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 6, 09:30 UTC
test Qwen/Qwen3-4B-Instruct-2507 for unbacked (#36064)
Signed-off-by: Laith Sakka <lsakka@meta.com>
Laith Sakka committed
Mar 19, 2026 at 14:28 UTC
112944fab91e63c5daaeed3c0d85478af4e13f50
1 file changed
+8
-2
tests/compile/test_dynamic_shapes_compilation.py
+8
-2
@@ -23,8 +23,14 @@ from vllm.utils.torch_utils import is_torch_equal_or_newer
23
24
def get_test_models():
25
"""Get list of models to test based on PyTorch version"""
26
- # TODO "Qwen/Qwen3-4B-Instruct-2507" fails Fix issue and support it.
27
- return ["gpt2", "Qwen/Qwen2-7B-Instruct", "meta-llama/Llama-3.1-8B"]
26
+ models = [
27
+ "gpt2",
28
+ "Qwen/Qwen2-7B-Instruct",
29
+ "meta-llama/Llama-3.1-8B",
30
+ ]
31
+ if is_torch_equal_or_newer("2.12.0"):
32
+ models.append("Qwen/Qwen3-4B-Instruct-2507")
33
+ return models
34
35
36
@pytest.mark.parametrize("model_name", get_test_models())