Fix issues with token counts

This commit is contained in:
Webifi 2023-06-08 15:09:14 -05:00
parent 0cc30a28dc
commit 567845be06
4 changed files with 55 additions and 26 deletions

View File

@ -22,8 +22,9 @@ html {
--chatContentPaddingTop: 20px;
--chatContentPaddingRight: 40px;
--chatContentPaddingBottom: 130px;
--chatContentPaddingBottom: 110px;
--chatContentPaddingLeft: 40px;
--runningTotalLineHeight: 28px;
--chatInputPaddingTop: 0px;
--chatInputPaddingRight: 60px;
@ -31,6 +32,7 @@ html {
--chatInputPaddingLeft: 60px;
--BgColor: var(-BgColorLight);
--running-totals: 0;
}
.navbar {
@ -371,7 +373,7 @@ aside.menu.main-menu .menu-expanse {
display: block;
position: fixed;
bottom: 0px;
height: var(--chatContentPaddingBottom) ;
height: calc(var(--chatContentPaddingBottom) + var(--runningTotalLineHeight) * var(--running-totals));
width: 100%;
background-image: linear-gradient(180deg,hsla(0,0%,100%,0) 13.94%, var(--BgColorLight) 54.73%);
}
@ -419,7 +421,7 @@ aside.menu.main-menu .menu-expanse {
padding:
var(--chatContentPaddingTop)
var(--chatContentPaddingRight)
var(--chatContentPaddingBottom)
calc(var(--chatContentPaddingBottom) + var(--runningTotalLineHeight) * var(--running-totals))
var(--chatContentPaddingLeft) ;
}

View File

@ -370,7 +370,7 @@
request.stream = opts.streaming
chatResponse.setPromptTokenCount(promptTokenCount)
chatResponse.setPromptTokenCount(promptTokenCount) // streaming needs this
const signal = controller.signal
@ -589,7 +589,7 @@
</script>
{#if chat}
<ChatSettingsModal chatId={chatId} bind:show={showSettingsModal} />
<div class="chat-page" style="--running-totals: {Object.entries(chat.usage || {}).length}">
<div class="chat-content">
<nav class="level chat-header">
<div class="level-left">
@ -676,4 +676,5 @@
{/each}
</div>
</Footer>
</div>
{/if}

View File

@ -1,7 +1,7 @@
<script context="module" lang="ts">
// TODO: Integrate API calls
import { addMessage, saveChatStore, subtractRunningTotal, updateRunningTotal } from './Storage.svelte'
import type { Chat, ChatCompletionOpts, Message, Response, Usage } from './Types.svelte'
import { addMessage, getLatestKnownModel, saveChatStore, setLatestKnownModel, subtractRunningTotal, updateRunningTotal } from './Storage.svelte'
import type { Chat, ChatCompletionOpts, Message, Model, Response, Usage } from './Types.svelte'
import { encode } from 'gpt-tokenizer'
import { v4 as uuidv4 } from 'uuid'
@ -20,6 +20,7 @@ export class ChatCompletionResponse {
private offsetTotals: Usage
private isFill: boolean = false
private didFill: boolean = false
private opts: ChatCompletionOpts
private chat: Chat
@ -28,6 +29,16 @@ export class ChatCompletionResponse {
private error: string
private model: Model
private lastModel: Model
private setModel = (model: Model) => {
if (!model) return
!this.model && setLatestKnownModel(this.chat.settings.model as Model, model)
this.lastModel = this.model || model
this.model = model
}
private finishResolver: (value: Message[]) => void
private errorResolver: (error: string) => void
private finishPromise = new Promise<Message[]>((resolve, reject) => {
@ -49,11 +60,11 @@ export class ChatCompletionResponse {
const exitingMessage = this.messages[i]
const message = exitingMessage || choice.message
if (exitingMessage) {
if (this.isFill && choice.message.content.match(/^'(t|ll|ve|m|d|re)[^a-z]/i)) {
if (!this.didFill && this.isFill && choice.message.content.match(/^'(t|ll|ve|m|d|re)[^a-z]/i)) {
// deal with merging contractions since we've added an extra space to your fill message
message.content.replace(/ $/, '')
}
this.isFill = false
this.didFill = true
message.content += choice.message.content
message.usage = message.usage || {
prompt_tokens: 0,
@ -61,8 +72,8 @@ export class ChatCompletionResponse {
total_tokens: 0
} as Usage
message.usage.completion_tokens += response.usage.completion_tokens
message.usage.prompt_tokens += response.usage.prompt_tokens
message.usage.total_tokens += response.usage.total_tokens
message.usage.prompt_tokens = response.usage.prompt_tokens + (this.offsetTotals?.prompt_tokens || 0)
message.usage.total_tokens += response.usage.total_tokens + (this.offsetTotals?.total_tokens || 0)
} else {
message.content = choice.message.content
message.usage = response.usage
@ -70,6 +81,7 @@ export class ChatCompletionResponse {
message.finish_reason = choice.finish_reason
message.role = choice.message.role
message.model = response.model
this.setModel(response.model)
this.messages[i] = message
if (this.opts.autoAddMessages) addMessage(this.chat.id, message)
})
@ -87,30 +99,30 @@ export class ChatCompletionResponse {
} as Message
choice.delta?.role && (message.role = choice.delta.role)
if (choice.delta?.content) {
if (this.isFill && choice.delta.content.match(/^'(t|ll|ve|m|d|re)[^a-z]/i)) {
if (!this.didFill && this.isFill && choice.delta.content.match(/^'(t|ll|ve|m|d|re)[^a-z]/i)) {
// deal with merging contractions since we've added an extra space to your fill message
message.content.replace(/([a-z]) $/i, '$1')
}
this.isFill = false
this.didFill = true
message.content += choice.delta.content
}
completionTokenCount += encode(message.content).length
message.usage = response.usage || {
prompt_tokens: this.promptTokenCount
} as Usage
message.model = response.model
this.setModel(response.model)
message.finish_reason = choice.finish_reason
message.streaming = choice.finish_reason === null && !this.finished
this.messages[i] = message
if (this.opts.autoAddMessages) addMessage(this.chat.id, message)
})
// total up the tokens
const totalTokens = this.promptTokenCount + completionTokenCount
const promptTokens = this.promptTokenCount + (this.offsetTotals?.prompt_tokens || 0)
const totalTokens = promptTokens + completionTokenCount
this.messages.forEach(m => {
if (m.usage) {
m.usage.completion_tokens = completionTokenCount
m.usage.total_tokens = totalTokens
}
m.usage = {
completion_tokens: completionTokenCount,
total_tokens: totalTokens,
prompt_tokens: promptTokens
} as Usage
if (this.opts.autoAddMessages) addMessage(this.chat.id, m)
})
const finished = !this.messages.find(m => m.streaming)
this.notifyMessageChange()
@ -167,12 +179,13 @@ export class ChatCompletionResponse {
this.messages.forEach(m => { m.streaming = false }) // make sure all are marked stopped
saveChatStore()
const message = this.messages[0]
const model = this.model || getLatestKnownModel(this.chat.settings.model as Model)
if (message) {
if (this.offsetTotals) {
if (this.isFill && this.lastModel === this.model && this.offsetTotals && model && message.usage) {
// Need to subtract some previous message totals before we add new combined message totals
subtractRunningTotal(this.chat.id, this.offsetTotals, this.chat.settings.model as any)
subtractRunningTotal(this.chat.id, this.offsetTotals, model)
}
updateRunningTotal(this.chat.id, message.usage as any, message.model as any)
updateRunningTotal(this.chat.id, message.usage as Usage, model)
} else {
// If no messages it's probably because of an error or user initiated abort.
// We could miss counting the cost of the prompts sent.
@ -186,7 +199,7 @@ export class ChatCompletionResponse {
completion_tokens: 0, // We have no idea if there are any to count
total_tokens: this.promptTokenCount
}
updateRunningTotal(this.chat.id, usage as any, this.chat.settings.model as any)
updateRunningTotal(this.chat.id, usage as Usage, model)
}
this.notifyFinish()
if (this.error) {

View File

@ -8,6 +8,7 @@
import { errorNotice } from './Util.svelte'
export const chatsStorage = persisted('chats', [] as Chat[])
export const latestModelMap = persisted('latestModelMap', {} as Record<Model, Model>) // What was returned when a model was requested
export const globalStorage = persisted('global', {} as GlobalSettings)
export const apiKeyStorage = persisted('apiKey', '' as string)
export let checkStateChange = writable(0) // Trigger for Chat
@ -462,5 +463,17 @@
}
return cname
}
export const getLatestKnownModel = (model:Model) => {
const modelMapStore = get(latestModelMap)
return modelMapStore[model] || model
}
export const setLatestKnownModel = (requestedModel:Model, responseModel:Model) => {
const modelMapStore = get(latestModelMap)
modelMapStore[requestedModel] = responseModel
latestModelMap.set(modelMapStore)
}
</script>