litellm/tests/translation_characterization/cases/params_sampling.json
mateo-berri 85585b63b0 test: add translation characterization corpus pinning v1 transform seams
Snapshot corpus for the translation v2 differential gate: 19 OpenAI-format
request cases x {anthropic, bedrock converse, bedrock invoke} through the
real v1 transform_request path (get_optional_params, validate_environment
included), recorded provider responses through transform_response, SSE and
event-stream replays through CustomStreamWrapper plus the real decoders, and
both /v1/messages surfaces (anthropic<->openai adapter round-trip and the
native AnthropicMessagesConfig quirks: max_tokens pop, reasoning_effort
rewrite, advisor-block stripping). 89 canonical-JSON snapshots, deterministic
via frozen uuid/fastuuid/time, regenerated with --snapshot-update or
SNAPSHOT_UPDATE=1. Wired into CI as a job in test-unit-llm-providers.yml and
a make test-characterization target
2026-06-11 18:48:10 +00:00

20 lines
309 B
JSON

{
"id": "params_sampling",
"messages": [
{
"content": "What is the capital of France?",
"role": "user"
}
],
"params": {
"max_tokens": 128,
"stop": [
"\n\n",
"END"
],
"temperature": 0.2,
"top_p": 0.9,
"user": "char-user-1234"
},
"skip": {}
}