人工智能大模型AI Agent代碼智能體自主智能體【免費(fèi)下載鏈接】openevolveOpen-source implementation of AlphaEvolve項(xiàng)目地址https://gitcode.com/gh_mirrors/op/openevolve點(diǎn)擊查看免費(fèi)下載OpenEvolve 是一個(gè)開源的 AlphaEvolve 實(shí)現(xiàn)它把 LLM 變成自主代碼優(yōu)化器通過(guò)進(jìn)化的方式反復(fù)改進(jìn)程序。本篇文章以倉(cāng)庫(kù)configs/目錄下的配置體系為核心系統(tǒng)講解如何從default_config.yaml全量模板出發(fā)理解每一個(gè)配置參數(shù)的含義與默認(rèn)值掌握島嶼模型Island-Based Evolution三個(gè)核心參數(shù)num_islands、migration_interval、migration_rate的調(diào)優(yōu)邏輯并基于island_examples.yaml的五大典型場(chǎng)景與early_stopping_example.yaml的早期停止示例寫出可直接運(yùn)行的實(shí)戰(zhàn)配置。讀完本文你將具備從零搭建 OpenEvolve 進(jìn)化任務(wù)配置、按問(wèn)題復(fù)雜度選擇合適的島嶼策略、并通過(guò)命令行或 Python API 一鍵啟動(dòng)的能力。configs 目錄OpenEvolve 配置文件的完整圖譜configs/目錄是 OpenEvolve 的配置中心共包含 4 個(gè) YAML 文件與一份導(dǎo)讀文檔configs/README.md它們分工明確文件定位configs/default_config.yaml主配置模板包含所有可用選項(xiàng)與合理的默認(rèn)值參數(shù)注釋最全適合作為自定義配置的起點(diǎn)configs/island_config_example.yaml島嶼模型實(shí)戰(zhàn)示例演示推薦的海島設(shè)置、平衡的遷移參數(shù)是一份可直接運(yùn)行的完整配置configs/island_examples.yaml五種不同場(chǎng)景的島嶼配置集合最大多樣性、聚焦探索、均衡、快速探索、大規(guī)模進(jìn)化并附參數(shù)選擇準(zhǔn)則configs/early_stopping_example.yaml早期停止功能示例演示如何在進(jìn)化的同時(shí)自動(dòng)判斷是否已無(wú)改進(jìn)空間其中default_config.yaml是靈魂文件它既是一份參數(shù)字典也是 OpenEvolve 各核心模塊LLM、Prompt、Database、Evaluator、EvolutionTrace配置入口的真實(shí)寫照與 openevolve/config.py 中的Config及其子 dataclassLLMConfig、PromptConfig、DatabaseConfig、EvaluatorConfig、EvolutionTraceConfig一一對(duì)應(yīng)。下面我們先逐模塊拆解這份全量模板。default_config.yaml全量配置模板逐模塊拆解通用設(shè)置General Settingsmax_iterations: 100 # Maximum number of evolution iterations checkpoint_interval: 10 # Save checkpoints every N iterations log_level: INFO # Logging level (DEBUG, INFO, WARNING, ERROR, CRITICAL) log_dir: null # Custom directory for logs (default: output_dir/logs) random_seed: 42 # Random seed for reproducibility (null random, 42 default)max_iterations進(jìn)化總迭代數(shù)上限。注意這是 YAML 模板中的示例值代碼層Config的默認(rèn)值為 10000見 openevolve/config.pyCLI 也可通過(guò)--iterations覆蓋checkpoint_interval每隔 N 次迭代保存一次檢查點(diǎn)配合 CLI 的--checkpoint參數(shù)可實(shí)現(xiàn)斷點(diǎn)續(xù)跑random_seed設(shè)置后控制器會(huì)同步 seed 全局random、numpy并對(duì)不同組件派生哈希種子保證可復(fù)現(xiàn)設(shè)為null則隨機(jī)。進(jìn)化設(shè)置Evolution Settingsdiff_based_evolution: true # Use diff-based evolution (true) or full rewrites (false) max_code_length: 10000 # Maximum allowed code length in characters enforce_evolve_blocks: false # Revert LLM edits outside EVOLVE-BLOCK-START/END markersdiff_based_evolutiontrue采用基于 SEARCH/REPLACE diff 的增量進(jìn)化false則使用全量重寫。Config中還維護(hù)diff_pattern正則默認(rèn) SEARCH\n(.*?)\n(.*?) REPLACE用于解析 LLM 返回的 diffenforce_evolve_blocks開啟后LLM 在EVOLVE-BLOCK-START/EVOLVE-BLOCK-END標(biāo)記之外的編輯會(huì)被回滾實(shí)現(xiàn)位于 openevolve/utils/code_utils.py 的enforce_evolve_blocks。早期停止設(shè)置Early Stoppingearly_stopping_patience: null # Stop after N iterations without improvement (null disabled) convergence_threshold: 0.001 # Minimum improvement required to reset patience counter early_stopping_metric: combined_score # Metric to track for early stopping詳見本文早期停止Early Stopping配置一節(jié)。LLM 配置llmllm: models: # List of available models with their weights - name: gemini-2.0-flash-lite weight: 0.8 - name: gemini-2.0-flash weight: 0.2 evaluator_models: # Models for LLM feedback - name: gemini-2.0-flash-lite weight: 0.8 - name: gemini-2.0-flash weight: 0.2 api_base: https://generativelanguage.googleapis.com/v1beta/openai/ # Base URL for API api_key: null # API key (defaults to OPENAI_API_KEY env variable) # api_key: ${GEMINI_API_KEY} # Reads API key from $GEMINI_API_KEY temperature: 0.7 # Higher more creative top_p: 0.95 # Top-p sampling parameter max_tokens: 4096 # Maximum tokens to generate timeout: 60 # Timeout for API requests in seconds retries: 3 # Number of retries for failed requests retry_delay: 5 # Delay between retries in seconds幾個(gè)值得注意的源碼級(jí)細(xì)節(jié)多模型加權(quán)集成models列表支持配置多個(gè)模型并賦予權(quán)重運(yùn)行時(shí)由LLMEnsemble按權(quán)重采樣見 openevolve/llm/ensemble.py。evaluator_models若未顯式配置會(huì)復(fù)用models列表openevolve/config.py${VAR}環(huán)境變量語(yǔ)法api_key支持api_key: ${GEMINI_API_KEY}形式配置加載時(shí)會(huì)通過(guò)_resolve_env_var完整匹配并展開若引用的環(huán)境變量未設(shè)置會(huì)拋出ValueErroropenevolve/config.py向后兼容語(yǔ)法除了models數(shù)組還兼容primary_model/primary_model_weight與secondary_model/secondary_model_weight的舊寫法LLMConfig.__post_init__會(huì)自動(dòng)把它們轉(zhuǎn)換為models列表?xiàng)l目openevolve/config.pyisland_config_example.yaml與 examples/circle_packing/config_phase_1.yaml 均使用了這種寫法provider 擴(kuò)展LLMModelConfig還支持provider字段openai默認(rèn)、claude_code、copilot_cli以及reasoning_effort、max_budget_usdClaude Code、max_ai_creditsCopilot CLI、manual_mode人機(jī)協(xié)同等進(jìn)階選項(xiàng)。Prompt 配置promptprompt: template_dir: null # Custom directory for prompt templates system_message: You are an expert coder helping to improve programs through evolution. island_system_messages: [] # Optional literal text or template names by island index evaluator_system_message: You are an expert code reviewer. num_top_programs: 3 # Number of top-performing programs to include num_diverse_programs: 2 # Number of diverse programs to include use_template_stochasticity: true # Use random variations in templates for diversity template_variations: # Different phrasings for parts of the template improvement_suggestion: - Heres how we could improve this code: - I suggest the following improvements: - We can enhance this code by: include_artifacts: true # Include execution outputs/errors in prompt max_artifact_bytes: 20480 # Maximum artifact size in bytes (20KB default) artifact_security_filter: true # Apply security filtering to artifacts suggest_simplification_after_chars: 500 include_changes_under_chars: 100 concise_implementation_max_lines: 10 comprehensive_implementation_min_lines: 50 diff_summary_max_line_len: 100 diff_summary_max_lines: 30template_dir自定義提示詞模板目錄。加載時(shí)若為相對(duì)路徑會(huì)相對(duì)配置文件所在目錄解析openevolve/config.py倉(cāng)庫(kù)默認(rèn)模板見 openevolve/prompts/defaults/island_system_messages按島嶼下標(biāo)為不同島嶼指定專屬系統(tǒng)消息字面量或模板名null/缺失則回退到system_message。它必須與database.num_islands配合條目數(shù)不能超過(guò)島嶼數(shù)否則Config.from_dict會(huì)拋出ValueErroropenevolve/config.py閾值類參數(shù)suggest_simplification_after_chars等用于特征提取與程序標(biāo)簽簡(jiǎn)潔實(shí)現(xiàn)/全面實(shí)現(xiàn)diff_summary_*控制 Previous Attempts 段落中 SEARCH/REPLACE 塊的展示長(zhǎng)度。Database 配置databasedatabase: db_path: null # Path to persist database (null in-memory only) in_memory: true log_prompts: true population_size: 1000 # Maximum number of programs to keep in memory archive_size: 100 # Size of elite archive num_islands: 5 # Number of islands (separate populations) migration_interval: 50 # Migrate between islands every N generations migration_rate: 0.1 # Fraction of top programs to migrate (0.1 10%) elite_selection_ratio: 0.1 # Ratio of elite programs to select exploration_ratio: 0.2 # Ratio of exploration vs exploitation exploitation_ratio: 0.7 # Ratio of exploitation vs random selection feature_dimensions: # MAP-Elites feature map dimensions - complexity # Code length (built-in) - diversity # Code diversity (built-in) feature_bins: 10 # Number of bins per dimension diversity_reference_size: 20 # Size of reference set for diversity calculationdb_path持久化路徑null時(shí)數(shù)據(jù)庫(kù)僅在內(nèi)存中運(yùn)行MAP-Elites 特征圖feature_dimensions默認(rèn)[complexity, diversity]兩者均為內(nèi)置特征代碼長(zhǎng)度與結(jié)構(gòu)多樣性也可混入評(píng)估器返回的自定義指標(biāo)如performance、correctness。注意自定義維度必須由評(píng)估器返回連續(xù)的原始數(shù)值OpenEvolve 內(nèi)部完成縮放與分箱評(píng)估器不應(yīng)返回預(yù)先算好的 bin 索引feature_bins每個(gè)維度的分箱數(shù)可以是整數(shù)所有維度一致或按維度指定的字典例如feature_bins: {complexity: 10, diversity: 15}從源碼看diversity_metric目前固定為edit_distancefeature_based 未實(shí)現(xiàn)選擇比例默認(rèn)elite_selection_ratio0.1、exploration_ratio0.2、exploitation_ratio0.7。Evaluator 配置evaluatorevaluator: timeout: 300 # Maximum evaluation time in seconds max_retries: 3 # Maximum number of retries for evaluation cascade_evaluation: true # Use cascade evaluation to filter bad solutions early cascade_thresholds: # Thresholds for advancing to next evaluation stage - 0.5 - 0.75 - 0.9 parallel_evaluations: 4 # Number of parallel evaluations use_llm_feedback: false # Use LLM to evaluate code quality (experimental) llm_feedback_weight: 0.1 # Weight for LLM feedback in final score適應(yīng)度計(jì)算優(yōu)先使用評(píng)估器返回的combined_score否則對(duì)所有指標(biāo)取平均但會(huì)排除database.feature_dimensions中列出的特征維度級(jí)聯(lián)評(píng)估cascade先跑廉價(jià)的第一階段達(dá)到閾值0.5才進(jìn)入下一階段用于盡早過(guò)濾劣質(zhì)解對(duì)應(yīng) openevolve/evaluator.py 中的_cascade_evaluate與run_stage1/2/3資源限制memory_limit_mb、cpu_limit與分布式評(píng)估distributed在注釋中明確標(biāo)注not yet implemented配置時(shí)不要依賴。Evolution Trace 配置evolution_traceevolution_trace: enabled: false # Enable evolution trace logging format: jsonl # Output format: jsonl, json, or hdf5 include_code: false # Include full program code in traces include_prompts: true # Include prompts and LLM responses output_path: null # Defaults to output_dir/evolution_trace.{format} buffer_size: 10 # Number of traces to buffer before writing compress: false # Compress output file (jsonl only)該模塊為強(qiáng)化學(xué)習(xí)訓(xùn)練與進(jìn)化分析記錄細(xì)粒度的進(jìn)化軌跡支持 jsonl/json/hdf5 三種格式與可選壓縮底層實(shí)現(xiàn)在 openevolve/evolution_trace.py。島嶼模型OpenEvolve 保持進(jìn)化多樣性的核心機(jī)制configs/README.md明確指出default_config.yaml的價(jià)值不僅在于參數(shù)齊全還在于內(nèi)置了島嶼式進(jìn)化參數(shù)Island-Based Evolution Parameters——這是 OpenEvolve 維持進(jìn)化多樣性的關(guān)鍵設(shè)計(jì)。島嶼模型的思想是不維護(hù)一個(gè)全局種群而是把種群拆成多個(gè)獨(dú)立進(jìn)化的子種群島嶼每個(gè)島嶼在自己的搜索區(qū)域深耕每隔若干代把各島嶼上的優(yōu)秀程序遷移到相鄰島嶼實(shí)現(xiàn)知識(shí)共享。這樣既避免了單一種群過(guò)早收斂到局部最優(yōu)又通過(guò)周期性交流保留了全局進(jìn)步的動(dòng)力。三個(gè)核心參數(shù)database: num_islands: 5 # Number of separate populations migration_interval: 50 # Migrate every N generations migration_rate: 0.1 # Fraction of top programs to migratenum_islands島嶼數(shù)量。島嶼越多獨(dú)立搜索方向越多、多樣性越強(qiáng)但收斂變慢推薦大多數(shù)問(wèn)題取 3–10migration_interval每隔多少代遷移一次。值越大各島嶼越獨(dú)立推薦 25–100 代migration_rate每次遷移時(shí)從每個(gè)島嶼頂部程序按適應(yīng)度降序中遷出的比例推薦 0.05–0.25%–20%。比例越高知識(shí)傳播越快。源碼級(jí)的遷移實(shí)現(xiàn)細(xì)節(jié)在 openevolve/database.py 中可以看到島嶼機(jī)制的真實(shí)實(shí)現(xiàn)輪轉(zhuǎn)調(diào)度next_island()以 round-robin 方式切換當(dāng)前進(jìn)化島嶼increment_island_generation()維護(hù)每個(gè)島嶼的代數(shù)計(jì)數(shù)器openevolve/database.py遷移判定should_migrate()比較最大島嶼代數(shù)與上次遷移代數(shù)的差值是否達(dá)到migration_intervalopenevolve/database.py遷移數(shù)量migrate_programs()將每個(gè)島嶼的程序按get_fitness_score優(yōu)先combined_score否則對(duì)非特征維度指標(biāo)取平均降序排列取max(1, int(len(island_programs) * migration_rate))個(gè)作為遷出對(duì)象openevolve/database.py環(huán)狀拓?fù)淠J(rèn)把遷出程序復(fù)制到相鄰島嶼(i1) % n與(i-1) % nopenevolve/database.py即鄰居共享而非全局廣播去重保護(hù)_migrate_one()會(huì)跳過(guò)目標(biāo)島嶼已存在相同代碼的程序并為遷移副本生成新 UUID、打上migrant: true標(biāo)記openevolve/database.py最少島嶼限制島嶼數(shù)小于 2 時(shí)直接跳過(guò)遷移openevolve/database.py。這些細(xì)節(jié)解釋了為什么migration_rate是比例而非數(shù)量遷移量隨島嶼種群規(guī)模動(dòng)態(tài)縮放小種群也能保證至少遷移 1 個(gè)程序。何時(shí)使用什么參數(shù)選擇速查configs/README.md與 configs/island_examples.yaml 給出的經(jīng)驗(yàn)法則問(wèn)題特征策略說(shuō)明復(fù)雜問(wèn)題易陷入局部最優(yōu)更多島嶼 更低遷移頻率增大搜索空間覆蓋延長(zhǎng)各島獨(dú)立演化時(shí)間簡(jiǎn)單問(wèn)題結(jié)構(gòu)清晰更少島嶼 更頻繁遷移加速知識(shí)共享快速收斂長(zhǎng)時(shí)運(yùn)行更多島嶼用島嶼多樣性對(duì)抗長(zhǎng)跑中的早熟收斂短時(shí)運(yùn)行更少島嶼更快收斂避免資源浪費(fèi)island_config_example.yaml一份可直接運(yùn)行的島嶼配置configs/island_config_example.yaml 是島嶼模型的標(biāo)準(zhǔn)答案它同時(shí)示范了兩套 LLM 配置語(yǔ)法與島嶼專屬系統(tǒng)消息的用法max_iterations: 1000 checkpoint_interval: 100 log_level: INFO llm: primary_model: gemini-2.0-flash-lite primary_model_weight: 0.8 secondary_model: gemini-2.0-flash secondary_model_weight: 0.2 temperature: 0.7 top_p: 0.95 max_tokens: 4096 database: population_size: 500 archive_size: 100 num_islands: 5 # 5 個(gè)獨(dú)立種群 migration_interval: 50 # 每 50 代遷移 migration_rate: 0.1 # 遷移頂部 10% 的程序 elite_selection_ratio: 0.1 exploration_ratio: 0.3 exploitation_ratio: 0.7 feature_bins: 10 prompt: system_message: Improve the program through evolution. island_system_messages: - Focus on simpler algorithms. - null - Focus on runtime performance. num_top_programs: 3 num_diverse_programs: 2 use_template_stochasticity: true evaluator: timeout: 300 max_retries: 3 cascade_evaluation: true parallel_evaluations: 4 diff_based_evolution: true allow_full_rewrites: false max_code_length: 10000亮點(diǎn)解讀島嶼專屬提示詞island_system_messages允許給不同島嶼定制演化風(fēng)格——島嶼 0 專注更簡(jiǎn)單的算法、島嶼 2 專注運(yùn)行時(shí)性能第二個(gè)條目為null表示該島嶼沿用system_message。這種分工進(jìn)化正是島嶼模型多樣性的延伸兼容語(yǔ)法primary_modelsecondary_model的寫法會(huì)在Config加載時(shí)被自動(dòng)展開為models權(quán)重列表主模型 0.8、次模型 0.2feature_bins 支持字典形式注釋中展示了feature_bins: {performance: 20, correctness: 10}的按維度分箱寫法。island_examples.yaml五大典型場(chǎng)景配置速查configs/island_examples.yaml 把島嶼參數(shù)壓縮成五個(gè)可直接套用的場(chǎng)景模板# 1. 最大多樣性適合復(fù)雜問(wèn)題、長(zhǎng)時(shí)運(yùn)行、避免局部最優(yōu) max_diversity: database: num_islands: 10 migration_interval: 25 migration_rate: 0.2 population_size: 1000 archive_size: 200 # 2. 聚焦探索適合結(jié)構(gòu)清晰的問(wèn)題、較短運(yùn)行 focused_exploration: database: num_islands: 3 migration_interval: 100 migration_rate: 0.05 population_size: 500 archive_size: 50 # 3. 均衡方案大多數(shù)問(wèn)題的默認(rèn)起點(diǎn) balanced: database: num_islands: 5 migration_interval: 50 migration_rate: 0.1 population_size: 1000 archive_size: 100 # 4. 快速探索小規(guī)模、原型驗(yàn)證 quick_exploration: database: num_islands: 3 migration_interval: 20 migration_rate: 0.15 population_size: 200 archive_size: 30 # 5. 大規(guī)模進(jìn)化復(fù)雜優(yōu)化、超長(zhǎng)進(jìn)化運(yùn)行 large_scale: database: num_islands: 15 migration_interval: 75 migration_rate: 0.08 population_size: 2000 archive_size: 300從 5 個(gè)場(chǎng)景可以觀察出清晰的權(quán)衡規(guī)律max_diversity用 10 個(gè)島嶼 高頻高比例遷移25 代 / 20%最大化搜索空間覆蓋focused_exploration用 3 個(gè)島嶼 低頻低比例遷移100 代 / 5%讓每個(gè)島嶼深入挖掘large_scale則把島嶼數(shù)推到 15、種群擴(kuò)到 2000同時(shí)用保守的 8% 遷移率避免大規(guī)模種群被過(guò)早同化。參數(shù)選取的本質(zhì)是在多樣性多島、少遷與收斂速度少島、多遷之間找平衡點(diǎn)。早期停止Early Stopping配置configs/early_stopping_example.yaml 演示了如何在進(jìn)化長(zhǎng)跑中自動(dòng)止損max_iterations: 1000 checkpoint_interval: 50 log_level: INFO # 連續(xù) 30 代無(wú)改進(jìn)即停止 early_stopping_patience: 30 # Stop after 30 iterations without improvement convergence_threshold: 0.01 # Minimum improvement of 0.01 required to reset patience early_stopping_metric: combined_score # Track the combined_score metric llm: models: - name: gpt-4o-mini weight: 1.0 api_base: https://api.openai.com/v1 temperature: 0.7 max_tokens: 4096 database: population_size: 50 num_islands: 3 migration_interval: 20 evaluator: timeout: 60 max_retries: 2 parallel_evaluations: 2 diff_based_evolution: true max_code_length: 8000三個(gè)早期停止參數(shù)的語(yǔ)義見 openevolve/config.pyearly_stopping_patience連續(xù) N 次迭代無(wú)改進(jìn)即停止null表示禁用。該示例中30表示30 代無(wú)改進(jìn)就收工可顯著節(jié)省 LLM API 調(diào)用成本convergence_threshold判定有改進(jìn)的最小幅度。只有當(dāng)early_stopping_metric的提升超過(guò)該閾值示例為 0.01才會(huì)重置耐心計(jì)數(shù)器若只是小幅抖動(dòng)則不重置early_stopping_metric跟蹤的指標(biāo)默認(rèn)combined_score可改為評(píng)估器返回的其他指標(biāo)名。該文件同樣是一個(gè)完整的可運(yùn)行配置gpt-4o-mini OpenAI API適合預(yù)算有限、想快速驗(yàn)證想法的場(chǎng)景。關(guān)于早期停止參數(shù)更細(xì)的行為與校驗(yàn)邏輯可參考 tests/test_early_stopping_config.py。使用方法把配置跑起來(lái)方式一命令行啟動(dòng)推薦快速上手先復(fù)制模板為自有配置cp configs/default_config.yaml my_config.yaml # Edit my_config.yaml for your specific needs然后通過(guò)倉(cāng)庫(kù)入口腳本openevolve-run.py啟動(dòng)python openevolve-run.py initial_program.py evaluator.py --config my_config.yaml --output my_outputCLI 支持的主要參數(shù)見 openevolve/cli.py參數(shù)說(shuō)明initial_program位置參數(shù)初始程序文件路徑evaluation_file位置參數(shù)含evaluate函數(shù)的評(píng)估文件路徑--config/-c配置文件路徑Y(jié)AML--output/-o結(jié)果輸出目錄--iterations/-i覆蓋配置中的最大迭代數(shù)--target-score/-t達(dá)到目標(biāo)分?jǐn)?shù)即停止--log-level/-l覆蓋日志級(jí)別--checkpoint指定檢查點(diǎn)目錄續(xù)跑--api-base、--primary-model、--secondary-model覆蓋 LLM API 與模型配置實(shí)際項(xiàng)目中這種用法非常常見例如 examples/arc_benchmark/run_evolution.sh 中的啟動(dòng)命令python ../../openevolve-run.py initial_program.py evaluator.py --config config.yaml --output $OUTPUT_DIR $方式二Python API 啟動(dòng)適合集成與腳本化configs/README.md給出的 API 用法如下from openevolve import OpenEvolve from openevolve.config import Config config Config() config.from_yaml(my_config.yaml) evolve OpenEvolve( initial_program_pathprogram.py, evaluation_fileevaluator.py, configconfig )此外倉(cāng)庫(kù)還提供了更簡(jiǎn)潔的函數(shù)式 APIrun_evolutionopenevolve/api.py它接受文件路徑 / 代碼字符串 / 代碼行列表三種形式的初始程序以及文件路徑 / Callable兩種形式的評(píng)估器config參數(shù)既可以是 YAML 路徑也可以是Config對(duì)象from openevolve import run_evolution result run_evolution( initial_programprogram.py, evaluatorevaluator.py, configmy_config.yaml, iterations200, ) print(result.best_score)配置加載的底層行為了解即可從 openevolve/config.py 的Config.from_yaml/from_dict實(shí)現(xiàn)可知YAML 通過(guò)yaml.safe_load讀取再用dacite.from_dict轉(zhuǎn)換為類型化的Configdataclass 結(jié)構(gòu)因此字段名必須與 dataclass 定義一致llm.temperature/top_p為null時(shí)會(huì)被自動(dòng)剔除避免 dacite 類型錯(cuò)誤頂層random_seed會(huì)自動(dòng)下放到database.random_seed兩處校驗(yàn)會(huì)直接拋出ValueErrorprompt.island_system_messages條目數(shù)超過(guò)database.num_islandsprompt.programs_as_changes_descriptiontrue但diff_based_evolutionfalsediff_pattern若提供會(huì)先嘗試編譯正則非法表達(dá)式會(huì)報(bào)錯(cuò)。從示例到實(shí)戰(zhàn)參考倉(cāng)庫(kù)里的真實(shí)配置如果想看島嶼參數(shù)之外的完整實(shí)戰(zhàn)配置倉(cāng)庫(kù)的 examples 目錄提供了大量真實(shí)任務(wù)的配置文件是絕佳的參考樣本examples/circle_packing/config_phase_1.yaml圓填充問(wèn)題n26使用 4 個(gè)島嶼、cascade_thresholds: [0.5, 0.75]、diff_based_evolution: false構(gòu)造型程序用全量重寫以及含領(lǐng)域知識(shí)的長(zhǎng) system messageexamples/arc_benchmark/config.yamlARC 基準(zhǔn)任務(wù)的配置配套post_evolution_eval.py做進(jìn)化后評(píng)估examples/algotune/AlgoTune 基準(zhǔn)下的多個(gè)算法優(yōu)化任務(wù)FFT、LU 分解、特征向量等每個(gè)任務(wù)都包含config.yaml、evaluator.py與initial_program.py三件套。綜合來(lái)看OpenEvolve 的配置哲學(xué)可以總結(jié)為三句話用default_config.yaml保證參數(shù)覆蓋完整用島嶼三參數(shù)num_islands/migration_interval/migration_rate控制多樣性與收斂的平衡用早期停止參數(shù)控制成本與收益的平衡。從復(fù)制模板、調(diào)整島嶼參數(shù)、按需開啟早期停止到用 CLI 或 API 啟動(dòng)任務(wù)這就是一條完整且可復(fù)現(xiàn)的 OpenEvolve 配置實(shí)戰(zhàn)路徑。贊分享人工智能大模型AI Agent代碼智能體自主智能體【免費(fèi)下載鏈接】openevolveOpen-source implementation of AlphaEvolve項(xiàng)目地址https://gitcode.com/gh_mirrors/op/openevolve點(diǎn)擊查看免費(fèi)下載相關(guān)推薦CAI模型設(shè)置完全指南如何配置默認(rèn)模型與單個(gè)代理CAICybersecurity AI是一個(gè)開源的人工智能平臺(tái)專為網(wǎng)絡(luò)安全測(cè)試和安全評(píng)估而設(shè)計(jì)。作為一名安全研究人員或滲透測(cè)試人員了解如何正確配置CAI人工智能AI Agent網(wǎng)絡(luò)安全滲透測(cè)試工具調(diào)用AI 評(píng)測(cè)2024最新three-globe入門教程5分鐘上手3D地球可視化開發(fā)2024最新three globe入門教程5分鐘上手3D地球可視化開發(fā) three globe是一款基于ThreeJS的WebGL地球數(shù)據(jù)可視化庫(kù)能幫助開發(fā)數(shù)據(jù)可視化3D渲染openclaw-cn models 命令完全指南模型發(fā)現(xiàn)、掃描、默認(rèn)模型與認(rèn)證配置openclaw cn models 命令完全指南模型發(fā)現(xiàn)、掃描、默認(rèn)模型與認(rèn)證配置 openclaw cn models 是中文社區(qū)版 OpenClawo人工智能AI Agent即時(shí)通訊后端本地部署語(yǔ)音上一篇Azure Community-Policy 與內(nèi)置策略對(duì)比何時(shí)選擇自定義解決方案下一篇解放雙手KeymouseGo自動(dòng)化操作從入門到精通指南創(chuàng)作聲明：本文部分內(nèi)容由AI輔助生成（AIGC），僅供參考