[CI/Build] Add new CI job to validate Hybrid Models for every PR (#20147)

Signed-off-by: Thomas Parnell <tpa@zurich.ibm.com>

.buildkite/test-pipeline.yaml

@@ -536,6 +536,17 @@ steps:
     - pip freeze | grep -E 'torch'
     - pytest -v -s models/language -m core_model
 
+- label: Language Models Test (Hybrid) # 35 min
+  mirror_hardwares: [amdexperimental]
+  torch_nightly: true
+  source_file_dependencies:
+  - vllm/
+  - tests/models/language/generation
+  commands:
+    # Install causal-conv1d for plamo2 models here, as it is not compatible with pip-compile.
+    - pip install 'git+https://github.com/Dao-AILab/causal-conv1d@v1.5.0.post8'
+    - pytest -v -s models/language/generation -m hybrid_model
+
 - label: Language Models Test (Extended Generation) # 1hr20min
   mirror_hardwares: [amdexperimental]
   optional: true
@@ -545,7 +556,7 @@ steps:
   commands:
     # Install causal-conv1d for plamo2 models here, as it is not compatible with pip-compile.
     - pip install 'git+https://github.com/Dao-AILab/causal-conv1d@v1.5.0.post8'
-    - pytest -v -s models/language/generation -m 'not core_model'
+    - pytest -v -s models/language/generation -m '(not core_model) and (not hybrid_model)'
 
 - label: Language Models Test (Extended Pooling)  # 36min
   mirror_hardwares: [amdexperimental]

pyproject.toml

@@ -150,6 +150,7 @@ skip_gitignore = true
 markers = [
     "skip_global_cleanup",
     "core_model: enable this model test in each PR instead of only nightly",
+    "hybrid_model: models that contain mamba layers (including pure SSM and hybrid architectures)",
     "cpu_model: enable this model test in CPU tests",
     "split: run this test as part of a split",
     "distributed: run this test only in distributed GPU tests",

tests/models/language/generation/test_hybrid.py

@@ -9,6 +9,9 @@ from vllm.sampling_params import SamplingParams
 
 from ...utils import check_logprobs_close, check_outputs_equal
 
+# Mark all tests as hybrid
+pytestmark = pytest.mark.hybrid_model
+
 # NOTE: The first model in each list is taken as the primary model,
 # meaning that it will be used in all tests in this file
 # The rest of the models will only be tested by test_models