Andreas/LibreChat
1
0
Fork 0
mirror of https://github.com/danny-avila/LibreChat.git synced 2025-12-17 08:50:15 +01:00
Code Issues Projects Releases Packages Wiki Activity Actions 7

🚀 feat: Assistants Streaming (#2159)

Browse source 
* chore: bump openai to 4.29.0 and npm audit fix

* chore: remove unnecessary stream field from ContentData

* feat: new enum and types for AssistantStreamEvent

* refactor(AssistantService): remove stream field and add conversationId to text ContentData
> - return `finalMessage` and `text` on run completion
> - move `processMessages` to services/Threads to avoid circular dependencies with new stream handling
> - refactor(processMessages/retrieveAndProcessFile): add new `client` field to differentiate new RunClient type

* WIP: new assistants stream handling

* chore: stores messages to StreamRunManager

* chore: add additional typedefs

* fix: pass req and openai to StreamRunManager

* fix(AssistantService): pass openai as client to `retrieveAndProcessFile`

* WIP: streaming tool i/o, handle in_progress and completed run steps

* feat(assistants): process required actions with streaming enabled

* chore: condense early return check for useSSE useEffect

* chore: remove unnecessary comments and only handle completed tool calls when not function

* feat: add TTL for assistants run abort cacheKey

* feat: abort stream runs

* fix(assistants): render streaming cursor

* fix(assistants): hide edit icon as functionality is not supported

* fix(textArea): handle pasting edge cases; first, when onChange events wouldn't fire; second, when textarea wouldn't resize

* chore: memoize Conversations

* chore(useTextarea): reverse args order

* fix: load default capabilities when an azure is configured to support assistants, but `assistants` endpoint is not configured

* fix(AssistantSelect): update form assistant model on assistant form select

* fix(actions): handle azure strict validation for function names to fix crud for actions

* chore: remove content data debug log as it fires in rapid succession

* feat: improve UX for assistant errors mid-request

* feat: add tool call localizations and replace any domain separators from azure action names

* refactor(chat): error out tool calls without outputs during handleError

* fix(ToolService): handle domain separators allowing Azure use of actions

* refactor(StreamRunManager): types and throw Error if tool submission fails
This commit is contained in:
Danny Avila 2024-03-21 22:42:25 -04:00 • committed by GitHub
parent ed64c76053
commit f427ad792a
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
39 changed files with 1503 additions and 330 deletions
Download patch file Download diff file Expand all files Collapse all files

5
api/server/middleware/abortRun.js
View file

@ -4,9 +4,10 @@ const { checkMessageGaps, recordUsage } = require('~/server/services/Threads');
const { getConvo } = require('~/models/Conversation');
const getLogStores = require('~/cache/getLogStores');
const { sendMessage } = require('~/server/utils');
// const spendTokens = require('~/models/spendTokens');
const { logger } = require('~/config');
const three_minutes = 1000 * 60 * 3;
async function abortRun(req, res) {
res.setHeader('Content-Type', 'application/json');
const { abortKey } = req.body;
@ -40,7 +41,7 @@ async function abortRun(req, res) {
const { openai } = await initializeClient({ req, res });
try {
await cache.set(cacheKey, 'cancelled');
await cache.set(cacheKey, 'cancelled', three_minutes);
const cancelledRun = await openai.beta.threads.runs.cancel(thread_id, run_id);
logger.debug('[abortRun] Cancelled run:', cancelledRun);
} catch (error) {