[Refactor] litellm/init.py: lazy load default encoding from client decorator (#18059)

litellm/_lazy_imports.py

@@ -1,10 +1,31 @@
-from typing import Any, cast
+from typing import Any, Optional, cast
 import sys
 
 def _get_litellm_globals() -> dict:
     """Helper to get the globals dictionary of the litellm module."""
     return sys.modules["litellm"].__dict__
 
+# Lazy loader for default encoding to avoid importing tiktoken at module import time
+_default_encoding: Optional[Any] = None
+
+
+def _get_default_encoding() -> Any:
+    """
+    Lazily load and cache the default OpenAI encoding.
+    
+    This avoids importing `litellm.litellm_core_utils.default_encoding` (and thus tiktoken)
+    at `litellm` import time. The encoding is cached after the first import.
+    
+    This is used internally by utils.py functions that need the encoding but shouldn't
+    trigger its import during module load.
+    """
+    global _default_encoding
+    if _default_encoding is None:
+        from litellm.litellm_core_utils.default_encoding import encoding
+
+        _default_encoding = encoding
+    return _default_encoding
+
 # Cost calculator names that support lazy loading via _lazy_import_cost_calculator
 COST_CALCULATOR_NAMES = (
     "completion_cost",

litellm/utils.py

@@ -1,3 +1,6 @@
+# from __future__ import annotations must be the first non-comment statement
+from __future__ import annotations
+
 # +-----------------------------------------------+
 # |                                               |
 # |           Give Feedback / Get Help            |
@@ -96,11 +99,11 @@ from litellm.litellm_core_utils.core_helpers import (
     process_response_headers,
 )
 from litellm.litellm_core_utils.credential_accessor import CredentialAccessor
-from litellm.litellm_core_utils.default_encoding import encoding
 from litellm.litellm_core_utils.dot_notation_indexing import (
     delete_nested_value,
     is_nested_path,
 )
+from litellm._lazy_imports import _get_default_encoding
 from litellm.litellm_core_utils.exception_mapping_utils import (
     _get_response_headers,
     exception_type,
@@ -260,12 +263,16 @@ from litellm.llms.base_llm.base_utils import (
     BaseLLMModelInfo,
     type_to_response_format_param,
 )
+
+if TYPE_CHECKING:
+    # Heavy types that are only needed for type checking; avoid importing
+    # their modules at runtime during `litellm` import.
+    from litellm.llms.base_llm.files.transformation import BaseFilesConfig
 from litellm.llms.base_llm.batches.transformation import BaseBatchesConfig
 from litellm.llms.base_llm.chat.transformation import BaseConfig
 from litellm.llms.base_llm.completion.transformation import BaseTextCompletionConfig
 from litellm.llms.base_llm.containers.transformation import BaseContainerConfig
 from litellm.llms.base_llm.embedding.transformation import BaseEmbeddingConfig
-from litellm.llms.base_llm.files.transformation import BaseFilesConfig
 from litellm.llms.base_llm.image_edit.transformation import BaseImageEditConfig
 from litellm.llms.base_llm.image_generation.transformation import (
     BaseImageGenerationConfig,
@@ -293,6 +300,7 @@ from .caching.caching import (
     RedisSemanticCache,
     S3Cache,
 )
+
 from .exceptions import (
     APIConnectionError,
     APIError,
@@ -1752,7 +1760,7 @@ def _select_tokenizer_helper(model: str) -> SelectTokenizerResponse:
 
 
 def _return_openai_tokenizer(model: str) -> SelectTokenizerResponse:
-    return {"type": "openai_tokenizer", "tokenizer": encoding}
+    return {"type": "openai_tokenizer", "tokenizer": _get_default_encoding()}
 
 
 def _return_huggingface_tokenizer(model: str) -> Optional[SelectTokenizerResponse]:
@@ -5842,7 +5850,7 @@ def prompt_token_calculator(model, messages):
         anthropic_obj = Anthropic()
         num_tokens = anthropic_obj.count_tokens(text)  # type: ignore
     else:
-        num_tokens = len(encoding.encode(text))
+        num_tokens = len(_get_default_encoding().encode(text))
     return num_tokens
 
 
@@ -8187,9 +8195,6 @@ def extract_duration_from_srt_or_vtt(srt_or_vtt_content: str) -> Optional[float]
     return max(durations) if durations else None
 
 
-import httpx
-
-
 def _add_path_to_api_base(api_base: str, ending_path: str) -> str:
     """
     Adds an ending path to an API base URL while preventing duplicate path segments.