Reenable OpenAI Tokenizer (#3062)

* k

* clean up test embeddings

* nit

* minor update to ensure consistency

* minor organizational update

* minor updates

---------

Co-authored-by: pablodanswer <pablo@danswer.ai>

backend/danswer/natural_language_processing/utils.py

@@ -35,23 +35,31 @@ class BaseTokenizer(ABC):
 class TiktokenTokenizer(BaseTokenizer):
     _instances: dict[str, "TiktokenTokenizer"] = {}
 
-    def __new__(cls, encoding_name: str = "cl100k_base") -> "TiktokenTokenizer":
-        if encoding_name not in cls._instances:
-            cls._instances[encoding_name] = super(TiktokenTokenizer, cls).__new__(cls)
-        return cls._instances[encoding_name]
+    def __new__(cls, model_name: str) -> "TiktokenTokenizer":
+        if model_name not in cls._instances:
+            cls._instances[model_name] = super(TiktokenTokenizer, cls).__new__(cls)
+        return cls._instances[model_name]
 
-    def __init__(self, encoding_name: str = "cl100k_base"):
+    def __init__(self, model_name: str):
         if not hasattr(self, "encoder"):
             import tiktoken
 
-            self.encoder = tiktoken.get_encoding(encoding_name)
+            self.encoder = tiktoken.encoding_for_model(model_name)
 
     def encode(self, string: str) -> list[int]:
-        # this returns no special tokens
+        # this ignores special tokens that the model is trained on, see encode_ordinary for details
         return self.encoder.encode_ordinary(string)
 
     def tokenize(self, string: str) -> list[str]:
-        return [self.encoder.decode([token]) for token in self.encode(string)]
+        encoded = self.encode(string)
+        decoded = [self.encoder.decode([token]) for token in encoded]
+
+        if len(decoded) != len(encoded):
+            logger.warning(
+                f"OpenAI tokenized length {len(decoded)} does not match encoded length {len(encoded)} for string: {string}"
+            )
+
+        return decoded
 
     def decode(self, tokens: list[int]) -> str:
         return self.encoder.decode(tokens)
@@ -74,22 +82,35 @@ class HuggingFaceTokenizer(BaseTokenizer):
         return self.encoder.decode(tokens)
 
 
-_TOKENIZER_CACHE: dict[str, BaseTokenizer] = {}
+_TOKENIZER_CACHE: dict[tuple[EmbeddingProvider | None, str | None], BaseTokenizer] = {}
 
 
-def _check_tokenizer_cache(tokenizer_name: str) -> BaseTokenizer:
+def _check_tokenizer_cache(
+    model_provider: EmbeddingProvider | None, model_name: str | None
+) -> BaseTokenizer:
     global _TOKENIZER_CACHE
 
-    if tokenizer_name not in _TOKENIZER_CACHE:
-        if tokenizer_name == "openai":
-            _TOKENIZER_CACHE[tokenizer_name] = TiktokenTokenizer("cl100k_base")
-            return _TOKENIZER_CACHE[tokenizer_name]
+    id_tuple = (model_provider, model_name)
+
+    if id_tuple not in _TOKENIZER_CACHE:
+        if model_provider in [EmbeddingProvider.OPENAI, EmbeddingProvider.AZURE]:
+            if model_name is None:
+                raise ValueError(
+                    "model_name is required for OPENAI and AZURE embeddings"
+                )
+
+            _TOKENIZER_CACHE[id_tuple] = TiktokenTokenizer(model_name)
+            return _TOKENIZER_CACHE[id_tuple]
+
         try:
-            logger.debug(f"Initializing HuggingFaceTokenizer for: {tokenizer_name}")
-            _TOKENIZER_CACHE[tokenizer_name] = HuggingFaceTokenizer(tokenizer_name)
+            if model_name is None:
+                model_name = DOCUMENT_ENCODER_MODEL
+
+            logger.debug(f"Initializing HuggingFaceTokenizer for: {model_name}")
+            _TOKENIZER_CACHE[id_tuple] = HuggingFaceTokenizer(model_name)
         except Exception as primary_error:
             logger.error(
-                f"Error initializing HuggingFaceTokenizer for {tokenizer_name}: {primary_error}"
+                f"Error initializing HuggingFaceTokenizer for {model_name}: {primary_error}"
             )
             logger.warning(
                 f"Falling back to default embedding model: {DOCUMENT_ENCODER_MODEL}"
@@ -98,7 +119,7 @@ def _check_tokenizer_cache(tokenizer_name: str) -> BaseTokenizer:
             try:
                 # Cache this tokenizer name to the default so we don't have to try to load it again
                 # and fail again
-                _TOKENIZER_CACHE[tokenizer_name] = HuggingFaceTokenizer(
+                _TOKENIZER_CACHE[id_tuple] = HuggingFaceTokenizer(
                     DOCUMENT_ENCODER_MODEL
                 )
             except Exception as fallback_error:
@@ -106,10 +127,10 @@ def _check_tokenizer_cache(tokenizer_name: str) -> BaseTokenizer:
                     f"Error initializing fallback HuggingFaceTokenizer: {fallback_error}"
                 )
                 raise ValueError(
-                    f"Failed to initialize tokenizer for {tokenizer_name} and fallback model"
+                    f"Failed to initialize tokenizer for {model_name} and fallback model"
                 ) from fallback_error
 
-    return _TOKENIZER_CACHE[tokenizer_name]
+    return _TOKENIZER_CACHE[id_tuple]
 
 
 _DEFAULT_TOKENIZER: BaseTokenizer = HuggingFaceTokenizer(DOCUMENT_ENCODER_MODEL)
@@ -118,11 +139,16 @@ _DEFAULT_TOKENIZER: BaseTokenizer = HuggingFaceTokenizer(DOCUMENT_ENCODER_MODEL)
 def get_tokenizer(
     model_name: str | None, provider_type: EmbeddingProvider | str | None
 ) -> BaseTokenizer:
-    # Currently all of the viable models use the same sentencepiece tokenizer
-    # OpenAI uses a different one but currently it's not supported due to quality issues
-    # the inconsistent chunking makes using the sentencepiece tokenizer default better for now
-    # LLM tokenizers are specified by strings
-    global _DEFAULT_TOKENIZER
+    if provider_type is not None:
+        if isinstance(provider_type, str):
+            try:
+                provider_type = EmbeddingProvider(provider_type)
+            except ValueError:
+                logger.debug(
+                    f"Invalid provider_type '{provider_type}'. Falling back to default tokenizer."
+                )
+                return _DEFAULT_TOKENIZER
+        return _check_tokenizer_cache(provider_type, model_name)
     return _DEFAULT_TOKENIZER
 
 

web/src/app/admin/embeddings/modals/ChangeCredentialsModal.tsx

@@ -11,6 +11,7 @@ import {
   LLM_PROVIDERS_ADMIN_URL,
 } from "../../configuration/llm/constants";
 import { mutate } from "swr";
+import { testEmbedding } from "../pages/utils";
 
 export function ChangeCredentialsModal({
   provider,
@@ -112,16 +113,15 @@ export function ChangeCredentialsModal({
     const normalizedProviderType = provider.provider_type
       .toLowerCase()
       .split(" ")[0];
+
     try {
-      const testResponse = await fetch("/api/admin/embedding/test-embedding", {
-        method: "POST",
-        headers: { "Content-Type": "application/json" },
-        body: JSON.stringify({
-          provider_type: normalizedProviderType,
-          api_key: apiKey,
-          api_url: apiUrl,
-          model_name: modelName,
-        }),
+      const testResponse = await testEmbedding({
+        provider_type: normalizedProviderType,
+        modelName,
+        apiKey,
+        apiUrl,
+        apiVersion: null,
+        deploymentName: null,
       });
 
       if (!testResponse.ok) {

web/src/app/admin/embeddings/modals/ProviderCreationModal.tsx

@@ -110,20 +110,27 @@ export function ProviderCreationModal({
     setErrorMsg("");
     try {
       const customConfig = Object.fromEntries(values.custom_config);
+      const providerType = values.provider_type.toLowerCase().split(" ")[0];
+      const isOpenAI = providerType === "openai";
+
+      const testModelName =
+        isOpenAI || isAzure ? "text-embedding-3-small" : values.model_name;
+
+      const testEmbeddingPayload = {
+        provider_type: providerType,
+        api_key: values.api_key,
+        api_url: values.api_url,
+        model_name: testModelName,
+        api_version: values.api_version,
+        deployment_name: values.deployment_name,
+      };
 
       const initialResponse = await fetch(
         "/api/admin/embedding/test-embedding",
         {
           method: "POST",
           headers: { "Content-Type": "application/json" },
-          body: JSON.stringify({
-            provider_type: values.provider_type.toLowerCase().split(" ")[0],
-            api_key: values.api_key,
-            api_url: values.api_url,
-            model_name: values.model_name,
-            api_version: values.api_version,
-            deployment_name: values.deployment_name,
-          }),
+          body: JSON.stringify(testEmbeddingPayload),
         }
       );
 

web/src/app/admin/embeddings/pages/utils.ts

@@ -8,3 +8,37 @@ export const deleteSearchSettings = async (search_settings_id: number) => {
   });
   return response;
 };
+
+export const testEmbedding = async ({
+  provider_type,
+  modelName,
+  apiKey,
+  apiUrl,
+  apiVersion,
+  deploymentName,
+}: {
+  provider_type: string;
+  modelName: string;
+  apiKey: string | null;
+  apiUrl: string | null;
+  apiVersion: string | null;
+  deploymentName: string | null;
+}) => {
+  const testModelName =
+    provider_type === "openai" ? "text-embedding-3-small" : modelName;
+
+  const testResponse = await fetch("/api/admin/embedding/test-embedding", {
+    method: "POST",
+    headers: { "Content-Type": "application/json" },
+    body: JSON.stringify({
+      provider_type: provider_type,
+      api_key: apiKey,
+      api_url: apiUrl,
+      model_name: testModelName,
+      api_version: apiVersion,
+      deployment_name: deploymentName,
+    }),
+  });
+
+  return testResponse;
+};