[Core] Encoder separation for Encode-Prefill-Decode Disaggregation (#4176)

### What this PR does / why we need it?
Support Encoder separation for Encode-Prefill-Decode Disaggregation

- vLLM version: v0.11.2
- vLLM main: https://github.com/vllm-project/vllm/commit/v0.11.2

Signed-off-by: amy-why-3459 <wuhaiyan17@huawei.com>

vllm_ascend/patch/platform/__init__.py

@@ -18,6 +18,7 @@ import os
 
 import vllm_ascend.patch.platform.patch_config  # noqa
 import vllm_ascend.patch.platform.patch_distributed  # noqa
+import vllm_ascend.patch.platform.patch_ec_connector  # noqa
 import vllm_ascend.patch.platform.patch_mamba_config  # noqa
 import vllm_ascend.patch.platform.patch_sched_yield  # noqa
 

vllm_ascend/patch/platform/patch_ec_connector.py

@@ -0,0 +1,32 @@
+import vllm.distributed.ec_transfer.ec_connector.shared_storage_connector
+from safetensors.torch import load_file
+from vllm.distributed.ec_transfer.ec_connector.base import ECConnectorMetadata
+from vllm.distributed.ec_transfer.ec_connector.shared_storage_connector import (
+    ECSharedStorageConnector, ECSharedStorageConnectorMetadata)
+from vllm.logger import logger
+
+
+class AscendECSharedStorageConnector(ECSharedStorageConnector):
+
+    def start_load_caches(self, encoder_cache, **kwargs) -> None:
+        metadata: ECConnectorMetadata = self._get_connector_metadata()
+        assert isinstance(metadata, ECSharedStorageConnectorMetadata)
+        assert encoder_cache is not None
+        if metadata is None:
+            logger.warning((
+                "In connector.start_load_caches, ",
+                "but the connector metadata is None",
+            ))
+            return
+        # Load the EC for each mm data
+        for mm_data in metadata.mm_datas:
+            if mm_data.mm_hash in encoder_cache:
+                continue
+            filename = self._generate_filename_debug(mm_data.mm_hash)
+            ec_cache = load_file(filename)["ec_cache"].npu()
+            encoder_cache[mm_data.mm_hash] = ec_cache
+            logger.debug("Success load encoder cache for hash %s",
+                         mm_data.mm_hash)
+
+
+vllm.distributed.ec_transfer.ec_connector.shared_storage_connector.ECSharedStorageConnector = AscendECSharedStorageConnector