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 7, 11:30 UTC
Fix deprecated v1 config tests (#35327)
Signed-off-by: Jesse Cai <jessecai@fb.com>
Jesse Cai committed
Mar 1, 2026 at 17:32 UTC
a60985b07eaf0fefa4c353ad2309b745707591ed
1 file changed
+1
-17
tests/quantization/test_torchao.py
+1
-17
@@ -20,7 +20,7 @@ TORCHAO_AVAILABLE = importlib.util.find_spec("torchao") is not None
20
@pytest.mark.skipif(not TORCHAO_AVAILABLE, reason="torchao is not available")
21
def test_pre_quantized_model(vllm_runner):
22
with vllm_runner(
23
- "drisspg/fp8-opt-125m",
23
+ "torchao-testing/opt-125m-Float8WeightOnlyConfig-v2-0.15.0",
24
quantization="torchao",
25
dtype="bfloat16",
26
enforce_eager=True,
@@ -52,22 +52,6 @@ def test_opt_125m_int8wo_model_loading_with_params(vllm_runner, pt_load_map_loca
52
assert output
53
54
55
-@pytest.mark.skipif(not TORCHAO_AVAILABLE, reason="torchao is not available")
56
-def test_opt_125m_int4wo_model_per_module_quant(vllm_runner):
57
- torch._dynamo.reset()
58
- model_name = "jerryzh168/opt-125m-int4wo-per-module"
59
- with vllm_runner(
60
- model_name=model_name,
61
- quantization="torchao",
62
- dtype="bfloat16",
63
- pt_load_map_location="cuda:0",
64
- enforce_eager=True,
65
- ) as llm:
66
- output = llm.generate_greedy(["The capital of France is"], max_tokens=4)
67
-
68
- assert output
69
-
70
-
55
@pytest.mark.skipif(not TORCHAO_AVAILABLE, reason="torchao is not available")
56
def test_qwenvl_int8wo_model_loading_with_params(vllm_runner):
57
torch._dynamo.reset()