Merge pull request #136 from devchat-ai/remove_litellm_package

use cl100k_base to count tokens

devchat/_cli/__init__.py

@@ -1,4 +1,3 @@
-from .optimize import *
 from .log import log
 from .prompt import prompt
 from .run import run

devchat/config.py

@@ -11,7 +11,7 @@
 class Client(str, Enum):
     OPENAI = "openai"
     ANTHROPIC = "anthropic"
-    LITELLM = "litellm"
+    GENERAL = "general"
 
 
 class ProviderConfig(BaseModel):
@@ -90,7 +90,7 @@ def _load_and_validate_config(self) -> ChatConfig:
                     data['providers'][provider] = OpenAIProviderConfig(**config)
                 elif config['client'] == "anthropic":
                     data['providers'][provider] = AnthropicProviderConfig(**config)
-                elif config['client'] == "litellm":
+                else:
                     data['providers'][provider] = ProviderConfig(**config)
         for model, config in data['models'].items():
             if 'provider' not in config:
@@ -101,7 +101,7 @@ def _load_and_validate_config(self) -> ChatConfig:
                     data['models'][model] = OpenAIModelConfig(**config)
                 elif provider.client == Client.ANTHROPIC:
                     data['models'][model] = AnthropicModelConfig(**config)
-                elif provider.client == Client.LITELLM:
+                else:
                     data['models'][model] = GeneralModelConfig(**config)
 
         return ChatConfig(**data)
@@ -146,7 +146,7 @@ def _create_sample_file(self):
                     api_key=""
                 ),
                 "general": ProviderConfig(
-                    client=Client.LITELLM
+                    client=Client.GENERAL
                 )
             },
             models={

devchat/utils.py

@@ -8,7 +8,6 @@
 import datetime
 import hashlib
 import tiktoken
-from litellm import token_counter
 
 
 encoding = tiktoken.get_encoding("cl100k_base")
@@ -207,11 +206,9 @@ def _count_tokens(encoding_tik: tiktoken.Encoding, string: str) -> int:
         return int(word_count / 0.75)
 
 
-def openai_message_tokens(message: dict, model: str) -> int:
+def openai_message_tokens(messages: dict, model: str) -> int: # pylint: disable=unused-argument
     """Returns the number of tokens used by a message."""
-    if "claude" in model:
-        return token_counter(model=model, text=str(message))
-    return len(encoding.encode(str(message)))
+    return len(encoding.encode(str(messages)))
 
 
 def openai_response_tokens(message: dict, model: str) -> int:

pyproject.toml

@@ -29,7 +29,6 @@ urllib3 = "<2.0"
 importlib-metadata = "^6.8.0"
 pyyaml = "^6.0.1"
 gitpython = "^3.1.32"
-litellm = "^0.1.538"
 anthropic = "^0.3.11"
 
 [tool.poetry.scripts]