test(zai): type the zai provider tests and mark their locals Final

This commit is contained in:
mateo-berri 2026-10-29 18:56:34 -07:00
parent 218e7e08ef
commit aaffadebef

View file

@ -4,8 +4,11 @@ Tests for Z.AI (Zhipu AI) provider - GLM models
import json
import math
from collections.abc import Mapping
from typing import Final
import pytest
import respx
import litellm
from litellm import completion
@ -13,7 +16,7 @@ from litellm.cost_calculator import cost_per_token
from litellm.llms.zai.chat.transformation import ZAIChatConfig
from litellm.types.utils import PromptTokensDetailsWrapper, Usage
JSON_SCHEMA_RESPONSE_FORMAT = {
JSON_SCHEMA_RESPONSE_FORMAT: Final = {
"type": "json_schema",
"json_schema": {
"name": "profile",
@ -112,7 +115,9 @@ def test_zai_sync_completion(respx_mock, zai_response, monkeypatch):
assert response.usage.total_tokens == 25
def test_zai_sends_response_format_and_maps_max_completion_tokens(respx_mock, zai_response, monkeypatch):
def test_zai_sends_response_format_and_maps_max_completion_tokens(
respx_mock: respx.MockRouter, zai_response: Mapping[str, object], monkeypatch: pytest.MonkeyPatch
) -> None:
monkeypatch.setenv("ZAI_API_KEY", "test-api-key")
monkeypatch.setattr(litellm, "disable_aiohttp_transport", True)
respx_mock.post("https://api.z.ai/api/paas/v4/chat/completions").respond(json=zai_response)
@ -124,7 +129,7 @@ def test_zai_sends_response_format_and_maps_max_completion_tokens(respx_mock, za
max_completion_tokens=64,
)
body = json.loads(respx_mock.calls[0].request.content)
body: Final = json.loads(respx_mock.calls[0].request.content)
assert body["response_format"] == {"type": "json_object"}
assert body["max_tokens"] == 64
assert "max_completion_tokens" not in body
@ -134,8 +139,8 @@ def test_zai_sends_response_format_and_maps_max_completion_tokens(respx_mock, za
"token_limits",
[{"max_tokens": 300, "max_completion_tokens": 15}, {"max_completion_tokens": 15, "max_tokens": 300}],
)
def test_zai_max_completion_tokens_wins_over_max_tokens(token_limits):
mapped = ZAIChatConfig().map_openai_params(token_limits, {}, "glm-5.3", drop_params=False)
def test_zai_max_completion_tokens_wins_over_max_tokens(token_limits: dict[str, int]) -> None:
mapped: Final = ZAIChatConfig().map_openai_params(token_limits, {}, "glm-5.3", drop_params=False)
assert mapped["max_tokens"] == 15
assert "max_completion_tokens" not in mapped
@ -146,12 +151,18 @@ def test_zai_max_completion_tokens_wins_over_max_tokens(token_limits):
[(1, True, False), (1, False, True), (0, True, True)],
)
def test_zai_drop_params_drops_max_completion_tokens_above_model_limit(
local_model_cost_map, respx_mock, zai_response, monkeypatch, over_limit_by, drop_params, forwarded
):
local_model_cost_map: None,
respx_mock: respx.MockRouter,
zai_response: Mapping[str, object],
monkeypatch: pytest.MonkeyPatch,
over_limit_by: int,
drop_params: bool,
forwarded: bool,
) -> None:
monkeypatch.setenv("ZAI_API_KEY", "test-api-key")
monkeypatch.setattr(litellm, "disable_aiohttp_transport", True)
respx_mock.post("https://api.z.ai/api/paas/v4/chat/completions").respond(json=zai_response)
requested = litellm.model_cost["zai/glm-4.5-flash"]["max_output_tokens"] + over_limit_by
requested: Final = litellm.model_cost["zai/glm-4.5-flash"]["max_output_tokens"] + over_limit_by
completion(
model="zai/glm-4.5-flash",
@ -160,19 +171,23 @@ def test_zai_drop_params_drops_max_completion_tokens_above_model_limit(
drop_params=drop_params,
)
body = json.loads(respx_mock.calls[0].request.content)
body: Final = json.loads(respx_mock.calls[0].request.content)
assert body.get("max_tokens") == (requested if forwarded else None)
@pytest.mark.parametrize("with_over_limit_max_completion_tokens", [False, True])
def test_zai_drop_params_keeps_max_tokens_above_model_limit(
local_model_cost_map, respx_mock, zai_response, monkeypatch, with_over_limit_max_completion_tokens
):
local_model_cost_map: None,
respx_mock: respx.MockRouter,
zai_response: Mapping[str, object],
monkeypatch: pytest.MonkeyPatch,
with_over_limit_max_completion_tokens: bool,
) -> None:
monkeypatch.setenv("ZAI_API_KEY", "test-api-key")
monkeypatch.setattr(litellm, "disable_aiohttp_transport", True)
respx_mock.post("https://api.z.ai/api/paas/v4/chat/completions").respond(json=zai_response)
over_limit = litellm.model_cost["zai/glm-4.5-flash"]["max_output_tokens"] + 1
over_limit_max_completion_tokens = (
over_limit: Final = litellm.model_cost["zai/glm-4.5-flash"]["max_output_tokens"] + 1
over_limit_max_completion_tokens: Final = (
{"max_completion_tokens": over_limit + 1} if with_over_limit_max_completion_tokens else {}
)
@ -184,30 +199,38 @@ def test_zai_drop_params_keeps_max_tokens_above_model_limit(
**over_limit_max_completion_tokens,
)
body = json.loads(respx_mock.calls[0].request.content)
body: Final = json.loads(respx_mock.calls[0].request.content)
assert body["max_tokens"] == over_limit
@pytest.mark.parametrize("pass_through", [{}, {"allowed_openai_params": ["response_format"]}])
def test_zai_forwards_json_schema_to_models_that_enforce_it(
local_model_cost_map, respx_mock, zai_response, monkeypatch, pass_through
):
local_model_cost_map: None,
respx_mock: respx.MockRouter,
zai_response: Mapping[str, object],
monkeypatch: pytest.MonkeyPatch,
pass_through: dict[str, list[str]],
) -> None:
monkeypatch.setenv("ZAI_API_KEY", "test-api-key")
monkeypatch.setattr(litellm, "disable_aiohttp_transport", True)
respx_mock.post("https://api.z.ai/api/paas/v4/chat/completions").respond(json=zai_response)
messages = [{"role": "user", "content": "Ann likes 7"}]
messages: Final = [{"role": "user", "content": "Ann likes 7"}]
completion(model="zai/glm-5.3", messages=messages, response_format=JSON_SCHEMA_RESPONSE_FORMAT, **pass_through)
body = json.loads(respx_mock.calls[0].request.content)
body: Final = json.loads(respx_mock.calls[0].request.content)
assert body["response_format"] == JSON_SCHEMA_RESPONSE_FORMAT
assert body["messages"] == messages
@pytest.mark.parametrize("pass_through", [{}, {"allowed_openai_params": ["response_format"]}])
def test_zai_puts_json_schema_in_prompt_for_models_that_ignore_it(
local_model_cost_map, respx_mock, zai_response, monkeypatch, pass_through
):
local_model_cost_map: None,
respx_mock: respx.MockRouter,
zai_response: Mapping[str, object],
monkeypatch: pytest.MonkeyPatch,
pass_through: dict[str, list[str]],
) -> None:
monkeypatch.setenv("ZAI_API_KEY", "test-api-key")
monkeypatch.setattr(litellm, "disable_aiohttp_transport", True)
respx_mock.post("https://api.z.ai/api/paas/v4/chat/completions").respond(json=zai_response)
@ -219,17 +242,17 @@ def test_zai_puts_json_schema_in_prompt_for_models_that_ignore_it(
**pass_through,
)
body = json.loads(respx_mock.calls[0].request.content)
body: Final = json.loads(respx_mock.calls[0].request.content)
assert body["response_format"] == {"type": "json_object"}
assert body["messages"][0] == {"role": "user", "content": "Ann likes 7"}
assert body["messages"][-1]["role"] == "user"
prompt = body["messages"][-1]["content"]
rendered_schema = prompt[prompt.index("{") : prompt.rindex("}") + 1]
prompt: Final = body["messages"][-1]["content"]
rendered_schema: Final = prompt[prompt.index("{") : prompt.rindex("}") + 1]
assert json.loads(rendered_schema) == JSON_SCHEMA_RESPONSE_FORMAT["json_schema"]["schema"]
def test_zai_glm_5v_turbo_bills_cached_input_at_cache_read_price(local_model_cost_map: None) -> None:
usage = Usage(
usage: Final = Usage(
prompt_tokens=1000,
completion_tokens=0,
total_tokens=1000,
@ -238,7 +261,7 @@ def test_zai_glm_5v_turbo_bills_cached_input_at_cache_read_price(local_model_cos
prompt_cost, _ = cost_per_token(model="glm-5v-turbo", custom_llm_provider="zai", usage_object=usage)
entry = litellm.model_cost["zai/glm-5v-turbo"]
entry: Final = litellm.model_cost["zai/glm-5v-turbo"]
assert math.isclose(
prompt_cost, 200 * entry["input_cost_per_token"] + 800 * entry["cache_read_input_token_cost"]
)
@ -250,12 +273,17 @@ def test_zai_cost_resolves_case_insensitively(local_model_cost_map: None, model:
model=model, custom_llm_provider="zai", prompt_tokens=1000, completion_tokens=1000
)
entry = litellm.model_cost[f"zai/{model.lower()}"]
entry: Final = litellm.model_cost[f"zai/{model.lower()}"]
assert math.isclose(prompt_cost, 1000 * entry["input_cost_per_token"])
assert math.isclose(completion_cost, 1000 * entry["output_cost_per_token"])
def test_zai_glm_5v_turbo_accepts_thinking(local_model_cost_map: None, respx_mock, zai_response, monkeypatch):
def test_zai_glm_5v_turbo_accepts_thinking(
local_model_cost_map: None,
respx_mock: respx.MockRouter,
zai_response: Mapping[str, object],
monkeypatch: pytest.MonkeyPatch,
) -> None:
monkeypatch.setenv("ZAI_API_KEY", "test-api-key")
monkeypatch.setattr(litellm, "disable_aiohttp_transport", True)
respx_mock.post("https://api.z.ai/api/paas/v4/chat/completions").respond(json=zai_response)
@ -266,5 +294,5 @@ def test_zai_glm_5v_turbo_accepts_thinking(local_model_cost_map: None, respx_moc
thinking={"type": "disabled"},
)
body = json.loads(respx_mock.calls[0].request.content)
body: Final = json.loads(respx_mock.calls[0].request.content)
assert body["thinking"] == {"type": "disabled"}