morgan
/
chatgpt-web
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

fix prompt continuation for petals

This commit is contained in:
Webifi 2023-08-19 19:16:18 -05:00
parent 54aac40788
commit 3b72f73080
2 changed files with 8 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
src/lib/providers/petals/models.svelte
View File

@ -45,7 +45,7 @@ const chatModelBase = {
return prompts.reduce((a, m) => { return prompts.reduce((a, m) => {
a += countMessageTokens(m, model, chat) a += countMessageTokens(m, model, chat)
return a return a
}, 0) + countTokens(model, getStartSequence(chat)) + ((prompts[prompts.length - 1] || {}).role !== 'assistant' ? countTokens(model, getLeadPrompt(chat)) : 0) }, 0) + countTokens(model, getStartSequence(chat)) + countTokens(model, getLeadPrompt(chat))
} }
} as ModelDetail } as ModelDetail

8
src/lib/providers/petals/request.svelte
View File

@ -62,6 +62,12 @@ export const chatRequest = async (
const buildMessage = (m: Message): string => { const buildMessage = (m: Message): string => {
return getRoleTag(m.role, model, chat) + m.content + getRoleEnd(m.role, model, chat) return getRoleTag(m.role, model, chat) + m.content + getRoleEnd(m.role, model, chat)
} }
const lastMessage = rMessages[rMessages.length - 1]
let doLead = true
if (lastMessage && lastMessage.role === 'assistant') {
lastMessage.content = leadPromptSequence + lastMessage.content
doLead = false
}
const inputArray = rMessages.reduce((a, m, i) => { const inputArray = rMessages.reduce((a, m, i) => {
let c = buildMessage(m) let c = buildMessage(m)
let replace = false let replace = false
@ -96,7 +102,7 @@ export const chatRequest = async (
} }
return a return a
}, [] as Message[]) }, [] as Message[])
const leadPrompt = (leadPromptSequence && ((inputArray[inputArray.length - 1] || {}) as Message).role !== 'assistant') ? delimiter + leadPromptSequence : '' const leadPrompt = (leadPromptSequence && doLead) ? delimiter + leadPromptSequence : ''
const fullPromptInput = getStartSequence(chat) + inputArray.map(m => m.content).join(delimiter) + leadPrompt const fullPromptInput = getStartSequence(chat) + inputArray.map(m => m.content).join(delimiter) + leadPrompt
let maxLen = Math.min(opts.maxTokens || chatSettings.max_tokens || maxTokens, maxTokens) let maxLen = Math.min(opts.maxTokens || chatSettings.max_tokens || maxTokens, maxTokens)