Fix summarize_page in a text-only context, and for unknown models. (#…

…5388)

WebSurfer's summarize_page was failing when the model was text-only, or
unknown.

python/packages/autogen-ext/src/autogen_ext/agents/web_surfer/_multimodal_web_surfer.py

@@ -65,6 +65,8 @@
 from ._types import InteractiveRegion, UserContent
 from .playwright_controller import PlaywrightController
 
+DEFAULT_CONTEXT_SIZE = 128000
+
 
 class MultimodalWebSurferConfig(BaseModel):
     name: str
@@ -855,35 +857,50 @@ async def _summarize_page(
         buffer = ""
         # for line in re.split(r"([\r\n]+)", page_markdown):
         for line in page_markdown.splitlines():
-            message = UserMessage(
-                # content=[
+            trial_message = UserMessage(
                 content=prompt + buffer + line,
-                #    ag_image,
-                # ],
                 source=self.name,
             )
 
-            remaining = self._model_client.remaining_tokens(messages + [message])
-            if remaining > self.SCREENSHOT_TOKENS:
-                buffer += line
-            else:
+            try:
+                remaining = self._model_client.remaining_tokens(messages + [trial_message])
+            except KeyError:
+                # Use the default if the model isn't found
+                remaining = DEFAULT_CONTEXT_SIZE - self._model_client.count_tokens(messages + [trial_message])
+
+            if self._model_client.model_info["vision"] and remaining <= 0:
                 break
 
+            if self._model_client.model_info["vision"] and remaining <= self.SCREENSHOT_TOKENS:
+                break
+
+            buffer += line
+
         # Nothing to do
         buffer = buffer.strip()
         if len(buffer) == 0:
             return "Nothing to summarize."
 
         # Append the message
-        messages.append(
-            UserMessage(
-                content=[
-                    prompt + buffer,
-                    ag_image,
-                ],
-                source=self.name,
+        if self._model_client.model_info["vision"]:
+            # Multimodal
+            messages.append(
+                UserMessage(
+                    content=[
+                        prompt + buffer,
+                        ag_image,
+                    ],
+                    source=self.name,
+                )
+            )
+        else:
+            # Text only
+            messages.append(
+                UserMessage(
+                    content=prompt + buffer,
+                    source=self.name,
+                )
             )
-        )
 
         # Generate the response
         response = await self._model_client.create(messages, cancellation_token=cancellation_token)