Modifying SPO's README.md

examples/spo/README.md

@@ -21,7 +21,7 @@ Configure LLM parameters in `config/config2.yaml` (see `examples/aflow/config2.e
 Create a Iteration template file `metagpt/ext/spo/settings/task_name.yaml`:
 ```yaml
 prompt: |
-  solve question.
+  Please solve the following problem.
 
 requirements: |
   ...
@@ -40,6 +40,14 @@ faq:
       ...
 ```
 
+Notes:
+- `prompt`: Initial prompt for iteration
+- `requirements`: Desired effects/outcomes (e.g., generate more thinking, use more humorous language)
+- `count`: Target word count for the generated prompt (e.g., 50). Set to None for no limit
+- `faq`: QA pairs used for iteration, can include appropriate number of pairs (typically 3)
+  - `question`: Questions from the dataset used for iteration
+  - `answer`: Corresponding answers. Can contain desired thinking patterns or responses instead of actual answers, or can be left empty. See `metagpt/ext/spo/settings/Navigate.yaml` for reference
+
 ### 3. Implement the Optimizer 🔧
 
 Use `metagpt/ext/spo/optimize.py` to execute:

examples/spo/optimize.py

@@ -7,32 +7,32 @@ def parse_args():
     parser = argparse.ArgumentParser(description='SPO Optimizer CLI')
 
     # LLM parameter
-    parser.add_argument('--optimize-model', type=str, default='claude-3-5-sonnet-20240620',
-                        help='Model for optimization')
-    parser.add_argument('--optimize-temperature', type=float, default=0.7,
-                        help='Temperature for optimization')
-    parser.add_argument('--evaluate-model', type=str, default='gpt-4o-mini',
-                        help='Model for evaluation')
-    parser.add_argument('--evaluate-temperature', type=float, default=0.3,
-                        help='Temperature for evaluation')
-    parser.add_argument('--execute-model', type=str, default='gpt-4o-mini',
-                        help='Model for execution')
-    parser.add_argument('--execute-temperature', type=float, default=0,
-                        help='Temperature for execution')
+    parser.add_argument('--opt-model', type=str, default='claude-3-5-sonnet-20240620',
+                       help='Model for optimization')
+    parser.add_argument('--opt-temp', type=float, default=0.7,
+                       help='Temperature for optimization')
+    parser.add_argument('--eval-model', type=str, default='gpt-4o-mini',
+                       help='Model for evaluation')
+    parser.add_argument('--eval-temp', type=float, default=0.3,
+                       help='Temperature for evaluation')
+    parser.add_argument('--exec-model', type=str, default='gpt-4o-mini',
+                       help='Model for execution')
+    parser.add_argument('--exec-temp', type=float, default=0,
+                       help='Temperature for execution')
 
     # Optimizer parameter
     parser.add_argument('--workspace', type=str, default='workspace',
-                        help='Path for optimized output')
+                       help='Path for optimized output')
     parser.add_argument('--initial-round', type=int, default=1,
-                        help='Initial round number')
+                       help='Initial round number')
     parser.add_argument('--max-rounds', type=int, default=10,
-                        help='Maximum number of rounds')
+                       help='Maximum number of rounds')
     parser.add_argument('--template', type=str, default='Poem.yaml',
-                        help='Template file name')
+                       help='Template file name')
     parser.add_argument('--name', type=str, default='Poem',
-                        help='Project name')
+                       help='Project name')
     parser.add_argument('--no-iteration', action='store_false', dest='iteration',
-                        help='Disable iteration mode')
+                       help='Disable iteration mode')
 
     return parser.parse_args()
 
@@ -42,16 +42,16 @@ def main():
 
     SPO_LLM.initialize(
         optimize_kwargs={
-            "model": args.optimize_model,
-            "temperature": args.optimize_temperature
+            "model": args.opt_model,
+            "temperature": args.opt_temp
         },
         evaluate_kwargs={
-            "model": args.evaluate_model,
-            "temperature": args.evaluate_temperature
+            "model": args.eval_model,
+            "temperature": args.eval_temp
         },
         execute_kwargs={
-            "model": args.execute_model,
-            "temperature": args.execute_temperature
+            "model": args.exec_model,
+            "temperature": args.exec_temp
         }
     )
 

metagpt/ext/spo/settings/Navigate.yaml

@@ -0,0 +1,47 @@
+prompt: |
+  Please think step by step.
+  Ensure the response concludes with the answer in the XML format: 
+  <answer>[Yes or No]</answer>.
+
+requirements: |
+  Must put the final answer at the end with XML. (<answer>(Yes or No)</answer>,such as <answer>Yes</answer>)
+  The provided prompt needs to adapt to all current types of questions.
+
+count: None
+
+faq:
+  - question: |
+        If you follow these instructions, do you return to the starting point? Always face forward. Take 7 steps left. Take 2 steps backward. Take 7 steps backward. Take 7 steps backward. Take 3 steps forward.
+        Options:
+        - Yes
+        - No
+
+    answer: |
+        A lot of thinking and analysis processes.
+        ...
+        Final Answer:
+        <answer>(Yes or No)</answer>
+
+  - question: |
+        If you follow these instructions, do you return to the starting point? Always face forward. Take 6 steps backward. Take 8 steps left. Take 3 steps right. Take 7 steps forward. Take 3 steps right. Take 9 steps right. Take 1 step backward. Take 7 steps left.
+        Options:
+        - Yes
+        - No
+
+    answer: |
+        A lot of thinking and analysis processes.
+        ...
+        Final Answer:
+        <answer>(Yes or No)</answer>
+
+  - question: |
+        If you follow these instructions, do you return to the starting point? Turn left. Turn left. Take 6 steps. Take 3 steps. Turn around. Take 1 step. Take 3 steps. Take 5 steps.
+        Options:
+        - Yes
+        - No
+
+    answer: |
+        A lot of thinking and analysis processes.
+        ...
+        Final Answer:
+        <answer>(Yes or No)</answer>