[Quantization] dtype fix for GGUF + fix BnB tests (#11159)

* update

* update

* update

* update

src/diffusers/loaders/single_file_model.py

@@ -282,6 +282,7 @@ class FromOriginalModelMixin:
         if quantization_config is not None:
             hf_quantizer = DiffusersAutoQuantizer.from_config(quantization_config)
             hf_quantizer.validate_environment()
+            torch_dtype = hf_quantizer.update_torch_dtype(torch_dtype)
 
         else:
             hf_quantizer = None

tests/quantization/bnb/test_mixed_int8.py

@@ -90,13 +90,16 @@ class Base8bitTests(unittest.TestCase):
 
     def get_dummy_inputs(self):
         prompt_embeds = load_pt(
-            "https://huggingface.co/datasets/hf-internal-testing/bnb-diffusers-testing-artifacts/resolve/main/prompt_embeds.pt"
+            "https://huggingface.co/datasets/hf-internal-testing/bnb-diffusers-testing-artifacts/resolve/main/prompt_embeds.pt",
+            map_location="cpu",
         )
         pooled_prompt_embeds = load_pt(
-            "https://huggingface.co/datasets/hf-internal-testing/bnb-diffusers-testing-artifacts/resolve/main/pooled_prompt_embeds.pt"
+            "https://huggingface.co/datasets/hf-internal-testing/bnb-diffusers-testing-artifacts/resolve/main/pooled_prompt_embeds.pt",
+            map_location="cpu",
         )
         latent_model_input = load_pt(
-            "https://huggingface.co/datasets/hf-internal-testing/bnb-diffusers-testing-artifacts/resolve/main/latent_model_input.pt"
+            "https://huggingface.co/datasets/hf-internal-testing/bnb-diffusers-testing-artifacts/resolve/main/latent_model_input.pt",
+            map_location="cpu",
         )
 
         input_dict_for_transformer = {

tests/quantization/gguf/test_gguf.py

@@ -57,7 +57,7 @@ class GGUFSingleFileTesterMixin:
             if isinstance(module, torch.nn.Linear) and hasattr(module.weight, "quant_type"):
                 assert module.weight.dtype == torch.uint8
                 if module.bias is not None:
-                    assert module.bias.dtype == torch.float32
+                    assert module.bias.dtype == self.torch_dtype
 
     def test_gguf_memory_usage(self):
         quantization_config = GGUFQuantizationConfig(compute_dtype=self.torch_dtype)