A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 14, 05:30 UTC
[CI/Build] Cleanup LoRA test (#26752)
Signed-off-by: Jee Jee Li <pandaleefree@gmail.com>
Jee Jee Li committed
Oct 14, 2025 at 20:06 UTC
fdd32750f0bce9edd05f85c3550d6ebc3b06931f
3 files changed
+3
-11
tests/lora/test_chatglm3_tp.py
-5
@@ -58,7 +58,6 @@ def test_chatglm3_lora(chatglm3_lora_files):
58
max_loras=4,
59
max_lora_rank=64,
60
trust_remote_code=True,
61
- enable_chunked_prefill=True,
61
)
62
63
output1 = do_sample(llm, chatglm3_lora_files, lora_id=1)
@@ -70,7 +69,6 @@ def test_chatglm3_lora(chatglm3_lora_files):
69
70
71
@multi_gpu_test(num_gpus=4)
73
-@create_new_process_for_each_test()
72
def test_chatglm3_lora_tp4(chatglm3_lora_files):
73
llm = vllm.LLM(
74
MODEL_PATH,
@@ -81,7 +79,6 @@ def test_chatglm3_lora_tp4(chatglm3_lora_files):
79
tensor_parallel_size=4,
80
trust_remote_code=True,
81
fully_sharded_loras=False,
84
- enable_chunked_prefill=True,
82
)
83
84
output1 = do_sample(llm, chatglm3_lora_files, lora_id=1)
@@ -93,7 +90,6 @@ def test_chatglm3_lora_tp4(chatglm3_lora_files):
90
91
92
@multi_gpu_test(num_gpus=4)
96
-@create_new_process_for_each_test()
93
def test_chatglm3_lora_tp4_fully_sharded_loras(chatglm3_lora_files):
94
# https://github.com/NVIDIA/nccl/issues/1790, set a lower value for
95
# gpu_memory_utilization here because NCCL >= 2.26.3 seems to use
@@ -107,7 +103,6 @@ def test_chatglm3_lora_tp4_fully_sharded_loras(chatglm3_lora_files):
103
tensor_parallel_size=4,
104
trust_remote_code=True,
105
fully_sharded_loras=True,
110
- enable_chunked_prefill=True,
106
gpu_memory_utilization=0.85,
107
)
108
output1 = do_sample(llm, chatglm3_lora_files, lora_id=1)
tests/lora/test_llama_tp.py
-3
@@ -113,7 +113,6 @@ def test_llama_lora(sql_lora_files):
113
114
115
@multi_gpu_test(num_gpus=4)
116
-@create_new_process_for_each_test()
116
def test_llama_lora_tp4(sql_lora_files):
117
llm = vllm.LLM(
118
MODEL_PATH,
@@ -127,7 +126,6 @@ def test_llama_lora_tp4(sql_lora_files):
126
127
128
@multi_gpu_test(num_gpus=4)
130
-@create_new_process_for_each_test()
129
def test_llama_lora_tp4_fully_sharded_loras(sql_lora_files):
130
llm = vllm.LLM(
131
MODEL_PATH,
@@ -142,7 +140,6 @@ def test_llama_lora_tp4_fully_sharded_loras(sql_lora_files):
140
141
142
@multi_gpu_test(num_gpus=2)
145
-@create_new_process_for_each_test()
143
def test_tp2_serialize_and_deserialize_lora(
144
tmp_path, sql_lora_files, sql_lora_huggingface_id
145
):
tests/lora/test_minicpmv_tp.py
+3
-3
@@ -8,7 +8,7 @@ from vllm.assets.image import ImageAsset
8
from vllm.lora.request import LoRARequest
9
from vllm.platforms import current_platform
10
11
-from ..utils import create_new_process_for_each_test
11
+from ..utils import multi_gpu_test
12
13
MODEL_PATH = "openbmb/MiniCPM-Llama3-V-2_5"
14
@@ -88,7 +88,7 @@ def test_minicpmv_lora(minicpmv_lora_files):
88
current_platform.is_rocm(),
89
reason="MiniCPM-V dependency xformers incompatible with ROCm",
90
)
91
-@create_new_process_for_each_test()
91
+@multi_gpu_test(num_gpus=4)
92
def test_minicpmv_tp4_wo_fully_sharded_loras(minicpmv_lora_files):
93
llm = vllm.LLM(
94
MODEL_PATH,
@@ -112,7 +112,7 @@ def test_minicpmv_tp4_wo_fully_sharded_loras(minicpmv_lora_files):
112
current_platform.is_rocm(),
113
reason="MiniCPM-V dependency xformers incompatible with ROCm",
114
)
115
-@create_new_process_for_each_test()
115
+@multi_gpu_test(num_gpus=4)
116
def test_minicpmv_tp4_fully_sharded_loras(minicpmv_lora_files):
117
llm = vllm.LLM(
118
MODEL_PATH,