DeepSeek 专家模式文件上传突然下线，长文档处理怎么办

# pdf_chunk_for_deepseek.py
# 依赖：pip install pdfplumber

import pdfplumber
import textwrap

PDF_PATH = "your_document.pdf"   # ← 改成你的文件路径
CHUNK_SIZE = 3000                 # 每块字符数，留足推理空间
OVERLAP = 200                     # 相邻块重叠字符，防止断句丢失上下文

def extract_text(pdf_path: str) -> str:
    """从 PDF 提取全部文本，按页拼接"""
    full_text = ""
    with pdfplumber.open(pdf_path) as pdf:
        for i, page in enumerate(pdf.pages):
            page_text = page.extract_text() or ""
            full_text += f"\n--- 第 {i+1} 页 ---\n{page_text}"
    return full_text

def chunk_text(text: str, size: int, overlap: int) -> list[str]:
    """按字符数切分，相邻块有重叠"""
    chunks = []
    start = 0
    while start < len(text):
        end = start + size
        chunk = text[start:end]
        chunks.append(chunk)
        start = end - overlap   # 回退 overlap 字符
    return chunks

def main():
    raw = extract_text(PDF_PATH)
    chunks = chunk_text(raw, CHUNK_SIZE, OVERLAP)

    for idx, chunk in enumerate(chunks, 1):
        header = f"【文档片段 {idx}/{len(chunks)}】以下是文档的第 {idx} 段内容：\n"
        print(header + chunk)
        print("\n" + "=" * 60 + "\n")

    print(f"共提取 {len(raw)} 字符，切分为 {len(chunks)} 块")
    print("将每块粘贴到 DeepSeek 专家模式对话中即可逐段分析")

if __name__ == "__main__":
    main()

pip install pdfplumber
python pdf_chunk_for_deepseek.py

# deepseek_api_chunk.py
# 依赖：pip install openai

from openai import OpenAI
import pdfplumber

API_KEY = "sk-xxx"                # ← 替换成你的 DeepSeek API Key
PDF_PATH = "your_document.pdf"
CHUNK_SIZE = 3000

client = OpenAI(api_key=API_KEY, base_url="https://api.deepseek.com")

def extract_text(pdf_path):
    with pdfplumber.open(pdf_path) as pdf:
        return "\n".join(p.extract_text() or "" for p in pdf.pages)

def ask_deepseek(chunks: list[str], question: str):
    """逐块送入，最后一块提问"""
    messages = []
    for i, chunk in enumerate(chunks, 1):
        messages.append({
            "role": "user",
            "content": f"文档片段 {i}/{len(chunks)}：\n{chunk}"
        })
        messages.append({
            "role": "assistant",
            "content": f"已收到片段 {i}，请继续。"
        })
    # 最后一条 user 消息替换为真正的提问
    messages[-2]["content"] = (
        f"文档片段 {len(chunks)}/{len(chunks)}：\n{chunks[-1]}\n\n"
        f"基于以上全部文档内容，请回答：{question}"
    )
    # 移除最后一条 assistant 占位
    messages.pop()

    resp = client.chat.completions.create(
        model="deepseek-reasoner",   # 专家模式对应 reasoner
        messages=messages,
    )
    return resp.choices[0].message.content

raw = extract_text(PDF_PATH)
chunks = [raw[i:i+CHUNK_SIZE] for i in range(0, len(raw), CHUNK_SIZE)]

answer = ask_deepseek(chunks, "请总结这份文档的核心结论，并列出 3 个关键风险点")
print(answer)