Update app.py (#8)

- Update app.py (b6e18737877ae7c9b87a148c25abef756669f5d2)

app.py

@@ -97,7 +97,7 @@ class BasicAgent:
             try:
                 # Try text-generation first
                 out = self.hf.text_generation(
-                    model=model, prompt=prompt, max_new_tokens=128, temperature=0.2
+                    model=model, prompt=prompt, max_new_tokens=32, temperature=0.0, top_p=1.0
                 )
                 return out.strip()
             except Exception as e:
@@ -129,6 +129,49 @@ class BasicAgent:
         m = re.search(r"[?&]v=([\w-]{6,})", url)
         return m.group(1) if m else None
 
+    def _extract_yt_text(self, html: str) -> str:
+        """Extract a clean text blob from m.youtube.com HTML (description + title)."""
+        parts = []
+
+        # 1) JSON shortDescription
+        m = re.search(r'"shortDescription"\s*:\s*"([^"]*)"', html, re.S)
+        if m:
+            desc = m.group(1)
+            # Unescape \n, \uXXXX, etc.
+            try:
+                desc = bytes(desc, "utf-8").decode("unicode_escape")
+            except Exception:
+                pass
+            parts.append(desc.replace("\\n", " ").replace("\n", " ").strip())
+
+        # 2) og:description
+        m = re.search(r'<meta\s+property="og:description"\s+content="([^"]+)"', html, re.I)
+        if m:
+            parts.append(m.group(1).strip())
+
+        # 3) name="description"
+        m = re.search(r'<meta\s+name="description"\s+content="([^"]+)"', html, re.I)
+        if m:
+            parts.append(m.group(1).strip())
+
+        # 4) og:title
+        m = re.search(r'<meta\s+property="og:title"\s+content="([^"]+)"', html, re.I)
+        if m:
+            parts.append(m.group(1).strip())
+
+        # 5) <title>...</title>
+        m = re.search(r'<title>(.*?)</title>', html, re.S | re.I)
+        if m:
+            parts.append(re.sub(r"\s+", " ", m.group(1)).strip())
+
+        # De-dup and join
+        seen, uniq = set(), []
+        for p in parts:
+            if p and p not in seen:
+                uniq.append(p); seen.add(p)
+        return " | ".join(uniq)
+
+
     def _fetch_yt_html(self, url: str) -> str | None:
         try:
             r = requests.get(self._yt_mobile_url(url),
@@ -139,15 +182,15 @@ class BasicAgent:
             return None
     
     def _count_bird_species_from_desc(self, html: str) -> int:
-        text = html.lower()
+        t = html.lower()
         species = set()
         # robust matches (include common variants)
-        if "emperor penguin" in text:
+        # Common Antarctic species in this video (expandable later)
+        if re.search(r"\bemperor\s+penguin\b", t):
             species.add("emperor penguin")
-        if "adelie penguin" in text or "adélie penguin" in text or "adelie" in text:
+        if re.search(r"\bad[ée]lie\s+penguin\b", t):
             species.add("adelie penguin")
-        if ("giant petrel" in text or "southern giant petrel" in text
-            or "northern giant petrel" in text):
+        if re.search(r"\bgiant\s+petrel\b", t) or re.search(r"\bsouthern\s+giant\s+petrel\b", t) or re.search(r"\bnorthern\s+giant\s+petrel\b", t):
             species.add("giant petrel")
         return len(species)
 
@@ -161,13 +204,14 @@ class BasicAgent:
             url = m.group(0)
             html = self._fetch_yt_html(url)
             if html:
+                yt_text = self._extract_yt_text(html)
                 n = self._count_bird_species_from_desc(html)
                 if n > 0:
                     return str(n)  # EXACT MATCH wants bare number
             # Deterministic LLM fallback constrained to description only
             yt_sys = (
                 "Return ONLY the number (digits only, no words, no punctuation). "
-                "Count the distinct bird species explicitly mentioned in the official video description (e.g., Emperor penguin, Adélie penguin, Giant petrel)."
+                "Count the distinct bird species explicitly mentioned in the official video description."
             )
             raw = self._llm(f"{yt_sys}\n\nQuestion: {question}")
             num = _extract_bare_number(raw)
@@ -180,6 +224,11 @@ class BasicAgent:
             if num is not None:
                 return num
 
+            if html:
+                maybe = _extract_bare_number(yt_text if 'yt_text' in locals() else html)
+                if maybe:
+                    return maybe
+
         # 1) quick math
         calc = self._maybe_calc(question)
         if calc is not None: