sets default dtype in vectorizers to float32

Author: justin-cechmanek

redisvl/utils/vectorize/base.py

@@ -81,13 +81,7 @@ def batchify(self, seq: list, size: int, preprocess: Optional[Callable] = None):
             else:
                 yield seq[pos : pos + size]
 
-    def _process_embedding(
-        self, embedding: List[float], as_buffer: bool, dtype: Optional[str]
-    ):
+    def _process_embedding(self, embedding: List[float], as_buffer: bool, dtype: str):
         if as_buffer:
-            if not dtype:
-                raise RuntimeError(
-                    "dtype is required if converting from float to byte string."
-                )
             return array_to_buffer(embedding, dtype)
         return embedding

redisvl/utils/vectorize/text/azureopenai.py

@@ -190,7 +190,7 @@ def embed_many(
         if len(texts) > 0 and not isinstance(texts[0], str):
             raise TypeError("Must pass in a list of str values to embed.")
 
-        dtype = kwargs.pop("dtype", None)
+        dtype = kwargs.pop("dtype", "float32")
 
         embeddings: List = []
         for batch in self.batchify(texts, batch_size, preprocess):
@@ -234,7 +234,7 @@ def embed(
         if preprocess:
             text = preprocess(text)
 
-        dtype = kwargs.pop("dtype", None)
+        dtype = kwargs.pop("dtype", "float32")
 
         result = self._client.embeddings.create(input=[text], model=self.model)
         return self._process_embedding(result.data[0].embedding, as_buffer, dtype)
@@ -274,7 +274,7 @@ async def aembed_many(
         if len(texts) > 0 and not isinstance(texts[0], str):
             raise TypeError("Must pass in a list of str values to embed.")
 
-        dtype = kwargs.pop("dtype", None)
+        dtype = kwargs.pop("dtype", "float32")
 
         embeddings: List = []
         for batch in self.batchify(texts, batch_size, preprocess):
@@ -320,7 +320,7 @@ async def aembed(
         if preprocess:
             text = preprocess(text)
 
-        dtype = kwargs.pop("dtype", None)
+        dtype = kwargs.pop("dtype", "float32")
 
         result = await self._aclient.embeddings.create(input=[text], model=self.model)
         return self._process_embedding(result.data[0].embedding, as_buffer, dtype)

redisvl/utils/vectorize/text/bedrock.py

@@ -145,7 +145,7 @@ def embed(
         response_body = json.loads(response["body"].read())
         embedding = response_body["embedding"]
 
-        dtype = kwargs.pop("dtype", None)
+        dtype = kwargs.pop("dtype", "float32")
         return self._process_embedding(embedding, as_buffer, dtype)
 
     @retry(
@@ -181,7 +181,7 @@ def embed_many(
             raise TypeError("Texts must be a list of strings")
 
         embeddings: List[List[float]] = []
-        dtype = kwargs.pop("dtype", None)
+        dtype = kwargs.pop("dtype", "float32")
 
         for batch in self.batchify(texts, batch_size, preprocess):
             # Process each text in the batch individually since Bedrock

redisvl/utils/vectorize/text/cohere.py

@@ -159,7 +159,7 @@ def embed(
         if preprocess:
             text = preprocess(text)
 
-        dtype = kwargs.pop("dtype", None)
+        dtype = kwargs.pop("dtype", "float32")
 
         embedding = self._client.embed(
             texts=[text], model=self.model, input_type=input_type
@@ -228,7 +228,7 @@ def embed_many(
                     See https://docs.cohere.com/reference/embed."
             )
 
-        dtype = kwargs.pop("dtype", None)
+        dtype = kwargs.pop("dtype", "float32")
 
         embeddings: List = []
         for batch in self.batchify(texts, batch_size, preprocess):

redisvl/utils/vectorize/text/custom.py

@@ -173,7 +173,7 @@ def embed(
         if preprocess:
             text = preprocess(text)
 
-        dtype = kwargs.pop("dtype", None)
+        dtype = kwargs.pop("dtype", "float32")
 
         result = self._embed_func(text, **kwargs)
         return self._process_embedding(result, as_buffer, dtype)
@@ -212,7 +212,7 @@ def embed_many(
         if not self._embed_many_func:
             raise NotImplementedError
 
-        dtype = kwargs.pop("dtype", None)
+        dtype = kwargs.pop("dtype", "float32")
 
         embeddings: List = []
         for batch in self.batchify(texts, batch_size, preprocess):
@@ -254,7 +254,7 @@ async def aembed(
         if preprocess:
             text = preprocess(text)
 
-        dtype = kwargs.pop("dtype", None)
+        dtype = kwargs.pop("dtype", "float32")
 
         result = await self._aembed_func(text, **kwargs)
         return self._process_embedding(result, as_buffer, dtype)
@@ -293,7 +293,7 @@ async def aembed_many(
         if not self._aembed_many_func:
             raise NotImplementedError
 
-        dtype = kwargs.pop("dtype", None)
+        dtype = kwargs.pop("dtype", "float32")
 
         embeddings: List = []
         for batch in self.batchify(texts, batch_size, preprocess):

redisvl/utils/vectorize/text/huggingface.py

@@ -100,7 +100,7 @@ def embed(
         if preprocess:
             text = preprocess(text)
 
-        dtype = kwargs.pop("dtype", None)
+        dtype = kwargs.pop("dtype", "float32")
 
         embedding = self._client.encode([text], **kwargs)[0]
         return self._process_embedding(embedding.tolist(), as_buffer, dtype)
@@ -136,7 +136,7 @@ def embed_many(
         if len(texts) > 0 and not isinstance(texts[0], str):
             raise TypeError("Must pass in a list of str values to embed.")
 
-        dtype = kwargs.pop("dtype", None)
+        dtype = kwargs.pop("dtype", "float32")
 
         embeddings: List = []
         for batch in self.batchify(texts, batch_size, preprocess):

redisvl/utils/vectorize/text/mistral.py

@@ -140,7 +140,7 @@ def embed_many(
         if len(texts) > 0 and not isinstance(texts[0], str):
             raise TypeError("Must pass in a list of str values to embed.")
 
-        dtype = kwargs.pop("dtype", None)
+        dtype = kwargs.pop("dtype", "float32")
 
         embeddings: List = []
         for batch in self.batchify(texts, batch_size, preprocess):
@@ -184,7 +184,7 @@ def embed(
         if preprocess:
             text = preprocess(text)
 
-        dtype = kwargs.pop("dtype", None)
+        dtype = kwargs.pop("dtype", "float32")
 
         result = self._client.embeddings(model=self.model, input=[text])
         return self._process_embedding(result.data[0].embedding, as_buffer, dtype)
@@ -224,7 +224,7 @@ async def aembed_many(
         if len(texts) > 0 and not isinstance(texts[0], str):
             raise TypeError("Must pass in a list of str values to embed.")
 
-        dtype = kwargs.pop("dtype", None)
+        dtype = kwargs.pop("dtype", "float32")
 
         embeddings: List = []
         for batch in self.batchify(texts, batch_size, preprocess):
@@ -268,7 +268,7 @@ async def aembed(
         if preprocess:
             text = preprocess(text)
 
-        dtype = kwargs.pop("dtype", None)
+        dtype = kwargs.pop("dtype", "float32")
 
         result = await self._aclient.embeddings(model=self.model, input=[text])
         return self._process_embedding(result.data[0].embedding, as_buffer, dtype)

redisvl/utils/vectorize/text/openai.py

@@ -144,7 +144,7 @@ def embed_many(
         if len(texts) > 0 and not isinstance(texts[0], str):
             raise TypeError("Must pass in a list of str values to embed.")
 
-        dtype = kwargs.pop("dtype", None)
+        dtype = kwargs.pop("dtype", "float32")
 
         embeddings: List = []
         for batch in self.batchify(texts, batch_size, preprocess):
@@ -188,7 +188,7 @@ def embed(
         if preprocess:
             text = preprocess(text)
 
-        dtype = kwargs.pop("dtype", None)
+        dtype = kwargs.pop("dtype", "float32")
 
         result = self._client.embeddings.create(input=[text], model=self.model)
         return self._process_embedding(result.data[0].embedding, as_buffer, dtype)
@@ -228,7 +228,7 @@ async def aembed_many(
         if len(texts) > 0 and not isinstance(texts[0], str):
             raise TypeError("Must pass in a list of str values to embed.")
 
-        dtype = kwargs.pop("dtype", None)
+        dtype = kwargs.pop("dtype", "float32")
 
         embeddings: List = []
         for batch in self.batchify(texts, batch_size, preprocess):
@@ -274,7 +274,7 @@ async def aembed(
         if preprocess:
             text = preprocess(text)
 
-        dtype = kwargs.pop("dtype", None)
+        dtype = kwargs.pop("dtype", "float32")
 
         result = await self._aclient.embeddings.create(input=[text], model=self.model)
         return self._process_embedding(result.data[0].embedding, as_buffer, dtype)

redisvl/utils/vectorize/text/vertexai.py

@@ -151,7 +151,7 @@ def embed_many(
         if len(texts) > 0 and not isinstance(texts[0], str):
             raise TypeError("Must pass in a list of str values to embed.")
 
-        dtype = kwargs.pop("dtype", None)
+        dtype = kwargs.pop("dtype", "float32")
 
         embeddings: List = []
         for batch in self.batchify(texts, batch_size, preprocess):
@@ -194,7 +194,7 @@ def embed(
         if preprocess:
             text = preprocess(text)
 
-        dtype = kwargs.pop("dtype", None)
+        dtype = kwargs.pop("dtype", "float32")
 
         result = self._client.get_embeddings([text])
         return self._process_embedding(result[0].values, as_buffer, dtype)