Evandro Fonseca Junior
a3f8226fda
Updated README.md from a detailed OllaHub project description to the standard Next.js getting started template, likely to initialize or reset the project as a Next.js application.
101 lines
2.2 KiB
TypeScript
101 lines
2.2 KiB
TypeScript
export type Device = "CPU" | "GPU";
|
|
|
|
export type ModelInfo = {
|
|
id: string;
|
|
name: string;
|
|
sizeGB: number;
|
|
quantization: string;
|
|
device: Device;
|
|
estCpuUsagePct?: number;
|
|
estVramGB?: number;
|
|
};
|
|
|
|
export const mockModels: ModelInfo[] = [
|
|
{
|
|
id: "llama3.1-8b-q4",
|
|
name: "llama3.1:8b-q4",
|
|
sizeGB: 4.8,
|
|
quantization: "Q4",
|
|
device: "CPU",
|
|
estCpuUsagePct: 85,
|
|
estVramGB: 0.0,
|
|
},
|
|
{
|
|
id: "llama3.1-8b-q8",
|
|
name: "llama3.1:8b-q8",
|
|
sizeGB: 8.9,
|
|
quantization: "Q8",
|
|
device: "GPU",
|
|
estCpuUsagePct: 20,
|
|
estVramGB: 8,
|
|
},
|
|
{
|
|
id: "mistral-7b-q4",
|
|
name: "mistral:7b-q4",
|
|
sizeGB: 4.1,
|
|
quantization: "Q4",
|
|
device: "CPU",
|
|
estCpuUsagePct: 75,
|
|
estVramGB: 0,
|
|
},
|
|
{
|
|
id: "mistral-7b-q8",
|
|
name: "mistral:7b-q8",
|
|
sizeGB: 7.9,
|
|
quantization: "Q8",
|
|
device: "GPU",
|
|
estCpuUsagePct: 18,
|
|
estVramGB: 8,
|
|
},
|
|
{
|
|
id: "qwen2.5-7b-q4",
|
|
name: "qwen2.5:7b-q4",
|
|
sizeGB: 4.2,
|
|
quantization: "Q4",
|
|
device: "CPU",
|
|
estCpuUsagePct: 80,
|
|
estVramGB: 0,
|
|
},
|
|
{
|
|
id: "qwen2.5-7b-q8",
|
|
name: "qwen2.5:7b-q8",
|
|
sizeGB: 8.1,
|
|
quantization: "Q8",
|
|
device: "GPU",
|
|
estCpuUsagePct: 22,
|
|
estVramGB: 8,
|
|
},
|
|
];
|
|
|
|
export function inferQuantization(nameOrDetails: string | unknown): string {
|
|
const s =
|
|
typeof nameOrDetails === "string" ? nameOrDetails.toLowerCase() : "";
|
|
if (s.includes("q8")) return "Q8";
|
|
if (s.includes("q5")) return "Q5";
|
|
if (s.includes("q4")) return "Q4";
|
|
return "Q?";
|
|
}
|
|
|
|
export function inferDevice(details?: unknown): Device {
|
|
const s = typeof details === "string" ? details.toLowerCase() : "";
|
|
if (s.includes("cuda") || s.includes("metal") || s.includes("gpu"))
|
|
return "GPU";
|
|
return "CPU";
|
|
}
|
|
|
|
export function estimateCpuVram(
|
|
info: Pick<ModelInfo, "sizeGB" | "quantization" | "device">,
|
|
) {
|
|
const q = info.quantization.toUpperCase();
|
|
let cpuPct = 50;
|
|
let vramGB = 0;
|
|
if (info.device === "GPU") {
|
|
vramGB = Math.max(4, Math.round(info.sizeGB));
|
|
cpuPct = 20;
|
|
} else {
|
|
// heurística simples: quantizações mais baixas usam mais CPU
|
|
cpuPct = q === "Q8" ? 60 : q === "Q5" ? 70 : 85;
|
|
}
|
|
return { cpuPct, vramGB };
|
|
}
|