* feat: tensorrt-llm-extension * fix: loading * feat: add download tensorrt llm runner Signed-off-by: James <james@jan.ai> * feat: update to rollupjs instead of webpack for monitoring extension Signed-off-by: James <james@jan.ai> * feat: move update nvidia info to monitor extension Signed-off-by: James <james@jan.ai> * allow download tensorrt Signed-off-by: James <james@jan.ai> * update Signed-off-by: James <james@jan.ai> * allow download tensor rt based on gpu setting Signed-off-by: James <james@jan.ai> * update downloaded models Signed-off-by: James <james@jan.ai> * feat: add extension compatibility * dynamic tensor rt engines Signed-off-by: James <james@jan.ai> * update models Signed-off-by: James <james@jan.ai> * chore: remove ts-ignore * feat: getting installation state from extension Signed-off-by: James <james@jan.ai> * chore: adding type for decompress Signed-off-by: James <james@jan.ai> * feat: update according Louis's comment Signed-off-by: James <james@jan.ai> * feat: add progress for installing extension Signed-off-by: James <james@jan.ai> * chore: remove args from extension installation * fix: model download does not work properly * fix: do not allow user to stop tensorrtllm inference * fix: extension installed style * fix: download tensorrt does not update state Signed-off-by: James <james@jan.ai> * chore: replace int4 by fl16 * feat: modal for installing extension Signed-off-by: James <james@jan.ai> * fix: start download immediately after press install Signed-off-by: James <james@jan.ai> * fix: error switching between engines * feat: rename inference provider to ai engine and refactor to core * fix: missing ulid * fix: core bundler * feat: add cancel extension installing Signed-off-by: James <james@jan.ai> * remove mocking for mac Signed-off-by: James <james@jan.ai> * fix: show models only when extension is ready * add tensorrt badge for model Signed-off-by: James <james@jan.ai> * fix: copy * fix: add compatible check (#2342) * fix: add compatible check Signed-off-by: James <james@jan.ai> * fix: copy * fix: font * fix: copy * fix: broken monitoring extension * chore: bump engine * fix: copy * fix: model copy * fix: copy * fix: model json --------- Signed-off-by: James <james@jan.ai> Co-authored-by: James <james@jan.ai> Co-authored-by: Louis <louis@jan.ai> * fix: vulkan support * fix: installation button padding * fix: empty script * fix: remove hard code string --------- Signed-off-by: James <james@jan.ai> Co-authored-by: James <james@jan.ai> Co-authored-by: NamH <NamNh0122@gmail.com>
99 lines
2.8 KiB
TypeScript
99 lines
2.8 KiB
TypeScript
import { useEffect, useRef } from 'react'
|
|
|
|
import { events, Model, ModelEvent } from '@janhq/core'
|
|
import { atom, useAtom, useAtomValue, useSetAtom } from 'jotai'
|
|
|
|
import { toaster } from '@/containers/Toast'
|
|
|
|
import { LAST_USED_MODEL_ID } from './useRecommendedModel'
|
|
|
|
import { downloadedModelsAtom } from '@/helpers/atoms/Model.atom'
|
|
import { activeThreadAtom } from '@/helpers/atoms/Thread.atom'
|
|
|
|
export const activeModelAtom = atom<Model | undefined>(undefined)
|
|
export const loadModelErrorAtom = atom<string | undefined>(undefined)
|
|
|
|
export const stateModelAtom = atom({
|
|
state: 'start',
|
|
loading: false,
|
|
model: '',
|
|
})
|
|
|
|
export function useActiveModel() {
|
|
const [activeModel, setActiveModel] = useAtom(activeModelAtom)
|
|
const activeThread = useAtomValue(activeThreadAtom)
|
|
const [stateModel, setStateModel] = useAtom(stateModelAtom)
|
|
const downloadedModels = useAtomValue(downloadedModelsAtom)
|
|
const setLoadModelError = useSetAtom(loadModelErrorAtom)
|
|
|
|
const downloadedModelsRef = useRef<Model[]>([])
|
|
|
|
useEffect(() => {
|
|
downloadedModelsRef.current = downloadedModels
|
|
}, [downloadedModels])
|
|
|
|
const startModel = async (modelId: string) => {
|
|
if (
|
|
(activeModel && activeModel.id === modelId) ||
|
|
(stateModel.model === modelId && stateModel.loading)
|
|
) {
|
|
console.debug(`Model ${modelId} is already initialized. Ignore..`)
|
|
return
|
|
}
|
|
|
|
let model = downloadedModelsRef?.current.find((e) => e.id === modelId)
|
|
|
|
// Switch between engines
|
|
if (model && activeModel && activeModel.engine !== model.engine) {
|
|
stopModel()
|
|
// TODO: Refactor inference provider would address this
|
|
await new Promise((res) => setTimeout(res, 1000))
|
|
}
|
|
|
|
// TODO: incase we have multiple assistants, the configuration will be from assistant
|
|
setLoadModelError(undefined)
|
|
|
|
setActiveModel(undefined)
|
|
|
|
setStateModel({ state: 'start', loading: true, model: modelId })
|
|
|
|
if (!model) {
|
|
toaster({
|
|
title: `Model ${modelId} not found!`,
|
|
description: `Please download the model first.`,
|
|
type: 'warning',
|
|
})
|
|
setStateModel(() => ({
|
|
state: 'start',
|
|
loading: false,
|
|
model: '',
|
|
}))
|
|
return
|
|
}
|
|
|
|
/// Apply thread model settings
|
|
if (activeThread?.assistants[0]?.model.id === modelId) {
|
|
model = {
|
|
...model,
|
|
settings: {
|
|
...model.settings,
|
|
...activeThread.assistants[0].model.settings,
|
|
},
|
|
}
|
|
}
|
|
|
|
localStorage.setItem(LAST_USED_MODEL_ID, model.id)
|
|
events.emit(ModelEvent.OnModelInit, model)
|
|
}
|
|
|
|
const stopModel = async () => {
|
|
if (activeModel) {
|
|
setActiveModel(undefined)
|
|
setStateModel({ state: 'stop', loading: true, model: activeModel.id })
|
|
events.emit(ModelEvent.OnModelStop, activeModel)
|
|
}
|
|
}
|
|
|
|
return { activeModel, startModel, stopModel, stateModel }
|
|
}
|