Merge pull request #25 from codelion/feat-implement-reread

Feat implement reread

Author: codelion

README.md

@@ -125,6 +125,7 @@ or your own code where you want to use the results from optillm. You can use it
 | CoT with Reflection     | `cot_reflection`   | Implements chain-of-thought reasoning with \<thinking\>, \<reflection> and \<output\> sections |
 | PlanSearch              | `plansearch`       | Implements a search algorithm over candidate plans for solving a problem in natural language   |
 | LEAP                    | `leap`             | Learns task-specific principles from few shot examples                                         |
+| ReRead                  | `re2`              | Implements rereading to improve reasoning by processing queries twice                          |
 
 ## Available Parameters
 
@@ -194,6 +195,7 @@ Authorization: Bearer your_secret_api_key
 
 ## References
 
+- [Re-Reading Improves Reasoning in Large Language Models](https://arxiv.org/abs/2309.06275)
 - [In-Context Principle Learning from Mistakes](https://arxiv.org/abs/2402.05403)
 - [Planning In Natural Language Improves LLM Search For Code Generation](https://arxiv.org/abs/2409.03733)
 - [Self-Consistency Improves Chain of Thought Reasoning in Language Models](https://arxiv.org/abs/2203.11171)

optillm.py

@@ -17,6 +17,7 @@
 from optillm.cot_reflection import cot_reflection
 from optillm.plansearch import plansearch
 from optillm.leap import leap
+from optillm.reread import re2_approach
 
 # Setup logging
 logging.basicConfig(level=logging.INFO, format='%(asctime)s - %(levelname)s - %(message)s')
@@ -69,7 +70,7 @@
 
 # List of known approaches
 known_approaches = ["mcts", "bon", "moa", "rto", "z3", "self_consistency", "pvg", "rstar",
-                    "cot_reflection", "plansearch", "leap"]
+                    "cot_reflection", "plansearch", "leap", "re2"]
 
 # Optional API key configuration to secure the proxy
 @app.before_request
@@ -149,6 +150,8 @@ def proxy():
             final_response = plansearch(system_prompt, initial_query, client, model, n=n)
         elif approach == 'leap':
             final_response = leap(system_prompt, initial_query, client, model)
+        elif approach == 're2':
+            final_response = re2_approach(system_prompt, initial_query, client, model, n=n)
         else:
             raise ValueError(f"Unknown approach: {approach}")
     except Exception as e:

optillm/reread.py

@@ -0,0 +1,43 @@
+import logging
+
+logger = logging.getLogger(__name__)
+
+def re2_approach(system_prompt, initial_query, client, model, n=1):
+    """
+    Implement the RE2 (Re-Reading) approach for improved reasoning in LLMs.
+    
+    Args:
+    system_prompt (str): The system prompt to be used.
+    initial_query (str): The initial user query.
+    client: The OpenAI client object.
+    model (str): The name of the model to use.
+    n (int): Number of completions to generate.
+    
+    Returns:
+    str or list: The generated response(s) from the model.
+    """
+    logger.info("Using RE2 approach for query processing")
+    
+    # Construct the RE2 prompt
+    re2_prompt = f"{initial_query}\nRead the question again: {initial_query}"
+    
+    messages = [
+        {"role": "system", "content": system_prompt},
+        {"role": "user", "content": re2_prompt}
+    ]
+    
+    try:
+        response = client.chat.completions.create(
+            model=model,
+            messages=messages,
+            n=n
+        )
+        
+        if n == 1:
+            return response.choices[0].message.content.strip()
+        else:
+            return [choice.message.content.strip() for choice in response.choices]
+    
+    except Exception as e:
+        logger.error(f"Error in RE2 approach: {str(e)}")
+        raise