enable determinism

tests/gptq/test_quantization.py

@@ -18,7 +18,7 @@
 
 import torch
 from parameterized import parameterized
-from transformers import AutoConfig, AutoModelForCausalLM, AutoTokenizer, set_seed
+from transformers import AutoConfig, AutoModelForCausalLM, AutoTokenizer, enable_full_determinism, set_seed
 from transformers.testing_utils import slow
 
 from optimum.gptq import GPTQQuantizer, load_quantized_model
@@ -75,7 +75,9 @@ def setUpClass(cls):
         Setup quantized model
         """
 
-        set_seed(42)
+        enable_full_determinism()
+        set_seed(42, deterministic=True)
+
         cls.model_fp16 = AutoModelForCausalLM.from_pretrained(
             cls.model_name, torch_dtype=torch.float16, device_map=cls.device_map_for_quantization
         )
@@ -130,14 +132,13 @@ def check_inference_correctness(self, model):
         Given that we are operating on small numbers + the testing model is relatively small, we might not get
         the same output across GPUs. So we'll generate few tokens (5-10) and check their output.
         """
-        set_seed(42)
+        enable_full_determinism()
+        set_seed(42, deterministic=True)
+
         input_ids = self.tokenizer(self.input_text, return_tensors="pt").input_ids.to(self.device_for_inference)
         output_ids = model.generate(input_ids, do_sample=False, min_new_tokens=10, max_new_tokens=10)
         output_text = self.tokenizer.decode(output_ids[0], skip_special_tokens=True)
 
-        # TODO: use pytest features to show what we're comparing
-        # Check the exactness of the result
-        print(output_text)
         self.assertIn(output_text, self.EXPECTED_OUTPUTS)
 
     def test_generate_quality(self):