BeatForge / README.md
Bilal140202
Add full model Colab link
55ec7e0
---
title: BeatForge
emoji: 🎚️
colorFrom: red
colorTo: blue
sdk: gradio
sdk_version: "5.24.0"
python_version: "3.10"
app_file: app.py
pinned: true
license: mit
short_description: Lyrics-to-music studio by Bilal Ansari.
tags:
- music
- audio
- gradio
- lyrics-to-music
- generative-audio
---
# BeatForge
BeatForge is a Hugging Face free-tier music generation studio by **Bilal Ansari**.
Website: [ansaribilal.com](https://ansaribilal.com)
The original Colab concept targets HeartMuLa 3B BF16 on a T4 GPU. This Space is built for Hugging Face CPU Basic, so it ships with a CPU-native procedural composition engine that turns structured lyrics and style tags into an original instrumental track. It keeps the same product shape: lyrics, tags, generation controls, branded output, and upgrade notes for the HeartMuLa/MusicGen backend path.
## Run The Full HeartMuLa 3B Model
The free Hugging Face Space runs the CPU-compatible BeatForge engine. For the full HeartMuLa 3B BF16 workflow, run the Colab notebook on a T4 GPU:
[Open BeatForge Full Model Colab](https://colab.research.google.com/drive/19CCLHrTGA0424VTWL5TLx4ELB42ph4j6)
## Why Not HeartMuLa 3B Directly On Free HF CPU?
Hugging Face CPU Basic provides 2 vCPU, 16 GB RAM, and no GPU. HeartMuLa 3B BF16 is a large GPU-oriented model, and the Colab setup depends on T4-class VRAM plus packages such as flash-attn. This Space avoids that fragile stack so it can actually build and run on the free tier.
## What This Version Generates
- Original instrumental tracks from lyrics and style tags
- Section-aware arrangements using `[Intro]`, `[Verse]`, `[Chorus]`, `[Bridge]`, and `[Outro]`
- Drums, bass, chords, melody, pads, and texture layers
- WAV or MP3 export
- Brand-forward UI with ansaribilal.com attribution
## Upgrade Path
For neural lyrics-to-music quality, upgrade the Space to GPU and replace the CPU engine with one of these backends:
- HeartMuLa 3B BF16 with lazy loading
- MusicGen Small/Medium with a GPU runtime or hosted endpoint
- A private inference endpoint exposed through `HF_TOKEN`
The UI is designed so the generation backend can be swapped without changing the user experience.