Skip to content

fix: in oai chat completion, use msg name field instead of prepended to msg content #754

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Closed
stevenksmith wants to merge 4 commits into from
Closed

fix: in oai chat completion, use msg name field instead of prepended to msg content #754

Changes from 1 commit
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
da682f4
fix: in oai chat completion, use msg name field instead of prepended …
Dec 13, 2023
88a89c4
refactor for readability
Dec 13, 2023
c4eefd5
fix: determine correct author name, fixing past impersonations & exam…
Dec 13, 2023
18ee08a
fix: make OpenAI name field follow API constraint
Dec 16, 2023
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
33 changes: 22 additions & 11 deletions srv/adapter/chat-completion.ts
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -249,15 +249,23 @@ export async function toChatCompletionPayload(

const line = all[i]

const obj: CompletionItem = {
role: 'assistant',
content: line.trim().replace(BOT_REPLACE, replyAs.name).replace(SELF_REPLACE, handle),
}

const isSystem = line.startsWith('System:')
const isUser = line.startsWith(handle)
const isBot = !isUser && !isSystem

const name = isBot ? replyAs.name : isUser ? handle : undefined

@khanonnie khanonnie Dec 13, 2023

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Is handle by this point accounting for impersonated characters? I'd assume so

@stevenksmith stevenksmith Dec 13, 2023

Copy link
Copy Markdown
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Good catch, this was breaking past impersonations (registering them as the main character in assistant message), and example messages (incorrect splitting). This is fixed in c4eefd5


const content = line
.trim()
.replace(BOT_REPLACE, replyAs.name)
.replace(SELF_REPLACE, handle)
.replace(name ? `${name}: ` : '', '')

const obj: CompletionItem = {
role: 'assistant',
content,
name,
}
const insert = inserts.get(distanceFromBottom)
if (insert) history.push({ role: 'system', content: insert })

Expand Down Expand Up @@ -291,7 +299,7 @@ export async function toChatCompletionPayload(
obj.role = 'user'
}

const length = await encoder()(obj.content)
const length = (await encoder()(obj.content)) + (obj.name ? 1 + (await encoder()(obj.name)) : 0)
if (tokens + length > maxBudget) {
--i
break
Expand Down Expand Up @@ -339,12 +347,15 @@ export async function splitSampleChat(opts: SplitSampleChatProps) {
? 'user'
: 'system'

const msg: CompletionItem = {
role: role,
content: sample.replace(BOT_REPLACE, char).replace(SELF_REPLACE, sender),
}
const name = role === 'assistant' ? char : role === 'user' ? sender : undefined

const content = sample
.replace(BOT_REPLACE, char)
.replace(SELF_REPLACE, sender)
.replace(name ? `${name}: ` : '', '')

const length = await encoder()(msg.content)
const msg: CompletionItem = { role, content, name }
const length = (await encoder()(msg.content)) + (msg.name ? 1 + (await encoder()(msg.name)) : 0)
if (budget && tokens + length > budget) break

additions.push(msg)
Expand Down