mirror of
https://github.com/BerriAI/litellm.git
synced 2025-04-26 03:04:13 +00:00
f47987e673
* anthropic_messages_handler v0 * fix /messages * working messages with router methods * test_anthropic_messages_handler_litellm_router_non_streaming * test_anthropic_messages_litellm_router_non_streaming_with_logging * AnthropicMessagesConfig * _handle_anthropic_messages_response_logging * working with /v1/messages endpoint * working /v1/messages endpoint * refactor to use router factory function * use aanthropic_messages * use BaseConfig for Anthropic /v1/messages * track api key, team on /v1/messages endpoint * fix get_logging_payload * BaseAnthropicMessagesTest * align test config * test_anthropic_messages_with_thinking * test_anthropic_streaming_with_thinking * fix - display anthropic url for debugging * test_bad_request_error_handling * test_anthropic_messages_router_streaming_with_bad_request * fix ProxyException * test_bad_request_error_handling_streaming * use provider_specific_header * test_anthropic_messages_with_extra_headers * test_anthropic_messages_to_wildcard_model * fix gcs pub sub test * standard_logging_payload * fix unit testing for anthopic /v1/messages support * fix pass through anthropic messages api * delete dead code * fix anthropic pass through response * revert change to spend tracking utils * fix get_litellm_metadata_from_kwargs * fix spend logs payload json * proxy_pass_through_endpoint_tests * TestAnthropicPassthroughBasic * fix pass through tests * test_async_vertex_proxy_route_api_key_auth * _handle_anthropic_messages_response_logging * vertex_credentials * test_set_default_vertex_config * test_anthropic_messages_litellm_router_non_streaming_with_logging * test_ageneric_api_call_with_fallbacks_basic * test__aadapter_completion
263 lines
11 KiB
Python
263 lines
11 KiB
Python
"""
|
|
This test ensures that the proxy can passthrough anthropic requests
|
|
"""
|
|
|
|
import pytest
|
|
import anthropic
|
|
import aiohttp
|
|
import asyncio
|
|
import json
|
|
|
|
|
|
@pytest.mark.asyncio
|
|
async def test_anthropic_basic_completion_with_headers():
|
|
print("making basic completion request to anthropic passthrough with aiohttp")
|
|
|
|
headers = {
|
|
"Authorization": f"Bearer sk-1234",
|
|
"Content-Type": "application/json",
|
|
"Anthropic-Version": "2023-06-01",
|
|
}
|
|
|
|
payload = {
|
|
"model": "claude-3-5-sonnet-20241022",
|
|
"max_tokens": 10,
|
|
"messages": [{"role": "user", "content": "Say 'hello test' and nothing else"}],
|
|
"litellm_metadata": {
|
|
"tags": ["test-tag-1", "test-tag-2"],
|
|
},
|
|
}
|
|
|
|
async with aiohttp.ClientSession() as session:
|
|
async with session.post(
|
|
"http://0.0.0.0:4000/anthropic/v1/messages", json=payload, headers=headers
|
|
) as response:
|
|
response_text = await response.text()
|
|
print(f"Response text: {response_text}")
|
|
|
|
response_json = await response.json()
|
|
response_headers = response.headers
|
|
print(
|
|
"non-streaming response",
|
|
json.dumps(response_json, indent=4, default=str),
|
|
)
|
|
reported_usage = response_json.get("usage", None)
|
|
anthropic_api_input_tokens = reported_usage.get("input_tokens", None)
|
|
anthropic_api_output_tokens = reported_usage.get("output_tokens", None)
|
|
litellm_call_id = response_headers.get("x-litellm-call-id")
|
|
|
|
print(f"LiteLLM Call ID: {litellm_call_id}")
|
|
|
|
# Wait for spend to be logged
|
|
await asyncio.sleep(15)
|
|
|
|
# Check spend logs for this specific request
|
|
async with session.get(
|
|
f"http://0.0.0.0:4000/spend/logs?request_id={litellm_call_id}",
|
|
headers={"Authorization": "Bearer sk-1234"},
|
|
) as spend_response:
|
|
print("text spend response")
|
|
print(f"Spend response: {spend_response}")
|
|
spend_data = await spend_response.json()
|
|
print(f"Spend data: {spend_data}")
|
|
assert spend_data is not None, "Should have spend data for the request"
|
|
|
|
log_entry = spend_data[
|
|
0
|
|
] # Get the first (and should be only) log entry
|
|
|
|
# Basic existence checks
|
|
assert spend_data is not None, "Should have spend data for the request"
|
|
assert isinstance(log_entry, dict), "Log entry should be a dictionary"
|
|
|
|
# Request metadata assertions
|
|
assert (
|
|
log_entry["request_id"] == litellm_call_id
|
|
), "Request ID should match"
|
|
assert (
|
|
log_entry["call_type"] == "pass_through_endpoint"
|
|
), "Call type should be pass_through_endpoint"
|
|
assert (
|
|
log_entry["api_base"] == "https://api.anthropic.com/v1/messages"
|
|
), "API base should be Anthropic's endpoint"
|
|
|
|
# Token and spend assertions
|
|
assert log_entry["spend"] > 0, "Spend value should not be None"
|
|
assert isinstance(
|
|
log_entry["spend"], (int, float)
|
|
), "Spend should be a number"
|
|
assert log_entry["total_tokens"] > 0, "Should have some tokens"
|
|
assert (
|
|
log_entry["prompt_tokens"] == anthropic_api_input_tokens
|
|
), f"Should have prompt tokens matching anthropic api. Expected {anthropic_api_input_tokens} but got {log_entry['prompt_tokens']}"
|
|
assert (
|
|
log_entry["completion_tokens"] == anthropic_api_output_tokens
|
|
), f"Should have completion tokens matching anthropic api. Expected {anthropic_api_output_tokens} but got {log_entry['completion_tokens']}"
|
|
assert (
|
|
log_entry["total_tokens"]
|
|
== log_entry["prompt_tokens"] + log_entry["completion_tokens"]
|
|
), "Total tokens should equal prompt + completion"
|
|
|
|
# Time assertions
|
|
assert all(
|
|
key in log_entry
|
|
for key in ["startTime", "endTime", "completionStartTime"]
|
|
), "Should have all time fields"
|
|
assert (
|
|
log_entry["startTime"] < log_entry["endTime"]
|
|
), "Start time should be before end time"
|
|
|
|
# Metadata assertions
|
|
assert (
|
|
str(log_entry["cache_hit"]).lower() != "true"
|
|
), "Cache should be off"
|
|
assert log_entry["request_tags"] == [
|
|
"test-tag-1",
|
|
"test-tag-2",
|
|
], "Tags should match input"
|
|
assert (
|
|
"user_api_key" in log_entry["metadata"]
|
|
), "Should have user API key in metadata"
|
|
|
|
assert "claude" in log_entry["model"]
|
|
assert log_entry["custom_llm_provider"] == "anthropic"
|
|
|
|
|
|
@pytest.mark.asyncio
|
|
async def test_anthropic_streaming_with_headers():
|
|
print("making streaming request to anthropic passthrough with aiohttp")
|
|
|
|
headers = {
|
|
"Authorization": f"Bearer sk-1234",
|
|
"Content-Type": "application/json",
|
|
"Anthropic-Version": "2023-06-01",
|
|
}
|
|
|
|
payload = {
|
|
"model": "claude-3-5-sonnet-20241022",
|
|
"max_tokens": 10,
|
|
"messages": [
|
|
{"role": "user", "content": "Say 'hello stream test' and nothing else"}
|
|
],
|
|
"stream": True,
|
|
"litellm_metadata": {
|
|
"tags": ["test-tag-stream-1", "test-tag-stream-2"],
|
|
"user": "test-user-1",
|
|
},
|
|
}
|
|
|
|
async with aiohttp.ClientSession() as session:
|
|
async with session.post(
|
|
"http://0.0.0.0:4000/anthropic/v1/messages", json=payload, headers=headers
|
|
) as response:
|
|
print("response status")
|
|
print(response.status)
|
|
assert response.status == 200, "Response should be successful"
|
|
response_headers = response.headers
|
|
print(f"Response headers: {response_headers}")
|
|
litellm_call_id = response_headers.get("x-litellm-call-id")
|
|
print(f"LiteLLM Call ID: {litellm_call_id}")
|
|
|
|
collected_output = []
|
|
async for line in response.content:
|
|
if line:
|
|
text = line.decode("utf-8").strip()
|
|
if text.startswith("data: "):
|
|
collected_output.append(text[6:]) # Remove 'data: ' prefix
|
|
|
|
print("Collected output:", "".join(collected_output))
|
|
anthropic_api_usage_chunks = []
|
|
for chunk in collected_output:
|
|
chunk_json = json.loads(chunk)
|
|
if "usage" in chunk_json:
|
|
anthropic_api_usage_chunks.append(chunk_json["usage"])
|
|
elif "message" in chunk_json and "usage" in chunk_json["message"]:
|
|
anthropic_api_usage_chunks.append(chunk_json["message"]["usage"])
|
|
|
|
print(
|
|
"anthropic_api_usage_chunks",
|
|
json.dumps(anthropic_api_usage_chunks, indent=4, default=str),
|
|
)
|
|
|
|
anthropic_api_input_tokens = sum(
|
|
[usage.get("input_tokens", 0) for usage in anthropic_api_usage_chunks]
|
|
)
|
|
anthropic_api_output_tokens = max(
|
|
[usage.get("output_tokens", 0) for usage in anthropic_api_usage_chunks]
|
|
)
|
|
print("anthropic_api_input_tokens", anthropic_api_input_tokens)
|
|
print("anthropic_api_output_tokens", anthropic_api_output_tokens)
|
|
|
|
# Wait for spend to be logged
|
|
await asyncio.sleep(20)
|
|
|
|
# Check spend logs for this specific request
|
|
async with session.get(
|
|
f"http://0.0.0.0:4000/spend/logs?request_id={litellm_call_id}",
|
|
headers={"Authorization": "Bearer sk-1234"},
|
|
) as spend_response:
|
|
spend_data = await spend_response.json()
|
|
print(f"Spend data: {spend_data}")
|
|
assert spend_data is not None, "Should have spend data for the request"
|
|
|
|
log_entry = spend_data[
|
|
0
|
|
] # Get the first (and should be only) log entry
|
|
|
|
# Basic existence checks
|
|
assert spend_data is not None, "Should have spend data for the request"
|
|
assert isinstance(log_entry, dict), "Log entry should be a dictionary"
|
|
|
|
# Request metadata assertions
|
|
assert (
|
|
log_entry["request_id"] == litellm_call_id
|
|
), "Request ID should match"
|
|
assert (
|
|
log_entry["call_type"] == "pass_through_endpoint"
|
|
), "Call type should be pass_through_endpoint"
|
|
# assert (
|
|
# log_entry["api_base"] == "https://api.anthropic.com/v1/messages"
|
|
# ), "API base should be Anthropic's endpoint"
|
|
|
|
# Token and spend assertions
|
|
assert log_entry["spend"] > 0, "Spend value should not be None"
|
|
assert isinstance(
|
|
log_entry["spend"], (int, float)
|
|
), "Spend should be a number"
|
|
assert log_entry["total_tokens"] > 0, "Should have some tokens"
|
|
assert (
|
|
log_entry["prompt_tokens"] == anthropic_api_input_tokens
|
|
), f"Should have prompt tokens matching anthropic api. Expected {anthropic_api_input_tokens} but got {log_entry['prompt_tokens']}"
|
|
assert (
|
|
log_entry["completion_tokens"] == anthropic_api_output_tokens
|
|
), f"Should have completion tokens matching anthropic api. Expected {anthropic_api_output_tokens} but got {log_entry['completion_tokens']}"
|
|
assert (
|
|
log_entry["total_tokens"]
|
|
== log_entry["prompt_tokens"] + log_entry["completion_tokens"]
|
|
), "Total tokens should equal prompt + completion"
|
|
|
|
# Time assertions
|
|
assert all(
|
|
key in log_entry
|
|
for key in ["startTime", "endTime", "completionStartTime"]
|
|
), "Should have all time fields"
|
|
assert (
|
|
log_entry["startTime"] < log_entry["endTime"]
|
|
), "Start time should be before end time"
|
|
|
|
# Metadata assertions
|
|
assert (
|
|
str(log_entry["cache_hit"]).lower() != "true"
|
|
), "Cache should be off"
|
|
assert log_entry["request_tags"] == [
|
|
"test-tag-stream-1",
|
|
"test-tag-stream-2",
|
|
], "Tags should match input"
|
|
assert (
|
|
"user_api_key" in log_entry["metadata"]
|
|
), "Should have user API key in metadata"
|
|
|
|
assert "claude" in log_entry["model"]
|
|
|
|
assert log_entry["end_user"] == "test-user-1"
|
|
assert log_entry["custom_llm_provider"] == "anthropic"
|