62 lines
1.9 KiB
JavaScript
62 lines
1.9 KiB
JavaScript
import { createRequire } from 'node:module';
|
|
import path from 'node:path';
|
|
|
|
const harnessPath = process.env.JARVIS_HARNESS_PATH || path.resolve(new URL('../vendor/agent-harness', import.meta.url).pathname);
|
|
const require = createRequire(import.meta.url);
|
|
process.env.QVAC_CONFIG_PATH ||= path.resolve(new URL('../qvac.config.json', import.meta.url).pathname);
|
|
const Agent = require(path.join(harnessPath, 'index.js'));
|
|
|
|
let loadPromise = null;
|
|
let ownerCount = 0;
|
|
|
|
export const QVAC_MASTER = Object.freeze({
|
|
configPath: process.env.QVAC_CONFIG_PATH,
|
|
model: process.env.JARVIS_QVAC_MODEL || 'qwen3.5-4b',
|
|
device: 'gpu',
|
|
gpuLayers: 99,
|
|
});
|
|
|
|
export async function acquireQvac() {
|
|
ownerCount += 1;
|
|
if (!loadPromise) {
|
|
const resources = await Agent.engine.resources();
|
|
const gpuVisible = (resources.gpus?.length || 0) > 0 ||
|
|
Boolean(resources.drivers?.vulkan || resources.drivers?.cuda || resources.drivers?.opencl || resources.gpu);
|
|
if (!gpuVisible) {
|
|
ownerCount = Math.max(0, ownerCount - 1);
|
|
throw new Error('Jarvis requires a QVAC-visible GPU backend; run npm run gpu-doctor');
|
|
}
|
|
loadPromise = Agent.engine.load({
|
|
model: QVAC_MASTER.model,
|
|
tools: true,
|
|
device: 'gpu',
|
|
gpu_layers: QVAC_MASTER.gpuLayers,
|
|
mmprojUseGpu: true,
|
|
}).then((loaded) => {
|
|
if (loaded.device !== 'gpu') {
|
|
throw new Error(`Jarvis requires GPU QVAC inference; loaded device was ${loaded.device}`);
|
|
}
|
|
return loaded;
|
|
}).catch((error) => {
|
|
loadPromise = null;
|
|
ownerCount = Math.max(0, ownerCount - 1);
|
|
throw error;
|
|
});
|
|
}
|
|
return loadPromise;
|
|
}
|
|
|
|
export function releaseQvac() { ownerCount = Math.max(0, ownerCount - 1); }
|
|
|
|
export async function closeQvac() {
|
|
if (ownerCount > 0) return;
|
|
loadPromise = null;
|
|
await Agent.engine.close();
|
|
}
|
|
|
|
export function qvacStatus() {
|
|
return { ...QVAC_MASTER, owners: ownerCount, loaded: Agent.engine.getLoaded() };
|
|
}
|
|
|
|
export { Agent };
|