adding require_hqq

Author: MekkCyber

src/transformers/testing_utils.py

@@ -87,6 +87,7 @@
     is_gguf_available,
     is_grokadamw_available,
     is_hadamard_available,
+    is_hqq_available,
     is_ipex_available,
     is_jieba_available,
     is_jinja_available,
@@ -1213,6 +1214,13 @@ def require_auto_gptq(test_case):
     return unittest.skipUnless(is_auto_gptq_available(), "test requires auto-gptq")(test_case)
 
 
+def require_hqq(test_case):
+    """
+    Decorator for hqq dependency
+    """
+    return unittest.skipUnless(is_hqq_available(), "test requires hqq")(test_case)
+
+
 def require_auto_awq(test_case):
     """
     Decorator for auto_awq dependency

src/transformers/utils/quantization_config.py

@@ -225,10 +225,9 @@ def __init__(
         if is_hqq_available():
             from hqq.core.quantize import BaseQuantizeConfig as HQQBaseQuantizeConfig
         else:
-
-            class HQQBaseQuantizeConfig:
-                def __init__(self, *args, **kwargs):
-                    pass
+            raise ImportError(
+                "A valid HQQ version (>=0.2.1) is not available. Please follow the instructions to install it: `https://github.com/mobiusml/hqq/`."
+            )
 
         for deprecated_key in ["quant_zero", "quant_scale", "offload_meta"]:
             if deprecated_key in kwargs:

tests/quantization/hqq/test_hqq.py

@@ -23,6 +23,7 @@
     require_torch_multi_gpu,
     slow,
     torch_device,
+    require_hqq,
 )
 from transformers.utils import is_hqq_available, is_torch_available
 
@@ -86,6 +87,7 @@ def check_forward(test_module, model, batch_size=1, context_size=1024):
 
 
 @require_torch_gpu
+@require_hqq
 class HqqConfigTest(unittest.TestCase):
     def test_to_dict(self):
         """
@@ -100,6 +102,7 @@ def test_to_dict(self):
 @slow
 @require_torch_gpu
 @require_accelerate
+@require_hqq
 class HQQTest(unittest.TestCase):
     def tearDown(self):
         cleanup()
@@ -122,6 +125,7 @@ def test_fp16_quantized_model(self):
 @require_torch_gpu
 @require_torch_multi_gpu
 @require_accelerate
+@require_hqq
 class HQQTestMultiGPU(unittest.TestCase):
     def tearDown(self):
         cleanup()
@@ -144,6 +148,7 @@ def test_fp16_quantized_model_multipgpu(self):
 @slow
 @require_torch_gpu
 @require_accelerate
+@require_hqq
 class HQQSerializationTest(unittest.TestCase):
     def tearDown(self):
         cleanup()