grix-connector/openclaw-plugin/skills/openclaw-memory-setup/scripts/bench_ollama_embeddings.js

Connect local AI coding agents (Claude, Codex, Gemini, Qwen, DeepSeek, Cursor, OpenCode, Pi, OpenHuman, Reasonix) to the Grix scheduling platform. Also serves as an OpenClaw plugin for Grix channel transport.

#!/usr/bin/env node
import{performance as f}from"node:perf_hooks";import{parseArgs as x}from"node:util";const _="Generate a stable semantic embedding for this short bilingual memory sample. It should be suitable for retrieval, recall, and near-duplicate detection.";function w(){return["Usage:","  node scripts/bench_ollama_embeddings.js [options] <model...>","","Options:","  --host <url>           Ollama host (default http://127.0.0.1:11434)","  --rounds <n>           Measured rounds per test after warmup (default 1)","  --batch-size <n>       Inputs per batch test (default 4)","  --timeout <seconds>    Request timeout (default 300)","  --single-text <text>   Input used for the single-input benchmark","  --skip-warmup          Skip warmup request per model","  --json                 Emit JSON results"].join(`
`)}function p(e,n){const s=Number(e);if(!Number.isFinite(s))throw new Error(`Invalid --${n}: ${String(e)}`);return s}function g(e){if(!e.length)throw new Error("median() requires at least one value");const n=[...e].sort((t,r)=>t-r),s=Math.floor(n.length/2);return n.length%2===0?(n[s-1]+n[s])/2:n[s]}function k(e){const n=e.filter(s=>typeof s=="number"&&Number.isFinite(s));return n.length?g(n):null}function v(e,n){return Array.from({length:n},(s,t)=>`${e} [sample ${t+1}]`)}async function $(e,n,s){const t=new AbortController,r=setTimeout(()=>t.abort(),s);try{return await fetch(e,{...n,signal:t.signal})}finally{clearTimeout(r)}}async function O(e){const n=e.host.replace(/\/+$/,"")+"/api/embed",s={model:e.model,input:e.inputs.length>1?e.inputs:e.inputs[0]},t=await $(n,{method:"POST",headers:{"Content-Type":"application/json"},body:JSON.stringify(s)},Math.max(1,e.timeoutSeconds)*1e3),r=await t.text();if(!t.ok)throw new Error(`HTTP ${t.status}: ${r}`);let u;try{u=JSON.parse(r)}catch{throw new Error(`Unexpected JSON from Ollama: ${r.slice(0,256)}`)}return u}async function b(e){const n=f.now(),s=await O(e),t=(f.now()-n)/1e3,r=s.total_duration,u=typeof r=="number"&&Number.isFinite(r)?r/1e9:null,a=s.embeddings;if(!Array.isArray(a))throw new Error(`Unexpected Ollama response for ${e.model}: ${JSON.stringify(s)}`);const o=a[0],i=Array.isArray(o)?o.length:0;return{wall_seconds:t,api_seconds:u,items:a.length,vector_dims:i}}async function E(e){e.skipWarmup||await b({host:e.host,model:e.model,inputs:[e.singleText],timeoutSeconds:e.timeoutSeconds});const n=[],s=[];for(let t=0;t<e.rounds;t+=1)n.push(await b({host:e.host,model:e.model,inputs:[e.singleText],timeoutSeconds:e.timeoutSeconds}));for(let t=0;t<e.rounds;t+=1)s.push(await b({host:e.host,model:e.model,inputs:e.batchInputs,timeoutSeconds:e.timeoutSeconds}));return{model:e.model,ok:!0,single_wall_seconds:g(n.map(t=>t.wall_seconds)),single_api_seconds:k(n.map(t=>t.api_seconds)),batch_wall_seconds:g(s.map(t=>t.wall_seconds)),batch_api_seconds:k(s.map(t=>t.api_seconds)),batch_size:e.batchInputs.length,vector_dims:n.at(-1)?.vector_dims??0,rounds:e.rounds}}function m(e){return e===null?"-":e.toFixed(2)}function T(e,n){const s=["Model","Single(s)",`Batch${n}(s)`,"API Single(s)",`API Batch${n}(s)`,"Dims","Status"],t=e.map(o=>o.ok?[o.model,m(o.single_wall_seconds),m(o.batch_wall_seconds),m(o.single_api_seconds),m(o.batch_api_seconds),String(o.vector_dims),"ok"]:[o.model,"-","-","-","-","-",o.error]),r=s.map((o,i)=>Math.max(o.length,...t.map(l=>(l[i]??"").length))),u=o=>{process.stdout.write(`${o.map((i,l)=>i.padEnd(r[l]??i.length)).join("  ")}
`)};u(s),u(r.map(o=>"-".repeat(o)));for(const o of t)u(o);const a=e.filter(o=>o.ok);if(a.length){const o=[...a].sort((i,l)=>i.batch_wall_seconds!==l.batch_wall_seconds?i.batch_wall_seconds-l.batch_wall_seconds:i.single_wall_seconds!==l.single_wall_seconds?i.single_wall_seconds-l.single_wall_seconds:i.model.localeCompare(l.model));process.stdout.write(`
Recommended by speed: ${o[0].model}
`)}}async function B(){const e=x({args:process.argv.slice(2),allowPositionals:!0,options:{host:{type:"string",default:"http://127.0.0.1:11434"},rounds:{type:"string",default:"1"},"batch-size":{type:"string",default:"4"},timeout:{type:"string",default:"300"},"single-text":{type:"string",default:_},"skip-warmup":{type:"boolean",default:!1},json:{type:"boolean",default:!1},help:{type:"boolean",default:!1}}});if(e.values.help)return process.stdout.write(`${w()}
`),0;const n=e.positionals.map(c=>String(c).trim()).filter(Boolean);if(!n.length)return process.stderr.write(`${w()}
`),1;const s=String(e.values.host??"").trim()||"http://127.0.0.1:11434",t=p(e.values.rounds,"rounds"),r=p(e.values["batch-size"],"batch-size"),u=p(e.values.timeout,"timeout"),a=String(e.values["single-text"]??_),o=!!e.values["skip-warmup"],i=!!e.values.json;if(t<1)throw new Error("--rounds must be at least 1");if(r<1)throw new Error("--batch-size must be at least 1");const l=v(a,r),d=[];for(const c of n)try{d.push(await E({host:s,model:c,singleText:a,batchInputs:l,rounds:t,timeoutSeconds:u,skipWarmup:o}))}catch(h){const S=h instanceof Error?h.message:String(h);d.push({model:c,ok:!1,error:S})}const y=d.some(c=>c.ok);return i?process.stdout.write(`${JSON.stringify({results:d},null,2)}
`):T(d,r),y?0:1}B().then(e=>{process.exitCode=e}).catch(e=>{const n=e instanceof Error?e.message:String(e);process.stderr.write(`${n}
`),process.exitCode=1});