#!/usr/bin/env node
import {parseArgs} from 'node:util';
import {writeFileSync} from 'node:fs';
try {
 const {values:o}=parseArgs({options:{endpoint:{type:'string'},model:{type:'string'},hardware:{type:'string'},requests:{type:'string',default:'10'},warmup:{type:'string',default:'1'},prompt:{type:'string',default:'Explain the purpose of a benchmark in one sentence.'},out:{type:'string',default:'orbitriva-run.json'},name:{type:'string',default:'Inference benchmark'},'max-tokens':{type:'string',default:'128'},help:{type:'boolean'}}});
 if(o.help){console.log('Usage: node benchmark.mjs --endpoint http://localhost:8000/v1 --model MODEL --hardware "Your hardware" [--requests 10] [--warmup 1] [--prompt TEXT] [--max-tokens 128] [--out run.json]\nOptional authentication: set MODEL_API_KEY in your local environment. Never upload it.');process.exit(0);}
 if(!o.endpoint||!o.model||!o.hardware)throw new Error('--endpoint, --model and --hardware are required. Use --help.');
 const endpoint=new URL(o.endpoint);if(!['http:','https:'].includes(endpoint.protocol)||endpoint.username||endpoint.password||endpoint.search||endpoint.hash)throw new Error('Endpoint must be an HTTP(S) base URL without credentials, query or fragment.');
 const requests=Number(o.requests),warmup=Number(o.warmup),maxTokens=Number(o['max-tokens']);
 if(!Number.isInteger(requests)||requests<1||requests>1000||!Number.isInteger(warmup)||warmup<0||warmup>100||!Number.isInteger(maxTokens)||maxTokens<1||maxTokens>32768)throw new Error('Requests: 1–1000. Warmup: 0–100. Max tokens: 1–32768.');
 if(process.env.MODEL_API_KEY&&endpoint.protocol==='http:'&&!['localhost','127.0.0.1','[::1]'].includes(endpoint.hostname))throw new Error('Use HTTPS to send an API key to a remote endpoint.');
 const samples=[];const url=endpoint.href.replace(/\/$/,'')+'/chat/completions';
 for(let i=0;i<warmup+requests;i++){
  const started=performance.now();const response=await fetch(url,{method:'POST',redirect:'error',headers:{'Content-Type':'application/json',...(process.env.MODEL_API_KEY?{Authorization:`Bearer ${process.env.MODEL_API_KEY}`}:{})},body:JSON.stringify({model:o.model,messages:[{role:'user',content:o.prompt}],stream:false,max_tokens:maxTokens}),signal:AbortSignal.timeout(120000)});
  const result=await response.json();const latencyMs=Math.round((performance.now()-started)*100)/100;
  if(!response.ok||!Array.isArray(result.choices)||!result.choices.length)throw new Error(`Request ${i+1} failed (HTTP ${response.status}). No run saved.`);
  if(i>=warmup){const outputTokens=result.usage?.completion_tokens;samples.push({latencyMs,...(Number.isSafeInteger(outputTokens)&&outputTokens>=0?{outputTokens}:{})});}
  console.error(`${i<warmup?'Warmup':'Measured'} ${i+1}/${warmup+requests}: ${latencyMs} ms`);
 }
 const result={name:o.name,model:o.model,hardware:o.hardware,source:'cli',notes:`Sequential, non-streaming HTTP benchmark; ${warmup} warmup request(s) excluded; max_tokens=${maxTokens}. Latency includes network and JSON parsing. Hardware is user-declared. Prompt and endpoint are not stored.`,samples};
 writeFileSync(o.out,JSON.stringify(result,null,2),{mode:0o600});console.log(`Saved ${samples.length} measured samples to ${o.out}. Import this JSON in Orbitriva.`);
}catch(e){console.error(e.message);process.exitCode=1;}