diff --git a/pyproject.toml b/pyproject.toml
index 82633a7158..6d5f06f579 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -55,15 +55,25 @@ optional-dependencies.quartus-report = [ "calmjs-parse", "tabulate" ]
 optional-dependencies.sr = [ "sympy>=1.13.1" ]
 optional-dependencies.testing = [
   "calmjs-parse",
-  "hgq>=0.2.3",
   "onnx>=1.4",
-  "pytest",
+  "pytest<9",
   "pytest-cov",
   "pytest-randomly",
   "qonnx",
   "tabulate",
   "torch",
 ]
+optional-dependencies.testing-keras2 = [
+  "hgq>=0.2.3",
+  "qkeras",
+  "tensorflow>=2.8,<=2.14.1",
+]
+optional-dependencies.testing-keras3 = [
+  "da4ml",
+  "hgq2>=0.0.1",
+  "keras>=3.10",
+  "tensorflow>=2.15",
+]
 urls.Homepage = "https://fastmachinelearning.org/hls4ml"
 scripts.hls4ml = "hls4ml.cli:main"
 entry-points.pytest_randomly.random_seeder = "hls4ml:reseed"
diff --git a/test/pytest/ci-template.yml b/test/pytest/ci-template.yml
index eb5facd1b4..cf4edfd30c 100644
--- a/test/pytest/ci-template.yml
+++ b/test/pytest/ci-template.yml
@@ -1,16 +1,19 @@
 .pytest:
   stage: test
-  image: gitlab-registry.cern.ch/fastmachinelearning/hls4ml-testing:0.6.1.base
+  image: gitlab-registry.cern.ch/fastmachinelearning/hls4ml-testing:0.6.3.base
   tags:
     - k8s-default
+  variables:
+    CONDA_ENV: "hls4ml-testing"
+    EXTRA_DEPS: "[da,testing,testing-keras2,sr,optimization]"
   before_script:
     - eval "$(conda shell.bash hook)"
-    - conda activate hls4ml-testing
+    - conda activate "$CONDA_ENV"
     - source /opt/intel/oneapi/setvars.sh --force
     - git config --global --add safe.directory /builds/fastmachinelearning/hls4ml
     - git submodule update --init --recursive hls4ml/templates/catapult/
     - if [ $EXAMPLEMODEL == 1 ]; then git submodule update --init example-models; fi
-    - pip install .[da,testing,sr,optimization]
+    - pip install .${EXTRA_DEPS}
   script:
     - cd test/pytest
     - pytest $PYTESTFILE -rA --cov-report xml --cov-report term --cov=hls4ml --junitxml=report.xml --randomly-seed=42 --randomly-dont-reorganize --randomly-dont-reset-seed
@@ -24,3 +27,9 @@
         path: test/pytest/coverage.xml
     paths:
       - test/pytest/hls4mlprj*.tar.gz
+
+.pytest-keras3-only:
+  extends: .pytest
+  variables:
+    CONDA_ENV: "hls4ml-testing-keras3"
+    EXTRA_DEPS: "[da,testing,testing-keras3,sr]"
diff --git a/test/pytest/generate_ci_yaml.py b/test/pytest/generate_ci_yaml.py
index 3f6a86a585..0b722c0582 100644
--- a/test/pytest/generate_ci_yaml.py
+++ b/test/pytest/generate_ci_yaml.py
@@ -12,7 +12,7 @@
 
 template = """
 pytest.{}:
-  extends: .pytest
+  extends: {}
   variables:
     PYTESTFILE: {}
     EXAMPLEMODEL: {}
@@ -26,6 +26,7 @@
 
 # Long-running tests will not be bundled with other tests
 LONGLIST = {'test_hgq_layers', 'test_hgq_players', 'test_qkeras', 'test_pytorch_api'}
+KERAS3_LIST = {'test_keras_v3_api', 'test_hgq2_mha', 'test_einsum_dense', 'test_qeinsum', 'test_multiout_onnx'}
 
 
 def path_to_name(test_path):
@@ -48,7 +49,7 @@ def uses_example_model(test_filename):
 
 def generate_test_yaml(test_root='.'):
     test_root = Path(test_root)
-    test_paths = [path for path in test_root.glob('**/test_*.py') if path.stem not in (BLACKLIST | LONGLIST)]
+    test_paths = [path for path in test_root.glob('**/test_*.py') if path.stem not in (BLACKLIST | LONGLIST | KERAS3_LIST)]
     need_example_models = [uses_example_model(path) for path in test_paths]
 
     idxs = list(range(len(need_example_models)))
@@ -61,7 +62,7 @@ def generate_test_yaml(test_root='.'):
         name = '+'.join(names)
         test_files = ' '.join([str(path.relative_to(test_root)) for path in batch_paths])
         batch_need_example_model = int(any([need_example_models[i] for i in batch_idxs]))
-        diff_yml = yaml.safe_load(template.format(name, test_files, batch_need_example_model))
+        diff_yml = yaml.safe_load(template.format(name, '.pytest', test_files, batch_need_example_model))
         if yml is None:
             yml = diff_yml
         else:
@@ -72,7 +73,22 @@ def generate_test_yaml(test_root='.'):
         name = path.stem.replace('test_', '')
         test_file = str(path.relative_to(test_root))
         needs_examples = uses_example_model(path)
-        diff_yml = yaml.safe_load(template.format(name, test_file, int(needs_examples)))
+        diff_yml = yaml.safe_load(template.format(name, '.pytest', test_file, int(needs_examples)))
+        yml.update(diff_yml)
+
+    keras3_paths = [path for path in test_root.glob('**/test_*.py') if path.stem in KERAS3_LIST]
+    keras3_need_examples = [uses_example_model(path) for path in keras3_paths]
+
+    k3_idxs = list(range(len(keras3_need_examples)))
+    k3_idxs = sorted(k3_idxs, key=lambda i: f'{keras3_need_examples[i]}_{path_to_name(keras3_paths[i])}')
+
+    for batch_idxs in batched(k3_idxs, n_test_files_per_yml):
+        batch_paths: list[Path] = [keras3_paths[i] for i in batch_idxs]
+        names = [path_to_name(path) for path in batch_paths]
+        name = 'keras3-' + '+'.join(names)
+        test_files = ' '.join([str(path.relative_to(test_root)) for path in batch_paths])
+        batch_need_example_model = int(any([keras3_need_examples[i] for i in batch_idxs]))
+        diff_yml = yaml.safe_load(template.format(name, '.pytest-keras3-only', test_files, batch_need_example_model))
         yml.update(diff_yml)
 
     return yml
diff --git a/test/pytest/test_hgq2_mha.py b/test/pytest/test_hgq2_mha.py
index bca97870f2..0214c8b59c 100644
--- a/test/pytest/test_hgq2_mha.py
+++ b/test/pytest/test_hgq2_mha.py
@@ -5,7 +5,6 @@
 
 if keras.__version__ < '3.0.0':
     pytest.skip('This test requires keras 3.0.0 or higher', allow_module_level=True)
-
 import numpy as np
 from hgq.config import QuantizerConfigScope
 from hgq.layers import QMultiHeadAttention
@@ -13,6 +12,9 @@
 
 from hls4ml.converters import convert_from_keras_model
 
+# Current hgq2 release rejects the parallelization_factor kwarg that hls4ml passes; skip until supported.
+pytest.skip('Skip until hgq2 supports parallelization_factor in QEinsumDense', allow_module_level=True)
+
 test_path = Path(__file__).parent