FIX low_cpu_mem_usage consolidates devices (#2113)

See: huggingface/diffusers#9510 (comment)

Right now, the low_cpu_mem_usage=True option does not consolidate the
devices. E.g. when the model is on GPU and the state_dict on CPU, the
adapter weight will be on CPU after loading, when it should be GPU. This
fix ensures that the devices are consolidated.

Author: BenjaminBossan

src/peft/utils/save_and_load.py

@@ -443,6 +443,10 @@ def renamed_dora_weights(k):
     )
     if low_cpu_mem_usage:
         load_result = model.load_state_dict(peft_model_state_dict, strict=False, assign=True)
+        # ensure that the correct device is set
+        for module in model.modules():
+            if hasattr(module, "_move_adapter_to_device_of_base_layer"):
+                module._move_adapter_to_device_of_base_layer(adapter_name)
     else:
         load_result = model.load_state_dict(peft_model_state_dict, strict=False)
 

tests/test_gpu_examples.py

@@ -55,8 +55,11 @@
     PromptEncoderConfig,
     TaskType,
     get_peft_model,
+    get_peft_model_state_dict,
+    inject_adapter_in_model,
     prepare_model_for_kbit_training,
     replace_lora_weights_loftq,
+    set_peft_model_state_dict,
 )
 from peft.tuners import boft
 from peft.utils import SAFETENSORS_WEIGHTS_NAME, infer_device
@@ -3226,3 +3229,51 @@ def test_p_tuning_exactly_reproducible_after_loading(self, tmp_path):
 
         torch.testing.assert_close(output_loaded, output_peft)
         torch.testing.assert_close(gen_loaded, gen_peft)
+
+
+@pytest.mark.skipif(not torch.cuda.is_available(), reason="test requires a GPU")
+@pytest.mark.single_gpu_tests
+class TestLowCpuMemUsageDifferentDevices:
+    """Test for the low CPU memory usage option for loading PEFT models.
+
+    There are already tests for this in test_initialization.py but here we want to specifically test diverging devices
+    for the model and state_dict.
+
+    """
+
+    model_id = "hf-internal-testing/tiny-random-OPTForCausalLM"
+
+    @pytest.mark.parametrize("device_model, device_sd", [("cpu", "cuda"), ("cuda", "cpu")])
+    def test_low_cpu_mem_usage_model_model_on_gpu_state_dict_on_cpu_works(self, device_model, device_sd):
+        inputs = {"input_ids": torch.randint(0, 100, (1, 10)), "attention_mask": torch.ones(1, 10)}
+        inputs = {k: v.to(device_model) for k, v in inputs.items()}
+
+        model = AutoModelForCausalLM.from_pretrained(self.model_id).to(device_model)
+        lora_config = LoraConfig(init_lora_weights=False, target_modules="all-linear")
+        model = get_peft_model(model, lora_config)
+        model.eval()
+        logits_not_low_cpu_mem = model(**inputs).logits
+
+        state_dict = get_peft_model_state_dict(model)
+        peft_model_state_dict = {}
+        # remap the state dict so that it can be correctly loaded, and move weights to the other device
+        prefix = "base_model.model."
+        for k, v in state_dict.items():
+            k = k[len(prefix) :]
+            peft_model_state_dict[k] = v.to(device_sd)
+
+        del model
+
+        model = AutoModelForCausalLM.from_pretrained(self.model_id).to(device_model)
+        model.eval()
+        inject_adapter_in_model(lora_config, model, low_cpu_mem_usage=True)
+        load_result = set_peft_model_state_dict(model, peft_model_state_dict, low_cpu_mem_usage=True)
+
+        # sanity check: all lora keys are matched
+        assert not any("lora" in k for k in load_result.missing_keys)
+        assert not any("lora" in k for k in load_result.unexpected_keys)
+
+        logits_low_cpu_mem = model(**inputs).logits
+
+        assert torch.allclose(logits_low_cpu_mem, logits_not_low_cpu_mem)
+        assert {p.device.type for p in model.parameters()} == {device_model}