mirror of
https://github.com/wassname/lora-lite.git
synced 2026-06-27 16:45:56 +08:00
wassname
8005423c47
Paper spreads LoRA-XS across all q/k/v/o + FFN linears, not down_proj only. Tried it (150 modules, 0.154M params): test 55.6 / valid 62.0, slightly below the down_proj row at 6x params, within single-seed noise. down_proj-only stays the table entry. result: outputs/metamath_gsm8k_alllinear/...__seed0/result.json Co-Authored-By: Claudypoo <288921227+claudypoo@users.noreply.github.com>
100 lines
5.4 KiB
Markdown
100 lines
5.4 KiB
Markdown
# lora-lite
|
|
|
|
Hackable PyTorch adapters for LoRA-family and small PEFT experiments.
|
|
|
|
## Hackable code
|
|
|
|
|
|
To keep it simple and hackable we make these choices:
|
|
|
|
- Simple forward hooks, no module replacement or custom modules.
|
|
- Simple code over fast performance
|
|
- No merge/unmerge
|
|
- Single test where we train on MetaMathQA and test on GSM8K for each variant
|
|
|
|
Take a look at [lora.py](src/lora_lite/variants/lora.py)
|
|
|
|
## Install
|
|
|
|
```bash
|
|
pip install -e git+https://github.com/wassname/lora-lite.git#egg=lora-lite
|
|
```
|
|
|
|
## Quickstart
|
|
|
|
```python
|
|
import torch, lora_lite as ll
|
|
|
|
model = MyTransformer()
|
|
cfg = ll.LoRAConfig(r=8, alpha=16, dtype=torch.bfloat16)
|
|
ll.attach(model, cfg)
|
|
|
|
opt = torch.optim.AdamW([p for p in model.parameters() if p.requires_grad], lr=1e-4)
|
|
# train...
|
|
|
|
ll.save(model, "adapter.safetensors")
|
|
ll.detach(model)
|
|
ll.load(model, "adapter.safetensors")
|
|
```
|
|
|
|
## Does it work?
|
|
|
|
```bash
|
|
just check # pytest + smoke + package build + metadata check
|
|
just bnb-smoke # required CUDA bitsandbytes 4bit/8bit smoke
|
|
just qwen-probe # Qwen/Qwen3-0.6B train/save-load probe
|
|
```
|
|
|
|
## Variants
|
|
|
|
Trained on a MetaMathQA subset, tested on GSM8K, all on `Qwen/Qwen3.5-0.8B-Base` targeting
|
|
`down_proj` in all 24 layers (2500 steps, effective batch 8 = 20k samples).
|
|
|
|
| Variant | r | test % | valid % | Params | +MACs/tok | fwd/bwd (ms) | init (s) |
|
|
| --------------------------------------------- | ---: | -----: | ------: | ------: | --------: | -----------: | -------: |
|
|
| [DoRA](https://arxiv.org/abs/2402.09353) | 32 | 60.2 | 68.0 | 3.56M | 3.54M | 161 / 556 | 0.16 |
|
|
| [LoRA](https://arxiv.org/abs/2106.09685) | 32 | 59.8 | 68.0 | 3.54M | 3.54M | 173 / 573 | 0.02 |
|
|
| [PiSSA](https://arxiv.org/abs/2404.02948) | 32 | 59.8 | 76.0 | 3.54M | 3.54M | 146 / 549 | 2.04 |
|
|
| [EVA](https://arxiv.org/abs/2410.07170) | 32 | 59.3 | 74.0 | 3.54M | 3.54M | 151 / 660 | 28.3 |
|
|
| [HRA](https://arxiv.org/abs/2405.17484) | 32 | 59.2 | 70.0 | 2.75M | 2.75M | 225 / 948 | 0.04 |
|
|
| [AntiPaSTO](https://arxiv.org/abs/2601.07473) | 256 | 57.2 | 60.0 | 0.015M | 28.3M | 165 / 596 | 2.0 |
|
|
| [LoRA-XS](https://arxiv.org/abs/2405.17604) | 32 | 56.8 | 68.0 | 0.025M | 3.56M | 162 / 575 | 2.22 |
|
|
| [IA3-FF](https://arxiv.org/pdf/2205.05638) | — | 56.3 | 62.0 | 0.086M | 0M | 140 / 510 | 0.01 |
|
|
| [DeLoRA](https://arxiv.org/abs/2503.18225) | 32 | 56.2 | 62.0 | 3.54M | 3.54M | 169 / 593 | 0.21 |
|
|
| [IA3](https://arxiv.org/pdf/2205.05638) | — | 52.3 | 62.0 | 0.006M | 0M | 161 / 515 | 0.01 |
|
|
|
|
r = adapter rank (— = not a low-rank method). test/valid % = GSM8K exact-match accuracy. Params =
|
|
trainable adapter params. +MACs/tok = added forward MACs per token (analytic, hardware-independent).
|
|
fwd/bwd = median ms over one batch. init = one-time calibration (EVA's PCA; ~0 for the rest). Peak
|
|
CUDA memory is ~9.8 GB for every row. Single seed, so accuracy differences within ~1.4pp (test
|
|
SE at n=1319) are noise.
|
|
|
|
Every row targets `down_proj` only, for an all-else-equal rank comparison. LoRA-XS is the one
|
|
method whose paper instead spreads across all q/k/v/o + FFN linears. Trying that here (150 modules,
|
|
0.154M params) did not help: test 55.6 / valid 62.0, slightly below the down_proj row at 6x the
|
|
params, within single-seed noise. So down_proj-only stays its table entry. Result:
|
|
`outputs/metamath_gsm8k_alllinear/Qwen--Qwen3.5-0.8B-Base__lora_xs__s2500__seed0/result.json`.
|
|
|
|
We validate our adapters the same way [PEFT](https://github.com/huggingface/peft/tree/main/method_comparison) does: train on a MetaMathQA subset and check meaningful GSM8K accuracy. See [this file](scripts/metamath_gsm8k_benchmark.py) for details.
|
|
|
|
AntiPaSTO is the novel row here: instead of adding trainable directions like LoRA, it freezes W's own top-r SVD and learns only a per-direction singular-value delta plus a block-diagonal Cayley rotation of that frozen basis. The singular directions stay interpretable and the adapter is tiny (15K params, ~230x smaller than LoRA's 3.54M) yet stays within noise of the full-rank adapters. The default rotates the input basis (V); rotating the output (U), both, or neither are `rotate_basis` ablation axes.
|
|
|
|
The full AntiPaSTO family (rotation-free gain core, the U/both rotation arms, contractive directional ablation [Arditi+ 2024](https://arxiv.org/abs/2406.11717), a low-rank mixing core, and CorDA/ASVD covariance-oriented bases [Yang+ 2024](https://arxiv.org/abs/2406.05223) / [Yuan+ 2023](https://arxiv.org/abs/2312.05821)) lives on the [`antipasto-variants`](https://github.com/wassname/lora-lite/tree/antipasto-variants) branch with its own ablation table. On GSM8K/down_proj none of those arms separated from this one (the covariance-oriented bases cost 34-120 s of init for no gain; the V/U/both rotation order flips between two seeds, so the basis is within noise), so main keeps one cheap arm: rotation of V.
|
|
|
|
|
|
## Developer docs
|
|
|
|
See [docs/developer_guide.md](docs/developer_guide.md) for the variant API, data-calibrated init, and save/load format.
|
|
|
|
## Citation
|
|
|
|
```bibtex
|
|
@misc{wassname2026loralite,
|
|
title = {LoRA-Lite: A Hackable Adapter Library for Research},
|
|
author = {Michael J. Clark},
|
|
year = {2026},
|
|
url = {https://github.com/wassname/lora-lite/}
|
|
}
|
|
```
|
|
|