| license: apache-2.0 | |
| language: | |
| - en | |
| base_model: | |
| - ACE-Step/ACE-Step-v1-3.5B | |
| # ACE Step LoRA Experimentation: Glitch Hop | |
| <!-- Provide a quick summary of what the model is/does. --> | |
| This is a Low-Rank Adaptation (LoRA) model of [ACE Step](https://huggingface.co/ACE-Step/ACE-Step-v1-3.5B) focused on a specific genre of Glitch Hop music. | |
| ## Triggers | |
| Any mix and match of the following words should do: | |
| ```neurohop, electronic, glitch hop, hip hop, dubstep, synth, vocal synth, vocal samples, talkbox, talkbox vocals``` | |
| Be sure to specify BPM as well as any normal trigger words aligning with the original ACE Step model. | |
| ## Training Details | |
| The LoRA was trained off of a privately curated dataset of 22 tracks produced in the Glitch Hop/Neurohop genre, using default config settings following the [ACE Step LoRA Training Outline](https://github.com/ace-step/ACE-Step/blob/main/TRAIN_INSTRUCTION.md) | |
| This the first ACE Step LoRA I've attempted, with the current model being the 3rd iteration total from the default training process (no specified maximum epochs) before the GPU I was using ran out of memory. | |
| Since the training was cut prematurely, it may not accurately produce satisfying results! Further experimentation, examples and paramater detailing will be done at a later date. In the mean time, feel free to share your results if you decided to utilize this LoRA fine-tune. | |
| ### Results | |
| TBD! | |