            "speedup": speedup,
+            "annotated_tests": annotated_tests,
+            "optimization_id": optimization_id,
+            "dependency_code": dependency_code,
+        }
+        logger.info("Generating explanation")
+        console.rule()
+        try:
+            response = self.make_ai_service_request("/explain", payload=payload, timeout=600)
+        except requests.exceptions.RequestException as e:
+            logger.exception(f"Error generating explanations: {e}")
+            ph("cli-optimize-error-caught", {"error": str(e)})
+            return ""
+
+        if response.status_code == 200:
+            explanation: str = response.json()["explanation"]
+            logger.info(f"New Explanation: {explanation}")
+            console.rule()
+            return explanation
+        try:
+            error = response.json()["error"]
+        except Exception:
+            error = response.text
+        logger.error(f"Error generating optimized candidates: {response.status_code} - {error}")
+        ph("cli-optimize-error-response", {"response_status_code": response.status_code, "error": error})
+        console.rule()
+        return ""
+
     def log_results(  # noqa: D417
         self,
         function_trace_id: str,

codeflash/optimization/function_optimizer.py

@@ -1120,11 +1120,35 @@ def find_and_process_best_optimization(
                     )
                     if concolic_test_str:
                         generated_tests_str += "\n\n" + concolic_test_str
-
+                    new_explanation_raw_str = self.aiservice_client.get_new_explanation(
+                        source_code=code_context.read_writable_code,
+                        dependency_code=code_context.read_only_context_code,
+                        trace_id=self.function_trace_id[:-4] + exp_type
+                        if self.experiment_id
+                        else self.function_trace_id,
+                        optimized_code=best_optimization.candidate.source_code,
+                        original_line_profiler_results=original_code_baseline.line_profile_results["str_out"],
+                        optimized_line_profiler_results=best_optimization.line_profiler_test_results["str_out"],
+                        original_code_runtime=humanize_runtime(original_code_baseline.runtime),
+                        optimized_code_runtime=humanize_runtime(best_optimization.runtime),
+                        speedup=f"{int(performance_gain(original_runtime_ns=original_code_baseline.runtime, optimized_runtime_ns=best_optimization.runtime) * 100)}%",
+                        annotated_tests=generated_tests_str,
+                        optimization_id=best_optimization.candidate.optimization_id,
+                    )
+                    new_explanation = Explanation(
+                        raw_explanation_message=new_explanation_raw_str or explanation.raw_explanation_message,
+                        winning_behavior_test_results=explanation.winning_behavior_test_results,
+                        winning_benchmarking_test_results=explanation.winning_benchmarking_test_results,
+                        original_runtime_ns=explanation.original_runtime_ns,
+                        best_runtime_ns=explanation.best_runtime_ns,
+                        function_name=explanation.function_name,
+                        file_path=explanation.file_path,
+                        benchmark_details=explanation.benchmark_details,
+                    )
                     check_create_pr(
                         original_code=original_code_combined,
                         new_code=new_code_combined,
-                        explanation=explanation,
+                        explanation=new_explanation,
                         existing_tests_source=existing_tests,
                         generated_original_test_source=generated_tests_str,
                         function_trace_id=self.function_trace_id[:-4] + exp_type