fixed plan testing prompt

vision_agent/agent/vision_agent_coder.py

@@ -138,6 +138,7 @@ def pick_plan(
  tool_info: str,
  model: LMM,
  code_interpreter: CodeInterpreter,
+ media: List[str],
  log_progress: Callable[[Dict[str, Any]], None],
  verbosity: int = 0,
  max_retries: int = 3,
@@ -155,7 +156,7 @@ def pick_plan(
 
  plan_str = format_plans(plans)
  prompt = TEST_PLANS.format(
- docstring=tool_info, plans=plan_str, previous_attempts=""
+ docstring=tool_info, plans=plan_str, previous_attempts="", media=media
  )
 
  code = extract_code(model(prompt))
@@ -186,6 +187,7 @@ def pick_plan(
  previous_attempts=PREVIOUS_FAILED.format(
  code=code, error=tool_output.text()
  ),
+ media=media,
  )
  code = extract_code(model(prompt))
  log_progress(
@@ -718,6 +720,7 @@ def chat_with_workflow(
  tool_infos["all"],
  self.coder,
  code_interpreter,
+ media_list,
  self.log_progress,
  verbosity=self.verbosity,
  )

vision_agent/agent/vision_agent_coder_prompts.py

@@ -66,8 +66,9 @@
 
 **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. Remove any array types from the printed dictionary.
-3. Print this final dictionary.
+2. Create a dictionary where the keys are the tool name and the values are the tool outputs. Remove numpy arrays from the printed dictionary.
+3. Your test case MUST run only on the given images which are {media}
+4. Print this final dictionary.
 
 **Example**:
 plan1: