Function planOffload

src/setup/offload-detect.ts:96–103

(opts: { baseUrl: string; model: string; vramBudgetGB?: number })

/** Gather hardware + model facts and compute an offload plan. Returns null when anything needed
 *  is unavailable (so the caller can say "couldn't auto-detect — set num_gpu manually"). */
export async function planOffload(opts: { baseUrl: string; model: string; vramBudgetGB?: number }): Promise<OffloadSuggestion | null> {
  const vramGB = opts.vramBudgetGB ?? detectVramGB();
  if (!vramGB) return null;
  const facts = await fetchOllamaModelFacts(opts.baseUrl, opts.model);
  if (!facts) return null;
  const plan = suggestGpuLayers({ modelSizeGB: facts.modelSizeGB, vramBudgetGB: vramGB, totalLayers: facts.totalLayers });
  return { plan, facts, vramGB };
}