|
| 1 | +import logging |
| 2 | +import traceback |
| 3 | +from os import getenv |
| 4 | +from urllib.parse import urlparse |
| 5 | + |
| 6 | +import streamlit as st |
| 7 | +import tiktoken |
| 8 | +from dotenv import load_dotenv |
| 9 | +from langchain_community.document_loaders import YoutubeLoader # Youtube用 |
| 10 | +from langchain_core.output_parsers import StrOutputParser |
| 11 | +from langchain_core.prompts import ChatPromptTemplate |
| 12 | +from langchain_core.runnables import RunnableLambda |
| 13 | +from langchain_openai import AzureChatOpenAI |
| 14 | +from langchain_text_splitters import RecursiveCharacterTextSplitter |
| 15 | + |
| 16 | +logger = logging.getLogger(__name__) |
| 17 | +load_dotenv() |
| 18 | + |
| 19 | + |
| 20 | +SUMMARIZE_PROMPT = """Please provide a clear 300 word summary of the following content in Japanese. |
| 21 | +
|
| 22 | +======== |
| 23 | +
|
| 24 | +{content} |
| 25 | +
|
| 26 | +======== |
| 27 | +""" |
| 28 | + |
| 29 | + |
| 30 | +def init_page(): |
| 31 | + st.set_page_config(page_title="Summarize YouTube", page_icon="💻") |
| 32 | + st.header("Summarize YouTube") |
| 33 | + st.sidebar.title("Options") |
| 34 | + |
| 35 | + |
| 36 | +def select_model(temperature=0): |
| 37 | + return AzureChatOpenAI( |
| 38 | + temperature=temperature, |
| 39 | + api_key=getenv("AZURE_OPENAI_API_KEY"), |
| 40 | + api_version=getenv("AZURE_OPENAI_API_VERSION"), |
| 41 | + azure_endpoint=getenv("AZURE_OPENAI_ENDPOINT"), |
| 42 | + model=getenv("AZURE_OPENAI_MODEL_CHAT"), |
| 43 | + ) |
| 44 | + |
| 45 | + |
| 46 | +def init_summarize_chain(): |
| 47 | + llm = select_model() |
| 48 | + prompt = ChatPromptTemplate.from_messages( |
| 49 | + [ |
| 50 | + ("user", SUMMARIZE_PROMPT), |
| 51 | + ] |
| 52 | + ) |
| 53 | + output_parser = StrOutputParser() |
| 54 | + return prompt | llm | output_parser |
| 55 | + |
| 56 | + |
| 57 | +def init_map_reduce_chain(): |
| 58 | + summarize_chain = init_summarize_chain() |
| 59 | + |
| 60 | + text_splitter = RecursiveCharacterTextSplitter.from_tiktoken_encoder( |
| 61 | + model_name="gpt-4o", # hard-coded for now |
| 62 | + chunk_size=16000, |
| 63 | + chunk_overlap=0, |
| 64 | + ) |
| 65 | + text_split = RunnableLambda(lambda x: [{"content": doc} for doc in text_splitter.split_text(x["content"])]) |
| 66 | + text_concat = RunnableLambda(lambda x: {"content": "\n".join(x)}) |
| 67 | + return text_split | summarize_chain.map() | text_concat | summarize_chain |
| 68 | + |
| 69 | + |
| 70 | +def init_chain(): |
| 71 | + summarize_chain = init_summarize_chain() |
| 72 | + map_reduce_chain = init_map_reduce_chain() |
| 73 | + |
| 74 | + def route(x): |
| 75 | + encoding = tiktoken.encoding_for_model("gpt-4o") |
| 76 | + token_count = len(encoding.encode(x["content"])) |
| 77 | + if token_count > 16000: |
| 78 | + return map_reduce_chain |
| 79 | + else: |
| 80 | + return summarize_chain |
| 81 | + |
| 82 | + chain = RunnableLambda(route) |
| 83 | + |
| 84 | + return chain |
| 85 | + |
| 86 | + |
| 87 | +def validate_url(url): |
| 88 | + """URLが有効かどうかを判定する関数""" |
| 89 | + try: |
| 90 | + result = urlparse(url) |
| 91 | + if result.netloc != "www.youtube.com": |
| 92 | + return False |
| 93 | + if not result.path.startswith("/watch"): |
| 94 | + return False |
| 95 | + return all([result.scheme, result.netloc]) |
| 96 | + except ValueError: |
| 97 | + return False |
| 98 | + |
| 99 | + |
| 100 | +def get_content(url): |
| 101 | + with st.spinner("Fetching Youtube ..."): |
| 102 | + loader = YoutubeLoader.from_youtube_url( |
| 103 | + url, |
| 104 | + add_video_info=True, # タイトルや再生数も取得できる |
| 105 | + language=["en", "ja"], # 英語→日本語の優先順位で字幕を取得 |
| 106 | + ) |
| 107 | + res = loader.load() # list of `Document` (page_content, metadata) |
| 108 | + try: |
| 109 | + if res: |
| 110 | + content = res[0].page_content |
| 111 | + title = res[0].metadata["title"] |
| 112 | + return f"Title: {title}\n\n{content}" |
| 113 | + else: |
| 114 | + return None |
| 115 | + except Exception as e: |
| 116 | + logger.error(f"An error occurred: {e}") |
| 117 | + st.write(traceback.format_exc()) |
| 118 | + return None |
| 119 | + |
| 120 | + |
| 121 | +def main(): |
| 122 | + init_page() |
| 123 | + chain = init_chain() |
| 124 | + if url := st.text_input("URL: ", key="input"): |
| 125 | + # clear text input |
| 126 | + is_valid_url = validate_url(url) |
| 127 | + if not is_valid_url: |
| 128 | + st.write("Please input valid url") |
| 129 | + else: |
| 130 | + if content := get_content(url): |
| 131 | + st.markdown("## Summary") |
| 132 | + st.write_stream(chain.stream({"content": content})) |
| 133 | + st.markdown("---") |
| 134 | + st.markdown("## Original Text") |
| 135 | + st.write(content) |
| 136 | + |
| 137 | + |
| 138 | +if __name__ == "__main__": |
| 139 | + main() |
0 commit comments