mirror of
https://github.com/NVIDIA/TensorRT-LLM.git
synced 2026-01-27 22:23:25 +08:00
38 lines
984 B
Python
38 lines
984 B
Python
import os
|
|
import sys
|
|
|
|
import pytest
|
|
import requests
|
|
from openai_server import RemoteOpenAIServer
|
|
|
|
sys.path.append(os.path.join(os.path.dirname(__file__), '..'))
|
|
from test_llm import get_model_path
|
|
|
|
from tensorrt_llm.version import __version__ as VERSION
|
|
|
|
|
|
@pytest.fixture(scope="module")
|
|
def model_name():
|
|
return "llama-models-v3/llama-v3-8b-instruct-hf"
|
|
|
|
|
|
@pytest.fixture(scope="module")
|
|
def server(model_name: str):
|
|
model_path = get_model_path(model_name)
|
|
args = ["--max_beam_width", "4"]
|
|
with RemoteOpenAIServer(model_path, args) as remote_server:
|
|
yield remote_server
|
|
|
|
|
|
def test_version(server: RemoteOpenAIServer):
|
|
version_url = server.url_for("version")
|
|
response = requests.get(version_url)
|
|
assert response.status_code == 200
|
|
assert response.json()["version"] == VERSION
|
|
|
|
|
|
def test_health(server: RemoteOpenAIServer):
|
|
health_url = server.url_for("health")
|
|
response = requests.get(health_url)
|
|
assert response.status_code == 200
|