fix(tests): use github-hosted PDF fixture for Anthropic Files API test

Anthropic's URL fetcher intermittently returns 400 'Unable to download
the file' for the Wikipedia URL the test was using. Point it at the
repo's existing tests/llm_translation/fixtures/dummy.pdf via raw
GitHub instead — small, deterministic, reliably fetchable.

With a stable URL the test no longer needs to be opted out of VCR;
remove it from the incompatible list so it can replay from cassette.
This commit is contained in:
mateo-berri 2026-05-01 14:07:32 -07:00
parent a47c4e7d1b
commit 225d01cb4f
2 changed files with 4 additions and 3 deletions

View file

@ -401,7 +401,10 @@ class BaseLLMChatTest(ABC):
{
"type": "file",
"file": {
"file_id": "https://upload.wikimedia.org/wikipedia/commons/2/20/Re_example.pdf"
# GitHub-hosted fixture (small, deterministic, reliably
# fetchable from Anthropic's egress) instead of a Wikipedia
# URL that intermittently returns 400 "Unable to download".
"file_id": "https://raw.githubusercontent.com/BerriAI/litellm/main/tests/llm_translation/fixtures/dummy.pdf"
},
},
]

View file

@ -63,8 +63,6 @@ _VCR_INCOMPATIBLE_NODEID_SUFFIXES = frozenset(
# Provider prompt-cache propagation isn't deterministic between two
# back-to-back calls; the test is flaky against the live provider.
"::test_prompt_caching",
# Wikipedia URL fetch through Anthropic Files API is flaky upstream.
"::test_async_pdf_handling_with_file_id",
# Bedrock Nova returns tool_call vs JSON nondeterministically; the
# base assertion expects JSON. Other providers' versions of this
# test are healthy, so we narrow with a class-name guard below.