fix(llm): Fix when use custom embedding providers. (#339)

* fix(llm): Fix when use custom embedding providers.

* fix validate.

* fix validate.

Author: lx200916

opencontext/llm/llm_client.py

@@ -266,14 +266,14 @@ async def _openai_chat_completion_stream_async(self, messages: List[Dict[str, An
 
     def _request_embedding(self, text: str, **kwargs) -> List[float]:
         try:
-            if self.provider == LLMProvider.DOUBAO.value:
+            if self.provider != LLMProvider.DOUBAO.value:
+                response = self.client.embeddings.create(model=self.model, input=[text])
+                embedding = response.data[0].embedding
+            else:
                 response = self.client.multimodal_embeddings.create(
                     model=self.model, input=[{"type": "text", "text": text}]
                 )
                 embedding = response.data.embedding
-            else:
-                response = self.client.embeddings.create(model=self.model, input=[text])
-                embedding = response.data[0].embedding
 
             # Record token usage
             if hasattr(response, "usage") and response.usage:
@@ -314,6 +314,7 @@ def _request_embedding(self, text: str, **kwargs) -> List[float]:
     async def _request_embedding_async(self, text: str, **kwargs) -> List[float]:
         try:
             if self.provider == LLMProvider.DOUBAO.value:
+                # Only ark has multimodal_embeddings
                 response = self.client.multimodal_embeddings.create(
                     model=self.model, input=[{"type": "text", "text": text}]
                 )