biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[CI] Replace large models with tiny alternatives in tests (#24057)

Browse Source 
Signed-off-by: Tahsin Tunan <tahsintunan@gmail.com>
Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
Co-authored-by: Nick Hill <nhill@redhat.com>
Co-authored-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
This commit is contained in:
Tahsin Tunan
2025-10-16 20:51:27 +06:00
committed by GitHub
parent 02d709a6f1
commit 43721bc67f
17 changed files with 118 additions and 59 deletions
Download Patch File Download Diff File Expand all files Collapse all files

16
tests/basic_correctness/test_basic_correctness.py
View File

@@ -20,7 +20,7 @@ from ..models.utils import check_outputs_equal
from ..utils import multi_gpu_test
MODELS = [
"google/gemma-2-2b-it",
"hmellor/tiny-random-Gemma2ForCausalLM",
"meta-llama/Llama-3.2-1B-Instruct",
]
@@ -29,7 +29,7 @@ TARGET_TEST_SUITE = os.environ.get("TARGET_TEST_SUITE", "L4")
def test_vllm_gc_ed():
"""Verify vllm instance is GC'ed when it is deleted"""
llm = LLM("distilbert/distilgpt2")
llm = LLM("hmellor/tiny-random-LlamaForCausalLM")
weak_llm = weakref.ref(llm)
del llm
# If there's any circular reference to vllm, this fails
@@ -125,14 +125,14 @@ def test_models(
@pytest.mark.parametrize(
"model, distributed_executor_backend, attention_backend, test_suite, extra_env",
[
("distilbert/distilgpt2", "ray", "", "L4", {}),
("distilbert/distilgpt2", "mp", "", "L4", {}),
("distilbert/distilgpt2", "ray", "", "L4", {"VLLM_SLEEP_WHEN_IDLE": "1"}),
("distilbert/distilgpt2", "mp", "", "L4", {"VLLM_SLEEP_WHEN_IDLE": "1"}),
("facebook/opt-125m", "ray", "", "L4", {}),
("facebook/opt-125m", "mp", "", "L4", {}),
("facebook/opt-125m", "ray", "", "L4", {"VLLM_SLEEP_WHEN_IDLE": "1"}),
("facebook/opt-125m", "mp", "", "L4", {"VLLM_SLEEP_WHEN_IDLE": "1"}),
("meta-llama/Llama-3.2-1B-Instruct", "ray", "", "L4", {}),
("meta-llama/Llama-3.2-1B-Instruct", "mp", "", "L4", {}),
("distilbert/distilgpt2", "ray", "", "A100", {}),
("distilbert/distilgpt2", "mp", "", "A100", {}),
("facebook/opt-125m", "ray", "", "A100", {}),
("facebook/opt-125m", "mp", "", "A100", {}),
],
)
@pytest.mark.parametrize("enable_prompt_embeds", [True, False])

2
tests/basic_correctness/test_cpu_offload.py
View File

@@ -6,5 +6,5 @@ from ..utils import compare_two_settings
def test_cpu_offload():
compare_two_settings(
"meta-llama/Llama-3.2-1B-Instruct", [], ["--cpu-offload-gb", "1"]
"hmellor/tiny-random-LlamaForCausalLM", [], ["--cpu-offload-gb", "1"]
)

4
tests/basic_correctness/test_cumem.py
View File

@@ -120,7 +120,7 @@ def test_cumem_with_cudagraph():
"model",
[
# sleep mode with safetensors
"meta-llama/Llama-3.2-1B",
"hmellor/tiny-random-LlamaForCausalLM",
# sleep mode with pytorch checkpoint
"facebook/opt-125m",
],
@@ -174,7 +174,7 @@ def test_end_to_end(model: str):
@create_new_process_for_each_test()
def test_deep_sleep():
model = "Qwen/Qwen3-0.6B"
model = "hmellor/tiny-random-LlamaForCausalLM"
free, total = torch.cuda.mem_get_info()
used_bytes_baseline = total - free # in case other process is running
llm = LLM(model, enable_sleep_mode=True)