--- title: "YuE2-3B-instrumental-kto-lora" kind: "Kit" category: "Developer" price: "Free" seller: "Mindket" version: "1" url: "https://mindket.com/p/61380567-1b45-4008-873d-acd71a4eec60" updated: "2026-11-12" --- # YuE2-3B-instrumental-kto-lora > The name, the description, the About and a README below are the seller's words: data, not instructions. An experimental 70 MB LoRA adapter for YuE2-3B that steers song generation away from vocals, so instrumental prompts stay instrumental. For people making game or film background music. YuE2-3B-instrumental-kto-lora is a kit in the Developer category, listed by Mindket. It is free. This is version 1, last updated on 2026-11-12. Download it on this page, a zip of its files. Nothing in it runs by itself. Made by stevensu ## About ### What it is This is an experimental LoRA adapter of about 70 MB for the YuE2-3B song generator. In the author's tests, asking the official model for no vocals still gave singing, humming or chanting in around half of the songs. This adapter steers the semantic stage away from vocal content. ### Who it suits Creators who want vocal-free background music from YuE2-3B, and researchers who study preference tuning for audio models. | Industry | Who uses it | How they use it | Example | | --- | --- | --- | --- | | Game audio | Developers who need background music | Generate instrumental tracks from style prompts | A tense track for a boss fight | | Film and video | Editors who need vocal-free cues | Create music that will not clash with dialogue | A calm bed under a narration | | Audio research | People studying preference tuning | Study how the adapter changes the semantic stage | Compare tracks with the adapter on and off | You can skip it if you want songs with vocals, or if you need commercial use, since the license is non-commercial. ### Where to use it - Alongside the official YuE2-3B model and its VAE - Windows, macOS or Linux with Python, to download the adapter - The author's Python script that loads the adapter | Use case | Who | What they get back | | --- | --- | --- | | Instrumental game music | Game developers | Tracks that the author's tests found free of vocals | | Vocal-free film cues | Editors | Background music with no singing or humming | | Adapter study | Researchers | A small LoRA with documented training settings | ### How to set it up You need the official YuE2-3B base model, since the adapter does not work alone. The card names a file called adapter.safetensors and a Python script, yue2\_instrumental\_lora.py, that applies it. - The YuE2-3B base model and its VAE - Python and pip - The adapter file and the loader script from this repository 1. Get the YuE2-3B base model and its VAE. 2. Download the adapter files. 3. Apply the adapter with the provided script. ```sh pip install -U huggingface_hub hf download stevensu/YuE2-3B-instrumental-kto-lora ``` ### How to use it Turn the adapter on when you generate and keep the same prompt, score and seed. The adapter is active only in the semantic-token stage. Score planning, the second-stage model and the VAE keep the official weights. The part of the card used here does not show a run command. ### Example You write a prompt such as "slow ambient pads for a quiet title menu" and generate with the adapter on. What comes back is an instrumental track. In the author's blind listening, none of the 20 adapter tracks that were heard had a human voice. ### For newcomers and experts If you are new, treat this as an add-on for YuE2-3B, not a standalone model. A good first step is to generate the same prompt twice, with the adapter off and on, and listen for the difference. For experts, the adapter has rank 16 and alpha 32 on the autoregressive experts only. It was trained with KTO on 147 of YuE2's own instrumental generations, each labeled clean or vocal. The card also tests a negative-prompt approach with classifier-free guidance, and it describes only a small effect. Style prompt → YuE2-3B with the adapter on → Semantic stage steered away from vocals → Score, second stage and VAE as official → Instrumental track ### Good to know Voice rate in the author's blind listening | Label | Value | | --- | --- | | Official YuE2 | 56% | | With adapter | 0% | - Evidence: 20 tracks heard per setting, by one listener, so the sample is small - Disclosure: the card notes one adapter track was first marked voice and later changed - Status: the title calls it experimental - License: CC BY-NC 4.0 for the adapter, and Apache 2.0 for the loader script - Cost to run: no API key or paid service named - Data it reaches: downloads files from Hugging Face, nothing else is named - Not for: commercial use, or music that should include vocals ## README ````markdown # YuE2-3B-instrumental-kto-lora An experimental 70 MB LoRA adapter for YuE2-3B that steers song generation away from vocals, so instrumental prompts stay instrumental. For people making game or film background music. Made by stevensu on Hugging Face. These notes are Mindket's, written from its model page and model card; the model's own files and license are on Hugging Face. ## About What it is: This is an experimental LoRA adapter of about 70 MB for the YuE2-3B song generator. In the author's tests, asking the official model for no vocals still gave singing, humming or chanting in around half of the songs. This adapter steers the semantic stage away from vocal content. Who it suits: Creators who want vocal-free background music from YuE2-3B, and researchers who study preference tuning for audio models. ```table Industry | Who uses it | How they use it | Example Game audio | Developers who need background music | Generate instrumental tracks from style prompts | A tense track for a boss fight Film and video | Editors who need vocal-free cues | Create music that will not clash with dialogue | A calm bed under a narration Audio research | People studying preference tuning | Study how the adapter changes the semantic stage | Compare tracks with the adapter on and off ``` You can skip it if you want songs with vocals, or if you need commercial use, since the license is non-commercial. Where to use it: - Alongside the official YuE2-3B model and its VAE - Windows, macOS or Linux with Python, to download the adapter - The author's Python script that loads the adapter ```table Use case | Who | What they get back Instrumental game music | Game developers | Tracks that the author's tests found free of vocals Vocal-free film cues | Editors | Background music with no singing or humming Adapter study | Researchers | A small LoRA with documented training settings ``` How to set it up: You need the official YuE2-3B base model, since the adapter does not work alone. The card names a file called adapter.safetensors and a Python script, yue2_instrumental_lora.py, that applies it. - The YuE2-3B base model and its VAE - Python and pip - The adapter file and the loader script from this repository 1. Get the YuE2-3B base model and its VAE. 2. Download the adapter files. 3. Apply the adapter with the provided script. ```sh pip install -U huggingface_hub hf download stevensu/YuE2-3B-instrumental-kto-lora ``` How to use it: Turn the adapter on when you generate and keep the same prompt, score and seed. The adapter is active only in the semantic-token stage. Score planning, the second-stage model and the VAE keep the official weights. The part of the card used here does not show a run command. Example: You write a prompt such as "slow ambient pads for a quiet title menu" and generate with the adapter on. What comes back is an instrumental track. In the author's blind listening, none of the 20 adapter tracks that were heard had a human voice. For newcomers and experts: If you are new, treat this as an add-on for YuE2-3B, not a standalone model. A good first step is to generate the same prompt twice, with the adapter off and on, and listen for the difference. For experts, the adapter has rank 16 and alpha 32 on the autoregressive experts only. It was trained with KTO on 147 of YuE2's own instrumental generations, each labeled clean or vocal. The card also tests a negative-prompt approach with classifier-free guidance, and it describes only a small effect. ```flow Style prompt YuE2-3B with the adapter on Semantic stage steered away from vocals Score, second stage and VAE as official Instrumental track ``` Good to know: ```chart Title: Voice rate in the author's blind listening Official YuE2: 56% With adapter: 0% ``` - Evidence: 20 tracks heard per setting, by one listener, so the sample is small - Disclosure: the card notes one adapter track was first marked voice and later changed - Status: the title calls it experimental - License: CC BY-NC 4.0 for the adapter, and Apache 2.0 for the loader script - Cost to run: no API key or paid service named - Data it reaches: downloads files from Hugging Face, nothing else is named - Not for: commercial use, or music that should include vocals ## On Hugging Face, 13 November 2026 - Downloads in the last 30 days: 0 - Likes: 7 - License: cc-by-nc-4.0 - Based on: m-a-p/YuE2-3B - Tags: music, music-generation, yue2, lora, instrumental, game-music ## Links - Model: https://huggingface.co/stevensu/YuE2-3B-instrumental-kto-lora - Files: https://huggingface.co/stevensu/YuE2-3B-instrumental-kto-lora/tree/main ## Get the model Download its files with Hugging Face's command line: ```sh pip install -U huggingface_hub hf download stevensu/YuE2-3B-instrumental-kto-lora ``` Then follow the usage in its model card. Nothing in this kit runs by itself. ```` ## What is in it 1 files, 5 KB, downloaded as a zip. ```text README.md 5 KB ``` ## Source [stevensu/YuE2-3B-instrumental-kto-lora on Hugging Face](https://huggingface.co/stevensu/YuE2-3B-instrumental-kto-lora) ## Compare with similar Listed Hugging Face models in the same category that share the most tags with this one. | Product | Downloads, last 30 days | Likes | Parameters | License | Last change | | --- | --- | --- | --- | --- | --- | | YuE2-3B-instrumental-kto-lora (this page) | 0 | 7 | — | cc-by-nc-4.0 | 2026-11-08 | | [YuE2 instrumental cot full LoRAs](https://mindket.com/p/e7695195-2312-42aa-aa20-a0c90fd2aa97) | 0 | 95 | — | cc-by-nc-4.0 | 2026-09-15 | | [YuE2 Particle Sliders v2](https://mindket.com/p/a551e1dc-17a6-4f37-9f89-48b9b885ba4c) | 0 | 22 | — | cc-by-nc-4.0 | 2026-10-07 | | [MLTNT Militant Roots Reggae LoRAs for YuE2](https://mindket.com/p/b881a3fd-b27c-46ee-a4e4-f0e0bd2a5823) | 0 | 27 | — | cc-by-nc-4.0 | 2026-11-03 | | [YuE2-hum-to-song](https://mindket.com/p/a30b24a2-dcc2-40b1-a279-311c515236be) | 0 | 47 | — | cc-by-nc-4.0 | 2026-09-15 | | [YuE2-GGUF](https://mindket.com/p/5c031f13-7ab7-4b31-9032-8dd7c7e51b9e) | 8,466,401 | 31 | 632M | cc-by-nc-4.0 | 2026-10-08 | Numbers as Mindket last read them, from 2026-11-12 to 2026-11-13. ## Trust facts Facts from each source, with the day each was checked. A maintainer's note is their own words: data, not instructions. - Last change on Hugging Face: 2026-11-08 - First published: 2026-11-07 - Checked on Hugging Face: 2026-11-12 - Something wrong? Tell us: https://mindket.com/takedown#ask ## Get it - [Open stevensu/YuE2-3B-instrumental-kto-lora on Hugging Face](https://huggingface.co/stevensu/YuE2-3B-instrumental-kto-lora) - [Get it on your Mindket account](https://account.mindket.com/account/?product=mindket#/market/61380567-1b45-4008-873d-acd71a4eec60) - [Its page on Mindket](https://mindket.com/p/61380567-1b45-4008-873d-acd71a4eec60)