Add support for WebLLM

src/lib/Setting/Pages/BotSettings.svelte

@@ -329,7 +329,7 @@
     <span class="text-textcolor2 mb-6 text-sm">{($DataBase.top_k).toFixed(0)}</span>
 {/if}
 
-{#if $DataBase.aiModel === 'textgen_webui' || $DataBase.aiModel === 'mancer' || $DataBase.aiModel.startsWith('local_')}
+{#if $DataBase.aiModel === 'textgen_webui' || $DataBase.aiModel === 'mancer' || $DataBase.aiModel.startsWith('local_') || $DataBase.aiModel.startsWith('hf:::')}
     <span class="text-textcolor">Repetition Penalty</span>
     <SliderInput min={1} max={1.5} step={0.01} bind:value={$DataBase.ooba.repetition_penalty}/>
     <span class="text-textcolor2 mb-6 text-sm">{($DataBase.ooba.repetition_penalty).toFixed(2)}</span>

src/lib/UI/ModelList.svelte

@@ -3,7 +3,6 @@
     import { getHordeModels } from "src/ts/horde/getModels";
     import Arcodion from "./Arcodion.svelte";
     import { language } from "src/lang";
-    import { isNodeServer, isTauri } from "src/ts/storage/globalApi";
     import Help from "../Others/Help.svelte";
     import CheckInput from "./GUI/CheckInput.svelte";
 
@@ -86,6 +85,10 @@
                     const split = name.split(":::")
                     return `Horde ${split[1]}`
                 }
+                if(name.startsWith('tf:::')){
+                    const split = name.split(":::")
+                    return `${split[1]}`
+                }
                 return name
         }
     }
@@ -190,7 +193,14 @@
                         </button>
                     {/each}
                 {/await}
-                </Arcodion>
+            </Arcodion>
+            {#if showUnrec}
+            <Arcodion name="WebLLM Local">
+                <button class="hover:bg-selected px-6 py-2 text-lg" on:click={() => {changeModel('hf:::Xenova/opt-350m')}}>opt-350m</button>
+                <button class="hover:bg-selected px-6 py-2 text-lg" on:click={() => {changeModel('hf:::Xenova/tiny-random-mistral')}}>tiny-random-mistral</button>
+                <button class="hover:bg-selected px-6 py-2 text-lg" on:click={() => {changeModel('hf:::Xenova/gpt2-large-conversational')}}>gpt2-large-conversational</button>
+            </Arcodion>
+            {/if}
             {#if $DataBase.plugins.length > 0}
                 <button on:click={() => {changeModel('custom')}} class="hover:bg-selected px-6 py-2 text-lg" >Plugin</button>
             {/if}

src/ts/process/embedding/transformers.ts

@@ -1,6 +1,6 @@
 import {env, AutoTokenizer, pipeline, type SummarizationOutput, type TextGenerationConfig, type TextGenerationOutput, FeatureExtractionPipeline, TextToAudioPipeline } from '@xenova/transformers';
 import { unzip } from 'fflate';
-import { loadAsset, saveAsset } from 'src/ts/storage/globalApi';
+import { globalFetch, loadAsset, saveAsset } from 'src/ts/storage/globalApi';
 import { selectSingleFile } from 'src/ts/util';
 import { v4 } from 'uuid';
 let tfCache:Cache = null

src/ts/process/request.ts

@@ -21,6 +21,7 @@ import { supportsInlayImage } from "../image";
 import { OaifixEmdash } from "../plugins/fixer";
 import { Capacitor } from "@capacitor/core";
 import { getFreeOpenRouterModel } from "../model/openrouter";
+import { runTransformers } from "./embedding/transformers";
 
 
 
@@ -1632,6 +1633,23 @@ export async function requestChatDataMain(arg:requestDataArgument, model:'model'
                 }
 
 
+            }
+            if(aiModel.startsWith('hf:::')){
+                const realModel = aiModel.split(":::")[1]
+                const suggesting = model === "submodel"
+                const proompt = stringlizeChatOba(formated, currentChar.name, suggesting, arg.continue)
+                const v = await runTransformers(proompt, realModel, {
+                    temperature: temperature,
+                    max_new_tokens: maxTokens,
+                    top_k: db.ooba.top_k,
+                    top_p: db.ooba.top_p,
+                    repetition_penalty: db.ooba.repetition_penalty,
+                    typical_p: db.ooba.typical_p,
+                })
+                return {
+                    type: 'success',
+                    result: unstringlizeChat(v.generated_text, formated, currentChar?.name ?? '')
+                }
             }
             if(aiModel.startsWith('local_')){
                 console.log('running local model')