phoenix/litellm-mirror
0
0
Fork 1
mirror of https://github.com/BerriAI/litellm.git synced 2025-04-25 02:34:29 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

Revert "fix: Anthropic prompt caching on GCP Vertex AI (#9605)" (#9670)

Browse source 
This reverts commit a8673246dc.
This commit is contained in:
Krish Dholakia 2025-03-31 17:13:55 -07:00 • committed by GitHub
parent b37cb642c1
commit 46b3dbde8f
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
2 changed files with 2 additions and 145 deletions
Download patch file Download diff file Expand all files Collapse all files

10
litellm/llms/anthropic/chat/transformation.py
View file

@ -141,15 +141,9 @@ class AnthropicConfig(BaseConfig):
if user_anthropic_beta_headers is not None:
betas.update(user_anthropic_beta_headers)
# Handle beta headers for Vertex AI
# We allow prompt caching beta header for Vertex, but exclude other beta headers that might cause issues
# Don't send any beta headers to Vertex, Vertex has failed requests when they are sent
if is_vertex_request is True:
vertex_safe_betas = set()
# Allow prompt caching beta header for Vertex
if "prompt-caching-2024-07-31" in betas:
vertex_safe_betas.add("prompt-caching-2024-07-31")
if len(vertex_safe_betas) > 0:
headers["anthropic-beta"] = ",".join(vertex_safe_betas)
pass
elif len(betas) > 0:
headers["anthropic-beta"] = ",".join(betas)

137
tests/litellm/llms/vertex_ai/test_vertex_anthropic_prompt_caching.py
View file

@ -1,137 +0,0 @@
import os
import sys
from unittest.mock import MagicMock, patch
import pytest
sys.path.insert(
0, os.path.abspath("../../../..")
) # Adds the parent directory to the system path
from litellm.llms.anthropic.chat.transformation import AnthropicConfig
def test_anthropic_prompt_caching_headers_for_vertex():
"""
Test that the prompt caching beta header is correctly set for Vertex AI requests
with Anthropic models when cache control is present in the messages.
"""
# Create an instance of AnthropicConfig
config = AnthropicConfig()
# Test case 1: Vertex request with prompt caching
# Create a message with cache control
messages = [
{
"role": "system",
"content": "You are a helpful assistant.",
"cache_control": {"type": "ephemeral"}
},
{
"role": "user",
"content": "Tell me about the solar system."
}
]
# Check if cache control is detected
is_cache_control_set = config.is_cache_control_set(messages=messages)
assert is_cache_control_set is True, "Cache control should be detected in messages"
# Generate headers for a Vertex AI request with prompt caching
headers = config.get_anthropic_headers(
api_key="test-api-key",
prompt_caching_set=is_cache_control_set,
is_vertex_request=True
)
# Verify that the anthropic-beta header is set with prompt-caching-2024-07-31
assert "anthropic-beta" in headers, "anthropic-beta header should be present"
assert "prompt-caching-2024-07-31" in headers["anthropic-beta"], "prompt-caching-2024-07-31 should be in the beta header"
# Test case 2: Vertex request without prompt caching
messages_without_cache = [
{
"role": "system",
"content": "You are a helpful assistant."
},
{
"role": "user",
"content": "Tell me about the solar system."
}
]
# Check if cache control is detected
is_cache_control_set = config.is_cache_control_set(messages=messages_without_cache)
assert is_cache_control_set is False, "Cache control should not be detected in messages"
# Generate headers for a Vertex AI request without prompt caching
headers = config.get_anthropic_headers(
api_key="test-api-key",
prompt_caching_set=is_cache_control_set,
is_vertex_request=True
)
# Verify that the anthropic-beta header is not set
assert "anthropic-beta" not in headers, "anthropic-beta header should not be present"
def test_anthropic_prompt_caching_with_content_blocks():
"""
Test that prompt caching is correctly detected when cache control is in content blocks.
"""
config = AnthropicConfig()
# Message with cache control in content blocks
messages = [
{
"role": "system",
"content": [
{
"type": "text",
"text": "You are a helpful assistant.",
"cache_control": {"type": "ephemeral"}
}
]
},
{
"role": "user",
"content": "Tell me about the solar system."
}
]
# Check if cache control is detected
is_cache_control_set = config.is_cache_control_set(messages=messages)
assert is_cache_control_set is True, "Cache control should be detected in content blocks"
# Generate headers for a Vertex AI request with prompt caching
headers = config.get_anthropic_headers(
api_key="test-api-key",
prompt_caching_set=is_cache_control_set,
is_vertex_request=True
)
# Verify that the anthropic-beta header is set with prompt-caching-2024-07-31
assert "anthropic-beta" in headers, "anthropic-beta header should be present"
assert "prompt-caching-2024-07-31" in headers["anthropic-beta"], "prompt-caching-2024-07-31 should be in the beta header"
def test_anthropic_vertex_other_beta_headers():
"""
Test that other beta headers are not included for Vertex AI requests.
"""
config = AnthropicConfig()
# Generate headers with multiple beta features
headers = config.get_anthropic_headers(
api_key="test-api-key",
prompt_caching_set=True,
computer_tool_used=True, # This should be excluded for Vertex
pdf_used=True, # This should be excluded for Vertex
is_vertex_request=True
)
# Verify that only prompt-caching is included in the beta header
assert "anthropic-beta" in headers, "anthropic-beta header should be present"
assert headers["anthropic-beta"] == "prompt-caching-2024-07-31", "Only prompt-caching should be in the beta header"
assert "computer-use-2024-10-22" not in headers["anthropic-beta"], "computer-use beta should not be included"
assert "pdfs-2024-09-25" not in headers["anthropic-beta"], "pdfs beta should not be included"