3 comments (3 comments)0 reactions (0 reactions)0 assignees (0 assignees)Python5,658 forks (5,658 forks)batch import
feature requesthelp wanted
Repository metrics
- Stars
- 64,271 stars (64,271 stars)
- PR merge metrics
- Avg merge 3d 15h (Avg merge 3d 15h)525 merged PRs in 30d (525 merged PRs in 30d)
Description
Not sure if this allign with unsloth but spectrum is actually a very great way to finetune models it selects layers with higher Signal to Noise Ratio while adhering great performance
A great model which use spectrum is Llama-3.1-Storm-8B
paper
github
Performance

Contributor guide
- Research direction
- Study the Spectrum paper (arXiv:2406.06623) and its implementation at github.com/cognitivecomputations/spectrum. Understand how it selects layers based on Signal to Noise Ratio. Then identify where to integrate this into unsloth's model training pipeline, likely during the layer freezing step. Implement a configuration option to enable Spectrum fine tuning.
- Tech stack
- python
- Domain
- machine learningai
- Issue type
- Feature
- Prerequisites
- PythonGit