apps/FastDeploy
1
0
Fork 0
mirror of https://github.com/PaddlePaddle/FastDeploy.git synced 2025-10-17 14:11:14 +08:00
Code Issues Actions 5 Packages Projects Releases Wiki Activity

load hadamard_block_size from config (#3797)

Browse Source
This commit is contained in:
Yuan Xiaolan
2025-09-05 17:07:58 +08:00
committed by GitHub
parent 41aee08982
commit 2cf55168ca
10 changed files with 60 additions and 30 deletions
Download Patch File Download Diff File Expand all files Collapse all files

12
tests/model_loader/test_w4a8_model.py
View File

@@ -23,10 +23,10 @@ from fastdeploy.entrypoints.llm import LLM
bash_path = os.getenv("MODEL_PATH")
FD_ENGINE_QUEUE_PORTS = [
[9961, 9962, 9963, 9964, 9965, 9966, 9967, 9968],
[9971, 9972, 9973, 9974, 9975, 9976, 9977, 9978],
[9981, 9982, 9983, 9984, 9985, 9986, 9987, 9988],
[9991, 9992, 9993, 9994, 9995, 9996, 9997, 9998],
[9961, 9962],
[9971, 9972],
[9981, 9982],
[9991, 9992],
]
@@ -49,7 +49,7 @@ def llm(request):
llm_instance = LLM(
model=model_path,
tensor_parallel_size=1,
data_parallel_size=8,
data_parallel_size=2,
max_model_len=8192,
num_gpu_blocks_override=1024,
engine_worker_queue_port=FD_ENGINE_QUEUE_PORTS[port_index],
@@ -58,7 +58,7 @@ def llm(request):
)
yield weakref.proxy(llm_instance)
except Exception as e:
pytest.skip(f"LLM initialization failed: {e}")
assert False, f"LLM initialization failed: {e}"
@pytest.mark.timeout(60)