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, 09:00 UTC
[CI] And PPL test for Qwen3.5. (#35853)
Signed-off-by: wang.yuqi <yuqi.wang@daocloud.io> Signed-off-by: wang.yuqi <noooop@126.com> Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>
wang.yuqi committed
Mar 3, 2026 at 21:15 UTC
fd4a90f337f7fe188581d71d4d3ec712767320c0
3 files changed
+16
-10
tests/models/language/generation_ppl_test/test_gemma.py
+3
-3
@@ -7,9 +7,9 @@ from tests.models.utils import GenerateModelInfo
7
from .ppl_utils import wikitext_ppl_test
8
9
MODELS = [
10
- GenerateModelInfo("google/gemma-2b"),
11
- GenerateModelInfo("google/gemma-2-2b"),
12
- GenerateModelInfo("google/gemma-3-4b-it"),
10
+ GenerateModelInfo("google/gemma-2b", hf_ppl=21.48524284362793),
11
+ GenerateModelInfo("google/gemma-2-2b", hf_ppl=102.59290313720703),
12
+ GenerateModelInfo("google/gemma-3-4b-it", hf_ppl=27.79648208618164),
13
]
14
15
tests/models/language/generation_ppl_test/test_gpt.py
+1
-1
@@ -6,7 +6,7 @@ from tests.models.utils import GenerateModelInfo
6
7
from .ppl_utils import wikitext_ppl_test
8
9
-MODELS = [GenerateModelInfo("openai-community/gpt2-large")]
9
+MODELS = [GenerateModelInfo("openai-community/gpt2-large", hf_ppl=19.457056045532227)]
10
11
12
@pytest.mark.parametrize("model_info", MODELS)
tests/models/language/generation_ppl_test/test_qwen.py
+12
-6
@@ -8,14 +8,20 @@ from tests.models.utils import GenerateModelInfo
8
from .ppl_utils import wikitext_ppl_test
9
10
MODELS = [
11
- GenerateModelInfo("Qwen/Qwen3-0.6B"),
12
- GenerateModelInfo("Qwen/Qwen3-0.6B-FP8"),
13
- # transformers:
14
- # Loading a GPTQ quantized model requires optimum, gptqmodel
15
- # GenerateModelInfo("Qwen/Qwen3-0.6B-GPTQ-Int8"),
11
+ # for Qwen3
12
+ GenerateModelInfo("Qwen/Qwen3-0.6B", hf_ppl=23.864173889160156),
13
+ GenerateModelInfo("Qwen/Qwen3-0.6B-FP8", hf_ppl=24.313045501708984),
14
+ # for Qwen3.5
15
+ GenerateModelInfo("Qwen/Qwen3.5-0.8B", hf_ppl=19.38858413696289),
16
]
17
18
19
@pytest.mark.parametrize("model_info", MODELS)
20
def test_ppl(hf_runner, vllm_runner, model_info: GenerateModelInfo):
21
- wikitext_ppl_test(hf_runner, vllm_runner, model_info)
21
+ vllm_extra_kwargs = {}
22
+ if model_info.name == "Qwen/Qwen3.5-0.8B":
23
+ vllm_extra_kwargs["language_model_only"] = True
24
+
25
+ wikitext_ppl_test(
26
+ hf_runner, vllm_runner, model_info, vllm_extra_kwargs=vllm_extra_kwargs
27
+ )