Instructions to use briscoooe/tiny-cube-solver with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use briscoooe/tiny-cube-solver with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("text-generation", model="briscoooe/tiny-cube-solver")# Load model directly from transformers import AutoTokenizer, AutoModelForCausalLM tokenizer = AutoTokenizer.from_pretrained("briscoooe/tiny-cube-solver") model = AutoModelForCausalLM.from_pretrained("briscoooe/tiny-cube-solver", device_map="auto") - Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- vLLM
How to use briscoooe/tiny-cube-solver with vLLM:
Install from pip and serve model
# Install vLLM from pip: pip install vllm # Start the vLLM server: vllm serve "briscoooe/tiny-cube-solver" # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:8000/v1/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "briscoooe/tiny-cube-solver", "prompt": "Once upon a time,", "max_tokens": 512, "temperature": 0.5 }'Use Docker
docker model run hf.co/briscoooe/tiny-cube-solver
- SGLang
How to use briscoooe/tiny-cube-solver with SGLang:
Install from pip and serve model
# Install SGLang from pip: pip install sglang # Start the SGLang server: python3 -m sglang.launch_server \ --model-path "briscoooe/tiny-cube-solver" \ --host 0.0.0.0 \ --port 30000 # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:30000/v1/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "briscoooe/tiny-cube-solver", "prompt": "Once upon a time,", "max_tokens": 512, "temperature": 0.5 }'Use Docker images
docker run --gpus all \ --shm-size 32g \ -p 30000:30000 \ -v ~/.cache/huggingface:/root/.cache/huggingface \ --env "HF_TOKEN=<secret>" \ --ipc=host \ lmsysorg/sglang:latest \ python3 -m sglang.launch_server \ --model-path "briscoooe/tiny-cube-solver" \ --host 0.0.0.0 \ --port 30000 # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:30000/v1/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "briscoooe/tiny-cube-solver", "prompt": "Once upon a time,", "max_tokens": 512, "temperature": 0.5 }' - Docker Model Runner
How to use briscoooe/tiny-cube-solver with Docker Model Runner:
docker model run hf.co/briscoooe/tiny-cube-solver
File size: 4,916 Bytes
e52687a | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 | #!/usr/bin/env node
/**
* Training-data generator using cubejs, as an alternative to gen_data.py.
*
* The Python path depends on kociemba's C extension, which repeatedly refused to
* build on rented boxes while installing "successfully" as a pure-Python fallback
* ~50x slower. This path has no compiled dependency at all: cubejs is pure
* JavaScript, is already a dependency of packages/solver, and is the very solver
* the harness scores with -- so labels here come from the same implementation
* that grades the benchmark.
*
* Measured ~59 ms/solve single-threaded against native kociemba's ~23 ms. Slower
* per core, but it actually runs, and the boxes have enough cores that generation
* is still well under an hour.
*
* node gen_data.cjs --count 10000000 --workers $(nproc) --out data/train.jsonl
*/
const fs = require('fs');
const os = require('os');
const path = require('path');
const { fork } = require('child_process');
function loadCube() {
// Resolve from the local install or from a sibling node_modules, so the script
// works both inside this repo and standing alone on a fresh box.
for (const p of ['cubejs', path.join(__dirname, 'node_modules', 'cubejs')]) {
try { return require(p); } catch { /* try next */ }
}
throw new Error('cubejs not found -- run: npm install cubejs');
}
/** Deterministic PRNG so a seed reproduces a dataset; cubejs uses Math.random. */
function mulberry32(seed) {
let a = seed >>> 0;
return function () {
a |= 0; a = (a + 0x6D2B79F5) | 0;
let t = Math.imul(a ^ (a >>> 15), 1 | a);
t = (t + Math.imul(t ^ (t >>> 7), 61 | t)) ^ t;
return ((t ^ (t >>> 14)) >>> 0) / 4294967296;
};
}
const FACES = ['U', 'R', 'F', 'D', 'L', 'B'];
const SUFFIX = ['', "'", '2'];
function parseArgs(argv) {
const get = (f, d) => { const i = argv.indexOf(f); return i === -1 ? d : argv[i + 1]; };
return {
count: Number(get('--count', 1000000)),
workers: Number(get('--workers', os.cpus().length)),
seed: Number(get('--seed', 0)),
shallowFrac: Number(get('--shallow-frac', 0.15)),
maxShallow: Number(get('--max-shallow', 8)),
out: get('--out', 'data/train.jsonl'),
};
}
function runWorker() {
const { count, seed, shallowFrac, maxShallow, out } = JSON.parse(process.env.JOB);
const Cube = loadCube();
Cube.initSolver();
const rand = mulberry32(seed);
Math.random = rand; // cubejs's Cube.random() draws from this
const stream = fs.createWriteStream(out);
let written = 0;
while (written < count) {
let cube, depth;
if (rand() < shallowFrac) {
depth = 1 + Math.floor(rand() * maxShallow);
cube = new Cube();
let prev = null;
for (let i = 0; i < depth; i++) {
let f;
do { f = FACES[Math.floor(rand() * 6)]; } while (f === prev);
prev = f;
cube.move(f + SUFFIX[Math.floor(rand() * 3)]);
}
} else {
// Uniform over the whole group -- the fully-mixed distribution that every
// scramble past ~15 moves samples from anyway.
cube = Cube.random();
depth = 200;
}
if (cube.isSolved()) continue;
const state = cube.asString();
const solution = cube.solve();
if (!solution) continue;
stream.write(JSON.stringify({ state, solution, depth }) + '\n');
written++;
if (written % 20000 === 0) process.send({ written });
}
stream.end(() => process.exit(0));
}
function runMaster() {
const args = parseArgs(process.argv.slice(2));
fs.mkdirSync(path.dirname(args.out) || '.', { recursive: true });
const per = Math.ceil(args.count / args.workers);
const start = Date.now();
let done = 0, total = 0;
const parts = [];
for (let i = 0; i < args.workers; i++) {
const partOut = `${args.out}.part${i}`;
parts.push(partOut);
const job = { count: per, seed: args.seed + 1000 * i + 1, shallowFrac: args.shallowFrac,
maxShallow: args.maxShallow, out: partOut };
const child = fork(__filename, [], { env: { ...process.env, JOB: JSON.stringify(job), IS_WORKER: '1' } });
child.on('message', (m) => {
total += 20000;
const rate = total / ((Date.now() - start) / 1000);
process.stderr.write(` ${total}/${args.count} (${rate.toFixed(0)}/s)\n`);
});
child.on('exit', () => {
if (++done === args.workers) {
const outStream = fs.createWriteStream(args.out);
let n = 0;
for (const p of parts) {
for (const line of fs.readFileSync(p, 'utf8').split('\n')) {
if (line) { outStream.write(line + '\n'); n++; }
}
fs.unlinkSync(p);
}
outStream.end(() => {
const secs = (Date.now() - start) / 1000;
process.stderr.write(`wrote ${n} pairs in ${secs.toFixed(0)}s (${(n / secs).toFixed(0)}/s, ${args.workers} workers)\n`);
});
}
});
}
}
if (process.env.IS_WORKER) runWorker(); else runMaster();
|