Merge branch 'main' into benchmarking-overhaul

Author: sayakpaul

tests/models/test_modeling_common.py

@@ -1060,7 +1060,7 @@ def test_deprecated_kwargs(self):
     @parameterized.expand([True, False])
     @torch.no_grad()
     @unittest.skipIf(not is_peft_available(), "Only with PEFT")
-    def test_save_load_lora_adapter(self, use_dora=False):
+    def test_lora_save_load_adapter(self, use_dora=False):
         import safetensors
         from peft import LoraConfig
         from peft.utils import get_peft_model_state_dict
@@ -1117,7 +1117,7 @@ def test_save_load_lora_adapter(self, use_dora=False):
         self.assertTrue(torch.allclose(outputs_with_lora, outputs_with_lora_2, atol=1e-4, rtol=1e-4))
 
     @unittest.skipIf(not is_peft_available(), "Only with PEFT")
-    def test_wrong_adapter_name_raises_error(self):
+    def test_lora_wrong_adapter_name_raises_error(self):
         from peft import LoraConfig
 
         from diffusers.loaders.peft import PeftAdapterMixin

tests/quantization/bnb/test_4bit.py

@@ -526,7 +526,7 @@ def test_moving_to_cpu_throws_warning(self):
         reason="Test will pass after https://github.com/huggingface/accelerate/pull/3223 is in a release.",
         strict=True,
     )
-    def test_pipeline_device_placement_works_with_nf4(self):
+    def test_pipeline_cuda_placement_works_with_nf4(self):
         transformer_nf4_config = BitsAndBytesConfig(
             load_in_4bit=True,
             bnb_4bit_quant_type="nf4",
@@ -560,7 +560,7 @@ def test_pipeline_device_placement_works_with_nf4(self):
         ).to(torch_device)
 
         # Check if inference works.
-        _ = pipeline_4bit("table", max_sequence_length=20, num_inference_steps=2)
+        _ = pipeline_4bit(self.prompt, max_sequence_length=20, num_inference_steps=2)
 
         del pipeline_4bit
 

tests/quantization/bnb/test_mixed_int8.py

@@ -492,7 +492,7 @@ def test_generate_quality_dequantize(self):
         self.assertTrue(max_diff < 1e-2)
 
         # 8bit models cannot be offloaded to CPU.
-        self.assertTrue(self.pipeline_8bit.transformer.device.type == "cuda")
+        self.assertTrue(self.pipeline_8bit.transformer.device.type == torch_device)
         # calling it again shouldn't be a problem
         _ = self.pipeline_8bit(
             prompt=self.prompt,
@@ -534,7 +534,7 @@ def test_pipeline_cuda_placement_works_with_mixed_int8(self):
         ).to(device)
 
         # Check if inference works.
-        _ = pipeline_8bit("table", max_sequence_length=20, num_inference_steps=2)
+        _ = pipeline_8bit(self.prompt, max_sequence_length=20, num_inference_steps=2)
 
         del pipeline_8bit