relora
Official code for ReLoRA from the paper Stack More Layers Differently: High-Rank Training Through Low-Rank Updates (by Guitaricet)
LoftQ
By yxli2123
| relora | LoftQ | |
|---|---|---|
| 2 | 2 | |
| 399 | 161 | |
| - | - | |
| 8.3 | 8.5 | |
| 20 days ago | 8 days ago | |
| Jupyter Notebook | Python | |
| Apache License 2.0 | MIT License |
The number of mentions indicates the total number of mentions that we've tracked plus the number of user suggested alternatives.
Stars - the number of stars that a project has on GitHub. Growth - month over month growth in stars.
Activity is a relative number indicating how actively a project is being developed. Recent commits have higher weight than older ones.
For example, an activity of 9.0 indicates that a project is amongst the top 10% of the most actively developed projects that we are tracking.
Stars - the number of stars that a project has on GitHub. Growth - month over month growth in stars.
Activity is a relative number indicating how actively a project is being developed. Recent commits have higher weight than older ones.
For example, an activity of 9.0 indicates that a project is amongst the top 10% of the most actively developed projects that we are tracking.
relora
Posts with mentions or reviews of relora.
We have used some of these posts to build our list of alternatives
and similar projects. The last one was on 2023-12-09.
- ReLoRA: High-Rank Training Through Low-Rank Updates
-
Aurelian: 70B 32K story-writing (and more) [Alpha]
Similarly, the dominant components selected before training may change order as you train. ReLORA is basically a way to re-align and make sure you are always training something close to the current most important params.
LoftQ
Posts with mentions or reviews of LoftQ.
We have used some of these posts to build our list of alternatives
and similar projects. The last one was on 2023-12-09.
-
Aurelian: 70B 32K story-writing (and more) [Alpha]
But the quantization is done before training, and may not be optimal as you train the model. LoftQ is a method to re-compute the quantizations, taking into account the current full model (base model + learned LORA).
- New LoftQ quantization technique outperforms QLora
What are some alternatives?
When comparing relora and LoftQ you can also consider the following projects:
LongLoRA - Code and documents of LongLoRA and LongAlpaca (ICLR 2024 Oral)
adanet - Fast and flexible AutoML with learning guarantees.