change

agent.py

@@ -1,3 +1,4 @@
+# v8
 # Develop an AI agent with LangGraph and LangChain 
 # to answer the questions in the "gaia-benchmark/GAIA" dataset.
 
@@ -157,10 +158,10 @@ def _convert_to_wav_mono16k(src_path: str) -> str:
     return out
 
 # ==== NEW: URL helpers ====
-_URL_RE = re.compile(r'https?://\S+')
-
 def _extract_urls(text: str) -> List[str]:
-    return _URL_RE.findall(text or "")
+    raw = re.findall(r'https?://\S+', text)
+    urls = [re.sub(r'[)\]\}>\'",;:.!?]+$', '', u) for u in raw]
+    return urls
 
 
 # ----------------------Tools ----------------------
@@ -362,49 +363,43 @@ def wikipedia_lookup(query: str, sentences: int = 4) -> Dict[str, Any]:
         return {"title":"", "url":"", "summary": f"[wikipedia import error: {e}]"}
 
 @tool
-def youtube_get_transcript(url_or_id: str, prefer_langs: List[str] | None = None) -> str:
+def youtube_get_transcript(url_or_id: str, prefer_langs: List[str] = ["en"] ) -> str:
     """
     Get YouTube transcript via API (no download). Returns plain text.
     """
     print('try to get youtube video transcript')
     try:
-        prefer_langs = prefer_langs or ["en", "en-US", "en-GB", "auto"]
         vid = url_or_id
-        print("vid: ", vid)
+        print("url_or_id: ", url_or_id)
         if "youtube.com" in url_or_id or "youtu.be" in url_or_id:
             u = urlparse(url_or_id)
             if u.netloc.endswith("youtu.be"):
                 vid = u.path.lstrip("/")
             else:
                 vid = parse_qs(u.query).get("v", [""])[0]
+
+        print("vid: ", vid)
         ytt_api = YouTubeTranscriptApi()
         trs_list = ytt_api.list(vid)
+
         # choose first matching language
         for lang in prefer_langs:
-            try:
-                trs = trs_list.find_transcript([lang])
-                chunks = trs.fetch()
-                print("transcript from youtube website?")
-                print(" ".join([c["text"] for c in chunks if c.get("text")]).strip())
-                return " ".join([c["text"] for c in chunks if c.get("text")]).strip()
-            except Exception:
-                continue
-        # fallback: first any transcript
-        trs = list(trs_list)[0]
-        chunks = trs.fetch()
-        print("transcript from youtube website?")
-        print(" ".join([c["text"] for c in chunks if c.get("text")]).strip())
-        return " ".join([c["text"] for c in chunks if c.get("text")]).strip()
+            trs = trs_list.find_transcript([lang])
+            # trs = list(trs_list)[0]
+            chunks = trs.fetch()
+            print("transcript from youtube website?")
+            transcript_en = " ".join([c.text for c in chunks]).strip()
+            return transcript_en
     except (TranscriptsDisabled, NoTranscriptFound):
         return "[no captions available]"
-    except Exception as e:
-        return f"[youtube transcript error: {e}]"
+
 
 @tool
 def youtube_transcribe_audio(url: str, model_size: str = "base") -> str:
     """
     Download YouTube audio (yt-dlp) and transcribe with Whisper.
     """
+    print("Download YouTube audio (yt-dlp) and transcribe with Whisper.")
     tmpdir = tempfile.mkdtemp(prefix="gaia_yt_")
     outfile = os.path.join(tmpdir, "%(id)s.%(ext)s")
 
@@ -422,6 +417,7 @@ def youtube_transcribe_audio(url: str, model_size: str = "base") -> str:
         # convert & transcribe
         wav = _convert_to_wav_mono16k(path)
         txt = transcribe_audio.invoke({"path": wav, "model_size": model_size})
+        print('caption by Whisper:', txt)
         return txt
     except Exception as e:
         return f"[youtube download/transcribe error: {e}]"
@@ -620,7 +616,8 @@ def solve_text_only_node(state: "AgentState") -> "AgentState":
     ev_text = _summarize_evidence(evidence) or "(none)"
 
     # LLM (text-only). Swap model as you like.
-    llm = ChatOpenAI(model="gpt-4o-mini", temperature=0)
+    # llm = ChatOpenAI(model="gpt-4o-mini", temperature=0)
+    llm = ChatOpenAI(model="gpt-4o", temperature=0)
 
     sys = SystemMessage(content=(
         "You solve GAIA tasks. Use careful step-by-step reasoning but keep it concise.\n"
@@ -799,7 +796,7 @@ def crawl_node(state: AgentState) -> AgentState:
         if _is_youtube(u):
             print("is_youtube? ", _is_youtube(u))
             cap = youtube_get_transcript.invoke({"url_or_id": u})
-            print('cap: ', cap)
+            print('caption: ', cap)
             if cap and not cap.startswith("[no captions"):
                 ev.append({"kind":"doc_text","text":cap,"path":None,
                            "meta":{"source":"youtube","title": title, "url":u,"mime":"text/plain"}})

app.py

@@ -1,3 +1,4 @@
+# v8
 import os
 import gradio as gr
 import requests