remove array types from printed tool results

vision_agent/agent/vision_agent.py

@@ -223,14 +223,16 @@ def pick_plan(
  user_req = chat[-1]["content"]
  context = USER_REQ.format(user_request=user_req)
  # because the tool picker model gets the image as well, we have to be careful with
- # how much text we send it, so we truncate the tool output to 500 characters
+ # how much text we send it, so we truncate the tool output to 20,000 characters
  prompt = PICK_PLAN.format(
  context=context,
  plans=format_plans(plans),
- tool_output=tool_output_str[:500],
+ tool_output=tool_output_str[:20_000],
  )
  chat[-1]["content"] = prompt
  best_plan = extract_json(model(chat))
+ if verbosity >= 1:
+ _LOGGER.info(f"Best plan:\n{best_plan}")
  return best_plan["best_plan"], tool_output_str
 
 

vision_agent/agent/vision_agent_prompts.py

@@ -66,7 +66,7 @@
 
 **Instructions**:
 1. Write a program to load the media and call each tool and save it's output.
-2. Create a dictionary where the keys are the tool name and the values are the tool outputs.
+2. Create a dictionary where the keys are the tool name and the values are the tool outputs. Remove any array types from the printed dictionary.
 3. Print this final dictionary.
 
 **Example**:
@@ -87,6 +87,7 @@
 owl_v2_out = owl_v2("person", image)
 florencev2_out = florencev2_object_detection(image)
 loca_out = loca_zero_shot_counting(image)
+loca_out = loca_out["count"]
 final_out = {{"owl_v2": owl_v2_out, "florencev2_object_detection": florencev2_out, "loca_zero_shot_counting": loca_out}}
 print(final_out)
 ```