Restored litellm token counter.

Author: emeryberger

src/coverup/llm.py

@@ -82,23 +82,12 @@ def compute_cost(usage: dict, model_name: str) -> float | None:
     return None
 
 _token_encoding_cache: dict[str, T.Any] = dict()
+
 def count_tokens(model_name: str, completion: dict):
     """Counts the number of tokens in a chat completion request."""
-    import tiktoken
-
-    if "anthropic" in model_name:
-        return 1
-
-    if not (encoding := _token_encoding_cache.get(model_name)):
-        model = model_name
-        if model_name.startswith('openai/'):
-            model = model_name[7:]
-
-        encoding = _token_encoding_cache[model_name] = tiktoken.encoding_for_model(model)
+    from litellm import token_counter
 
-    count = 0
-    for m in completion['messages']:
-        count += len(encoding.encode(m['content']))
+    count = token_counter(model=model_name, messages=completion['messages'])
 
     return count
 
@@ -312,4 +301,4 @@ async def chat(self, messages: list, *, ctx: T.Optional[object] = None) -> dict
                 })
 
         self._log_msg(ctx, f"Too many function call requests, giving up")
-        return None
+        return None