| input_interface: | |
| - "action" | |
| - "content" | |
| _target_: flow_modules.aiflows.FunSearchFlowModule.SamplerFlow.instantiate_from_default_config | |
| name: SamplerFlowModule | |
| description: A flow that queries an LLM model to generate prompts for the Sampler flow | |
| input_interface_initialized: | |
| - header | |
| - code | |
| input_interface_non_initialized: | |
| - header | |
| - code | |
| output_interface: | |
| - "api_output" | |
| - "from" | |
| backend: | |
| api_infos: ??? | |
| model_name: gpt-4 | |
| frequency_penalty: 0 | |
| max_tokens: 2000 | |
| n: 1 | |
| presence_penalty: 0 | |
| stream: false | |
| temperature: 1.0 | |
| top_p: 0.4 | |
| previous_messages: | |
| last_k: 1 | |
| first_k: 1 | |
| system_message_prompt_template: | |
| template: |2- | |
| Role: You are an expert coder and algorithm designer who provides solutions. | |
| Goal: Write evolve functions that optimize the score of an evaluation function (provided). Aim to achieve higher scores than previous functions. Write a new function that optimizes the evaluation function and scores better than the given functions. Develop non-trivial functions and actual algorithms. | |
| Resources: | |
| You will be given code related to evolve and functions. | |
| You will be given a few functions attempting to optimize the evaluation function. | |
| The function you're evolving (the last one passed) is the only one you have to write. Make sure to include only nested functions or recursive calls. | |
| Naming: | |
| Evaluation function is named {{evaluate_name}}. | |
| Evolve functions have a naming structure similar to: {{evolve_name}}_v. | |
| Important Constraints: | |
| Only write one evolve function solution: Write only the uncompleted function given to you (i.e., only {{evolve_name}}_v{i}, where {i} is the function with the largest number in the prompt, and i is at most {{artifacts_per_prompt}}). Stop writing once you're done with your first solution. | |
| Output must be a function that is parsable by the AST library: Write code only within functions. No textual comments, no code blocks (like ```python), or no explanations. Provide only Python code encapsulated in a single function. Your output should be parsable by the AST Python library. Do not include starting or ending comments like "Here's an improved version." | |
| Follow the naming structure: Keep the function name as it is in the input. | |
| Keep the same function format: Maintain the same input and output format in your solution. | |
| Self-contained solution: You can't call other functions (only recursive calls or calls to nested functions). | |
| Do not use the evaluate function in the solution: Avoid using the evaluate function in your solution. | |
| partial_variables: | |
| evaluate_name: ??? | |
| evolve_name: ??? | |
| artifacts_per_prompt: ??? | |
| init_human_message_prompt_template: | |
| _target_: aiflows.prompt_template.JinjaPrompt | |
| template: |2- | |
| functions for evaluation: | |
| {{header}} | |
| function to evolve and to evaluate: | |
| {{code}} | |
| input_variables: ["header","code"] | |
| partial_variables: {} | |
| human_message_prompt_template: | |
| _target_: aiflows.prompt_template.JinjaPrompt | |
| template: |2- | |
| full code: | |
| {{header}} | |
| function to evolve and to evaluate: | |
| {{code}} | |
| input_variables: ["header","code"] | |
| partial_variables: {} | |