mirror of
https://github.com/BerriAI/litellm.git
synced 2026-10-28 06:23:07 +00:00
c57d273aa6
Streamed transcription runs unstreamed when a deployment post_call guardrail applies, so the guardrail still sees the transcript. Azure transcription maps v1, latest and preview to the default dated version on the deployments route for every model. The SDK client secret keeps session.model ahead of model, realtime WebSockets connect to api_base again, realtime metadata keeps only guardrails, and managed agents need an explicit transcription model. Reverts the two unrelated streamed Responses error codes and the duplicate cached_tokens_details field, and the gpt-transcribe format check only reads the request provider's cost map entry.
72 lines
2.8 KiB
Python
72 lines
2.8 KiB
Python
from collections.abc import Mapping
|
|
from typing import Any, Final, cast
|
|
|
|
import pytest
|
|
from pydantic import TypeAdapter
|
|
|
|
from litellm.proxy.common_utils.realtime_utils import realtime_request_body
|
|
from litellm.proxy.proxy_server import _realtime_query_params_template
|
|
|
|
|
|
@pytest.fixture(autouse=True)
|
|
def clear_realtime_caches():
|
|
realtime_request_body.cache_clear()
|
|
_realtime_query_params_template.cache_clear()
|
|
yield
|
|
realtime_request_body.cache_clear()
|
|
_realtime_query_params_template.cache_clear()
|
|
|
|
|
|
def test_realtime_request_body_returns_immutable_bytes():
|
|
cached_body = realtime_request_body("gpt-4o")
|
|
|
|
with pytest.raises(TypeError):
|
|
cast(Any, cached_body)[0] = ord("x")
|
|
|
|
|
|
def test_realtime_query_params_template_returns_immutable_tuples():
|
|
cached_tuple = _realtime_query_params_template("gpt-4o", "intent-a")
|
|
|
|
with pytest.raises(TypeError):
|
|
cast(Any, cached_tuple)[0] = ("model", "mutated")
|
|
|
|
|
|
def test_realtime_request_body_caches_each_model_separately():
|
|
gpt4o_body_first = realtime_request_body("gpt-4o")
|
|
gpt4o_body_second = realtime_request_body("gpt-4o")
|
|
gpt4o_mini_body = realtime_request_body("gpt-4o-mini")
|
|
|
|
assert gpt4o_body_first is gpt4o_body_second
|
|
assert gpt4o_body_first == b'{"model": "gpt-4o"}'
|
|
assert gpt4o_mini_body == b'{"model": "gpt-4o-mini"}'
|
|
assert gpt4o_body_first is not gpt4o_mini_body
|
|
|
|
|
|
@pytest.mark.parametrize("model", (None, "requested", 'model"with\\quotes', "model\nwith-newline"))
|
|
def test_realtime_request_body_preserves_model_names(model: str | None) -> None:
|
|
body: Final = TypeAdapter(Mapping[str, object]).validate_json(realtime_request_body(model))
|
|
|
|
assert body["model"] == model
|
|
|
|
|
|
def test_realtime_query_params_template_caches_each_pair_separately():
|
|
params_with_intent_first = _realtime_query_params_template("gpt-4o", "intent-a")
|
|
params_with_intent_second = _realtime_query_params_template("gpt-4o", "intent-a")
|
|
params_without_intent = _realtime_query_params_template("gpt-4o", None)
|
|
params_transcription_without_model = _realtime_query_params_template(None, "transcription")
|
|
|
|
assert params_with_intent_first is params_with_intent_second
|
|
assert params_with_intent_first == (("model", "gpt-4o"), ("intent", "intent-a"))
|
|
assert params_without_intent == (("model", "gpt-4o"),)
|
|
assert params_transcription_without_model == (("intent", "transcription"),)
|
|
assert params_with_intent_first is not params_without_intent
|
|
|
|
|
|
def test_realtime_query_params_dict_copies_do_not_leak_state():
|
|
params_dict_one = dict(_realtime_query_params_template("gpt-4o", "intent-a"))
|
|
params_dict_one["new"] = "value"
|
|
|
|
params_dict_two = dict(_realtime_query_params_template("gpt-4o", "intent-a"))
|
|
|
|
assert "new" not in params_dict_two
|
|
assert params_dict_two == {"model": "gpt-4o", "intent": "intent-a"}
|