fix sumarization

app/utils/llm_summarizer.py

@@ -24,13 +24,16 @@ class OpenAICompatSummarizer:
         timeout: Optional[int] = None,
         max_items_per_request: Optional[int] = None,
         max_chars_per_item: Optional[int] = None,
+        max_chars_total: Optional[int] = None,
     ):
         self.api_base = (api_base or os.getenv("LLM_API_BASE") or "https://researchengineering-agi.hf.space").rstrip("/")
         self.api_key = api_key if api_key is not None else os.getenv("LLM_API_KEY", "")
         self.model = model or os.getenv("LLM_MODEL", "gpt-4o-mini")
-        self.timeout = timeout or int(os.getenv("LLM_TIMEOUT", "120"))
-        self.max_items_per_request = max_items_per_request or int(os.getenv("LLM_SUMMARY_BATCH", "8"))
-        self.max_chars_per_item = max_chars_per_item or int(os.getenv("LLM_SUMMARY_MAX_CHARS", "1200"))
+        self.timeout = timeout or int(os.getenv("LLM_TIMEOUT", "300"))
+        # Conservative defaults to avoid large token bursts on slow servers.
+        self.max_items_per_request = max_items_per_request or int(os.getenv("LLM_SUMMARY_BATCH", "2"))
+        self.max_chars_per_item = max_chars_per_item or int(os.getenv("LLM_SUMMARY_MAX_CHARS", "600"))
+        self.max_chars_total = max_chars_total or int(os.getenv("LLM_SUMMARY_MAX_CHARS_TOTAL", "1200"))
         self.enabled = os.getenv("ENABLE_AI_SUMMARIZATION", "true").lower() in {"1", "true", "yes"}
         self.sleep_seconds = float(os.getenv("LLM_SUMMARY_SLEEP_SECONDS", "0"))
 
@@ -54,6 +57,8 @@ class OpenAICompatSummarizer:
         chunks = self._chunked(candidates, self.max_items_per_request)
         for idx, chunk in enumerate(chunks, start=1):
             texts = [text for _, text in chunk]
+            if self.max_chars_total > 0:
+                texts = self._truncate_to_total(texts, self.max_chars_total)
             summaries = self._summarize_chunk(texts, source=source)
             if not summaries:
                 continue
@@ -68,25 +73,11 @@ class OpenAICompatSummarizer:
 
     def _build_input_text(self, item: Dict) -> str:
         title = str(item.get("title", "")).strip()
-        summary = str(item.get("summary_raw", item.get("summary", ""))).strip()
-        extra = str(item.get("content", item.get("text", item.get("description", "")))).strip()
-
-        parts = []
         if title:
-            parts.append(f"Title: {title}")
-        if summary and summary != title:
-            parts.append(f"Summary: {summary}")
-        if extra and extra not in summary and extra not in title:
-            parts.append(f"Details: {extra}")
-
-        combined = "\n".join(parts).strip()
-        if not combined:
-            return ""
-
-        if len(combined) > self.max_chars_per_item:
-            combined = combined[: self.max_chars_per_item].rstrip()
-
-        return combined
+            if len(title) > self.max_chars_per_item:
+                title = title[: self.max_chars_per_item].rstrip()
+            return f"Title: {title}"
+        return ""
 
     def _summarize_chunk(self, texts: List[str], source: Optional[str] = None) -> List[str]:
         system_prompt = (
@@ -154,3 +145,18 @@ class OpenAICompatSummarizer:
         if size <= 0:
             return [items]
         return [items[i : i + size] for i in range(0, len(items), size)]
+
+    def _truncate_to_total(self, texts: List[str], max_total: int) -> List[str]:
+        if max_total <= 0:
+            return texts
+        truncated = []
+        total = 0
+        for text in texts:
+            if total >= max_total:
+                break
+            remaining = max_total - total
+            if len(text) > remaining:
+                text = text[:remaining].rstrip()
+            truncated.append(text)
+            total += len(text)
+        return truncated