updated max context window default to 100K tokens

Author: AndreFCruz

llm_api_client/_version.py

@@ -1,6 +1,4 @@
 """Module to expose version information.
-
-Resilient to running from source without an installed distribution.
 """
 from importlib import metadata
 

llm_api_client/api_client.py

@@ -23,12 +23,12 @@
 # Default max context window tokens
 DEFAULT_MAX_CONTEXT_TOKENS_ENV_VAR = "DEFAULT_MAX_CONTEXT_TOKENS"
 try:
-    DEFAULT_MAX_CONTEXT_TOKENS = int(os.getenv(DEFAULT_MAX_CONTEXT_TOKENS_ENV_VAR, "20000"))
+    DEFAULT_MAX_CONTEXT_TOKENS = int(os.getenv(DEFAULT_MAX_CONTEXT_TOKENS_ENV_VAR, "100000"))
 except ValueError:
     logging.getLogger(__name__).warning(
         f"Environment variable {DEFAULT_MAX_CONTEXT_TOKENS_ENV_VAR} must be an integer. "
-        "Falling back to 20,000 tokens.")
-    DEFAULT_MAX_CONTEXT_TOKENS = 20_000
+        "Falling back to 100,000 tokens.")
+    DEFAULT_MAX_CONTEXT_TOKENS = 100_000
 
 
 class APIClient:

tests/test_api_client_additional.py

@@ -11,10 +11,10 @@ def test_get_max_context_tokens_fallback_on_error(mock_get_model_info):
     mock_get_model_info.side_effect = RuntimeError("boom")
     client = APIClient()
 
-    # When litellm raises, the client should fall back to default env value (20_000)
+    # When litellm raises, the client should fall back to default env value (100_000)
     max_tokens = client.get_max_context_tokens("any-model")
     assert isinstance(max_tokens, int)
-    assert max_tokens >= 10_000  # conservative lower bound; default is 20k
+    assert max_tokens >= 100_000
 
 
 @patch("litellm.token_counter")