Revert "changes"

This reverts commit fa88bc9632.

litellm/__init__.py

@@ -306,7 +306,7 @@ _key_management_settings: KeyManagementSettings = KeyManagementSettings()
 #### PII MASKING ####
 output_parse_pii: bool = False
 #############################################
-from litellm.litellm_core_utils.get_model_cost_map import get_model_cost_map, get_locally_cached_model_cost_map
+from litellm.litellm_core_utils.get_model_cost_map import get_model_cost_map
 
 model_cost = get_model_cost_map(url=model_cost_map_url)
 custom_prompt_dict: Dict[str, dict] = {}

litellm/litellm_core_utils/get_model_cost_map.py

@@ -8,13 +8,16 @@ export LITELLM_LOCAL_MODEL_COST_MAP=True
 ```
 """
 
-from functools import cache
 import os
 
 import httpx
 
-@cache
+
-def get_locally_cached_model_cost_map():
+def get_model_cost_map(url: str):
+    if (
+        os.getenv("LITELLM_LOCAL_MODEL_COST_MAP", False)
+        or os.getenv("LITELLM_LOCAL_MODEL_COST_MAP", False) == "True"
+    ):
         import importlib.resources
         import json
 
@@ -24,14 +27,6 @@ def get_locally_cached_model_cost_map():
             content = json.load(f)
             return content
 
-
-def get_model_cost_map(url: str):
-    if (
-        os.getenv("LITELLM_LOCAL_MODEL_COST_MAP", False)
-        or os.getenv("LITELLM_LOCAL_MODEL_COST_MAP", False) == "True"
-    ):
-        return get_locally_cached_model_cost_map()
-
     try:
         response = httpx.get(
             url, timeout=5

model_prices_and_context_window.json

@@ -6,7 +6,7 @@
         "input_cost_per_token": 0.0000,
         "output_cost_per_token": 0.000,
         "litellm_provider": "one of https://docs.litellm.ai/docs/providers",
-        "mode": "one of: chat, embedding, completion, image_generation, audio_transcription, audio_speech, image_generation, moderation, moderations, rerank",
+        "mode": "one of chat, embedding, completion, image_generation, audio_transcription, audio_speech",
         "supports_function_calling": true,
         "supports_parallel_function_calling": true,
         "supports_vision": true,

tests/code_coverage_tests/bedrock_pricing.py

@@ -191,7 +191,8 @@ def _check_if_model_name_in_pricing(
     input_cost_per_1k_tokens: str,
     output_cost_per_1k_tokens: str,
 ):
-    litellm.model_cost = litellm.get_locally_cached_model_cost_map()
+    os.environ["LITELLM_LOCAL_MODEL_COST_MAP"] = "True"
+    litellm.model_cost = litellm.get_model_cost_map(url="")
 
     for model, value in litellm.model_cost.items():
         if model.startswith(bedrock_model_name):

tests/litellm_utils_tests/test_utils.py

@@ -907,7 +907,8 @@ def test_supports_response_schema(model, expected_bool):
     Should be true for gemini-1.5-pro on google ai studio / vertex ai AND predibase models
     Should be false otherwise
     """
-    litellm.model_cost = litellm.get_locally_cached_model_cost_map()
+    os.environ["LITELLM_LOCAL_MODEL_COST_MAP"] = "True"
+    litellm.model_cost = litellm.get_model_cost_map(url="")
 
     from litellm.utils import supports_response_schema
 
@@ -1065,7 +1066,8 @@ def test_async_http_handler_force_ipv4(mock_async_client):
     "model, expected_bool", [("gpt-3.5-turbo", False), ("gpt-4o-audio-preview", True)]
 )
 def test_supports_audio_input(model, expected_bool):
-    litellm.model_cost = litellm.get_locally_cached_model_cost_map()
+    os.environ["LITELLM_LOCAL_MODEL_COST_MAP"] = "True"
+    litellm.model_cost = litellm.get_model_cost_map(url="")
 
     from litellm.utils import supports_audio_input, supports_audio_output
 
@@ -1163,7 +1165,8 @@ def test_models_by_provider():
     """
     Make sure all providers from model map are in the valid providers list
     """
-    litellm.model_cost = litellm.get_locally_cached_model_cost_map()
+    os.environ["LITELLM_LOCAL_MODEL_COST_MAP"] = "True"
+    litellm.model_cost = litellm.get_model_cost_map(url="")
 
     from litellm import models_by_provider
 
@@ -1481,7 +1484,8 @@ def test_get_valid_models_default(monkeypatch):
 
 
 def test_supports_vision_gemini():
-    litellm.model_cost = litellm.get_locally_cached_model_cost_map()
+    os.environ["LITELLM_LOCAL_MODEL_COST_MAP"] = "True"
+    litellm.model_cost = litellm.get_model_cost_map(url="")
     from litellm.utils import supports_vision
 
     assert supports_vision("gemini-1.5-pro") is True

tests/llm_translation/base_embedding_unit_tests.py

@@ -84,7 +84,8 @@ class BaseLLMEmbeddingTest(ABC):
         litellm.set_verbose = True
         from litellm.utils import supports_embedding_image_input
 
-        litellm.model_cost = litellm.get_locally_cached_model_cost_map()
+        os.environ["LITELLM_LOCAL_MODEL_COST_MAP"] = "True"
+        litellm.model_cost = litellm.get_model_cost_map(url="")
 
         base_embedding_call_args = self.get_base_embedding_call_args()
         if not supports_embedding_image_input(base_embedding_call_args["model"], None):

tests/llm_translation/base_llm_unit_tests.py

@@ -342,7 +342,8 @@ class BaseLLMChatTest(ABC):
         from pydantic import BaseModel
         from litellm.utils import supports_response_schema
 
-        litellm.model_cost = litellm.get_locally_cached_model_cost_map()
+        os.environ["LITELLM_LOCAL_MODEL_COST_MAP"] = "True"
+        litellm.model_cost = litellm.get_model_cost_map(url="")
 
         class TestModel(BaseModel):
             first_response: str
@@ -381,14 +382,16 @@ class BaseLLMChatTest(ABC):
         from pydantic import BaseModel
         from litellm.utils import supports_response_schema
 
-        litellm.model_cost = litellm.get_locally_cached_model_cost_map()
+        os.environ["LITELLM_LOCAL_MODEL_COST_MAP"] = "True"
+        litellm.model_cost = litellm.get_model_cost_map(url="")
 
     @pytest.mark.flaky(retries=6, delay=1)
     def test_json_response_nested_pydantic_obj(self):
         from pydantic import BaseModel
         from litellm.utils import supports_response_schema
 
-        litellm.model_cost = litellm.get_locally_cached_model_cost_map()
+        os.environ["LITELLM_LOCAL_MODEL_COST_MAP"] = "True"
+        litellm.model_cost = litellm.get_model_cost_map(url="")
 
         class CalendarEvent(BaseModel):
             name: str
@@ -435,7 +438,8 @@ class BaseLLMChatTest(ABC):
         from litellm.utils import supports_response_schema
         from litellm.llms.base_llm.base_utils import type_to_response_format_param
 
-        litellm.model_cost = litellm.get_locally_cached_model_cost_map()
+        os.environ["LITELLM_LOCAL_MODEL_COST_MAP"] = "True"
+        litellm.model_cost = litellm.get_model_cost_map(url="")
 
         class CalendarEvent(BaseModel):
             name: str
@@ -556,7 +560,8 @@ class BaseLLMChatTest(ABC):
         litellm.set_verbose = True
         from litellm.utils import supports_vision
 
-        litellm.model_cost = litellm.get_locally_cached_model_cost_map()
+        os.environ["LITELLM_LOCAL_MODEL_COST_MAP"] = "True"
+        litellm.model_cost = litellm.get_model_cost_map(url="")
 
         base_completion_call_args = self.get_base_completion_call_args()
         if not supports_vision(base_completion_call_args["model"], None):
@@ -610,7 +615,8 @@ class BaseLLMChatTest(ABC):
         litellm.set_verbose = True
         from litellm.utils import supports_vision
 
-        litellm.model_cost = litellm.get_locally_cached_model_cost_map()
+        os.environ["LITELLM_LOCAL_MODEL_COST_MAP"] = "True"
+        litellm.model_cost = litellm.get_model_cost_map(url="")
 
         image_url = "https://upload.wikimedia.org/wikipedia/commons/thumb/d/dd/Gfp-wisconsin-madison-the-nature-boardwalk.jpg/2560px-Gfp-wisconsin-madison-the-nature-boardwalk.jpg"
 
@@ -650,7 +656,8 @@ class BaseLLMChatTest(ABC):
         litellm.set_verbose = True
         from litellm.utils import supports_prompt_caching
 
-        litellm.model_cost = litellm.get_locally_cached_model_cost_map()
+        os.environ["LITELLM_LOCAL_MODEL_COST_MAP"] = "True"
+        litellm.model_cost = litellm.get_model_cost_map(url="")
 
         base_completion_call_args = self.get_base_completion_call_args()
         if not supports_prompt_caching(base_completion_call_args["model"], None):
@@ -766,7 +773,8 @@ class BaseLLMChatTest(ABC):
             litellm._turn_on_debug()
             from litellm.utils import supports_function_calling
 
-            litellm.model_cost = litellm.get_locally_cached_model_cost_map()
+            os.environ["LITELLM_LOCAL_MODEL_COST_MAP"] = "True"
+            litellm.model_cost = litellm.get_model_cost_map(url="")
 
             base_completion_call_args = self.get_base_completion_call_args()
             if not supports_function_calling(base_completion_call_args["model"], None):
@@ -864,7 +872,8 @@ class BaseLLMChatTest(ABC):
     async def test_completion_cost(self):
         from litellm import completion_cost
 
-        litellm.model_cost = litellm.get_locally_cached_model_cost_map()
+        os.environ["LITELLM_LOCAL_MODEL_COST_MAP"] = "True"
+        litellm.model_cost = litellm.get_model_cost_map(url="")
 
         litellm.set_verbose = True
         response = await self.async_completion_function(

tests/llm_translation/base_rerank_unit_tests.py

@@ -87,7 +87,8 @@ class BaseLLMRerankTest(ABC):
     @pytest.mark.parametrize("sync_mode", [True, False])
     async def test_basic_rerank(self, sync_mode):
         litellm._turn_on_debug()
-        litellm.model_cost = litellm.get_locally_cached_model_cost_map()
+        os.environ["LITELLM_LOCAL_MODEL_COST_MAP"] = "True"
+        litellm.model_cost = litellm.get_model_cost_map(url="")
         rerank_call_args = self.get_base_rerank_call_args()
         custom_llm_provider = self.get_custom_llm_provider()
         if sync_mode is True:

tests/llm_translation/test_anthropic_completion.py

@@ -491,7 +491,8 @@ class TestAnthropicCompletion(BaseLLMChatTest):
         from pydantic import BaseModel
         from litellm.utils import supports_response_schema
 
-        litellm.model_cost = litellm.get_locally_cached_model_cost_map()
+        os.environ["LITELLM_LOCAL_MODEL_COST_MAP"] = "True"
+        litellm.model_cost = litellm.get_model_cost_map(url="")
 
         class RFormat(BaseModel):
             question: str

tests/llm_translation/test_bedrock_completion.py

@@ -1975,7 +1975,8 @@ def test_bedrock_converse_route():
 
 def test_bedrock_mapped_converse_models():
     litellm.set_verbose = True
-    litellm.model_cost = litellm.get_locally_cached_model_cost_map()
+    os.environ["LITELLM_LOCAL_MODEL_COST_MAP"] = "True"
+    litellm.model_cost = litellm.get_model_cost_map(url="")
     litellm.add_known_models()
     litellm.completion(
         model="bedrock/us.amazon.nova-pro-v1:0",
@@ -2107,7 +2108,8 @@ def test_bedrock_supports_tool_call(model, expected_supports_tool_call):
 
 class TestBedrockConverseChatCrossRegion(BaseLLMChatTest):
     def get_base_completion_call_args(self) -> dict:
-        litellm.model_cost = litellm.get_locally_cached_model_cost_map()
+        os.environ["LITELLM_LOCAL_MODEL_COST_MAP"] = "True"
+        litellm.model_cost = litellm.get_model_cost_map(url="")
         litellm.add_known_models()
         return {
             "model": "bedrock/us.anthropic.claude-3-5-sonnet-20241022-v2:0",
@@ -2135,7 +2137,8 @@ class TestBedrockConverseChatCrossRegion(BaseLLMChatTest):
         """
         Test if region models info is correctly used for cost calculation. Using the base model info for cost calculation.
         """
-        litellm.model_cost = litellm.get_locally_cached_model_cost_map()
+        os.environ["LITELLM_LOCAL_MODEL_COST_MAP"] = "True"
+        litellm.model_cost = litellm.get_model_cost_map(url="")
         bedrock_model = "us.anthropic.claude-3-5-sonnet-20241022-v2:0"
         litellm.model_cost.pop(bedrock_model, None)
         model = f"bedrock/{bedrock_model}"
@@ -2152,7 +2155,8 @@ class TestBedrockConverseChatCrossRegion(BaseLLMChatTest):
 
 class TestBedrockConverseChatNormal(BaseLLMChatTest):
     def get_base_completion_call_args(self) -> dict:
-        litellm.model_cost = litellm.get_locally_cached_model_cost_map()
+        os.environ["LITELLM_LOCAL_MODEL_COST_MAP"] = "True"
+        litellm.model_cost = litellm.get_model_cost_map(url="")
         litellm.add_known_models()
         return {
             "model": "bedrock/anthropic.claude-3-5-sonnet-20240620-v1:0",
@@ -2321,7 +2325,8 @@ def test_bedrock_nova_topk(top_k_param):
 def test_bedrock_cross_region_inference(monkeypatch):
     from litellm.llms.custom_httpx.http_handler import HTTPHandler
 
-    litellm.model_cost = litellm.get_locally_cached_model_cost_map()
+    monkeypatch.setenv("LITELLM_LOCAL_MODEL_COST_MAP", "True")
+    litellm.model_cost = litellm.get_model_cost_map(url="")
     litellm.add_known_models()
 
     litellm.set_verbose = True

tests/llm_translation/test_openai_o1.py

@@ -29,7 +29,8 @@ async def test_o1_handle_system_role(model):
     from openai import AsyncOpenAI
     from litellm.utils import supports_system_messages
 
-    litellm.model_cost = litellm.get_locally_cached_model_cost_map()
+    os.environ["LITELLM_LOCAL_MODEL_COST_MAP"] = "True"
+    litellm.model_cost = litellm.get_model_cost_map(url="")
 
     litellm.set_verbose = True
 
@@ -82,7 +83,8 @@ async def test_o1_handle_tool_calling_optional_params(
     from litellm.utils import ProviderConfigManager
     from litellm.types.utils import LlmProviders
 
-    litellm.model_cost = litellm.get_locally_cached_model_cost_map()
+    os.environ["LITELLM_LOCAL_MODEL_COST_MAP"] = "True"
+    litellm.model_cost = litellm.get_model_cost_map(url="")
 
     config = ProviderConfigManager.get_provider_chat_config(
         model=model, provider=LlmProviders.OPENAI
@@ -188,7 +190,8 @@ class TestOpenAIO3(BaseOSeriesModelsTest, BaseLLMChatTest):
 
 def test_o1_supports_vision():
     """Test that o1 supports vision"""
-    litellm.model_cost = litellm.get_locally_cached_model_cost_map()
+    os.environ["LITELLM_LOCAL_MODEL_COST_MAP"] = "True"
+    litellm.model_cost = litellm.get_model_cost_map(url="")
     for k, v in litellm.model_cost.items():
         if k.startswith("o1") and v.get("litellm_provider") == "openai":
             assert v.get("supports_vision") is True, f"{k} does not support vision"

tests/llm_translation/test_rerank.py

@@ -274,7 +274,8 @@ class TestLogger(CustomLogger):
 
 @pytest.mark.asyncio()
 async def test_rerank_custom_callbacks():
-    litellm.model_cost = litellm.get_locally_cached_model_cost_map()
+    os.environ["LITELLM_LOCAL_MODEL_COST_MAP"] = "True"
+    litellm.model_cost = litellm.get_model_cost_map(url="")
 
     custom_logger = TestLogger()
     litellm.callbacks = [custom_logger]

tests/llm_translation/test_together_ai.py

@@ -42,7 +42,8 @@ class TestTogetherAI(BaseLLMChatTest):
     def test_get_supported_response_format_together_ai(
         self, model: str, expected_bool: bool
     ) -> None:
-        litellm.model_cost = litellm.get_locally_cached_model_cost_map()
+        os.environ["LITELLM_LOCAL_MODEL_COST_MAP"] = "True"
+        litellm.model_cost = litellm.get_model_cost_map(url="")
         optional_params = litellm.get_supported_openai_params(
             model, custom_llm_provider="together_ai"
         )

tests/local_testing/test_amazing_vertex_completion.py

@@ -1433,7 +1433,8 @@ async def test_gemini_pro_json_schema_args_sent_httpx(
     enforce_validation,
 ):
     load_vertex_ai_credentials()
-    litellm.model_cost = litellm.get_locally_cached_model_cost_map()
+    os.environ["LITELLM_LOCAL_MODEL_COST_MAP"] = "True"
+    litellm.model_cost = litellm.get_model_cost_map(url="")
 
     litellm.set_verbose = True
     messages = [{"role": "user", "content": "List 5 cookie recipes"}]
@@ -1553,7 +1554,8 @@ async def test_gemini_pro_json_schema_args_sent_httpx_openai_schema(
     from pydantic import BaseModel
 
     load_vertex_ai_credentials()
-    litellm.model_cost = litellm.get_locally_cached_model_cost_map()
+    os.environ["LITELLM_LOCAL_MODEL_COST_MAP"] = "True"
+    litellm.model_cost = litellm.get_model_cost_map(url="")
 
     litellm.set_verbose = True
 

tests/local_testing/test_completion_cost.py

@@ -634,7 +634,8 @@ def test_gemini_completion_cost(above_128k, provider):
     """
     Check if cost correctly calculated for gemini models based on context window
     """
-    litellm.model_cost = litellm.get_locally_cached_model_cost_map()
+    os.environ["LITELLM_LOCAL_MODEL_COST_MAP"] = "True"
+    litellm.model_cost = litellm.get_model_cost_map(url="")
     if provider == "gemini":
         model_name = "gemini-1.5-flash-latest"
     else:
@@ -689,7 +690,8 @@ def _count_characters(text):
 
 
 def test_vertex_ai_completion_cost():
-    litellm.model_cost = litellm.get_locally_cached_model_cost_map()
+    os.environ["LITELLM_LOCAL_MODEL_COST_MAP"] = "True"
+    litellm.model_cost = litellm.get_model_cost_map(url="")
 
     text = "The quick brown fox jumps over the lazy dog."
     characters = _count_characters(text=text)
@@ -724,7 +726,8 @@ def test_vertex_ai_medlm_completion_cost():
             model=model, messages=messages, custom_llm_provider="vertex_ai"
         )
 
-    litellm.model_cost = litellm.get_locally_cached_model_cost_map()
+    os.environ["LITELLM_LOCAL_MODEL_COST_MAP"] = "True"
+    litellm.model_cost = litellm.get_model_cost_map(url="")
 
     model = "vertex_ai/medlm-medium"
     messages = [{"role": "user", "content": "Test MedLM completion cost."}]
@@ -743,7 +746,8 @@ def test_vertex_ai_claude_completion_cost():
     from litellm import Choices, Message, ModelResponse
     from litellm.utils import Usage
 
-    litellm.model_cost = litellm.get_locally_cached_model_cost_map()
+    os.environ["LITELLM_LOCAL_MODEL_COST_MAP"] = "True"
+    litellm.model_cost = litellm.get_model_cost_map(url="")
 
     litellm.set_verbose = True
     input_tokens = litellm.token_counter(
@@ -792,7 +796,8 @@ def test_vertex_ai_embedding_completion_cost(caplog):
     """
     Relevant issue - https://github.com/BerriAI/litellm/issues/4630
     """
-    litellm.model_cost = litellm.get_locally_cached_model_cost_map()
+    os.environ["LITELLM_LOCAL_MODEL_COST_MAP"] = "True"
+    litellm.model_cost = litellm.get_model_cost_map(url="")
 
     text = "The quick brown fox jumps over the lazy dog."
     input_tokens = litellm.token_counter(
@@ -834,7 +839,8 @@ def test_vertex_ai_embedding_completion_cost(caplog):
 #     from test_amazing_vertex_completion import load_vertex_ai_credentials
 
 #     load_vertex_ai_credentials()
-#     litellm.model_cost = litellm.get_locally_cached_model_cost_map()
+#     os.environ["LITELLM_LOCAL_MODEL_COST_MAP"] = "True"
+#     litellm.model_cost = litellm.get_model_cost_map(url="")
 
 #     text = "The quick brown fox jumps over the lazy dog."
 #     input_tokens = litellm.token_counter(
@@ -861,7 +867,8 @@ def test_vertex_ai_embedding_completion_cost(caplog):
 
 def test_completion_azure_ai():
     try:
-        litellm.model_cost = litellm.get_locally_cached_model_cost_map()
+        os.environ["LITELLM_LOCAL_MODEL_COST_MAP"] = "True"
+        litellm.model_cost = litellm.get_model_cost_map(url="")
 
         litellm.set_verbose = True
         response = litellm.completion(
@@ -967,7 +974,8 @@ def test_vertex_ai_mistral_predict_cost(usage):
 
 @pytest.mark.parametrize("model", ["openai/tts-1", "azure/tts-1"])
 def test_completion_cost_tts(model):
-    litellm.model_cost = litellm.get_locally_cached_model_cost_map()
+    os.environ["LITELLM_LOCAL_MODEL_COST_MAP"] = "True"
+    litellm.model_cost = litellm.get_model_cost_map(url="")
 
     cost = completion_cost(
         model=model,
@@ -1163,7 +1171,8 @@ def test_completion_cost_azure_common_deployment_name():
     ],
 )
 def test_completion_cost_prompt_caching(model, custom_llm_provider):
-    litellm.model_cost = litellm.get_locally_cached_model_cost_map()
+    os.environ["LITELLM_LOCAL_MODEL_COST_MAP"] = "True"
+    litellm.model_cost = litellm.get_model_cost_map(url="")
 
     from litellm.utils import Choices, Message, ModelResponse, Usage
 
@@ -1264,7 +1273,8 @@ def test_completion_cost_prompt_caching(model, custom_llm_provider):
     ],
 )
 def test_completion_cost_databricks(model):
-    litellm.model_cost = litellm.get_locally_cached_model_cost_map()
+    os.environ["LITELLM_LOCAL_MODEL_COST_MAP"] = "True"
+    litellm.model_cost = litellm.get_model_cost_map(url="")
     model, messages = model, [{"role": "user", "content": "What is 2+2?"}]
 
     resp = litellm.completion(model=model, messages=messages)  # works fine
@@ -1281,7 +1291,8 @@ def test_completion_cost_databricks(model):
     ],
 )
 def test_completion_cost_databricks_embedding(model):
-    litellm.model_cost = litellm.get_locally_cached_model_cost_map()
+    os.environ["LITELLM_LOCAL_MODEL_COST_MAP"] = "True"
+    litellm.model_cost = litellm.get_model_cost_map(url="")
     resp = litellm.embedding(model=model, input=["hey, how's it going?"])  # works fine
 
     print(resp)
@@ -1308,7 +1319,8 @@ def test_get_model_params_fireworks_ai(model, base_model):
     ["fireworks_ai/llama-v3p1-405b-instruct", "fireworks_ai/mixtral-8x7b-instruct"],
 )
 def test_completion_cost_fireworks_ai(model):
-    litellm.model_cost = litellm.get_locally_cached_model_cost_map()
+    os.environ["LITELLM_LOCAL_MODEL_COST_MAP"] = "True"
+    litellm.model_cost = litellm.get_model_cost_map(url="")
 
     messages = [{"role": "user", "content": "Hey, how's it going?"}]
     resp = litellm.completion(model=model, messages=messages)  # works fine
@@ -1325,7 +1337,8 @@ def test_cost_azure_openai_prompt_caching():
     )
     from litellm import get_model_info
 
-    litellm.model_cost = litellm.get_locally_cached_model_cost_map()
+    os.environ["LITELLM_LOCAL_MODEL_COST_MAP"] = "True"
+    litellm.model_cost = litellm.get_model_cost_map(url="")
 
     model = "azure/o1-mini"
 
@@ -1416,7 +1429,8 @@ def test_cost_azure_openai_prompt_caching():
 
 
 def test_completion_cost_vertex_llama3():
-    litellm.model_cost = litellm.get_locally_cached_model_cost_map()
+    os.environ["LITELLM_LOCAL_MODEL_COST_MAP"] = "True"
+    litellm.model_cost = litellm.get_model_cost_map(url="")
 
     from litellm.utils import Choices, Message, ModelResponse, Usage
 
@@ -1456,7 +1470,8 @@ def test_cost_openai_prompt_caching():
     from litellm.utils import Choices, Message, ModelResponse, Usage
     from litellm import get_model_info
 
-    litellm.model_cost = litellm.get_locally_cached_model_cost_map()
+    os.environ["LITELLM_LOCAL_MODEL_COST_MAP"] = "True"
+    litellm.model_cost = litellm.get_model_cost_map(url="")
 
     model = "gpt-4o-mini-2024-07-18"
 
@@ -1546,7 +1561,8 @@ def test_cost_openai_prompt_caching():
 def test_completion_cost_azure_ai_rerank(model):
     from litellm import RerankResponse, rerank
 
-    litellm.model_cost = litellm.get_locally_cached_model_cost_map()
+    os.environ["LITELLM_LOCAL_MODEL_COST_MAP"] = "True"
+    litellm.model_cost = litellm.get_model_cost_map(url="")
 
     response = RerankResponse(
         id="b01dbf2e-63c8-4981-9e69-32241da559ed",
@@ -1577,7 +1593,8 @@ def test_completion_cost_azure_ai_rerank(model):
 def test_together_ai_embedding_completion_cost():
     from litellm.utils import Choices, EmbeddingResponse, Message, ModelResponse, Usage
 
-    litellm.model_cost = litellm.get_locally_cached_model_cost_map()
+    os.environ["LITELLM_LOCAL_MODEL_COST_MAP"] = "True"
+    litellm.model_cost = litellm.get_model_cost_map(url="")
     response = EmbeddingResponse(
         model="togethercomputer/m2-bert-80M-8k-retrieval",
         data=[
@@ -2434,7 +2451,8 @@ def test_completion_cost_params_gemini_3():
 
     from litellm.llms.vertex_ai.cost_calculator import cost_per_character
 
-    litellm.model_cost = litellm.get_locally_cached_model_cost_map()
+    os.environ["LITELLM_LOCAL_MODEL_COST_MAP"] = "True"
+    litellm.model_cost = litellm.get_model_cost_map(url="")
 
     response = ModelResponse(
         id="chatcmpl-61043504-4439-48be-9996-e29bdee24dc3",
@@ -2503,7 +2521,8 @@ def test_completion_cost_params_gemini_3():
 # @pytest.mark.flaky(retries=3, delay=1)
 @pytest.mark.parametrize("stream", [False])  # True,
 async def test_test_completion_cost_gpt4o_audio_output_from_model(stream):
-    litellm.model_cost = litellm.get_locally_cached_model_cost_map()
+    os.environ["LITELLM_LOCAL_MODEL_COST_MAP"] = "True"
+    litellm.model_cost = litellm.get_model_cost_map(url="")
     from litellm.types.utils import (
         Choices,
         Message,
@@ -2600,7 +2619,8 @@ def test_completion_cost_model_response_cost(response_model, custom_llm_provider
     """
     from litellm import ModelResponse
 
-    litellm.model_cost = litellm.get_locally_cached_model_cost_map()
+    os.environ["LITELLM_LOCAL_MODEL_COST_MAP"] = "True"
+    litellm.model_cost = litellm.get_model_cost_map(url="")
 
     litellm.set_verbose = True
     response = {
@@ -2700,7 +2720,8 @@ def test_select_model_name_for_cost_calc():
 def test_moderations():
     from litellm import moderation
 
-    litellm.model_cost = litellm.get_locally_cached_model_cost_map()
+    os.environ["LITELLM_LOCAL_MODEL_COST_MAP"] = "True"
+    litellm.model_cost = litellm.get_model_cost_map(url="")
     litellm.add_known_models()
 
     assert "omni-moderation-latest" in litellm.model_cost
@@ -2753,7 +2774,8 @@ def test_bedrock_cost_calc_with_region():
 
     from litellm import ModelResponse
 
-    litellm.model_cost = litellm.get_locally_cached_model_cost_map()
+    os.environ["LITELLM_LOCAL_MODEL_COST_MAP"] = "True"
+    litellm.model_cost = litellm.get_model_cost_map(url="")
 
     litellm.add_known_models()
 
@@ -2952,7 +2974,9 @@ async def test_cost_calculator_with_custom_pricing_router(model_item, custom_pri
 def test_json_valid_model_cost_map():
     import json
 
-    model_cost = litellm.get_locally_cached_model_cost_map()
+    os.environ["LITELLM_LOCAL_MODEL_COST_MAP"] = "True"
+
+    model_cost = litellm.get_model_cost_map(url="")
 
     try:
         # Attempt to serialize and deserialize the JSON

tests/local_testing/test_embedding.py

@@ -115,7 +115,8 @@ def test_openai_embedding_3():
 @pytest.mark.asyncio
 async def test_openai_azure_embedding_simple(model, api_base, api_key, sync_mode):
     try:
-        litellm.model_cost = litellm.get_locally_cached_model_cost_map()
+        os.environ["LITELLM_LOCAL_MODEL_COST_MAP"] = "True"
+        litellm.model_cost = litellm.get_model_cost_map(url="")
         # litellm.set_verbose = True
         if sync_mode:
             response = embedding(
@@ -197,7 +198,8 @@ def _azure_ai_image_mock_response(*args, **kwargs):
 @pytest.mark.asyncio
 async def test_azure_ai_embedding_image(model, api_base, api_key, sync_mode):
     try:
-        litellm.model_cost = litellm.get_locally_cached_model_cost_map()
+        os.environ["LITELLM_LOCAL_MODEL_COST_MAP"] = "True"
+        litellm.model_cost = litellm.get_model_cost_map(url="")
         input = base64_image
         if sync_mode:
             client = HTTPHandler()

tests/local_testing/test_get_model_info.py

@@ -58,14 +58,16 @@ def test_get_model_info_shows_correct_supports_vision():
 
 
 def test_get_model_info_shows_assistant_prefill():
-    litellm.model_cost = litellm.get_locally_cached_model_cost_map()
+    os.environ["LITELLM_LOCAL_MODEL_COST_MAP"] = "True"
+    litellm.model_cost = litellm.get_model_cost_map(url="")
     info = litellm.get_model_info("deepseek/deepseek-chat")
     print("info", info)
     assert info.get("supports_assistant_prefill") is True
 
 
 def test_get_model_info_shows_supports_prompt_caching():
-    litellm.model_cost = litellm.get_locally_cached_model_cost_map()
+    os.environ["LITELLM_LOCAL_MODEL_COST_MAP"] = "True"
+    litellm.model_cost = litellm.get_model_cost_map(url="")
     info = litellm.get_model_info("deepseek/deepseek-chat")
     print("info", info)
     assert info.get("supports_prompt_caching") is True
@@ -114,7 +116,8 @@ def test_get_model_info_gemini():
     """
     Tests if ALL gemini models have 'tpm' and 'rpm' in the model info
     """
-    litellm.model_cost = litellm.get_locally_cached_model_cost_map()
+    os.environ["LITELLM_LOCAL_MODEL_COST_MAP"] = "True"
+    litellm.model_cost = litellm.get_model_cost_map(url="")
 
     model_map = litellm.model_cost
     for model, info in model_map.items():
@@ -124,7 +127,8 @@ def test_get_model_info_gemini():
 
 
 def test_get_model_info_bedrock_region():
-    litellm.model_cost = litellm.get_locally_cached_model_cost_map()
+    os.environ["LITELLM_LOCAL_MODEL_COST_MAP"] = "True"
+    litellm.model_cost = litellm.get_model_cost_map(url="")
     args = {
         "model": "us.anthropic.claude-3-5-sonnet-20241022-v2:0",
         "custom_llm_provider": "bedrock",
@@ -208,7 +212,8 @@ def test_model_info_bedrock_converse(monkeypatch):
 
     This ensures they are automatically routed to the converse endpoint.
     """
-    litellm.model_cost = litellm.get_locally_cached_model_cost_map()
+    monkeypatch.setenv("LITELLM_LOCAL_MODEL_COST_MAP", "True")
+    litellm.model_cost = litellm.get_model_cost_map(url="")
     try:
         # Load whitelist models from file
         with open("whitelisted_bedrock_models.txt", "r") as file:
@@ -226,7 +231,8 @@ def test_model_info_bedrock_converse_enforcement(monkeypatch):
     """
     Test the enforcement of the whitelist by adding a fake model and ensuring the test fails.
     """
-    litellm.model_cost = litellm.get_locally_cached_model_cost_map()
+    monkeypatch.setenv("LITELLM_LOCAL_MODEL_COST_MAP", "True")
+    litellm.model_cost = litellm.get_model_cost_map(url="")
 
     # Add a fake unwhitelisted model
     litellm.model_cost["fake.bedrock-chat-model"] = {
@@ -317,7 +323,8 @@ def test_get_model_info_bedrock_models():
     """
     from litellm.llms.bedrock.common_utils import BedrockModelInfo
 
-    litellm.model_cost = litellm.get_locally_cached_model_cost_map()
+    os.environ["LITELLM_LOCAL_MODEL_COST_MAP"] = "True"
+    litellm.model_cost = litellm.get_model_cost_map(url="")
 
     for k, v in litellm.model_cost.items():
         if v["litellm_provider"] == "bedrock":

tests/local_testing/test_router_utils.py

@@ -178,7 +178,8 @@ async def test_update_kwargs_before_fallbacks(call_type):
 
 
 def test_router_get_model_info_wildcard_routes():
-    litellm.model_cost = litellm.get_locally_cached_model_cost_map()
+    os.environ["LITELLM_LOCAL_MODEL_COST_MAP"] = "True"
+    litellm.model_cost = litellm.get_model_cost_map(url="")
     router = Router(
         model_list=[
             {
@@ -199,7 +200,8 @@ def test_router_get_model_info_wildcard_routes():
 
 @pytest.mark.asyncio
 async def test_router_get_model_group_usage_wildcard_routes():
-    litellm.model_cost = litellm.get_locally_cached_model_cost_map()
+    os.environ["LITELLM_LOCAL_MODEL_COST_MAP"] = "True"
+    litellm.model_cost = litellm.get_model_cost_map(url="")
     router = Router(
         model_list=[
             {
@@ -295,7 +297,8 @@ async def test_call_router_callbacks_on_failure():
 
 @pytest.mark.asyncio
 async def test_router_model_group_headers():
-    litellm.model_cost = litellm.get_locally_cached_model_cost_map()
+    os.environ["LITELLM_LOCAL_MODEL_COST_MAP"] = "True"
+    litellm.model_cost = litellm.get_model_cost_map(url="")
     from litellm.types.utils import OPENAI_RESPONSE_HEADERS
 
     router = Router(
@@ -327,7 +330,8 @@ async def test_router_model_group_headers():
 
 @pytest.mark.asyncio
 async def test_get_remaining_model_group_usage():
-    litellm.model_cost = litellm.get_locally_cached_model_cost_map()
+    os.environ["LITELLM_LOCAL_MODEL_COST_MAP"] = "True"
+    litellm.model_cost = litellm.get_model_cost_map(url="")
     from litellm.types.utils import OPENAI_RESPONSE_HEADERS
 
     router = Router(

tests/test_model_prices_and_context_window_schema.py

@@ -1,111 +0,0 @@
-import litellm
-from jsonschema import validate
-
-def test_model_prices_and_context_window_json_is_valid():
-    '''
-    Validates the `model_prices_and_context_window.json` file.
-
-    If this test fails after you update the json, you need to update the schema or correct the change you made.
-    '''
-
-    INTENDED_SCHEMA = {
-        "type": "object",
-        "additionalProperties": {
-            "type": "object",
-            "properties": {
-                "cache_creation_input_audio_token_cost": {"type": "number"},
-                "cache_creation_input_token_cost": {"type": "number"},
-                "cache_read_input_token_cost": {"type": "number"},
-                "deprecation_date": {"type": "string"},
-                "input_cost_per_audio_per_second": {"type": "number"},
-                "input_cost_per_audio_per_second_above_128k_tokens": {"type": "number"},
-                "input_cost_per_audio_token": {"type": "number"},
-                "input_cost_per_character": {"type": "number"},
-                "input_cost_per_character_above_128k_tokens": {"type": "number"},
-                "input_cost_per_image": {"type": "number"},
-                "input_cost_per_image_above_128k_tokens": {"type": "number"},
-                "input_cost_per_pixel": {"type": "number"},
-                "input_cost_per_query": {"type": "number"},
-                "input_cost_per_request": {"type": "number"},
-                "input_cost_per_second": {"type": "number"},
-                "input_cost_per_token": {"type": "number"},
-                "input_cost_per_token_above_128k_tokens": {"type": "number"},
-                "input_cost_per_token_batch_requests": {"type": "number"},
-                "input_cost_per_token_batches": {"type": "number"},
-                "input_cost_per_token_cache_hit": {"type": "number"},
-                "input_cost_per_video_per_second": {"type": "number"},
-                "input_cost_per_video_per_second_above_128k_tokens": {"type": "number"},
-                "input_dbu_cost_per_token": {"type": "number"},
-                "litellm_provider": {"type": "string"},
-                "max_audio_length_hours": {"type": "number"},
-                "max_audio_per_prompt": {"type": "number"},
-                "max_document_chunks_per_query": {"type": "number"},
-                "max_images_per_prompt": {"type": "number"},
-                "max_input_tokens": {"type": "number"},
-                "max_output_tokens": {"type": "number"},
-                "max_pdf_size_mb": {"type": "number"},
-                "max_query_tokens": {"type": "number"},
-                "max_tokens": {"type": "number"},
-                "max_tokens_per_document_chunk": {"type": "number"},
-                "max_video_length": {"type": "number"},
-                "max_videos_per_prompt": {"type": "number"},
-                "metadata": {"type": "object"},
-                "mode": {
-                    "type": "string",
-                    "enum": [
-                        "audio_speech",
-                        "audio_transcription",
-                        "chat",
-                        "completion",
-                        "embedding",
-                        "image_generation",
-                        "moderation",
-                        "moderations",
-                        "rerank"
-                    ],
-                },
-                "output_cost_per_audio_token": {"type": "number"},
-                "output_cost_per_character": {"type": "number"},
-                "output_cost_per_character_above_128k_tokens": {"type": "number"},
-                "output_cost_per_image": {"type": "number"},
-                "output_cost_per_pixel": {"type": "number"},
-                "output_cost_per_second": {"type": "number"},
-                "output_cost_per_token": {"type": "number"},
-                "output_cost_per_token_above_128k_tokens": {"type": "number"},
-                "output_cost_per_token_batches": {"type": "number"},
-                "output_db_cost_per_token": {"type": "number"},
-                "output_dbu_cost_per_token": {"type": "number"},
-                "output_vector_size": {"type": "number"},
-                "rpd": {"type": "number"},
-                "rpm": {"type": "number"},
-                "source": {"type": "string"},
-                "supports_assistant_prefill": {"type": "boolean"},
-                "supports_audio_input": {"type": "boolean"},
-                "supports_audio_output": {"type": "boolean"},
-                "supports_embedding_image_input": {"type": "boolean"},
-                "supports_function_calling": {"type": "boolean"},
-                "supports_image_input": {"type": "boolean"},
-                "supports_parallel_function_calling": {"type": "boolean"},
-                "supports_pdf_input": {"type": "boolean"},
-                "supports_prompt_caching": {"type": "boolean"},
-                "supports_response_schema": {"type": "boolean"},
-                "supports_system_messages": {"type": "boolean"},
-                "supports_tool_choice": {"type": "boolean"},
-                "supports_video_input": {"type": "boolean"},
-                "supports_vision": {"type": "boolean"},
-                "tool_use_system_prompt_tokens": {"type": "number"},
-                "tpm": {"type": "number"},
-            },
-            "additionalProperties": False,
-        },
-    }
-
-    actual_json = litellm.get_locally_cached_model_cost_map()
-    assert isinstance(actual_json, dict)
-    temporarily_removed = actual_json.pop('sample_spec', None) # remove the sample, whose schema is inconsistent with the real data
-
-    validate(actual_json, INTENDED_SCHEMA)
-
-    if temporarily_removed is not None:
-        # put back the sample spec that we removed
-        actual_json.update({'sample_spec': temporarily_removed})