(args)
| 38 | |
| 39 | |
| 40 | def main(args): |
| 41 | load_api_key(toml_file_path="secrets.toml") |
| 42 | lm_configs = STORMWikiLMConfigs() |
| 43 | gemini_kwargs = { |
| 44 | "api_key": os.getenv("GOOGLE_API_KEY"), |
| 45 | "temperature": 1.0, |
| 46 | "top_p": 0.9, |
| 47 | } |
| 48 | |
| 49 | # STORM is a LM system so different components can be powered by different models. |
| 50 | # For a good balance between cost and quality, you can choose a cheaper/faster model for conv_simulator_lm |
| 51 | # which is used to split queries, synthesize answers in the conversation. We recommend using stronger models |
| 52 | # for outline_gen_lm which is responsible for organizing the collected information, and article_gen_lm |
| 53 | # which is responsible for generating sections with citations. |
| 54 | # To check out available Google models, see: |
| 55 | # https://ai.google.dev/gemini-api/docs/get-started/tutorial?lang=python#list_models |
| 56 | conv_simulator_lm = GoogleModel( |
| 57 | model="models/gemini-1.5-flash", max_tokens=500, **gemini_kwargs |
| 58 | ) |
| 59 | question_asker_lm = GoogleModel( |
| 60 | model="models/gemini-1.5-flash", max_tokens=500, **gemini_kwargs |
| 61 | ) |
| 62 | outline_gen_lm = GoogleModel( |
| 63 | model="models/gemini-1.5-pro-exp-0801", max_tokens=400, **gemini_kwargs |
| 64 | ) |
| 65 | article_gen_lm = GoogleModel( |
| 66 | model="models/gemini-1.5-pro-exp-0801", max_tokens=700, **gemini_kwargs |
| 67 | ) |
| 68 | article_polish_lm = GoogleModel( |
| 69 | model="models/gemini-1.5-pro-exp-0801", max_tokens=4000, **gemini_kwargs |
| 70 | ) |
| 71 | |
| 72 | lm_configs.set_conv_simulator_lm(conv_simulator_lm) |
| 73 | lm_configs.set_question_asker_lm(question_asker_lm) |
| 74 | lm_configs.set_outline_gen_lm(outline_gen_lm) |
| 75 | lm_configs.set_article_gen_lm(article_gen_lm) |
| 76 | lm_configs.set_article_polish_lm(article_polish_lm) |
| 77 | |
| 78 | engine_args = STORMWikiRunnerArguments( |
| 79 | output_dir=args.output_dir, |
| 80 | max_conv_turn=args.max_conv_turn, |
| 81 | max_perspective=args.max_perspective, |
| 82 | search_top_k=args.search_top_k, |
| 83 | max_thread_num=args.max_thread_num, |
| 84 | ) |
| 85 | |
| 86 | # STORM is a knowledge curation system which consumes information from the retrieval module. |
| 87 | # Currently, the information source is the Internet and we use search engine API as the retrieval module. |
| 88 | match args.retriever: |
| 89 | case "bing": |
| 90 | rm = BingSearch( |
| 91 | bing_search_api=os.getenv("BING_SEARCH_API_KEY"), |
| 92 | k=engine_args.search_top_k, |
| 93 | ) |
| 94 | case "you": |
| 95 | rm = YouRM(ydc_api_key=os.getenv("YDC_API_KEY"), k=engine_args.search_top_k) |
| 96 | case "brave": |
| 97 | rm = BraveRM( |
no test coverage detected