[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>

tests/models/language/generation_ppl_test/test_gemma.py

@@ -7,9 +7,9 @@ from tests.models.utils import GenerateModelInfo
 from .ppl_utils import wikitext_ppl_test
 
 MODELS = [
-    GenerateModelInfo("google/gemma-2b"),
-    GenerateModelInfo("google/gemma-2-2b"),
-    GenerateModelInfo("google/gemma-3-4b-it"),
+    GenerateModelInfo("google/gemma-2b", hf_ppl=21.48524284362793),
+    GenerateModelInfo("google/gemma-2-2b", hf_ppl=102.59290313720703),
+    GenerateModelInfo("google/gemma-3-4b-it", hf_ppl=27.79648208618164),
 ]
 
 

tests/models/language/generation_ppl_test/test_gpt.py

@@ -6,7 +6,7 @@ from tests.models.utils import GenerateModelInfo
 
 from .ppl_utils import wikitext_ppl_test
 
-MODELS = [GenerateModelInfo("openai-community/gpt2-large")]
+MODELS = [GenerateModelInfo("openai-community/gpt2-large", hf_ppl=19.457056045532227)]
 
 
 @pytest.mark.parametrize("model_info", MODELS)

tests/models/language/generation_ppl_test/test_qwen.py

@@ -8,14 +8,20 @@ from tests.models.utils import GenerateModelInfo
 from .ppl_utils import wikitext_ppl_test
 
 MODELS = [
-    GenerateModelInfo("Qwen/Qwen3-0.6B"),
-    GenerateModelInfo("Qwen/Qwen3-0.6B-FP8"),
-    # transformers:
-    # Loading a GPTQ quantized model requires optimum, gptqmodel
-    # GenerateModelInfo("Qwen/Qwen3-0.6B-GPTQ-Int8"),
+    # for Qwen3
+    GenerateModelInfo("Qwen/Qwen3-0.6B", hf_ppl=23.864173889160156),
+    GenerateModelInfo("Qwen/Qwen3-0.6B-FP8", hf_ppl=24.313045501708984),
+    # for Qwen3.5
+    GenerateModelInfo("Qwen/Qwen3.5-0.8B", hf_ppl=19.38858413696289),
 ]
 
 
 @pytest.mark.parametrize("model_info", MODELS)
 def test_ppl(hf_runner, vllm_runner, model_info: GenerateModelInfo):
-    wikitext_ppl_test(hf_runner, vllm_runner, model_info)
+    vllm_extra_kwargs = {}
+    if model_info.name == "Qwen/Qwen3.5-0.8B":
+        vllm_extra_kwargs["language_model_only"] = True
+
+    wikitext_ppl_test(
+        hf_runner, vllm_runner, model_info, vllm_extra_kwargs=vllm_extra_kwargs
+    )