[EPLB][Bugfix] EPLB support fp/bf16 (#5531)

### What this PR does / why we need it?
EPLB support dtype of fp/bf16.
### Does this PR introduce _any_ user-facing change?

### How was this patch tested?
w8a8_dynamic Baseline:
| dataset | version | metric | mode | vllm-api-general-chat |
|----- | ----- | ----- | ----- | -----|
| aime2024 | 604a78 | accuracy | gen | 86.67 |

w8a8_dynamic eplb:
| dataset | version | metric | mode | vllm-api-general-chat |
|----- | ----- | ----- | ----- | -----|
| aime2024 | 604a78 | accuracy | gen | 86.67 |

The fp16 conversation is normal.
The fp16 test is in progress.

Baseline fp16
| dataset | version | metric | mode | vllm-api-general-chat |
|----- | ----- | ----- | ----- | -----|
| aime2024 | 604a78 | accuracy | gen | 86.67 |

eplb fp16
| dataset | version | metric | mode | vllm-api-general-chat |
|----- | ----- | ----- | ----- | -----|
| aime2024 | 604a78 | accuracy | gen | 83.33 |

- vLLM version: v0.13.0
- vLLM main:
45c1ca1ca1

Signed-off-by: shenchuxiaofugui <1311027364@qq.com>

tests/ut/eplb/adaptor/test_abstract_adaptor.py

@@ -1,61 +0,0 @@
-import pytest
-
-from vllm_ascend.eplb.adaptor.abstract_adaptor import EplbAdaptor
-
-
-class DummyAdaptor(EplbAdaptor):
-
-    def __init__(self, **kwargs):
-        super().__init__(**kwargs)
-        self.args = kwargs
-
-    def get_rank_expert_workload(self):
-        return "workload"
-
-    def do_update_expert_map(self, layer_id, updated_expert_map):
-        return {"layer_id": layer_id, "map": updated_expert_map}
-
-    def do_update_expert_weight(self, layer_id, local_expert_to_replace,
-                                buffer_tensor_id):
-        return {
-            "layer_id": layer_id,
-            "replace": local_expert_to_replace,
-            "buffer": buffer_tensor_id,
-        }
-
-
-def test_base_class_methods_raise():
-    adaptor = EplbAdaptor()
-    with pytest.raises(NotImplementedError):
-        adaptor.get_rank_expert_workload()
-    with pytest.raises(NotImplementedError):
-        adaptor.do_update_expert_map(1, {})
-    with pytest.raises(NotImplementedError):
-        adaptor.do_update_expert_weight(1, "x", "y")
-
-
-def test_dummy_adaptor_init_and_args():
-    adaptor = DummyAdaptor(test_arg=123)
-    assert adaptor.args["test_arg"] == 123
-
-
-def test_get_rank_expert_workload():
-    adaptor = DummyAdaptor()
-    result = adaptor.get_rank_expert_workload()
-    assert result == "workload"
-
-
-def test_do_update_expert_map():
-    adaptor = DummyAdaptor()
-    updated = {"expert": 1}
-    result = adaptor.do_update_expert_map(2, updated)
-    assert result["layer_id"] == 2
-    assert result["map"] == updated
-
-
-def test_do_update_expert_weight():
-    adaptor = DummyAdaptor()
-    result = adaptor.do_update_expert_weight(1, "expertA", "bufferX")
-    assert result["layer_id"] == 1
-    assert result["replace"] == "expertA"
-    assert result["buffer"] == "bufferX"

tests/ut/eplb/adaptor/test_vllm_adaptor.py

@@ -0,0 +1,39 @@
+import unittest
+from unittest.mock import MagicMock, patch
+
+import torch
+
+from vllm_ascend.eplb.adaptor.vllm_adaptor import VllmEplbAdaptor
+from transformers import DeepseekV2Config
+
+
+class TestVllmAdaptor(unittest.TestCase):
+    def setUp(self):
+        n_routed_experts = 256
+        mock_model = MagicMock()
+        mock_model.model.named_parameters.return_value = dict()
+        config = DeepseekV2Config(n_routed_experts=n_routed_experts)
+        mock_model.config = config
+        mock_model.get_expert_map.return_value = [i for i in range(n_routed_experts)]
+        mock_model.get_log2phy_map.return_value = [i for i in range(n_routed_experts)]
+        self.model = mock_model
+
+        self.mock_rank = patch("vllm_ascend.eplb.adaptor.vllm_adaptor.dist.get_rank", return_value=0).start()
+        self.mock_size = patch("vllm_ascend.eplb.adaptor.vllm_adaptor.dist.get_world_size", return_value=4).start()
+
+    @patch("torch.empty_like", return_value=torch.zeros(16, 32))
+    def test_init_fp16(self, mock_func):
+        self.model.quant_config = None
+        VllmEplbAdaptor(self.model)
+
+    @patch("torch.empty_like", return_value=torch.zeros(16, 32))
+    def test_init_w8a8(self, mock_func):
+        VllmEplbAdaptor(self.model)
+
+    def tearDown(self):
+        self.mock_rank.stop()
+        self.mock_size.stop()
+
+if __name__ == "__main__":
+    unittest.main()
+