fix test_exporters_cli_int4_with_local_model_and_default_config

Author: IlyasMoutawwakil

optimum/intel/openvino/configuration.py

@@ -90,8 +90,8 @@ class OVQuantizationMethod(str, Enum):
     "bigscience/bloomz-560m": {
         "bits": 4,
         "sym": True,
-        "group_size": 64,
-        "ratio": 0.8,
+        "group_size": 128,
+        "ratio": 1.0,
         "dataset": "wikitext2",
         "quant_method": OVQuantizationMethod.AWQ,
     },

tests/openvino/test_exporters_cli.py

@@ -17,7 +17,7 @@
 from tempfile import TemporaryDirectory
 
 from parameterized import parameterized
-from transformers import AutoModelForCausalLM
+from transformers import AutoModelForCausalLM, AutoTokenizer
 from utils_tests import (
     _ARCHITECTURES_TO_EXPECTED_INT8,
     MODEL_NAMES,
@@ -253,10 +253,12 @@ def test_exporters_cli_int4(self, task: str, model_type: str, option: str, expec
 
     def test_exporters_cli_int4_with_local_model_and_default_config(self):
         with TemporaryDirectory() as tmpdir:
-            pt_model = AutoModelForCausalLM.from_pretrained(MODEL_NAMES["bloom"])
-            # overload for matching with default configuration
-            pt_model.config._name_or_path = "bigscience/bloomz-7b1"
+            model_id = "bigscience/bloomz-560m"
+            tokenizer = AutoTokenizer.from_pretrained(model_id)
+            pt_model = AutoModelForCausalLM.from_pretrained(model_id)
+            tokenizer.save_pretrained(tmpdir)
             pt_model.save_pretrained(tmpdir)
+
             subprocess.run(
                 f"optimum-cli export openvino --model {tmpdir} --task text-generation-with-past --weight-format int4 {tmpdir}",
                 shell=True,
@@ -267,16 +269,23 @@ def test_exporters_cli_int4_with_local_model_and_default_config(self):
             rt_info = model.model.get_rt_info()
             self.assertTrue("nncf" in rt_info)
             self.assertTrue("weight_compression" in rt_info["nncf"])
-            default_config = _DEFAULT_4BIT_CONFIGS["bigscience/bloomz-7b1"]
             model_weight_compression_config = rt_info["nncf"]["weight_compression"]
-            sym = default_config.pop("sym", False)
+
+            default_config = _DEFAULT_4BIT_CONFIGS[model_id]
             bits = default_config.pop("bits", None)
             self.assertEqual(bits, 4)
 
-            mode = f'int{bits}_{"sym" if sym else "asym"}'
-            default_config["mode"] = mode
+            sym = default_config.pop("sym", False)
+            default_config["mode"] = f'int{bits}_{"sym" if sym else "asym"}'
+
+            quant_method = default_config.pop("quant_method", None)
+            default_config["awq"] = quant_method == "awq"
+            default_config["gptq"] = quant_method == "gptq"
+
+            default_config.pop("dataset", None)
+
             for key, value in default_config.items():
-                self.assertTrue(key in model_weight_compression_config)
+                self.assertIn(key, model_weight_compression_config)
                 self.assertEqual(
                     model_weight_compression_config[key].value,
                     str(value),