Don't remap auto device_map and raise RuntimeError

If device_map=auto and modules are selected for disk offload, don't attempt to map them to any other device. Raise a runtime error when a GGUF model is configured to map any modules to disk.

src/transformers/modeling_utils.py

@@ -4241,27 +4241,14 @@ def from_pretrained(
             if hf_quantizer is not None:
                 hf_quantizer.validate_environment(device_map=device_map)
 
-            if gguf_path:
-                remapped_devices = set()
-                for name, device in device_map.items():
-                    if device == "disk":
-                        device_map[name] = "cpu"
-                        remapped_devices.add(name)
-                if len(remapped_devices) > 0:
-                    logger.warning(
-                        "Accelerate has auto-mapped modules to disk but disk offload is not supported for "
-                        "models loaded from GGUF files. Remapping modules to the cpu: "
-                        ", ".join(remapped_devices)
-                    )
-
         elif device_map is not None:
             model.tie_weights()
             tied_params = find_tied_parameters(model)
             # check if we don't have tied param in different devices
             check_tied_parameters_on_same_device(tied_params, device_map)
 
         if gguf_path and device_map is not None and "disk" in device_map.values():
-            raise NotImplementedError(
+            raise RuntimeError(
                 "One or more modules is configured to be mapped to disk. Disk offload is not supported for models "
                 "loaded from GGUF files."
             )

tests/quantization/ggml/test_ggml.py

@@ -223,7 +223,7 @@ def test_gguf_errors_disk_offload(self):
         from collections import OrderedDict
 
         q2_k_gguf_model_id = self.gguf_filename.format(quant_type=QuantType.Q2_K.name)
-        with self.assertRaises(NotImplementedError):
+        with self.assertRaises(RuntimeError):
             AutoModelForCausalLM.from_pretrained(
                 self.gguf_model_id,
                 device_map=OrderedDict(