mirror of
https://github.com/tiennm99/litellm.git
synced 2026-07-04 19:08:13 +00:00
4c82dd9b27
* fix(user_api_key_auth.py): fix else block Fixes https://github.com/BerriAI/litellm/issues/11170 * refactor(ollama/chat): refactor to base config pattern easier to maintain fixes * fix(ollama/chat): support tool call parsing on streaming Closes https://github.com/BerriAI/litellm/issues/11104 * test: update import location * fix: cleanup unused import * fix: fix ruff check error * test: update import * test: update test on ci * ci: cleanup * fix: fix chekc * fix: fix api key check order * test: fix import * ci: fix script * test: fix imports * fix: fix tests
815 lines
30 KiB
Python
815 lines
30 KiB
Python
import json
|
|
import os
|
|
import sys
|
|
from unittest.mock import patch
|
|
|
|
import pytest
|
|
from jsonschema import validate
|
|
|
|
sys.path.insert(
|
|
0, os.path.abspath("../..")
|
|
) # Adds the parent directory to the system path
|
|
|
|
import litellm
|
|
from litellm.types.utils import LlmProviders
|
|
from litellm.utils import (
|
|
ProviderConfigManager,
|
|
get_llm_provider,
|
|
get_optional_params_image_gen,
|
|
)
|
|
|
|
# Adds the parent directory to the system path
|
|
|
|
|
|
def test_get_optional_params_image_gen():
|
|
from litellm.llms.azure.image_generation import AzureGPTImageGenerationConfig
|
|
|
|
provider_config = AzureGPTImageGenerationConfig()
|
|
optional_params = get_optional_params_image_gen(
|
|
model="gpt-image-1",
|
|
response_format="b64_json",
|
|
n=3,
|
|
custom_llm_provider="azure",
|
|
drop_params=True,
|
|
provider_config=provider_config,
|
|
)
|
|
assert optional_params is not None
|
|
assert "response_format" not in optional_params
|
|
assert optional_params["n"] == 3
|
|
|
|
|
|
def test_all_model_configs():
|
|
from litellm.llms.vertex_ai.vertex_ai_partner_models.ai21.transformation import (
|
|
VertexAIAi21Config,
|
|
)
|
|
from litellm.llms.vertex_ai.vertex_ai_partner_models.llama3.transformation import (
|
|
VertexAILlama3Config,
|
|
)
|
|
|
|
assert (
|
|
"max_completion_tokens"
|
|
in VertexAILlama3Config().get_supported_openai_params(model="llama3")
|
|
)
|
|
assert VertexAILlama3Config().map_openai_params(
|
|
{"max_completion_tokens": 10}, {}, "llama3", drop_params=False
|
|
) == {"max_tokens": 10}
|
|
|
|
assert "max_completion_tokens" in VertexAIAi21Config().get_supported_openai_params(
|
|
model="jamba-1.5-mini@001"
|
|
)
|
|
assert VertexAIAi21Config().map_openai_params(
|
|
{"max_completion_tokens": 10}, {}, "jamba-1.5-mini@001", drop_params=False
|
|
) == {"max_tokens": 10}
|
|
|
|
from litellm.llms.fireworks_ai.chat.transformation import FireworksAIConfig
|
|
|
|
assert "max_completion_tokens" in FireworksAIConfig().get_supported_openai_params(
|
|
model="llama3"
|
|
)
|
|
assert FireworksAIConfig().map_openai_params(
|
|
model="llama3",
|
|
non_default_params={"max_completion_tokens": 10},
|
|
optional_params={},
|
|
drop_params=False,
|
|
) == {"max_tokens": 10}
|
|
|
|
from litellm.llms.nvidia_nim.chat.transformation import NvidiaNimConfig
|
|
|
|
assert "max_completion_tokens" in NvidiaNimConfig().get_supported_openai_params(
|
|
model="llama3"
|
|
)
|
|
assert NvidiaNimConfig().map_openai_params(
|
|
model="llama3",
|
|
non_default_params={"max_completion_tokens": 10},
|
|
optional_params={},
|
|
drop_params=False,
|
|
) == {"max_tokens": 10}
|
|
|
|
from litellm.llms.ollama.chat.transformation import OllamaChatConfig
|
|
|
|
assert "max_completion_tokens" in OllamaChatConfig().get_supported_openai_params(
|
|
model="llama3"
|
|
)
|
|
assert OllamaChatConfig().map_openai_params(
|
|
model="llama3",
|
|
non_default_params={"max_completion_tokens": 10},
|
|
optional_params={},
|
|
drop_params=False,
|
|
) == {"num_predict": 10}
|
|
|
|
from litellm.llms.predibase.chat.transformation import PredibaseConfig
|
|
|
|
assert "max_completion_tokens" in PredibaseConfig().get_supported_openai_params(
|
|
model="llama3"
|
|
)
|
|
assert PredibaseConfig().map_openai_params(
|
|
model="llama3",
|
|
non_default_params={"max_completion_tokens": 10},
|
|
optional_params={},
|
|
drop_params=False,
|
|
) == {"max_new_tokens": 10}
|
|
|
|
from litellm.llms.codestral.completion.transformation import (
|
|
CodestralTextCompletionConfig,
|
|
)
|
|
|
|
assert (
|
|
"max_completion_tokens"
|
|
in CodestralTextCompletionConfig().get_supported_openai_params(model="llama3")
|
|
)
|
|
assert CodestralTextCompletionConfig().map_openai_params(
|
|
model="llama3",
|
|
non_default_params={"max_completion_tokens": 10},
|
|
optional_params={},
|
|
drop_params=False,
|
|
) == {"max_tokens": 10}
|
|
|
|
from litellm.llms.volcengine import VolcEngineConfig
|
|
|
|
assert "max_completion_tokens" in VolcEngineConfig().get_supported_openai_params(
|
|
model="llama3"
|
|
)
|
|
assert VolcEngineConfig().map_openai_params(
|
|
model="llama3",
|
|
non_default_params={"max_completion_tokens": 10},
|
|
optional_params={},
|
|
drop_params=False,
|
|
) == {"max_tokens": 10}
|
|
|
|
from litellm.llms.ai21.chat.transformation import AI21ChatConfig
|
|
|
|
assert "max_completion_tokens" in AI21ChatConfig().get_supported_openai_params(
|
|
"jamba-1.5-mini@001"
|
|
)
|
|
assert AI21ChatConfig().map_openai_params(
|
|
model="jamba-1.5-mini@001",
|
|
non_default_params={"max_completion_tokens": 10},
|
|
optional_params={},
|
|
drop_params=False,
|
|
) == {"max_tokens": 10}
|
|
|
|
from litellm.llms.azure.chat.gpt_transformation import AzureOpenAIConfig
|
|
|
|
assert "max_completion_tokens" in AzureOpenAIConfig().get_supported_openai_params(
|
|
model="gpt-3.5-turbo"
|
|
)
|
|
assert AzureOpenAIConfig().map_openai_params(
|
|
model="gpt-3.5-turbo",
|
|
non_default_params={"max_completion_tokens": 10},
|
|
optional_params={},
|
|
api_version="2022-12-01",
|
|
drop_params=False,
|
|
) == {"max_completion_tokens": 10}
|
|
|
|
from litellm.llms.bedrock.chat.converse_transformation import AmazonConverseConfig
|
|
|
|
assert (
|
|
"max_completion_tokens"
|
|
in AmazonConverseConfig().get_supported_openai_params(
|
|
model="anthropic.claude-3-sonnet-20240229-v1:0"
|
|
)
|
|
)
|
|
assert AmazonConverseConfig().map_openai_params(
|
|
model="anthropic.claude-3-sonnet-20240229-v1:0",
|
|
non_default_params={"max_completion_tokens": 10},
|
|
optional_params={},
|
|
drop_params=False,
|
|
) == {"maxTokens": 10}
|
|
|
|
from litellm.llms.codestral.completion.transformation import (
|
|
CodestralTextCompletionConfig,
|
|
)
|
|
|
|
assert (
|
|
"max_completion_tokens"
|
|
in CodestralTextCompletionConfig().get_supported_openai_params(model="llama3")
|
|
)
|
|
assert CodestralTextCompletionConfig().map_openai_params(
|
|
model="llama3",
|
|
non_default_params={"max_completion_tokens": 10},
|
|
optional_params={},
|
|
drop_params=False,
|
|
) == {"max_tokens": 10}
|
|
|
|
from litellm import AmazonAnthropicClaude3Config, AmazonAnthropicConfig
|
|
|
|
assert (
|
|
"max_completion_tokens"
|
|
in AmazonAnthropicClaude3Config().get_supported_openai_params(
|
|
model="anthropic.claude-3-sonnet-20240229-v1:0"
|
|
)
|
|
)
|
|
|
|
assert AmazonAnthropicClaude3Config().map_openai_params(
|
|
non_default_params={"max_completion_tokens": 10},
|
|
optional_params={},
|
|
model="anthropic.claude-3-sonnet-20240229-v1:0",
|
|
drop_params=False,
|
|
) == {"max_tokens": 10}
|
|
|
|
assert (
|
|
"max_completion_tokens"
|
|
in AmazonAnthropicConfig().get_supported_openai_params(model="")
|
|
)
|
|
|
|
assert AmazonAnthropicConfig().map_openai_params(
|
|
non_default_params={"max_completion_tokens": 10},
|
|
optional_params={},
|
|
model="",
|
|
drop_params=False,
|
|
) == {"max_tokens_to_sample": 10}
|
|
|
|
from litellm.llms.databricks.chat.transformation import DatabricksConfig
|
|
|
|
assert "max_completion_tokens" in DatabricksConfig().get_supported_openai_params()
|
|
|
|
assert DatabricksConfig().map_openai_params(
|
|
model="databricks/llama-3-70b-instruct",
|
|
drop_params=False,
|
|
non_default_params={"max_completion_tokens": 10},
|
|
optional_params={},
|
|
) == {"max_tokens": 10}
|
|
|
|
from litellm.llms.vertex_ai.vertex_ai_partner_models.anthropic.transformation import (
|
|
VertexAIAnthropicConfig,
|
|
)
|
|
|
|
assert (
|
|
"max_completion_tokens"
|
|
in VertexAIAnthropicConfig().get_supported_openai_params(
|
|
model="claude-3-5-sonnet-20240620"
|
|
)
|
|
)
|
|
|
|
assert VertexAIAnthropicConfig().map_openai_params(
|
|
non_default_params={"max_completion_tokens": 10},
|
|
optional_params={},
|
|
model="claude-3-5-sonnet-20240620",
|
|
drop_params=False,
|
|
) == {"max_tokens": 10}
|
|
|
|
from litellm.llms.gemini.chat.transformation import GoogleAIStudioGeminiConfig
|
|
from litellm.llms.vertex_ai.gemini.vertex_and_google_ai_studio_gemini import (
|
|
VertexGeminiConfig,
|
|
)
|
|
|
|
assert "max_completion_tokens" in VertexGeminiConfig().get_supported_openai_params(
|
|
model="gemini-1.0-pro"
|
|
)
|
|
|
|
assert VertexGeminiConfig().map_openai_params(
|
|
model="gemini-1.0-pro",
|
|
non_default_params={"max_completion_tokens": 10},
|
|
optional_params={},
|
|
drop_params=False,
|
|
) == {"max_output_tokens": 10}
|
|
|
|
assert (
|
|
"max_completion_tokens"
|
|
in GoogleAIStudioGeminiConfig().get_supported_openai_params(
|
|
model="gemini-1.0-pro"
|
|
)
|
|
)
|
|
|
|
assert GoogleAIStudioGeminiConfig().map_openai_params(
|
|
model="gemini-1.0-pro",
|
|
non_default_params={"max_completion_tokens": 10},
|
|
optional_params={},
|
|
drop_params=False,
|
|
) == {"max_output_tokens": 10}
|
|
|
|
assert "max_completion_tokens" in VertexGeminiConfig().get_supported_openai_params(
|
|
model="gemini-1.0-pro"
|
|
)
|
|
|
|
assert VertexGeminiConfig().map_openai_params(
|
|
model="gemini-1.0-pro",
|
|
non_default_params={"max_completion_tokens": 10},
|
|
optional_params={},
|
|
drop_params=False,
|
|
) == {"max_output_tokens": 10}
|
|
|
|
|
|
def test_anthropic_web_search_in_model_info():
|
|
os.environ["LITELLM_LOCAL_MODEL_COST_MAP"] = "True"
|
|
litellm.model_cost = litellm.get_model_cost_map(url="")
|
|
|
|
supported_models = [
|
|
"anthropic/claude-3-7-sonnet-20250219",
|
|
"anthropic/claude-3-5-sonnet-latest",
|
|
"anthropic/claude-3-5-sonnet-20241022",
|
|
"anthropic/claude-3-5-haiku-20241022",
|
|
"anthropic/claude-3-5-haiku-latest",
|
|
]
|
|
for model in supported_models:
|
|
from litellm.utils import get_model_info
|
|
|
|
model_info = get_model_info(model)
|
|
assert model_info is not None
|
|
assert (
|
|
model_info["supports_web_search"] is True
|
|
), f"Model {model} should support web search"
|
|
assert (
|
|
model_info["search_context_cost_per_query"] is not None
|
|
), f"Model {model} should have a search context cost per query"
|
|
|
|
|
|
def test_cohere_embedding_optional_params():
|
|
from litellm import get_optional_params_embeddings
|
|
|
|
optional_params = get_optional_params_embeddings(
|
|
model="embed-v4.0",
|
|
custom_llm_provider="cohere",
|
|
input="Hello, world!",
|
|
input_type="search_query",
|
|
dimensions=512,
|
|
)
|
|
assert optional_params is not None
|
|
|
|
|
|
def test_aaamodel_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": {
|
|
"supports_computer_use": {"type": "boolean"},
|
|
"cache_creation_input_audio_token_cost": {"type": "number"},
|
|
"cache_creation_input_token_cost": {"type": "number"},
|
|
"cache_read_input_token_cost": {"type": "number"},
|
|
"cache_read_input_audio_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_token_above_200k_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_8s_interval": {"type": "number"},
|
|
"input_cost_per_video_per_second_above_15s_interval": {
|
|
"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",
|
|
"rerank",
|
|
"responses",
|
|
],
|
|
},
|
|
"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_above_200k_tokens": {"type": "number"},
|
|
"output_cost_per_token_batches": {"type": "number"},
|
|
"output_cost_per_reasoning_token": {"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"},
|
|
"supports_web_search": {"type": "boolean"},
|
|
"supports_reasoning": {"type": "boolean"},
|
|
"tool_use_system_prompt_tokens": {"type": "number"},
|
|
"tpm": {"type": "number"},
|
|
"supported_endpoints": {
|
|
"type": "array",
|
|
"items": {
|
|
"type": "string",
|
|
"enum": [
|
|
"/v1/responses",
|
|
"/v1/embeddings",
|
|
"/v1/chat/completions",
|
|
"/v1/completions",
|
|
"/v1/images/generations",
|
|
"/v1/images/variations",
|
|
"/v1/images/edits",
|
|
"/v1/batch",
|
|
"/v1/audio/transcriptions",
|
|
"/v1/audio/speech",
|
|
],
|
|
},
|
|
},
|
|
"search_context_cost_per_query": {
|
|
"type": "object",
|
|
"properties": {
|
|
"search_context_size_low": {"type": "number"},
|
|
"search_context_size_medium": {"type": "number"},
|
|
"search_context_size_high": {"type": "number"},
|
|
},
|
|
"additionalProperties": False,
|
|
},
|
|
"supported_modalities": {
|
|
"type": "array",
|
|
"items": {
|
|
"type": "string",
|
|
"enum": ["text", "audio", "image", "video"],
|
|
},
|
|
},
|
|
"supported_output_modalities": {
|
|
"type": "array",
|
|
"items": {
|
|
"type": "string",
|
|
"enum": ["text", "image", "audio", "code"],
|
|
},
|
|
},
|
|
"supports_native_streaming": {"type": "boolean"},
|
|
},
|
|
"additionalProperties": False,
|
|
},
|
|
}
|
|
|
|
prod_json = "./model_prices_and_context_window.json"
|
|
# prod_json = "../../model_prices_and_context_window.json"
|
|
with open(prod_json, "r") as model_prices_file:
|
|
actual_json = json.load(model_prices_file)
|
|
assert isinstance(actual_json, dict)
|
|
actual_json.pop(
|
|
"sample_spec", None
|
|
) # remove the sample, whose schema is inconsistent with the real data
|
|
|
|
validate(actual_json, INTENDED_SCHEMA)
|
|
|
|
|
|
def test_get_model_info_gemini():
|
|
"""
|
|
Tests if ALL gemini models have 'tpm' and 'rpm' in the model info
|
|
"""
|
|
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():
|
|
if (
|
|
model.startswith("gemini/")
|
|
and not "gemma" in model
|
|
and not "learnlm" in model
|
|
):
|
|
assert info.get("tpm") is not None, f"{model} does not have tpm"
|
|
assert info.get("rpm") is not None, f"{model} does not have rpm"
|
|
|
|
|
|
def test_openai_models_in_model_info():
|
|
os.environ["LITELLM_LOCAL_MODEL_COST_MAP"] = "True"
|
|
litellm.model_cost = litellm.get_model_cost_map(url="")
|
|
|
|
model_map = litellm.model_cost
|
|
violated_models = []
|
|
for model, info in model_map.items():
|
|
if (
|
|
info.get("litellm_provider") == "openai"
|
|
and info.get("supports_vision") is True
|
|
):
|
|
if info.get("supports_pdf_input") is not True:
|
|
violated_models.append(model)
|
|
assert (
|
|
len(violated_models) == 0
|
|
), f"The following models should support pdf input: {violated_models}"
|
|
|
|
|
|
def test_supports_tool_choice_simple_tests():
|
|
"""
|
|
simple sanity checks
|
|
"""
|
|
assert litellm.utils.supports_tool_choice(model="gpt-4o") == True
|
|
assert (
|
|
litellm.utils.supports_tool_choice(
|
|
model="bedrock/anthropic.claude-3-sonnet-20240229-v1:0"
|
|
)
|
|
== True
|
|
)
|
|
assert (
|
|
litellm.utils.supports_tool_choice(
|
|
model="anthropic.claude-3-sonnet-20240229-v1:0"
|
|
)
|
|
is True
|
|
)
|
|
|
|
assert (
|
|
litellm.utils.supports_tool_choice(
|
|
model="anthropic.claude-3-sonnet-20240229-v1:0",
|
|
custom_llm_provider="bedrock_converse",
|
|
)
|
|
is True
|
|
)
|
|
|
|
assert (
|
|
litellm.utils.supports_tool_choice(model="us.amazon.nova-micro-v1:0") is False
|
|
)
|
|
assert (
|
|
litellm.utils.supports_tool_choice(model="bedrock/us.amazon.nova-micro-v1:0")
|
|
is False
|
|
)
|
|
assert (
|
|
litellm.utils.supports_tool_choice(
|
|
model="us.amazon.nova-micro-v1:0", custom_llm_provider="bedrock_converse"
|
|
)
|
|
is False
|
|
)
|
|
|
|
assert litellm.utils.supports_tool_choice(model="perplexity/sonar") is False
|
|
|
|
|
|
def test_check_provider_match():
|
|
"""
|
|
Test the _check_provider_match function for various provider scenarios
|
|
"""
|
|
# Test bedrock and bedrock_converse cases
|
|
model_info = {"litellm_provider": "bedrock"}
|
|
assert litellm.utils._check_provider_match(model_info, "bedrock") is True
|
|
assert litellm.utils._check_provider_match(model_info, "bedrock_converse") is True
|
|
|
|
# Test bedrock_converse provider
|
|
model_info = {"litellm_provider": "bedrock_converse"}
|
|
assert litellm.utils._check_provider_match(model_info, "bedrock") is True
|
|
assert litellm.utils._check_provider_match(model_info, "bedrock_converse") is True
|
|
|
|
# Test non-matching provider
|
|
model_info = {"litellm_provider": "bedrock"}
|
|
assert litellm.utils._check_provider_match(model_info, "openai") is False
|
|
|
|
|
|
# Models that should be skipped during testing
|
|
OLD_PROVIDERS = ["aleph_alpha", "palm"]
|
|
SKIP_MODELS = [
|
|
"azure/mistral",
|
|
"azure/command-r",
|
|
"jamba",
|
|
"deepinfra",
|
|
"mistral.",
|
|
"groq/llama-guard-3-8b",
|
|
"groq/gemma2-9b-it",
|
|
]
|
|
|
|
# Bedrock models to block - organized by type
|
|
BEDROCK_REGIONS = ["ap-northeast-1", "eu-central-1", "us-east-1", "us-west-2"]
|
|
BEDROCK_COMMITMENTS = ["1-month-commitment", "6-month-commitment"]
|
|
BEDROCK_MODELS = {
|
|
"anthropic.claude-v1",
|
|
"anthropic.claude-v2",
|
|
"anthropic.claude-v2:1",
|
|
"anthropic.claude-instant-v1",
|
|
}
|
|
|
|
# Generate block_list dynamically
|
|
block_list = set()
|
|
for region in BEDROCK_REGIONS:
|
|
for commitment in BEDROCK_COMMITMENTS:
|
|
for model in BEDROCK_MODELS:
|
|
block_list.add(f"bedrock/{region}/{commitment}/{model}")
|
|
block_list.add(f"bedrock/{region}/{model}")
|
|
|
|
# Add Cohere models
|
|
for commitment in BEDROCK_COMMITMENTS:
|
|
block_list.add(f"bedrock/*/{commitment}/cohere.command-text-v14")
|
|
block_list.add(f"bedrock/*/{commitment}/cohere.command-light-text-v14")
|
|
|
|
print("block_list", block_list)
|
|
|
|
|
|
@pytest.mark.asyncio
|
|
async def test_supports_tool_choice():
|
|
"""
|
|
Test that litellm.utils.supports_tool_choice() returns the correct value
|
|
for all models in model_prices_and_context_window.json.
|
|
|
|
The test:
|
|
1. Loads model pricing data
|
|
2. Iterates through each model
|
|
3. Checks if tool_choice support matches the model's supported parameters
|
|
"""
|
|
# Load model prices
|
|
litellm._turn_on_debug()
|
|
# path = "../../model_prices_and_context_window.json"
|
|
path = "./model_prices_and_context_window.json"
|
|
with open(path, "r") as f:
|
|
model_prices = json.load(f)
|
|
litellm.model_cost = model_prices
|
|
config_manager = ProviderConfigManager()
|
|
|
|
for model_name, model_info in model_prices.items():
|
|
print(f"testing model: {model_name}")
|
|
|
|
# Skip certain models
|
|
if (
|
|
model_name == "sample_spec"
|
|
or model_info.get("mode") != "chat"
|
|
or any(skip in model_name for skip in SKIP_MODELS)
|
|
or any(provider in model_name for provider in OLD_PROVIDERS)
|
|
or model_info["litellm_provider"] in OLD_PROVIDERS
|
|
or model_name in block_list
|
|
or "azure/eu" in model_name
|
|
or "azure/us" in model_name
|
|
or "codestral" in model_name
|
|
or "o1" in model_name
|
|
or "o3" in model_name
|
|
or "mistral" in model_name
|
|
):
|
|
continue
|
|
|
|
try:
|
|
model, provider, _, _ = get_llm_provider(model=model_name)
|
|
except Exception as e:
|
|
print(f"\033[91mERROR for {model_name}: {e}\033[0m")
|
|
continue
|
|
|
|
# Get provider config and supported params
|
|
print("LLM provider", provider)
|
|
provider_enum = LlmProviders(provider)
|
|
config = config_manager.get_provider_chat_config(model, provider_enum)
|
|
print("config", config)
|
|
|
|
if config:
|
|
supported_params = config.get_supported_openai_params(model)
|
|
print("supported_params", supported_params)
|
|
else:
|
|
raise Exception(f"No config found for {model_name}, provider: {provider}")
|
|
|
|
# Check tool_choice support
|
|
supports_tool_choice_result = litellm.utils.supports_tool_choice(
|
|
model=model_name, custom_llm_provider=provider
|
|
)
|
|
tool_choice_in_params = "tool_choice" in supported_params
|
|
|
|
assert (
|
|
supports_tool_choice_result == tool_choice_in_params
|
|
), f"Tool choice support mismatch for {model_name}. supports_tool_choice() returned: {supports_tool_choice_result}, tool_choice in supported params: {tool_choice_in_params}\nConfig: {config}"
|
|
|
|
|
|
def test_supports_computer_use_utility():
|
|
"""
|
|
Tests the litellm.utils.supports_computer_use utility function.
|
|
"""
|
|
from litellm.utils import supports_computer_use
|
|
|
|
# Ensure LITELLM_LOCAL_MODEL_COST_MAP is set for consistent test behavior,
|
|
# as supports_computer_use relies on get_model_info.
|
|
# This also requires litellm.model_cost to be populated.
|
|
original_env_var = os.getenv("LITELLM_LOCAL_MODEL_COST_MAP")
|
|
original_model_cost = getattr(litellm, "model_cost", None)
|
|
|
|
os.environ["LITELLM_LOCAL_MODEL_COST_MAP"] = "True"
|
|
litellm.model_cost = litellm.get_model_cost_map(url="") # Load with local/backup
|
|
|
|
try:
|
|
# Test a model known to support computer_use from backup JSON
|
|
supports_cu_anthropic = supports_computer_use(
|
|
model="anthropic/claude-3-7-sonnet-20250219"
|
|
)
|
|
assert supports_cu_anthropic is True
|
|
|
|
# Test a model known not to have the flag or set to false (defaults to False via get_model_info)
|
|
supports_cu_gpt = supports_computer_use(model="gpt-3.5-turbo")
|
|
assert supports_cu_gpt is False
|
|
finally:
|
|
# Restore original environment and model_cost to avoid side effects
|
|
if original_env_var is None:
|
|
del os.environ["LITELLM_LOCAL_MODEL_COST_MAP"]
|
|
else:
|
|
os.environ["LITELLM_LOCAL_MODEL_COST_MAP"] = original_env_var
|
|
|
|
if original_model_cost is not None:
|
|
litellm.model_cost = original_model_cost
|
|
elif hasattr(litellm, "model_cost"):
|
|
delattr(litellm, "model_cost")
|
|
|
|
|
|
def test_get_model_info_shows_supports_computer_use():
|
|
"""
|
|
Tests if 'supports_computer_use' is correctly retrieved by get_model_info.
|
|
We'll use 'claude-3-7-sonnet-20250219' as it's configured
|
|
in the backup JSON to have supports_computer_use: True.
|
|
"""
|
|
os.environ["LITELLM_LOCAL_MODEL_COST_MAP"] = "True"
|
|
# Ensure litellm.model_cost is loaded, relying on the backup mechanism if primary fails
|
|
# as per previous debugging.
|
|
litellm.model_cost = litellm.get_model_cost_map(url="")
|
|
|
|
# This model should have 'supports_computer_use': True in the backup JSON
|
|
model_known_to_support_computer_use = "claude-3-7-sonnet-20250219"
|
|
info = litellm.get_model_info(model_known_to_support_computer_use)
|
|
print(f"Info for {model_known_to_support_computer_use}: {info}")
|
|
|
|
# After the fix in utils.py, this should now be present and True
|
|
assert info.get("supports_computer_use") is True
|
|
|
|
# Optionally, test a model known NOT to support it, or where it's undefined (should default to False)
|
|
# For example, if "gpt-3.5-turbo" doesn't have it defined, it should be False.
|
|
model_known_not_to_support_computer_use = "gpt-3.5-turbo"
|
|
info_gpt = litellm.get_model_info(model_known_not_to_support_computer_use)
|
|
print(f"Info for {model_known_not_to_support_computer_use}: {info_gpt}")
|
|
assert (
|
|
info_gpt.get("supports_computer_use") is False
|
|
) # Expecting False due to the default in ModelInfoBase
|
|
|
|
|
|
@pytest.mark.parametrize(
|
|
"model, custom_llm_provider",
|
|
[
|
|
("gpt-3.5-turbo", "openai"),
|
|
("anthropic.claude-3-7-sonnet-20250219-v1:0", "bedrock"),
|
|
("gemini-1.5-pro", "vertex_ai"),
|
|
],
|
|
)
|
|
def test_pre_process_non_default_params(model, custom_llm_provider):
|
|
from pydantic import BaseModel
|
|
|
|
from litellm.utils import pre_process_non_default_params
|
|
|
|
class ResponseFormat(BaseModel):
|
|
x: str
|
|
y: str
|
|
|
|
passed_params = {
|
|
"model": "gpt-3.5-turbo",
|
|
"response_format": ResponseFormat,
|
|
}
|
|
special_params = {}
|
|
processed_non_default_params = pre_process_non_default_params(
|
|
model=model,
|
|
passed_params=passed_params,
|
|
special_params=special_params,
|
|
custom_llm_provider=custom_llm_provider,
|
|
additional_drop_params=None,
|
|
)
|
|
print(processed_non_default_params)
|
|
assert processed_non_default_params == {
|
|
"response_format": {
|
|
"type": "json_schema",
|
|
"json_schema": {
|
|
"schema": {
|
|
"properties": {
|
|
"x": {"title": "X", "type": "string"},
|
|
"y": {"title": "Y", "type": "string"},
|
|
},
|
|
"required": ["x", "y"],
|
|
"title": "ResponseFormat",
|
|
"type": "object",
|
|
"additionalProperties": False,
|
|
},
|
|
"name": "ResponseFormat",
|
|
"strict": True,
|
|
},
|
|
}
|
|
}
|