Spaces:
Sleeping
Sleeping
| # algoforge_prime/core/evolution_engine.py | |
| print("DEBUG: Importing core.evolution_engine") | |
| from core.llm_clients import call_huggingface_api, call_gemini_api, LLMResponse | |
| from prompts.system_prompts import get_system_prompt | |
| # from core.evaluation_engine import EvaluationResultOutput # For type hinting if needed | |
| print("DEBUG: core.evolution_engine - Imports successful") | |
| def evolve_solution( | |
| original_solution_text: str, | |
| evaluation_output_obj, # This is an EvaluationResultOutput object | |
| problem_description: str, | |
| problem_type: str, | |
| llm_client_config: dict | |
| ) -> str: | |
| print(f"DEBUG: evolution_engine.py - Evolving solution. Problem type: {problem_type}") | |
| system_p_evolve = get_system_prompt("evolution_general") | |
| try: | |
| # Use the method from EvaluationResultOutput to get formatted critique and test results | |
| critique_and_test_feedback = evaluation_output_obj.get_display_critique() | |
| original_score = evaluation_output_obj.combined_score | |
| except AttributeError: # Fallback if evaluation_output_obj is not as expected | |
| critique_and_test_feedback = "Detailed evaluation feedback was not available or malformed." | |
| original_score = 0 # Or try to get it from evaluation_output_obj if it's just a simple dict | |
| if hasattr(evaluation_output_obj, 'score'): original_score = evaluation_output_obj.score | |
| elif isinstance(evaluation_output_obj, dict) and 'score' in evaluation_output_obj: original_score = evaluation_output_obj['score'] | |
| user_p_evolve = ( | |
| f"Original Problem Context: \"{problem_description}\"\n\n" | |
| f"The solution to be evolved achieved a combined score of {original_score}/10.\n" | |
| f"Here is the original solution text:\n```python\n{original_solution_text}\n```\n\n" | |
| f"Here is the comprehensive evaluation it received (including LLM critique AND automated test feedback/errors if run):\n'''\n{critique_and_test_feedback}\n'''\n\n" | |
| f"Your Task: Based on ALL the information above (solution, LLM critique, and crucially any test execution results/errors mentioned in the evaluation), " | |
| f"evolve the provided solution to make it demonstrably superior. " | |
| f"**Your HIGHEST PRIORITY is to fix any reported execution errors or failed tests.** " | |
| f"Then, address other critique points like efficiency, clarity, or completeness. " | |
| f"Output ONLY the *complete, raw, evolved Python code block*. Do not include explanations outside the code block unless explicitly part of the solution's comments." | |
| ) | |
| llm_response_obj = None | |
| if llm_client_config["type"] == "hf": | |
| llm_response_obj = call_huggingface_api(user_p_evolve, llm_client_config["model_id"], llm_client_config["temp"], llm_client_config["max_tokens"], system_p_evolve) | |
| elif llm_client_config["type"] == "google_gemini": | |
| llm_response_obj = call_gemini_api(user_p_evolve, llm_client_config["model_id"], llm_client_config["temp"], llm_client_config["max_tokens"], system_p_evolve) | |
| else: | |
| return f"ERROR (Evolution): Unknown LLM client type '{llm_client_config['type']}'" | |
| if llm_response_obj.success: | |
| # Optional: basic cleanup of the LLM output if it tends to add markdown | |
| from core.utils import basic_text_cleanup # Assuming you have this | |
| return basic_text_cleanup(llm_response_obj.text) | |
| else: | |
| return f"ERROR (Evolution with {llm_response_obj.model_id_used}): {llm_response_obj.error}" | |
| print("DEBUG: core.evolution_engine - Module fully defined.") |