Fix huggingface_hub API upgrade issue

comps/cores/mega/orchestrator.py

@@ -274,7 +274,7 @@ async def execute(
                         headers={"Content-type": "application/json", "Authorization": f"Bearer {access_token}"},
                         proxies={"http": None},
                         stream=True,
-                        timeout=1000,
+                        timeout=2000,
                     )
                 else:
                     response = requests.post(
@@ -285,7 +285,7 @@ async def execute(
                         },
                         proxies={"http": None},
                         stream=True,
-                        timeout=1000,
+                        timeout=2000,
                     )
 
             downstream = runtime_graph.downstream(cur_node)
@@ -317,6 +317,7 @@ def generate():
                                                 "Authorization": f"Bearer {access_token}",
                                             },
                                             proxies={"http": None},
+                                            timeout=2000,
                                         )
                                     else:
                                         res = requests.post(
@@ -326,6 +327,7 @@ def generate():
                                                 "Content-type": "application/json",
                                             },
                                             proxies={"http": None},
+                                            timeout=2000,
                                         )
                                     res_json = res.json()
                                     if "text" in res_json:

comps/embeddings/deployment/docker_compose/compose.yaml

@@ -36,6 +36,7 @@ services:
       https_proxy: ${https_proxy}
       TEI_EMBEDDING_ENDPOINT: ${TEI_EMBEDDING_ENDPOINT}
       EMBEDDING_COMPONENT_NAME: "OPEA_TEI_EMBEDDING"
+      EMBEDDING_MODEL_ID: ${EMBEDDING_MODEL_ID}
     depends_on:
       tei-embedding-serving:
         condition: service_healthy

comps/embeddings/src/integrations/ovms.py

@@ -10,7 +10,7 @@
 
 from comps import CustomLogger, OpeaComponent, OpeaComponentRegistry, ServiceType
 from comps.cores.mega.utils import get_access_token
-from comps.cores.proto.api_protocol import EmbeddingRequest, EmbeddingResponse
+from comps.cores.proto.api_protocol import EmbeddingRequest, EmbeddingResponse, EmbeddingResponseData
 
 logger = CustomLogger("opea_ovms_embedding")
 logflag = os.getenv("LOGFLAG", False)
@@ -45,7 +45,7 @@ def _initialize_client(self) -> AsyncInferenceClient:
         )
         headers = {"Authorization": f"Bearer {access_token}"} if access_token else {}
         return AsyncInferenceClient(
-            model=MODEL_ID,
+            model=f"{self.base_url}/embed",
             token=os.getenv("HUGGINGFACEHUB_API_TOKEN"),
             headers=headers,
         )
@@ -69,18 +69,13 @@ async def invoke(self, input: EmbeddingRequest) -> EmbeddingResponse:
                 raise ValueError("Invalid input format: Only string or list of strings are supported.")
         else:
             raise TypeError("Unsupported input type: input must be a string or list of strings.")
-        response = await self.client.post(
-            json={
-                "input": texts,
-                "encoding_format": input.encoding_format,
-                "model": self.client.model,
-                "user": input.user,
-            },
-            model=f"{self.base_url}/v3/embeddings",
-            task="text-embedding",
-        )
-        embeddings = json.loads(response.decode())
-        return EmbeddingResponse(**embeddings)
+        # feature_extraction return np.ndarray
+        response = await self.client.feature_extraction(text=texts, model=MODEL_ID)
+        # Convert np.ndarray to a list of lists (embedding)
+        data = [EmbeddingResponseData(index=i, embedding=embedding.tolist()) for i, embedding in enumerate(response)]
+        # Construct the EmbeddingResponse
+        response = EmbeddingResponse(data=data)
+        return response
 
     def check_health(self) -> bool:
         """Checks the health of the embedding service.

comps/embeddings/src/integrations/tei.py

@@ -10,13 +10,14 @@
 
 from comps import CustomLogger, OpeaComponent, OpeaComponentRegistry, ServiceType
 from comps.cores.mega.utils import get_access_token
-from comps.cores.proto.api_protocol import EmbeddingRequest, EmbeddingResponse
+from comps.cores.proto.api_protocol import EmbeddingRequest, EmbeddingResponse, EmbeddingResponseData
 
 logger = CustomLogger("opea_tei_embedding")
 logflag = os.getenv("LOGFLAG", False)
 TOKEN_URL = os.getenv("TOKEN_URL")
 CLIENTID = os.getenv("CLIENTID")
 CLIENT_SECRET = os.getenv("CLIENT_SECRET")
+MODEL_ID = os.getenv("EMBEDDING_MODEL_ID", "BAAI/bge-base-en-v1.5")
 
 
 @OpeaComponentRegistry.register("OPEA_TEI_EMBEDDING")
@@ -44,7 +45,7 @@ def _initialize_client(self) -> AsyncInferenceClient:
         )
         headers = {"Authorization": f"Bearer {access_token}"} if access_token else {}
         return AsyncInferenceClient(
-            model=f"{self.base_url}/v1/embeddings",
+            model=f"{self.base_url}/embed",
             token=os.getenv("HUGGINGFACEHUB_API_TOKEN"),
             headers=headers,
         )
@@ -68,13 +69,13 @@ async def invoke(self, input: EmbeddingRequest) -> EmbeddingResponse:
                 raise ValueError("Invalid input format: Only string or list of strings are supported.")
         else:
             raise TypeError("Unsupported input type: input must be a string or list of strings.")
-        response = await self.client.post(
-            json={"input": texts, "encoding_format": input.encoding_format, "model": input.model, "user": input.user},
-            model=f"{self.base_url}/v1/embeddings",
-            task="text-embedding",
-        )
-        embeddings = json.loads(response.decode())
-        return EmbeddingResponse(**embeddings)
+        # feature_extraction return np.ndarray
+        response = await self.client.feature_extraction(text=texts, model=MODEL_ID)
+        # Convert np.ndarray to a list of lists (embedding)
+        data = [EmbeddingResponseData(index=i, embedding=embedding.tolist()) for i, embedding in enumerate(response)]
+        # Construct the EmbeddingResponse
+        response = EmbeddingResponse(data=data)
+        return response
 
     def check_health(self) -> bool:
         """Checks the health of the embedding service.

comps/rerankings/src/requirements.txt

@@ -2,6 +2,7 @@ aiohttp
 docarray[full]
 fastapi
 httpx
+huggingface-hub==0.30.2
 opentelemetry-api
 opentelemetry-exporter-otlp
 opentelemetry-sdk