use qwen3-32B

basic_agent.py

@@ -24,21 +24,21 @@ class BasicAgent:
           self.model_id = model_id
         else:
           #self.model_id = "HuggingFaceTB/SmolLM2-1.7B-Instruct"
-          self.model_id = "Qwen/Qwen3-4B-FP8"
+          #self.model_id = "Qwen/Qwen3-4B-FP8"
+          self.model_id = "Qwen/Qwen3-32B"
 
         # Run locally.
-        self.model = smolagents.TransformersModel(
-          model_id=self.model_id,
-          max_new_tokens=8000,
-          )
+        #self.model = smolagents.TransformersModel(
+        #  model_id=self.model_id,
+        #  max_new_tokens=32000,
+        #  )
 
-        #self.model = smolagents.HfApiModel(
-        #    max_tokens=8000,
-        #    temperature=0.1,
-        #    model_id=self.model_id,
-        #    custom_role_conversions=None,
-        #    )
-        ddg = smolagents.DuckDuckGoSearchTool()
+        self.model = smolagents.HfApiModel(
+            max_tokens=32000,
+            temperature=0.3,
+            model_id=self.model_id,
+            custom_role_conversions=None,
+            )
         self.tools = [smolagents.DuckDuckGoSearchTool(), smolagents.VisitWebpageTool(), smolagents.FinalAnswerTool()]
 
         self.search_agent = smolagents.CodeAgent(

basic_agent_test.py

@@ -8,6 +8,6 @@ LOG = logging.getLogger(__name__)
 
 ba = basic_agent.BasicAgent()
 
-answer = ba("What color is the sky?")
+answer = ba("Who is the 47th president of the united states? If necessary, use a web search to get the most up to date information.")
 
 LOG.warning(f"{answer=}")

requirements.txt

@@ -1,3 +1,4 @@
+compressed-tensors
 gradio
 huggingface_hub[hf_xet]
 requests