Add support for Koboldcpp tokenization endpoint

public/index.html

@@ -2245,7 +2245,7 @@
                                 <option value="3">Sentencepiece (LLaMA)</option>
                                 <option value="4">NerdStash (NovelAI Clio)</option>
                                 <option value="5">NerdStash v2 (NovelAI Kayra)</option>
-                                <option value="6">API (WebUI)</option>
+                                <option value="6">API (WebUI / koboldcpp)</option>
                             </select>
                         </div>
                         <div class="range-block">

public/scripts/tokenizers.js

@@ -24,6 +24,15 @@ const gpt3 = new GPT3BrowserTokenizer({ type: 'gpt3' });
 
 let tokenCache = {};
 
+/**
+ * Guesstimates the token count for a string.
+ * @param {string} str String to tokenize.
+ * @returns {number} Token count.
+ */
+export function guesstimate(str) {
+    return Math.ceil(str.length / CHARACTERS_PER_TOKEN_RATIO);
+}
+
 async function loadTokenCache() {
     try {
         console.debug('Chat Completions: loading token cache')
@@ -89,7 +98,7 @@ export function getTokenCount(str, padding = undefined) {
     function calculate(type) {
         switch (type) {
             case tokenizers.NONE:
-                return Math.ceil(str.length / CHARACTERS_PER_TOKEN_RATIO) + padding;
+                return guesstimate(str) + padding;
             case tokenizers.GPT3:
                 return gpt3.encode(str).bpe.length + padding;
             case tokenizers.CLASSIC:
@@ -291,8 +300,16 @@ function getTokenCacheObject() {
     return tokenCache[String(chatId)];
 }
 
+/**
+ * Counts token using the remote server API.
+ * @param {string} endpoint API endpoint.
+ * @param {string} str String to tokenize.
+ * @param {number} padding Number of padding tokens.
+ * @returns {number} Token count with padding.
+ */
 function countTokensRemote(endpoint, str, padding) {
     let tokenCount = 0;
+
     jQuery.ajax({
         async: false,
         type: 'POST',
@@ -301,9 +318,25 @@ function countTokensRemote(endpoint, str, padding) {
         dataType: "json",
         contentType: "application/json",
         success: function (data) {
+            if (typeof data.count === 'number') {
                 tokenCount = data.count;
+            } else {
+                tokenCount = guesstimate(str);
+                console.error("Error counting tokens");
+
+                if (!sessionStorage.getItem('tokenizationWarningShown')) {
+                    toastr.warning(
+                        "Your selected API doesn't support the tokenization endpoint. Using estimated counts.",
+                        "Error counting tokens",
+                        { timeOut: 10000, preventDuplicates: true },
+                    );
+
+                    sessionStorage.setItem('tokenizationWarningShown', String(true));
+                }
+            }
         }
     });
+
     return tokenCount + padding;
 }
 

server.js

@@ -3842,11 +3842,19 @@ app.post("/tokenize_via_api", jsonParser, async function (request, response) {
 
         if (main_api == 'textgenerationwebui' && request.body.use_mancer) {
             args.headers = Object.assign(args.headers, get_mancer_headers());
+            const data = await postAsync(api_server + "/v1/token-count", args);
+            return response.send({ count: data['results'][0]['tokens'] });
         }
 
-        const data = await postAsync(api_server + "/v1/token-count", args);
+        else if (main_api == 'kobold') {
-        console.log(data);
+            const data = await postAsync(api_server + "/extra/tokencount", args);
-        return response.send({ count: data['results'][0]['tokens'] });
+            const count = data['value'];
+            return response.send({ count: count });
+        }
+
+        else {
+            return response.send({ error: true });
+        }
     } catch (error) {
         console.log(error);
         return response.send({ error: true });