dea29ff8c4
* fix: update logic to pull llama-cpp engine * chore: bump cortex version * fix: chat macos * chore: bump llama.cpp * chore: download windows * chore: download linux * chore: bump cortex to 1.0.13-rc6 * fix: stop cortex before exit * chore: bum llama.cpp engine to b5350 --------- Co-authored-by: vansangpfiev <sang@jan.ai>
104 lines
2.9 KiB
TypeScript
104 lines
2.9 KiB
TypeScript
import { GpuSetting, log } from '@janhq/core'
|
|
|
|
// Supported run modes
|
|
enum RunMode {
|
|
Cuda = 'cuda',
|
|
CPU = 'cpu',
|
|
}
|
|
|
|
// Supported instruction sets
|
|
const instructionBinaryNames = ['noavx', 'avx', 'avx2', 'avx512']
|
|
|
|
/**
|
|
* The GPU runMode that will be set - either 'vulkan', 'cuda', or empty for cpu.
|
|
* @param settings
|
|
* @returns
|
|
*/
|
|
|
|
const gpuRunMode = (settings?: GpuSetting): RunMode => {
|
|
return settings.gpus?.some(
|
|
(gpu) =>
|
|
gpu.activated &&
|
|
gpu.additional_information &&
|
|
gpu.additional_information.driver_version
|
|
)
|
|
? RunMode.Cuda
|
|
: RunMode.CPU
|
|
}
|
|
|
|
/**
|
|
* The OS & architecture that the current process is running on.
|
|
* @returns win, mac-x64, mac-arm64, or linux
|
|
*/
|
|
const os = (settings?: GpuSetting): string => {
|
|
return PLATFORM === 'win32'
|
|
? 'win'
|
|
: PLATFORM === 'darwin'
|
|
? settings?.cpu?.arch === 'arm64'
|
|
? 'macos-arm64'
|
|
: 'macos-x64'
|
|
: 'linux'
|
|
}
|
|
|
|
/**
|
|
* The CUDA version that will be set - either 'cu12.0' or 'cu11.7'.
|
|
* @param settings
|
|
* @returns
|
|
*/
|
|
const cudaVersion = (settings?: GpuSetting): 'cu12.0' | 'cu11.7' | undefined => {
|
|
return settings.gpus?.some((gpu) => gpu.version.includes('12'))
|
|
? 'cu12.0'
|
|
: 'cu11.7'
|
|
}
|
|
|
|
/**
|
|
* The CPU instructions that will be set - either 'avx512', 'avx2', 'avx', or 'noavx'.
|
|
* @returns
|
|
*/
|
|
|
|
/**
|
|
* Find which variant to run based on the current platform.
|
|
*/
|
|
export const engineVariant = async (
|
|
gpuSetting?: GpuSetting
|
|
): Promise<string> => {
|
|
const platform = os(gpuSetting)
|
|
|
|
// There is no need to append the variant extension for mac
|
|
if (platform.startsWith('mac')) return platform
|
|
|
|
const runMode = gpuRunMode(gpuSetting)
|
|
// Only Nvidia GPUs have addition_information set and activated by default
|
|
let engineVariant =
|
|
!gpuSetting?.vulkan ||
|
|
!gpuSetting.gpus?.length ||
|
|
gpuSetting.gpus.some((e) => e.additional_information && e.activated)
|
|
? [
|
|
platform,
|
|
...(runMode === RunMode.Cuda
|
|
? // For cuda we only need to check if the cpu supports avx2 or noavx - since other binaries are not shipped with the extension
|
|
[
|
|
gpuSetting.cpu?.instructions.includes('avx2') ||
|
|
gpuSetting.cpu?.instructions.includes('avx512')
|
|
? 'avx2'
|
|
: 'noavx',
|
|
runMode,
|
|
cudaVersion(gpuSetting),
|
|
'x64',
|
|
]
|
|
: // For cpu only we need to check all available supported instructions
|
|
[
|
|
(gpuSetting.cpu?.instructions ?? ['noavx']).find((e) =>
|
|
instructionBinaryNames.includes(e.toLowerCase())
|
|
) ?? 'noavx',
|
|
'x64',
|
|
]),
|
|
].filter(Boolean)
|
|
: [platform, 'vulkan', 'x64']
|
|
|
|
let engineVariantString = engineVariant.join('-')
|
|
|
|
log(`[CORTEX]: Engine variant: ${engineVariantString}`)
|
|
return engineVariantString
|
|
}
|