Skip to content
Other

Mooziq LM: Tiny models for browser-based lo-fi music generation

2 min read

Mooziq LM is a family of ultra-small transformers that generate lo-fi music as symbols (chords, melodies, basslines, and song structure) fast enough to run in real time directly within web browsers using pure JavaScript. Developed by Varun Lohade, these models power the Unlimited Lofi Music project to stream endless, device-generated playlists. The architecture uses decoder-only transformers with KV caching, running without heavy libraries like ONNX. This development will appeal to web developers, generative music hobbyists, and researchers studying compact language models.

Key features

  • A family of miniature models ranging from 114K to several million parameters, available in fp16, int8, and int4 weights.
  • Runs in browsers or Node.js via a lightweight JavaScript engine implementation (transformer.js).
  • Supports multiple musical styles, including Jazz Hop, Steady, Flow, and Folk.
  • Trained on public music datasets and generations from open models without proprietary systems.

What you can do

  • Integrate generative background music into web applications, websites, or browser games.
  • Run local lo-fi track generation on consumer devices without server requests.
  • Use the architecture for research in symbolic music generation and constrained decoding.

Example

import {loadModel} from './transformer.js';
const meta = await (await fetch('B483K/model.json')).json();
const bytes = await (await fetch('B483K/model.bin')).arrayBuffer();
const model = loadModel(meta, bytes);
model.reset();
let logits;
for (const t of [meta.BOS, meta.JAZZHOP, meta.MIN, meta.BAR]) logits = model.step(t);
Read moreDetails
Loss benchmarks: B483K (held-out loss ACE — 1.326, public domain — 1.133), S8-int8 (1.304 / 1.038), S4-int4 (1.301 / 1.025).
Hardware requirements: minimal VRAM and memory usage, runs in-browser.
Limitations: vocabulary is restricted to diatonic seventh chords, an eighth-note grid, and 4/4 time; new styles have limited training data.
License: MIT.
Discuss Telegram 0

Will it run?

Runs in Q8

Fits without quantisation.

  • Q8weights 0.0 GB · ready files0.5 GB
  • FP8 / INT8recommendedweights 0.0 GB0.5 GB
  • 4-bitweights 0.0 GB0.5 GB

Tip: go 8-bit.FP8 / INT8 is within about a percent of full precision in quality, takes half the memory and runs faster on RTX 40/50 and H100. Full precision only pays off for fine-tuning.

What to get for this model

Massed ComputeOpen Massed Compute

Your own GPU: FP8 / INT8 — from RTX 2060 · Q8 — from RTX 2060

Massed Compute hourly prices, November 2026. The discounted price applies with the coupon at checkout.

Estimate: weights plus working memory. Long videos, high resolutions and long contexts need more.

Run it on Massed Compute

The script downloads the model to the rented machine. How to run it is in the model card.

  1. 1. Rent a machine with a fitting GPU

    Open Massed Compute
  2. 2. Paste the script into the machine's terminal

    Connect over SSH (the Massed Compute panel shows the command) or open a terminal on the machine's desktop.

    HF_TOKEN / CIVITAI_TOKENThe script starts with token lines: paste yours between the quotes. The Hugging Face token is for gated models, the Civitai one for Civitai models.

    # orseliyas/mooziq-lm
    mkdir -p ~/models
    curl -LsSf https://astral.sh/uv/install.sh | sh
    source "$HOME/.local/bin/env"
    uv tool install "huggingface_hub[cli]"
    hf download orseliyas/mooziq-lm --local-dir ~/models/mooziq-lm
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to mooziq-lm
  3. 3. Use it

    The model is in ~/models. How to run it is in the model card (link above).

Stop the machine in the panel when you are done: it bills by the minute while it runs.

How to speed it up

Picked for this model type. Choose your hardware above to see what fits you exactly.

  • FlashAttention 44.0.0b40from November 18, 2026new release

    fastest exact attention on B200

    Kernels for Blackwell (beta). Exact attention without quantisation — fine for everything, LLMs included.

    How to enable: pip install flash-attn-4; vLLM and SGLang pick it up themselves.

    Earlier generations (2)
    • FlashAttention 3 — The Hopper (H100/H200) version with FP8.
    • FlashAttention 2 — The standard for RTX 30/40/50 and A100: exact attention without the memory cost of long contexts.
  • torch.compile2.15.0from October 28, 2026new release

    10–30% for free

    Compiles the model for your GPU. Slower first run, faster afterwards.

    How to enable: model = torch.compile(model); in ComfyUI the TorchCompileModel node.

Model family

New posts about versions of mooziq-lm (quants, finetunes, LoRAs) will show up under the bell.

No known relatives of this model yet. All derivatives on Hugging Face

About the model

orseliyas/mooziq-lm
Author
orseliyas
Category
Other
Released
Posted

On NeyroLearn

Hugging Face

00

Ask AI about this model

Answered by Gemini 4 · the chat is kept only in this tab

Comments