fix: allow configuring embeddings encoding format

README.md

@@ -204,6 +204,7 @@ The following environment variables are required to run the application:
     - google_genai: "gemini-embedding-001"
 - `EMBEDDINGS_CHUNK_SIZE`: (Optional) The chunk size used by the OpenAI and Azure embeddings clients to limit the number of inputs per request. Default value is `200`.
 - `EMBEDDINGS_DIMENSIONS`: (Optional) Output vector size to request from the embedding model. Only honored by the `openai` and `azure` providers, and only supported by `text-embedding-3-*` models. Leave unset to use the model's native dimensionality (1536 for `text-embedding-3-small`, 3072 for `text-embedding-3-large`). Setting a smaller value (e.g. `512`, `1024`) trades some retrieval quality for lower storage cost and faster similarity search. Note: do not change this on an existing collection — all vectors in a `pgvector` column must share the same dimensionality.
+- `EMBEDDINGS_ENCODING_FORMAT`: (Optional) Embedding response format for the `openai` and `azure` providers. Accepted values are `float` and `base64`. Leave unset to preserve the provider SDK default. Set to `float` for OpenAI-compatible APIs that do not support Base64 embedding responses.
 - `RAG_AZURE_OPENAI_API_VERSION`: (Optional) Default is `2023-05-15`. The version of the Azure OpenAI API.
 - `RAG_AZURE_OPENAI_API_KEY`: (Optional) The API key for Azure OpenAI service.
     - Note: `AZURE_OPENAI_API_KEY` will work but `RAG_AZURE_OPENAI_API_KEY` will override it in order to not conflict with LibreChat setting.

app/config.py

@@ -223,6 +223,13 @@ async def dispatch(self, request, call_next):
 GOOGLE_APPLICATION_CREDENTIALS = get_env_variable("GOOGLE_APPLICATION_CREDENTIALS", "")
 env_value = get_env_variable("RAG_CHECK_EMBEDDING_CTX_LENGTH", "True").lower()
 RAG_CHECK_EMBEDDING_CTX_LENGTH = True if env_value == "true" else False
+EMBEDDINGS_ENCODING_FORMAT = get_env_variable("EMBEDDINGS_ENCODING_FORMAT", None)
+if EMBEDDINGS_ENCODING_FORMAT is not None:
+    EMBEDDINGS_ENCODING_FORMAT = EMBEDDINGS_ENCODING_FORMAT.lower()
+    if EMBEDDINGS_ENCODING_FORMAT not in ("float", "base64"):
+        raise ValueError(
+            "EMBEDDINGS_ENCODING_FORMAT must be either 'float' or 'base64'"
+        )
 
 # Only parse RAG_DISTANCE_THRESHOLD when it will actually be applied (pgvector).
 # Under atlas-mongo the setting is documented as ignored, so parsing it
@@ -251,6 +258,10 @@ def init_embeddings(provider, model, dimensions=None):
         )
         if dimensions is not None:
             kwargs["dimensions"] = dimensions
+        if EMBEDDINGS_ENCODING_FORMAT is not None:
+            kwargs["model_kwargs"] = {
+                "encoding_format": EMBEDDINGS_ENCODING_FORMAT
+            }
         return OpenAIEmbeddings(**kwargs)
     elif provider == EmbeddingsProvider.AZURE:
         from langchain_openai import AzureOpenAIEmbeddings
@@ -265,6 +276,10 @@ def init_embeddings(provider, model, dimensions=None):
         )
         if dimensions is not None:
             kwargs["dimensions"] = dimensions
+        if EMBEDDINGS_ENCODING_FORMAT is not None:
+            kwargs["model_kwargs"] = {
+                "encoding_format": EMBEDDINGS_ENCODING_FORMAT
+            }
         return AzureOpenAIEmbeddings(**kwargs)
     elif provider == EmbeddingsProvider.HUGGINGFACE:
         from langchain_huggingface import HuggingFaceEmbeddings