Spaces:
Running
on
CPU Upgrade
Running
on
CPU Upgrade
File size: 4,222 Bytes
525a9ab 85504aa 525a9ab 85504aa 525a9ab 85504aa 525a9ab 85504aa 525a9ab 85504aa |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 |
import re
from collections import OrderedDict
CITATION_RE = re.compile(r"\[(\d+)\]")
def is_unknown_answer(txt: str) -> bool:
"""Detect 'no answer' / 'reformulate' replies."""
s = (txt or "").lower()
patterns = [
"Je suis navré, je n'ai pas trouvé la réponse",
"Je ne sais pas",
"Je ne comprends pas la question"
"Pourriez-vous reformuler",
"je n'ai pas trouvé d'information pertinente",
]
return any(p in s for p in patterns)
def _extract_cited_indices(text: str, k: int) -> list[int]:
"""Renvoie les indices (1..k) réellement cités dans le texte, sans doublon, ordonnés."""
seen = OrderedDict()
for m in CITATION_RE.finditer(text or ""):
try:
n = int(m.group(1))
if 1 <= n <= k and n not in seen:
seen[n] = True
except Exception:
pass
return list(seen.keys())
def linkify_text_with_sources(text: str, passages: list[dict]) -> str:
"""
Convertit [1], [2]… en vrais liens Markdown vers les sources.
"""
import re
mapping = {}
for i, h in enumerate(passages, start=1):
p = h.get("payload", h) or {}
title = p.get("title") or p.get("url") or f"Source {i}"
url = p.get("url")
mapping[i] = (title, url)
def _sub(m):
idx = int(m.group(1))
title, url = mapping.get(idx, (None, None))
if url:
# simple lien markdown [1](url)
return f"[_[{idx}]_]({url} \"{title}\")"
return m.group(0)
return re.sub(r"\[(\d+)\]", _sub, text)
def _group_sources_md(passages: list[dict], used_idxs: list[int]) -> str:
"""
Construit le markdown groupé :
### 📚 Sources (N)
1. [Titre](url) _(extrait #1, 3)_
2. [Autre](url2) _(extrait #2)_
"""
if not passages:
return "### 📚 Sources (0)\n_(aucune)_"
# Utiliser les indices cités si dispo, sinon tomber sur 1..len(passages)
if not used_idxs:
used_idxs = list(range(1, len(passages) + 1))
# Groupe par (url ou titre normalisé)
groups = [] # [(key, title, url, [idxs])]
key_to_pos = {}
for idx in used_idxs:
p = passages[idx-1]
pl = p.get("payload", p) or {}
title = (pl.get("title") or pl.get("url") or f"Source {idx}").strip()
url = pl.get("url")
key = (url or "").strip().lower() or title.lower()
if key not in key_to_pos:
key_to_pos[key] = len(groups)
groups.append([key, title, url, []])
groups[key_to_pos[key]][3].append(idx)
# Ordonner chaque liste d'indices et construire le markdown
lines = [f"### 📚 Sources ({len(groups)})"] if len(groups) > 1 else [f"### 📚 Source"]
for i, (_, title, url, idxs) in enumerate(groups, start=1):
idxs = sorted(idxs)
idx_txt = ", ".join(map(str, idxs))
label = "extrait" if len(idxs) == 1 else "extraits"
suffix = f" _({label} # {idx_txt})_"
if url:
lines.append(f"{i}. [{title}]({url}){suffix}")
else:
lines.append(f"{i}. {title}{suffix}")
return "\n".join(lines)
# def sources_markdown(passages: list[dict]) -> str:
# if not passages:
# return "### Sources\n_(aucune)_"
# lines = [f"### 📚 Sources ({len(passages)})"]
# for i, h in enumerate(passages, start=1):
# p = h.get("payload", h) or {}
# title = (p.get("title") or p.get("url") or f"Source {i}").strip()
# url = p.get("url")
# score = h.get("score")
# # snippet = (p.get("text") or "").strip().replace("\n", " ")
# # # on coupe le snippet pour pas que ce soit trop long
# # if len(snippet) > 180:
# # snippet = snippet[:180] + "…"
# # ligne principale
# if url:
# line = f"{i}. [{title}]({url})"
# else:
# line = f"{i}. {title}"
# # on ajoute le score et snippet en italique, plus discrets
# if isinstance(score, (int, float)):
# line += f" _(score {score:.3f})_"
# # if snippet:
# # line += f"\n > {snippet}"
# lines.append(line)
# return "\n".join(lines)
|