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] Fix model nightly tests (#26466)
Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
Cyrus Leung committed
Oct 9, 2025 at 14:44 UTC
0f29dca9884f7fc2e762998a3692fe40125651bd
5 files changed
+9
-7
tests/models/language/generation/test_common.py
+1
-1
@@ -100,7 +100,7 @@ AITER_MODEL_LIST = [
100
"allenai/OLMoE-1B-7B-0924-Instruct",
101
marks=[pytest.mark.cpu_model],
102
),
103
- pytest.param("swiss-ai/Apertus-8B-2509"), # apertus
103
+ pytest.param("swiss-ai/Apertus-8B-Instruct-2509"), # apertus
104
],
105
)
106
@pytest.mark.parametrize("max_tokens", [32])
tests/models/language/pooling/test_token_classification.py
+1
-1
@@ -67,4 +67,4 @@ def test_modernbert_models(
67
for hf_output, vllm_output in zip(hf_outputs, vllm_outputs):
68
hf_output = torch.tensor(hf_output).cpu().float()
69
vllm_output = torch.tensor(vllm_output).cpu().float()
70
- assert torch.allclose(hf_output, vllm_output, 1e-2)
70
+ assert torch.allclose(hf_output, vllm_output, atol=1e-2)
tests/models/multimodal/generation/test_common.py
+1
@@ -749,6 +749,7 @@ VLM_TEST_SETTINGS = {
749
max_num_seqs=2,
750
auto_cls=AutoModelForImageTextToText,
751
hf_output_post_proc=model_utils.smolvlm_trunc_hf_output,
752
+ num_logprobs=10,
753
),
754
"tarsier": VLMTestInfo(
755
models=["omni-research/Tarsier-7b"],
tests/models/multimodal/pooling/test_clip.py
+5
-3
@@ -45,14 +45,16 @@ def _run_test(
45
46
all_outputs = []
47
for inputs in all_inputs:
48
+ inputs = hf_model.wrap_device(inputs)
49
+
50
if "pixel_values" in inputs:
49
- inputs.pop("input_ids")
51
pooled_output = hf_model.model.get_image_features(
51
- **hf_model.wrap_device(inputs)
52
+ pixel_values=inputs.pixel_values,
53
).squeeze(0)
54
else:
55
pooled_output = hf_model.model.get_text_features(
55
- **hf_model.wrap_device(inputs)
56
+ input_ids=inputs.input_ids,
57
+ attention_mask=inputs.attention_mask,
58
).squeeze(0)
59
60
all_outputs.append(pooled_output.tolist())
tests/models/registry.py
+1
-2
@@ -172,9 +172,8 @@ class _HfExamplesInfo:
172
_TEXT_GENERATION_EXAMPLE_MODELS = {
173
# [Decoder-only]
174
"ApertusForCausalLM": _HfExamplesInfo(
175
- "swiss-ai/Apertus-8B-2509",
175
+ "swiss-ai/Apertus-8B-Instruct-2509",
176
min_transformers_version="4.56.0",
177
- trust_remote_code=True,
177
),
178
"AquilaModel": _HfExamplesInfo("BAAI/AquilaChat-7B", trust_remote_code=True),
179
"AquilaForCausalLM": _HfExamplesInfo("BAAI/AquilaChat2-7B", trust_remote_code=True),