Direct model loading from object storage with Runai Model Streamer (#17948)

Signed-off-by: Noa Neria <noa@run.ai>
This commit is contained in:
Noa Neria
2026-04-01 18:41:22 -07:00
committed by GitHub
parent ae3b207dfd
commit 8d9145d97e
14 changed files with 659 additions and 17 deletions
@@ -0,0 +1,50 @@
import unittest
import sglang as sgl
from sglang.srt.environ import temp_set_env
from sglang.test.ci.ci_register import register_cuda_ci
from sglang.test.test_utils import CustomTestCase
register_cuda_ci(est_time=120, suite="stage-b-test-1-gpu-small")
TEST_GCS_MODEL = "gs://vertex-model-garden-public-us/codegemma/codegemma-2b/"
PROMPTS = [
"Hello, my name is",
"The president of the United States is",
"The capital of France is",
"The future of AI is",
]
class TestRunaiModelLoader(CustomTestCase):
@classmethod
def setUpClass(cls):
with temp_set_env(
GOOGLE_CLOUD_PROJECT="fake-project",
RUNAI_STREAMER_GCS_USE_ANONYMOUS_CREDENTIALS="true",
CLOUD_STORAGE_EMULATOR_ENDPOINT="https://storage.googleapis.com",
):
cls.engine = sgl.Engine(
model_path=TEST_GCS_MODEL,
load_format="runai_streamer",
cuda_graph_max_bs=1,
max_total_tokens=64,
)
@classmethod
def tearDownClass(cls):
if hasattr(cls, "engine") and cls.engine:
cls.engine.shutdown()
def test_generate_produces_output(self):
outputs = self.engine.generate(PROMPTS)
self.assertEqual(len(outputs), len(PROMPTS))
for i, output in enumerate(outputs):
text = output["text"]
self.assertIsInstance(text, str)
self.assertGreater(len(text), 0, f"Prompt {i} produced empty output")
if __name__ == "__main__":
unittest.main()
+57
View File
@@ -0,0 +1,57 @@
import unittest
from pathlib import Path
from sglang.srt.configs.load_config import LoadFormat
from sglang.srt.utils.runai_utils import ObjectStorageModel, is_runai_obj_uri
from sglang.test.ci.ci_register import register_cpu_ci
from sglang.test.test_utils import CustomTestCase
register_cpu_ci(est_time=5, suite="stage-a-test-cpu")
class TestRunaiUtils(CustomTestCase):
def test_is_runai_obj_uri_s3(self):
self.assertTrue(is_runai_obj_uri("s3://bucket/model/"))
self.assertTrue(is_runai_obj_uri("S3://Bucket/Model/"))
def test_is_runai_obj_uri_gs(self):
self.assertTrue(is_runai_obj_uri("gs://bucket/model/"))
self.assertTrue(is_runai_obj_uri("GS://Bucket/Model/"))
def test_is_runai_obj_uri_az(self):
self.assertTrue(is_runai_obj_uri("az://container/model/"))
self.assertTrue(is_runai_obj_uri("AZ://Container/Model/"))
def test_is_runai_obj_uri_local_paths(self):
self.assertFalse(is_runai_obj_uri("/path/to/model"))
self.assertFalse(is_runai_obj_uri("./relative/path"))
self.assertFalse(is_runai_obj_uri("meta-llama/Llama-3.2-1B"))
def test_is_runai_obj_uri_other_schemes(self):
self.assertFalse(is_runai_obj_uri("http://example.com/model"))
self.assertFalse(is_runai_obj_uri("https://example.com/model"))
self.assertFalse(is_runai_obj_uri("ftp://example.com/model"))
def test_is_runai_obj_uri_pathlib(self):
self.assertFalse(is_runai_obj_uri(Path("/local/model")))
def test_get_path_deterministic(self):
path1 = ObjectStorageModel.get_path("s3://bucket/model/")
path2 = ObjectStorageModel.get_path("s3://bucket/model/")
self.assertEqual(path1, path2)
def test_get_path_different_uris(self):
path1 = ObjectStorageModel.get_path("s3://bucket/model-a/")
path2 = ObjectStorageModel.get_path("s3://bucket/model-b/")
self.assertNotEqual(path1, path2)
def test_get_path_contains_model_streamer(self):
path = ObjectStorageModel.get_path("s3://bucket/model/")
self.assertIn("model_streamer", path)
def test_load_format_enum(self):
self.assertEqual(LoadFormat.RUNAI_STREAMER.value, "runai_streamer")
if __name__ == "__main__":
unittest.main()