forked from phoenix/litellm-mirror
2e5c46ef6d
* fix(litellm_logging.py): ensure cache hits are scrubbed if 'turn_off_message_logging' is enabled * fix(sagemaker.py): fix streaming to raise error immediately Fixes https://github.com/BerriAI/litellm/issues/6054 * (fixes) gcs bucket key based logging (#6044) * fixes for gcs bucket logging * fix StandardCallbackDynamicParams * fix - gcs logging when payload is not serializable * add test_add_callback_via_key_litellm_pre_call_utils_gcs_bucket * working success callbacks * linting fixes * fix linting error * add type hints to functions * fixes for dynamic success and failure logging * fix for test_async_chat_openai_stream * fix handle case when key based logging vars are set as os.environ/ vars * fix prometheus track cooldown events on custom logger (#6060) * (docs) add 1k rps load test doc (#6059) * docs 1k rps load test * docs load testing * docs load testing litellm * docs load testing * clean up load test doc * docs prom metrics for load testing * docs using prometheus on load testing * doc load testing with prometheus * (fixes) docs + qa - gcs key based logging (#6061) * fixes for required values for gcs bucket * docs gcs bucket logging * bump: version 1.48.12 → 1.48.13 * ci/cd run again * bump: version 1.48.13 → 1.48.14 * update load test doc * (docs) router settings - on litellm config (#6037) * add yaml with all router settings * add docs for router settings * docs router settings litellm settings * (feat) OpenAI prompt caching models to model cost map (#6063) * add prompt caching for latest models * add cache_read_input_token_cost for prompt caching models * fix(litellm_logging.py): check if param is iterable Fixes https://github.com/BerriAI/litellm/issues/6025#issuecomment-2393929946 * fix(factory.py): support passing an 'assistant_continue_message' to prevent bedrock error Fixes https://github.com/BerriAI/litellm/issues/6053 * fix(databricks/chat): handle streaming responses * fix(factory.py): fix linting error * fix(utils.py): unify anthropic + deepseek prompt caching information to openai format Fixes https://github.com/BerriAI/litellm/issues/6069 * test: fix test * fix(types/utils.py): support all openai roles Fixes https://github.com/BerriAI/litellm/issues/6052 * test: fix test --------- Co-authored-by: Ishaan Jaff <ishaanjaffer0324@gmail.com>
81 lines
2.8 KiB
Python
81 lines
2.8 KiB
Python
"""Asserts that prompt caching information is correctly returned for Anthropic, OpenAI, and Deepseek"""
|
|
|
|
import io
|
|
import os
|
|
import sys
|
|
|
|
sys.path.insert(0, os.path.abspath("../.."))
|
|
|
|
import litellm
|
|
import pytest
|
|
|
|
|
|
@pytest.mark.parametrize(
|
|
"model",
|
|
[
|
|
"anthropic/claude-3-5-sonnet-20240620",
|
|
"openai/gpt-4o",
|
|
"deepseek/deepseek-chat",
|
|
],
|
|
)
|
|
def test_prompt_caching_model(model):
|
|
for _ in range(2):
|
|
response = litellm.completion(
|
|
model=model,
|
|
messages=[
|
|
# System Message
|
|
{
|
|
"role": "system",
|
|
"content": [
|
|
{
|
|
"type": "text",
|
|
"text": "Here is the full text of a complex legal agreement"
|
|
* 400,
|
|
"cache_control": {"type": "ephemeral"},
|
|
}
|
|
],
|
|
},
|
|
# marked for caching with the cache_control parameter, so that this checkpoint can read from the previous cache.
|
|
{
|
|
"role": "user",
|
|
"content": [
|
|
{
|
|
"type": "text",
|
|
"text": "What are the key terms and conditions in this agreement?",
|
|
"cache_control": {"type": "ephemeral"},
|
|
}
|
|
],
|
|
},
|
|
{
|
|
"role": "assistant",
|
|
"content": "Certainly! the key terms and conditions are the following: the contract is 1 year long for $10/mo",
|
|
},
|
|
# The final turn is marked with cache-control, for continuing in followups.
|
|
{
|
|
"role": "user",
|
|
"content": [
|
|
{
|
|
"type": "text",
|
|
"text": "What are the key terms and conditions in this agreement?",
|
|
"cache_control": {"type": "ephemeral"},
|
|
}
|
|
],
|
|
},
|
|
],
|
|
temperature=0.2,
|
|
max_tokens=10,
|
|
)
|
|
|
|
print("response=", response)
|
|
print("response.usage=", response.usage)
|
|
|
|
assert "prompt_tokens_details" in response.usage
|
|
assert response.usage.prompt_tokens_details.cached_tokens > 0
|
|
|
|
# assert "cache_read_input_tokens" in response.usage
|
|
# assert "cache_creation_input_tokens" in response.usage
|
|
|
|
# # Assert either a cache entry was created or cache was read - changes depending on the anthropic api ttl
|
|
# assert (response.usage.cache_read_input_tokens > 0) or (
|
|
# response.usage.cache_creation_input_tokens > 0
|
|
# )
|