Create regexs.py

regexs.py

@@ -0,0 +1,54 @@
+import re
+
+
+def extract_last_ai_text(messages: list) -> str:
+    """Récupère le dernier message assistant avec du texte (évite fin de run ambiguë)."""
+    from langchain_core.messages import AIMessage
+
+    for m in reversed(messages):
+        if not isinstance(m, AIMessage):
+            continue
+        c = m.content
+        if isinstance(c, list):
+            parts = []
+            for block in c:
+                if isinstance(block, dict) and block.get("type") == "text":
+                    parts.append(block.get("text", ""))
+                elif isinstance(block, str):
+                    parts.append(block)
+            c = "".join(parts)
+        if isinstance(c, str) and c.strip():
+            return c
+    for m in reversed(messages):
+        c = getattr(m, "content", None)
+        if isinstance(c, list):
+            continue
+        if isinstance(c, str) and c.strip():
+            return c
+    return ""
+
+
+def strip_final_answer_prefix(text: str) -> str:
+    """Retire le préfixe 'FINAL ANSWER:' pour ne garder que la valeur brute à soumettre."""
+    marker = "FINAL ANSWER:"
+    stripped = text.strip()
+    if stripped.upper().startswith(marker):
+        return stripped[len(marker):].strip()
+    return stripped
+
+
+def normalize_gaia_answer(raw: str) -> str:
+    """normalise la sortie du LLM en un format uniforme FINAL ANSWER: <valeur>"""
+    if not raw:
+        return "FINAL ANSWER: "
+    text = raw.strip()
+    marker = "FINAL ANSWER:"
+    idx = text.lower().rfind(marker.lower())
+    if idx != -1:
+        tail = text[idx + len(marker) :].strip()
+        first = tail.split("\n", 1)[0].strip() if tail else ""
+        first = re.sub(r"[*_`]+", "", first).strip()
+        return f"{marker} {first}".strip() if first else f"{marker} "
+    first_line = text.split("\n", 1)[0].strip()
+    first_line = re.sub(r"[*_`]+", "", first_line).strip()
+    return f"{marker} {first_line}"