Spectral Surgery — Instruction Following
Collection
Instruction-following LoRA and Spectral Surgery adapters evaluated on IFEval. • 12 items • Updated
Llama-3.1-8B-Instruct + Instruction Following + Spectral Surgery
This repository contains a Spectral Surgery adapter derived from the instruction-following LoRA checkpoint of Llama-3.1-8B-Instruct.
Base Model
meta-llama/Llama-3.1-8B-Instruct
Source LoRA
- Sequence length: 1024
- Global batch size: 128
- LoRA rank: 16
- Seed: 42
Spectral Surgery
- Fast HNS steps: 8
- Stable HNS steps: 2
Evaluation
Evaluation on IFEval.
| Model | Prompt Strict | Prompt Loose | Instruction Strict | Instruction Loose |
|---|---|---|---|---|
| Base | 65.80% | 72.83% | 75.54% | 81.18% |
| LoRA SFT | 65.99% | 70.98% | 74.82% | 79.26% |
| Spectral Surgery HNS 8+2 | 69.50% | 73.94% | 77.94% | 82.13% |
Relative to the vanilla LoRA checkpoint, Spectral Surgery HNS 8+2 improves:
- Prompt Strict by 3.51 percentage points
- Prompt Loose by 2.96 percentage points
- Instruction Strict by 3.12 percentage points
- Instruction Loose by 2.87 percentage points
The edited adapter also exceeds the base model on all four IFEval metrics.
- Downloads last month
- 17
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support
Model tree for tianzl66/Llama-3.1-8B-Instruct-InstructionFollowing-SpectralSurgery-HNS8p2
Base model
meta-llama/Llama-3.1-8B Finetuned
meta-llama/Llama-3.1-8B-Instruct