hypav2 enhancment 2

src/lib/Setting/Pages/OtherBotSettings.svelte

@@ -276,7 +276,7 @@
     <span class="text-textcolor mt-4">{language.type}</span>
 
     <SelectInput value={
-        $DataBase.supaMemoryType === 'hypaV2' ? 'hypaV2' :
+        $DataBase.hypav2 ? 'hypaV2' :
         $DataBase.supaMemoryType !== 'none' ? 'supaMemory' :
         $DataBase.hanuraiEnable ? 'hanuraiMemory' : 'none'
     } on:change={(v) => {
@@ -284,15 +284,19 @@
         const value = v.target.value
         if (value === 'supaMemory'){
             $DataBase.supaMemoryType = 'distilbart'
+            $DataBase.hypav2 = false
             $DataBase.hanuraiEnable = false
         } else if (value === 'hanuraiMemory'){
             $DataBase.supaMemoryType = 'none'
+            $DataBase.hypav2 = false
             $DataBase.hanuraiEnable = true
         } else if (value === 'hypaV2') {
             $DataBase.supaMemoryType = 'hypaV2'
+            $DataBase.hypav2= true
             $DataBase.hanuraiEnable = false
         } else {
             $DataBase.supaMemoryType = 'none'
+            $DataBase.hypav2 = false
             $DataBase.hanuraiEnable = false
         }
     }}>
@@ -309,26 +313,27 @@
         <div class="flex">
             <Check bind:check={$DataBase.hanuraiSplit} name="Text Spliting"/>
         </div>
-    {:else if $DataBase.supaMemoryType === 'hypaV2'}
+    {:else if $DataBase.hypav2}
         <span class="mb-2 text-textcolor2 text-sm text-wrap break-words max-w-full">{language.hypaV2Desc}</span>
         <span class="text-textcolor mt-4">{language.SuperMemory} {language.model}</span>
         <SelectInput className="mt-2 mb-2" bind:value={$DataBase.supaMemoryType}>
-            <OptionInput value="distilbart" >distilbart-cnn-6-6 (Free/Local)</OptionInput>
+            <OptionInput value="distilbart">distilbart-cnn-6-6 (Free/Local)</OptionInput>
-            <OptionInput value="instruct35" >OpenAI 3.5 Turbo Instruct</OptionInput>
+            <OptionInput value="instruct35">OpenAI 3.5 Turbo Instruct</OptionInput>
-            <OptionInput value="subModel" >{language.submodel}</OptionInput>
+            <OptionInput value="subModel">{language.submodel}</OptionInput>
         </SelectInput>
         <span class="text-textcolor">{language.SuperMemory} Prompt</span>
         <TextInput size="sm" marginBottom bind:value={$DataBase.supaMemoryPrompt} placeholder="Leave it blank to use default"/>
         <span class="text-textcolor">{language.HypaMemory} Model</span>
         <SelectInput className="mt-2 mb-2" bind:value={$DataBase.hypaModel}>
-            <OptionInput value="MiniLM" >MiniLM-L6-v2 (Free / Local)</OptionInput>
+            <OptionInput value="MiniLM">MiniLM-L6-v2 (Free / Local)</OptionInput>
-            <OptionInput value="ada" >OpenAI Ada (Davinci / Curie Only)</OptionInput>
+            <OptionInput value="nomic">Nomic (Free / Local)</OptionInput>
+            <OptionInput value="ada">OpenAI Ada (Davinci / Curie Only)</OptionInput>
         </SelectInput>
         <span class="text-textcolor">{language.hypaChunkSize}</span>
         <NumberInput size="sm" marginBottom bind:value={$DataBase.hypaChunkSize} min={100} />
         <span class="text-textcolor">{language.hypaAllocatedTokens}</span>
         <NumberInput size="sm" marginBottom bind:value={$DataBase.hypaAllocatedTokens} min={100} />
-    {:else if $DataBase.supaMemoryType !== 'none'}
+    {:else if ($DataBase.supaMemoryType !== 'none' && $DataBase.hypav2 === false)}
         <span class="mb-2 text-textcolor2 text-sm text-wrap break-words max-w-full">{language.supaDesc}</span>
         <span class="text-textcolor mt-4">{language.SuperMemory} {language.model}</span>
         <SelectInput className="mt-2 mb-2" bind:value={$DataBase.supaMemoryType}>

src/ts/process/memory/hypav2.ts

@@ -4,210 +4,266 @@ import type { ChatTokenizer } from "src/ts/tokenizer";
 import { get } from "svelte/store";
 import { requestChatData } from "../request";
 import { HypaProcesser } from "./hypamemory";
+import { globalFetch } from "src/ts/storage/globalApi";
+import { runSummarizer } from "../transformers";
 
-export interface HypaV2Data{
+export interface HypaV2Data {
     chunks: {
-        text:string
+        text: string;
-        targetId:string
+        targetId: string;
-    }[]
+    }[];
     mainChunks: {
-        text:string
+        text: string;
-        targetId:string
+        targetId: string;
-    }[]
+    }[];
 }
 
+async function summarize(stringlizedChat: string): Promise<{ success: boolean; data: string }> {
+    const db = get(DataBase);
 
+    if (db.supaMemoryType === 'distilbart') {
+        try {
+            const sum = await runSummarizer(stringlizedChat);
+            return { success: true, data: sum };
+        } catch (error) {
+            return {
+                success: false,
+                data: "SupaMemory: Summarizer: " + `${error}`
+            };
+        }
+    }
+
+    const supaPrompt = db.supaMemoryPrompt === '' ?
+        "[Summarize the ongoing role story, It must also remove redundancy and unnecessary text and content from the output to reduce tokens for gpt3 and other sublanguage models]\n"
+        : db.supaMemoryPrompt;
+
+    let result = '';
+
+    if (db.supaMemoryType !== 'subModel') {
+        const promptbody = stringlizedChat + '\n\n' + supaPrompt + "\n\nOutput:";
+
+        const da = await globalFetch("https://api.openai.com/v1/completions", {
+            headers: {
+                "Content-Type": "application/json",
+                "Authorization": "Bearer " + db.supaMemoryKey
+            },
+            method: "POST",
+            body: {
+                "model": db.supaMemoryType === 'curie' ? "text-curie-001"
+                    : db.supaMemoryType === 'instruct35' ? 'gpt-3.5-turbo-instruct'
+                        : "text-davinci-003",
+                "prompt": promptbody,
+                "max_tokens": 600,
+                "temperature": 0
+            }
+        });
+
+        try {
+            if (!da.ok) {
+                return {
+                    success: false,
+                    data: "SupaMemory: HTTP: " + JSON.stringify(da.data)
+                };
+            }
+
+            result = (await da.data)?.choices[0]?.text?.trim();
+
+            if (!result) {
+                return {
+                    success: false,
+                    data: "SupaMemory: HTTP: " + JSON.stringify(da.data)
+                };
+            }
+
+            return { success: true, data: result };
+        } catch (error) {
+            return {
+                success: false,
+                data: "SupaMemory: HTTP: " + error
+            };
+        }
+    } else {
+        const promptbody: OpenAIChat[] = [
+            {
+                role: "user",
+                content: stringlizedChat
+            },
+            {
+                role: "system",
+                content: supaPrompt
+            }
+        ];
+        const da = await requestChatData({
+            formated: promptbody,
+            bias: {},
+            useStreaming: false,
+            noMultiGen: true
+        }, 'submodel');
+        if (da.type === 'fail' || da.type === 'streaming' || da.type === 'multiline') {
+            return {
+                success: false,
+                data: "SupaMemory: HTTP: " + da.result
+            };
+        }
+        result = da.result;
+    }
+    return { success: true, data: result };
+}
 
 export async function hypaMemoryV2(
-    chats:OpenAIChat[],
+    chats: OpenAIChat[],
-    currentTokens:number,
+    currentTokens: number,
-    maxContextTokens:number,
+    maxContextTokens: number,
-    room:Chat,
+    room: Chat,
-    char:character|groupChat,
+    char: character | groupChat,
-    tokenizer:ChatTokenizer,
+    tokenizer: ChatTokenizer,
-    arg:{asHyper?:boolean} = {}
+    arg: { asHyper?: boolean, summaryModel?: string, summaryPrompt?: string, hypaModel?: string } = {}
-): Promise<{ currentTokens: number; chats: OpenAIChat[]; error?:string; memory?:HypaV2Data;}>{
+): Promise<{ currentTokens: number; chats: OpenAIChat[]; error?: string; memory?: HypaV2Data; }> {
 
-    const db = get(DataBase)
+    const db = get(DataBase);
 
-    const data:HypaV2Data = room.hypaV2Data ?? {
+    const data: HypaV2Data = room.hypaV2Data ?? {
-        chunks:[],
+        chunks: [],
-        mainChunks:[]
+        mainChunks: []
-    }
+    };
-    
-    //this is for the prompt
 
-    let allocatedTokens = db.hypaAllocatedTokens
+    let allocatedTokens = db.hypaAllocatedTokens;
-    let chunkSize = db.hypaChunkSize
+    let chunkSize = db.hypaChunkSize;
-    currentTokens += allocatedTokens
+    currentTokens += allocatedTokens;
-    currentTokens += 50 //this is for the template prompt
+    currentTokens += 50; // this is for the template prompt
-    let mainPrompt = ""
+    let mainPrompt = "";
 
-    while(data.mainChunks.length > 0){
+    while (data.mainChunks.length > 0) {
-        const chunk = data.mainChunks[0]
+        const chunk = data.mainChunks[0];
-        const ind = chats.findIndex(e => e.memo === chunk.targetId)
+        const ind = chats.findIndex(e => e.memo === chunk.targetId);
-        if(ind === -1){
+        if (ind === -1) {
-            data.mainChunks.shift()
+            data.mainChunks.shift();
-            continue
+            continue;
         }
 
-        const removedChats = chats.splice(0, ind)
+        const removedChats = chats.splice(0, ind);
-        for(const chat of removedChats){
+        for (const chat of removedChats) {
-            currentTokens -= await tokenizer.tokenizeChat(chat)
+            currentTokens -= await tokenizer.tokenizeChat(chat);
         }
-        chats = chats.slice(ind)
+        chats = chats.slice(ind);
-        mainPrompt = chunk.text
+        mainPrompt = chunk.text;
-        const mpToken = await tokenizer.tokenizeChat({role:'system', content:mainPrompt})
+        const mpToken = await tokenizer.tokenizeChat({ role: 'system', content: mainPrompt });
-        allocatedTokens -= mpToken
+        allocatedTokens -= mpToken;
-        break
+        break;
     }
 
-    while(currentTokens >= maxContextTokens){
+    while (currentTokens >= maxContextTokens) {
-    
+        let idx = 0;
-        let idx = 0
+        let targetId = '';
-        let targetId = ''
+        const halfData: OpenAIChat[] = [];
-        const halfData:OpenAIChat[] = []
 
-        let halfDataTokens = 0
+        let halfDataTokens = 0;
-        while(halfDataTokens < chunkSize){
+        while (halfDataTokens < chunkSize) {
-            const chat = chats[idx]
+            const chat = chats[idx];
-            if(!chat){
+            if (!chat) {
-                break
+                break;
             }
-            halfDataTokens += await tokenizer.tokenizeChat(chat)
+            halfDataTokens += await tokenizer.tokenizeChat(chat);
-            halfData.push(chat)
+            halfData.push(chat);
-            idx++
+            idx++;
-            targetId = chat.memo
+            targetId = chat.memo;
         }
 
-        async function summary(stringlizedChat:string):Promise<{
+        const stringlizedChat = halfData.map(e => `${e.role}: ${e.content}`).join('\n');
-            success:boolean
-            data:string
-        }>{
-            const promptbody:OpenAIChat[] = [
-                {
-                    role: "user",
-                    content: stringlizedChat
-                },
-                {
-                    role: "system",
-                    content: "Summarize this roleplay scene in a coherent narrative format for future reference. Summarize what happened, focusing on events and interactions between them. If someone or something is new or changed, include a brief characterization of them."
-                }
-            ]
-            const da = await requestChatData({
-                formated: promptbody,
-                bias: {},
-                useStreaming: false,
-                noMultiGen: true
-            }, 'model')
-            if(da.type === 'fail' || da.type === 'streaming' || da.type === 'multiline'){
-                return {
-                    data: "Hypamemory HTTP: " + da.result,
-                    success: false
-                }
-            }
-            return {
-                data: da.result,
-                success: true
-            }
-        }
 
-        const stringlizedChat = halfData.map(e => `${e.role}: ${e.content}`).join('\n')
+        const summaryData = await summarize(stringlizedChat);
 
-        const summaryData = await summary(stringlizedChat)
+        if (!summaryData.success) {
-
-        if(!summaryData.success){
             return {
                 currentTokens: currentTokens,
                 chats: chats,
                 error: summaryData.data
-            }
+            };
         }
 
-        const summaryDataToken = await tokenizer.tokenizeChat({role:'system', content:summaryData.data})
+        const summaryDataToken = await tokenizer.tokenizeChat({ role: 'system', content: summaryData.data });
-        mainPrompt += `\n\n${summaryData.data}`
+        mainPrompt += `\n\n${summaryData.data}`;
-        currentTokens -= halfDataTokens
+        currentTokens -= halfDataTokens;
-        allocatedTokens -= summaryDataToken
+        allocatedTokens -= summaryDataToken;
 
         data.mainChunks.unshift({
             text: mainPrompt,
             targetId: targetId
-        })
+        });
 
-        if(allocatedTokens < 1500){
+        if (allocatedTokens < 1500) {
-            const summarizedMp = await summary(mainPrompt)
+            const summarizedMp = await summarize(mainPrompt);
-            const mpToken = await tokenizer.tokenizeChat({role:'system', content:mainPrompt})
+            const mpToken = await tokenizer.tokenizeChat({ role: 'system', content: mainPrompt });
-            const summaryToken = await tokenizer.tokenizeChat({role:'system', content:summarizedMp.data})
+            const summaryToken = await tokenizer.tokenizeChat({ role: 'system', content: summarizedMp.data });
 
-            allocatedTokens -= summaryToken
+            allocatedTokens -= summaryToken;
-            allocatedTokens += mpToken
+            allocatedTokens += mpToken;
 
-            const splited = mainPrompt.split('\n\n').map(e => e.trim()).filter(e => e.length > 0)
+            const splited = mainPrompt.split('\n\n').map(e => e.trim()).filter(e => e.length > 0);
 
             data.chunks.push(...splited.map(e => ({
                 text: e,
                 targetId: targetId
-            })))
+            })));
 
-            data.mainChunks[0].text = mainPrompt
+            data.mainChunks[0].text = mainPrompt;
         }
     }
-    
+
-    const processer = new HypaProcesser("nomic")
+    const processer = new HypaProcesser(db.hypaModel);
 
     await processer.addText(data.chunks.filter(v => {
-        return v.text.trim().length > 0
+        return v.text.trim().length > 0;
     }).map((v) => {
-        return "search_document: " + v.text.trim()
+        return "search_document: " + v.text.trim();
-    }))
+    }));
 
-    let scoredResults:{[key:string]:number} = {}
+    let scoredResults: { [key: string]: number } = {};
-    for(let i=0;i<3;i++){
+    for (let i = 0; i < 3; i++) {
-        const pop = chats[chats.length - i - 1]
+        const pop = chats[chats.length - i - 1];
-        if(!pop){
+        if (!pop) {
-            break
+            break;
         }
-        const searched = await processer.similaritySearchScored(`search_query: ${pop.content}`)
+        const searched = await processer.similaritySearchScored(`search_query: ${pop.content}`);
-        for(const result of searched){
+        for (const result of searched) {
-            const score = result[1]/(i+1)
+            const score = result[1] / (i + 1);
-            if(scoredResults[result[0]]){
+            if (scoredResults[result[0]]) {
-                scoredResults[result[0]] += score
+                scoredResults[result[0]] += score;
-            }else{
+            } else {
-                scoredResults[result[0]] = score
+                scoredResults[result[0]] = score;
             }
         }
     }
 
-    const scoredArray = Object.entries(scoredResults).sort((a,b) => b[1] - a[1])
+    const scoredArray = Object.entries(scoredResults).sort((a, b) => b[1] - a[1]);
 
-    let chunkResultPrompts = ""
+    let chunkResultPrompts = "";
-    while(allocatedTokens > 0){
+    while (allocatedTokens > 0) {
-        const target = scoredArray.shift()
+        const target = scoredArray.shift();
-        if(!target){
+        if (!target) {
-            break
+            break;
         }
         const tokenized = await tokenizer.tokenizeChat({
             role: 'system',
             content: target[0].substring(14)
-        })
+        });
-        if(tokenized > allocatedTokens){
+        if (tokenized > allocatedTokens) {
-            break
+            break;
         }
-        chunkResultPrompts += target[0].substring(14) + '\n\n'
+        chunkResultPrompts += target[0].substring(14) + '\n\n';
-        allocatedTokens -= tokenized
+        allocatedTokens -= tokenized;
     }
 
-
+    const fullResult = `<Past Events Summary>${mainPrompt}</Past Events Summary>\n<Past Events Details>${chunkResultPrompts}</Past Events Details>`;
-    const fullResult = `<Past Events Summary>${mainPrompt}</Past Events Summary>\n<Past Events Details>${chunkResultPrompts}</Past Events Details>`
 
     chats.unshift({
         role: "system",
         content: fullResult,
         memo: "supaMemory"
-    })
+    });
     return {
         currentTokens: currentTokens,
         chats: chats,
         memory: data
-    }
+    };
-}
+}

src/ts/storage/database.ts

@@ -568,6 +568,7 @@ export interface Database{
     useAdditionalAssetsPreview:boolean,
     usePlainFetch:boolean
     hypaMemory:boolean
+    hypav2:boolean
     proxyRequestModel:string
     ooba:OobaSettings
     ainconfig: AINsettings