EngineX-Hygon/sglang
5
0
Fork 0
Code Issues Pull Requests Actions 7 Projects Releases Wiki Activity

chore: update vllm to 0.5.4 (#966)

Browse Source
This commit is contained in:
Yineng Zhang
2024-08-07 19:15:41 +08:00
committed by GitHub
parent a01ddd9605
commit c31f084c71
14 changed files with 15 additions and 18 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
test/srt/models/test_causal_models.py
View File

@@ -18,9 +18,7 @@ import torch
from sglang.test.runners import DEFAULT_PROMPTS, HFRunner, SRTRunner
MODELS = [
# (model_name, tp_size)
("meta-llama/Meta-Llama-3.1-8B-Instruct", 1),
# ("meta-llama/Meta-Llama-3.1-8B-Instruct", 2),
]
TORCH_DTYPES = [torch.float16]
@@ -51,7 +49,7 @@ class TestCausalModels(unittest.TestCase):
hf_logprobs = torch.Tensor(hf_outputs.top_input_logprobs[i])
srt_logprobs = torch.Tensor(srt_outputs.top_input_logprobs[i])
tolerance = 2e-2
tolerance = 3e-2
assert torch.all(
abs(hf_logprobs - srt_logprobs) < tolerance
), f"prefill logprobs not all close"

2
test/srt/run_suite.py
View File

@@ -20,7 +20,7 @@ if __name__ == "__main__":
arg_parser.add_argument(
"--timeout-per-file",
type=int,
default=1000,
default=2000,
help="The time limit for running one file in seconds.",
)
arg_parser.add_argument(

2
test/srt/test_chunked_prefill.py
View File

@@ -11,7 +11,7 @@ class TestAccuracy(unittest.TestCase):
@classmethod
def setUpClass(cls):
cls.model = DEFAULT_MODEL_NAME_FOR_TEST
cls.base_url = f"http://localhost:8157"
cls.base_url = "http://127.0.0.1:8157"
cls.process = popen_launch_server(
cls.model,
cls.base_url,

2
test/srt/test_eval_accuracy.py
View File

@@ -11,7 +11,7 @@ class TestAccuracy(unittest.TestCase):
@classmethod
def setUpClass(cls):
cls.model = DEFAULT_MODEL_NAME_FOR_TEST
cls.base_url = f"http://localhost:8157"
cls.base_url = "http://127.0.0.1:8157"
cls.process = popen_launch_server(cls.model, cls.base_url, timeout=300)
@classmethod

2
test/srt/test_openai_server.py
View File

@@ -14,7 +14,7 @@ class TestOpenAIServer(unittest.TestCase):
@classmethod
def setUpClass(cls):
cls.model = DEFAULT_MODEL_NAME_FOR_TEST
cls.base_url = f"http://localhost:8157"
cls.base_url = "http://127.0.0.1:8157"
cls.api_key = "sk-123456"
cls.process = popen_launch_server(
cls.model, cls.base_url, timeout=300, api_key=cls.api_key

2
test/srt/test_srt_endpoint.py
View File

@@ -13,7 +13,7 @@ class TestSRTEndpoint(unittest.TestCase):
@classmethod
def setUpClass(cls):
cls.model = DEFAULT_MODEL_NAME_FOR_TEST
cls.base_url = f"http://localhost:{8157}"
cls.base_url = "http://127.0.0.1:8157"
cls.process = popen_launch_server(cls.model, cls.base_url, timeout=300)
@classmethod

2
test/srt/test_torch_compile.py
View File

@@ -11,7 +11,7 @@ class TestAccuracy(unittest.TestCase):
@classmethod
def setUpClass(cls):
cls.model = DEFAULT_MODEL_NAME_FOR_TEST
cls.base_url = f"http://localhost:8157"
cls.base_url = "http://127.0.0.1:8157"
cls.process = popen_launch_server(
cls.model, cls.base_url, timeout=300, other_args=["--enable-torch-compile"]
)

2
test/srt/test_vision_openai_server.py
View File

@@ -13,7 +13,7 @@ class TestOpenAIVisionServer(unittest.TestCase):
@classmethod
def setUpClass(cls):
cls.model = "liuhaotian/llava-v1.6-vicuna-7b"
cls.base_url = "http://localhost:8157"
cls.base_url = "http://127.0.0.1:8157"
cls.api_key = "sk-123456"
cls.process = popen_launch_server(
cls.model,