commit code

llm/generate_daily_article.py

@@ -0,0 +1,110 @@
+import json
+from datetime import datetime, timedelta, timezone
+import re
+from typing import Any, Dict, List, Optional
+
+from openai import OpenAI
+
+from config.settings import settings
+from llm import prompt as prompts
+from utils.logger import logger
+from llm.prompts.daily_article_prompt import PROMPT_DAILY_ARTICLE
+
+
+BASE_URL = "https://dashscope.aliyuncs.com/compatible-mode/v1"
+MODEL = "deepseek-v3.2"
+
+
+def _make_client() -> OpenAI:
+    return OpenAI(api_key=settings.DASHSCOPE_API_KEY, base_url=BASE_URL)
+
+
+def _call_model(system_prompt: Optional[str], user_prompt: str, stream: bool = False, enable_search: bool = False) -> Any:
+    client = _make_client()
+    messages = []
+    if system_prompt:
+        messages.append({"role": "system", "content": system_prompt})
+    messages.append({"role": "user", "content": user_prompt})
+
+    # Non-streaming call for simplicity
+    resp = client.chat.completions.create(model=MODEL, messages=messages, stream=stream, extra_body={"enable_search": enable_search})
+    # When stream=False the SDK typically returns a full object; content location may vary.
+    # We'll try common access patterns.
+    try:
+        # OpenAI-compatible: resp.choices[0].message.content
+        return resp.choices[0].message.content
+    except Exception:
+        try:
+            # fallback: resp.choices[0].text
+            return resp.choices[0].text
+        except Exception:
+            # As last resort, return raw resp
+            return resp
+
+
+def _extract_json(text: str) -> str:
+    """Attempt to extract the first JSON object/array from text."""
+    if not isinstance(text, str):
+        raise ValueError("Expected text to be str")
+    # Find first '[' or '{'
+    start_idx = None
+    for i, ch in enumerate(text):
+        if ch in "[{":
+            start_idx = i
+            break
+    if start_idx is None:
+        raise ValueError("No JSON object/array found in text")
+
+    # Try to find a matching closing bracket by scanning and counting
+    stack = []
+    for j in range(start_idx, len(text)):
+        ch = text[j]
+        if ch in "{[":
+            stack.append(ch)
+        elif ch in "]}":
+            if not stack:
+                continue
+            opening = stack.pop()
+            if (opening == "{" and ch != "}") or (opening == "[" and ch != "]"):
+                # mismatched, continue
+                continue
+            if not stack:
+                return text[start_idx : j + 1]
+
+    # Fallback: try regex to capture last '}' or ']' occurrence
+    m = re.search(r"(\{.*\}|\[.*\])", text, re.S)
+    if m:
+        return m.group(1)
+    raise ValueError("Could not extract JSON from model output")
+
+
+def _parse_json_safe(text: str) -> Any:
+    try:
+        return json.loads(text)
+    except Exception:
+        # try to extract JSON substring
+        jtext = _extract_json(text)
+        return json.loads(jtext)
+
+
+def generate_daily_article() -> List[Dict[str, Any]]:
+    """Call PROMPT_DAILY_ARTICLE to generate a daily article."""
+    logger.debug(f"prompt for generate_daily_article:\n{PROMPT_DAILY_ARTICLE}")
+
+    content = _call_model(system_prompt=None, user_prompt=PROMPT_DAILY_ARTICLE, enable_search=True)
+    logger.debug(f"raw output from generate_daily_article:\n{content}")
+    if isinstance(content, (dict, list)):
+        return content
+    text = content if isinstance(content, str) else str(content)
+    data = _parse_json_safe(text)
+    logger.debug(f"result for generate_daily_article:\n{data}")
+    return data
+
+
+
+
+
+if __name__ == "__main__":
+    content = generate_daily_article()
+    article = content["阶段4_今日文章"]["文章正文"]
+    print(article)