hermes
c6fd9a5c77
halogen-qwen3.8-flash-next sometimes uses typographic quotes (“ ”) as JSON string delimiters, which breaks strict json.loads. Add a scanner that normalizes curly-delimited strings to straight quotes while preserving curly quotes used as content inside straight-quoted strings.
132 lines
4.4 KiB
Python
132 lines
4.4 KiB
Python
"""Call an OpenAI-compatible chat completions endpoint to generate jokes."""
|
|
import json
|
|
import re
|
|
|
|
import httpx
|
|
|
|
from .config import settings
|
|
|
|
SYSTEM_PROMPT = (
|
|
"You are a comedian. You will receive the text of Wikipedia's featured article "
|
|
"of the day. Write exactly 5 short, clean, family-friendly jokes inspired by "
|
|
"facts from the article. Vary the style (one-liners, puns, observational). "
|
|
"The jokes must be understandable on their own without reading the article. "
|
|
"Respond with ONLY a JSON array of 5 strings and nothing else. "
|
|
"Example: [\"joke one\", \"joke two\", \"joke three\", \"joke four\", \"joke five\"]"
|
|
)
|
|
|
|
|
|
def _try_parse(candidate: str):
|
|
try:
|
|
return json.loads(candidate)
|
|
except json.JSONDecodeError:
|
|
return None
|
|
|
|
|
|
def _normalize_quotes(text: str) -> str:
|
|
"""Rewrite LLM quote soup into valid JSON.
|
|
|
|
Handles models that use curly quotes (“ ”) as string DELIMITERS and also
|
|
as content inside straight-quoted strings. A scanner tracks which quote
|
|
character opened the current string so content quotes are preserved
|
|
(escaped) instead of breaking the structure.
|
|
"""
|
|
out: list[str] = []
|
|
in_str: str | None = None # '"' or '“'
|
|
i = 0
|
|
n = len(text)
|
|
while i < n:
|
|
c = text[i]
|
|
if in_str is None:
|
|
if c == '"':
|
|
in_str = '"'
|
|
out.append(c)
|
|
elif c == "\u201c": # “ opens a string
|
|
in_str = "\u201c"
|
|
out.append('"')
|
|
else:
|
|
out.append(c)
|
|
else:
|
|
if c == "\\" and i + 1 < n: # keep escape pairs intact
|
|
out.append(c)
|
|
out.append(text[i + 1])
|
|
i += 2
|
|
continue
|
|
if in_str == '"':
|
|
# Straight-delimited: curly quotes are just content.
|
|
out.append(c)
|
|
if c == '"':
|
|
in_str = None
|
|
else: # curly-delimited string
|
|
if c == "\u201d": # ” closes it
|
|
in_str = None
|
|
out.append('"')
|
|
elif c == '"': # raw straight quote inside -> escape
|
|
out.append('\\"')
|
|
else:
|
|
out.append(c)
|
|
i += 1
|
|
return "".join(out)
|
|
|
|
|
|
def _repair_and_parse(text: str):
|
|
"""Parse JSON, repairing common LLM quote mistakes if strict parse fails."""
|
|
try:
|
|
return json.loads(text)
|
|
except json.JSONDecodeError:
|
|
pass
|
|
data = _try_parse(_normalize_quotes(text))
|
|
if data is not None:
|
|
return data
|
|
raise ValueError(f"Could not parse jokes JSON even after repair: {text[:200]!r}")
|
|
|
|
|
|
def _extract_json_array(text: str) -> list[str]:
|
|
"""Pull a JSON array of strings out of a possibly messy LLM response."""
|
|
text = text.strip()
|
|
# Strip markdown fences if present.
|
|
fence = re.search(r"```(?:json)?\s*(.*?)```", text, re.DOTALL)
|
|
if fence:
|
|
text = fence.group(1).strip()
|
|
# Find the outermost [...] span.
|
|
start = text.find("[")
|
|
end = text.rfind("]")
|
|
if start == -1 or end == -1 or end <= start:
|
|
raise ValueError(f"No JSON array found in LLM response: {text[:200]!r}")
|
|
data = _repair_and_parse(text[start : end + 1])
|
|
if not isinstance(data, list):
|
|
raise ValueError("Parsed JSON is not a list")
|
|
jokes = [str(j).strip() for j in data if str(j).strip()]
|
|
if len(jokes) < 5:
|
|
raise ValueError(f"Only {len(jokes)} jokes returned, expected 5")
|
|
return jokes[:5]
|
|
|
|
|
|
def generate_jokes(article_title: str, article_extract: str) -> list[str]:
|
|
url = f"{settings.openai_base_url}/chat/completions"
|
|
headers = {"Content-Type": "application/json"}
|
|
if settings.openai_api_key:
|
|
headers["Authorization"] = f"Bearer {settings.openai_api_key}"
|
|
|
|
user_prompt = (
|
|
f"Today's Wikipedia featured article: \"{article_title}\"\n\n"
|
|
f"Article text:\n{article_extract}\n\n"
|
|
"Now write exactly 5 jokes as a JSON array of 5 strings."
|
|
)
|
|
|
|
body = {
|
|
"model": settings.openai_model,
|
|
"messages": [
|
|
{"role": "system", "content": SYSTEM_PROMPT},
|
|
{"role": "user", "content": user_prompt},
|
|
],
|
|
"temperature": 0.9,
|
|
"max_tokens": 800,
|
|
}
|
|
|
|
resp = httpx.post(url, headers=headers, json=body, timeout=120)
|
|
resp.raise_for_status()
|
|
payload = resp.json()
|
|
content = payload["choices"][0]["message"]["content"]
|
|
return _extract_json_array(content)
|