update tests

huggingface · Jul 2, 2024 · 19499a0 · 19499a0
1 parent 4da2038
commit 19499a0
Show file tree

Hide file tree

Showing 3 changed files with 17 additions and 7 deletions.
diff --git a/tests/openvino/test_exporters_cli.py b/tests/openvino/test_exporters_cli.py
@@ -41,7 +41,7 @@
 )
 from optimum.intel.openvino.configuration import _DEFAULT_4BIT_CONFIGS
 from optimum.intel.openvino.utils import _HEAD_TO_AUTOMODELS
-from optimum.intel.utils.import_utils import is_openvino_tokenizers_available
+from optimum.intel.utils.import_utils import is_openvino_tokenizers_available, is_transformers_version
 
 
 class OVCLIExportTestCase(unittest.TestCase):
@@ -90,20 +90,26 @@ class OVCLIExportTestCase(unittest.TestCase):
         ("text-generation-with-past", "opt125m", "int4_asym_g128", 62, 86),
         ("text-generation-with-past", "opt125m", "int4_sym_g64", 62, 86),
         ("text-generation-with-past", "opt125m", "int4_asym_g64", 62, 86),
-        ("text-generation-with-past", "llama_awq", "int4 --ratio 1.0 --sym --group-size 16 --all-layers", 0, 32),
+        (
+            "text-generation-with-past",
+            "llama_awq",
+            "int4 --ratio 1.0 --sym --group-size 8 --all-layers",
+            0,
+            34 if is_transformers_version(">=", "4.39") else 32,
+        ),
         (
             "text-generation-with-past",
             "llama_awq",
             "int4 --ratio 1.0 --sym --group-size 16 --awq --dataset wikitext2 --num-samples 100 "
             "--sensitivity-metric max_activation_variance",
-            4,
+            6 if is_transformers_version(">=", "4.39") else 4,
             28,
         ),
         (
             "text-generation-with-past",
             "llama_awq",
             "int4 --ratio 1.0 --sym --group-size 16 --scale-estimation --dataset wikitext2 --num-samples 100 ",
-            4,
+            6 if is_transformers_version(">=", "4.39") else 4,
             28,
         ),
     ]

diff --git a/tests/openvino/test_modeling.py b/tests/openvino/test_modeling.py
@@ -933,15 +933,18 @@ def test_beam_search(self, model_arch):
             do_sample=False,
             eos_token_id=None,
         )
+
         beam_sample_gen_config = GenerationConfig(
             max_new_tokens=10,
             min_new_tokens=10,
             num_beams=4,
             do_sample=True,
             eos_token_id=None,
-            top_k=1,
         )
 
+        if model_arch == "minicpm":
+            beam_sample_gen_config.top_k = 1
+
         group_beam_search_gen_config = GenerationConfig(
             max_new_tokens=10,
             min_new_tokens=10,
@@ -963,6 +966,7 @@ def test_beam_search(self, model_arch):
         )
 
         gen_configs = [
+            beam_sample_gen_config,
             beam_search_gen_config,
             beam_sample_gen_config,
             group_beam_search_gen_config,

diff --git a/tests/openvino/test_quantization.py b/tests/openvino/test_quantization.py
@@ -233,7 +233,7 @@ class OVWeightCompressionTest(unittest.TestCase):
                 quant_method=QuantizationMethod.AWQ,
                 scale_estimation=True,
             ),
-            16,
+            18 if is_transformers_version(">=", "4.39") else 16,
         ),
         (
             OVModelForCausalLM,
@@ -247,7 +247,7 @@ class OVWeightCompressionTest(unittest.TestCase):
                 dataset="c4",
                 quant_method="awq",
             ),
-            16,
+            18 if is_transformers_version(">=", "4.39") else 16,
         ),
     )