add caching

.gitignore

@@ -348,3 +348,4 @@ out
 data/mini-husky/basic/
 experiments/transformer-vs-recursive/results/
 experiments/transformer-vs-recursive/figures/
+projects/ai-math-autoformalization/python/.chat_completion_cache/cache.db

projects/ai-math-autoformalization/python/utils.py

@@ -0,0 +1,60 @@
+from openai import OpenAI
+import hashlib
+from diskcache import Cache
+
+cache = Cache(".chat_completion_cache")
+
+class ChatCompletionAPI:
+    def __init__(self, type, model=None):
+        if type == "local":
+            self.client = OpenAI(base_url="http://172.28.6.60:8000/v1") # DO NOT CHANGE THIS LINE
+            self.model = "/local1/models/models--meta-llama--Llama-3.1-70B-Instruct/snapshots/945c8663693130f8be2ee66210e062158b2a9693"  # DO NOT CHANGE THIS LINE
+        elif type == "openai":
+            self.client = OpenAI()
+            self.model = model
+        else:
+            raise ValueError("Invalid type")
+
+    def _generate_cache_key(self, model, messages):
+        messages_str = "".join([msg["role"] + msg["content"] for msg in messages])
+        key_base = f"{model}:{messages_str}"
+        cache_key = hashlib.md5(key_base.encode()).hexdigest()
+        return cache_key
+
+    def _fetch_completion(self, model, messages, use_cache):
+        if use_cache:
+            cache_key = self._generate_cache_key(model, messages)
+            cached_result = cache.get(cache_key)
+            if cached_result:
+                return cached_result
+
+        completion = self.client.chat.completions.create(
+            model=model,
+            messages=messages
+        )
+        result = {
+            "role": completion.choices[0].message.role,
+            "content": completion.choices[0].message.content,
+        }
+        cache.set(cache_key, result)
+        return result
+
+    def chat_completion(self, messages, use_cache=True):
+        return self._fetch_completion(self.model, messages, use_cache)
+
+if __name__ == "__main__":
+    api = ChatCompletionAPI("local")
+
+    messages = [
+        {"role": "system", "content": "You are a helpful assistant."},
+        {"role": "user", "content": "What is the capital of France?"},
+    ]
+    for _ in range(3):
+        completion = api.chat_completion(messages)
+        print(completion["content"])
+
+        messages.append(completion)
+        messages.append({
+            "role": "user",
+            "content": "Tell me more fun facts about this city.",
+        })