Skip to content

The fastest, litest AI Gateway. Rust core with Python SDK. Call 100+ LLM APIs in OpenAI (or native) format with cost tracking, guardrails, load balancing, and logging [Bedrock, Azure, OpenAI, Anthropic, OpenAI, VertexAI, vLLM, Nvidia NIM]

7.2 KB
Raw Blame
"""
Tests for litellm.acount_tokens() public API.
"""

import asyncio
import json
import os
from unittest.mock import AsyncMock, patch

import httpx
import pytest
import respx

import litellm
from litellm.types.utils import TokenCountResponse


def test_acount_tokens_routes_to_openai():
    """Test that acount_tokens routes to OpenAI token counter for openai/ models."""
    with patch(
        "litellm.llms.openai.responses.count_tokens.token_counter.openai_count_tokens_handler.handle_count_tokens_request",
        new_callable=AsyncMock,
        return_value={"input_tokens": 15},
    ):
        result = asyncio.run(
            litellm.acount_tokens(
                model="openai/gpt-4o",
                messages=[{"role": "user", "content": "Hello, how are you?"}],
                api_key="sk-test-key",
            )
        )

        assert result.total_tokens == 15
        assert result.tokenizer_type == "openai_api"
        assert result.request_model == "openai/gpt-4o"


def test_acount_tokens_routes_to_anthropic():
    """Test that acount_tokens routes to Anthropic token counter for anthropic/ models."""
    with patch(
        "litellm.llms.anthropic.count_tokens.token_counter.anthropic_count_tokens_handler.handle_count_tokens_request",
        new_callable=AsyncMock,
        return_value={"input_tokens": 20},
    ):
        result = asyncio.run(
            litellm.acount_tokens(
                model="anthropic/claude-3-5-sonnet-20241022",
                messages=[{"role": "user", "content": "Hello Claude!"}],
                api_key="sk-ant-test-key",
            )
        )

        assert result.total_tokens == 20
        assert result.tokenizer_type == "anthropic_api"
        assert result.request_model == "anthropic/claude-3-5-sonnet-20241022"


def test_acount_tokens_fallback_to_local():
    """Test that unsupported providers fall back to local tiktoken counting."""
    result = asyncio.run(
        litellm.acount_tokens(
            model="together_ai/meta-llama/Llama-3-8b-chat-hf",
            messages=[{"role": "user", "content": "Hello"}],
        )
    )

    assert result.total_tokens > 0
    assert result.tokenizer_type == "local_tokenizer"


def test_acount_tokens_with_tools():
    """Test that tools are passed through to the token counter."""
    tools = [
        {
            "type": "function",
            "function": {
                "name": "get_weather",
                "description": "Get weather info",
                "parameters": {
                    "type": "object",
                    "properties": {"city": {"type": "string"}},
                },
            },
        }
    ]

    with patch(
        "litellm.llms.openai.responses.count_tokens.token_counter.openai_count_tokens_handler.handle_count_tokens_request",
        new_callable=AsyncMock,
        return_value={"input_tokens": 30},
    ) as mock_handler:
        result = asyncio.run(
            litellm.acount_tokens(
                model="openai/gpt-4o",
                messages=[{"role": "user", "content": "What's the weather?"}],
                tools=tools,
                api_key="sk-test-key",
            )
        )

        assert result.total_tokens == 30
        mock_handler.assert_called_once()
        call_kwargs = mock_handler.call_args
        assert list(call_kwargs.kwargs.get("tools") or ()) == tools


def test_acount_tokens_with_system():
    """Test that system messages are passed through."""
    with patch(
        "litellm.llms.openai.responses.count_tokens.token_counter.openai_count_tokens_handler.handle_count_tokens_request",
        new_callable=AsyncMock,
        return_value={"input_tokens": 25},
    ):
        result = asyncio.run(
            litellm.acount_tokens(
                model="openai/gpt-4o",
                messages=[{"role": "user", "content": "Hello"}],
                system="You are a helpful assistant.",
                api_key="sk-test-key",
            )
        )

        assert result.total_tokens == 25


def test_acount_tokens_api_error_falls_back():
    """Test that API errors in token counting return error response."""
    from litellm.llms.openai.common_utils import OpenAIError

    with patch(
        "litellm.llms.openai.responses.count_tokens.token_counter.openai_count_tokens_handler.handle_count_tokens_request",
        new_callable=AsyncMock,
        side_effect=OpenAIError(status_code=401, message="Invalid API key"),
    ):
        result = asyncio.run(
            litellm.acount_tokens(
                model="openai/gpt-4o",
                messages=[{"role": "user", "content": "Hello"}],
                api_key="sk-bad-key",
            )
        )

        # Should fall back to local tokenizer when provider API errors
        assert result.error is False
        assert result.tokenizer_type == "local_tokenizer"
        assert result.total_tokens > 0


def test_acount_tokens_no_api_key_falls_back(monkeypatch):
    """Test that missing API key falls back to local counting."""
    monkeypatch.delenv("OPENAI_API_KEY", raising=False)
    result = asyncio.run(
        litellm.acount_tokens(
            model="openai/gpt-4o",
            messages=[{"role": "user", "content": "Hello"}],
        )
    )

    # Should fall back to local tokenizer since no API key
    assert result.total_tokens > 0
    assert result.tokenizer_type == "local_tokenizer"




@pytest.fixture
def httpx_transport_clients(monkeypatch):
    monkeypatch.setattr(litellm, "disable_aiohttp_transport", True)
    for name in ("ANTHROPIC_API_BASE", "ANTHROPIC_BASE_URL"):
        monkeypatch.delenv(name, raising=False)
    litellm.in_memory_llm_clients_cache.flush_cache()
    yield
    litellm.in_memory_llm_clients_cache.flush_cache()


def test_acount_tokens_sends_chat_tool_use_to_anthropic_in_the_shape_claude_inference_sends(httpx_transport_clients):
    schema = {"type": "object", "properties": {"city": {"type": "string"}}}
    with respx.mock(assert_all_called=True) as router:
        route = router.post("https://api.anthropic.com/v1/messages/count_tokens").mock(
            return_value=httpx.Response(200, json={"input_tokens": 52})
        )
        result = asyncio.run(
            litellm.acount_tokens(
                model="anthropic/claude-opus-5-5",
                messages=[
                    {"role": "user", "content": "Weather in SF?"},
                    {
                        "role": "assistant",
                        "content": None,
                        "tool_calls": [
                            {"id": "call_1", "type": "function", "function": {"name": "get_weather", "arguments": "{}"}}
                        ],
                    },
                    {"role": "tool", "tool_call_id": "call_1", "content": "sunny"},
                ],
                tools=[{"type": "function", "function": {"name": "get_weather", "parameters": schema}}],
                api_key="sk-ant-test-key",
            )
        )

    assert (result.error, result.total_tokens) == (False, 52)
    sent = json.loads(route.calls.last.request.content)
    assert sent["tools"] == [{"name": "get_weather", "input_schema": schema, "type": "custom"}]
    assert [message["role"] for message in sent["messages"]] == ["user", "assistant", "user"]
    assert sent["messages"][2]["content"] == [{"type": "tool_result", "tool_use_id": "call_1", "content": "sunny"}]