|
--- |
|
base_model: |
|
- meta-llama/Llama-2-7b-chat-hf |
|
- allenai/tulu-2-dpo-7b |
|
library_name: transformers |
|
tags: |
|
- mergekit |
|
- merge |
|
license: afl-3.0 |
|
--- |
|
# svd_franken_merge1 |
|
|
|
This is a merge of pre-trained language models created using [mergekit](https://github.com/cg123/mergekit). |
|
|
|
## Merge Details |
|
### Merge Method |
|
|
|
This model was merged using the svd_franken_merge merge method using [meta-llama/Llama-2-7b-chat-hf](https://huggingface.co/meta-llama/Llama-2-7b-chat-hf) as a base. |
|
|
|
### Models Merged |
|
|
|
The following models were included in the merge: |
|
* [allenai/tulu-2-dpo-7b](https://huggingface.co/allenai/tulu-2-dpo-7b) |
|
|
|
### Configuration |
|
|
|
The following YAML configuration was used to produce this model: |
|
|
|
```yaml |
|
models: |
|
- model: allenai/tulu-2-dpo-7b |
|
parameters: |
|
weight: 1.0 |
|
# - model: EleutherAI/llemma_7b |
|
# parameters: |
|
# weight: 1.0 |
|
merge_method: svd_franken_merge |
|
base_model: meta-llama/Llama-2-7b-chat-hf |
|
parameters: |
|
probabilistic: True # use probabilistic SVD algorithm (maybe fastest but little inaccurate) instead of the non probabilistic SVD algorithm (slowest but accurate) |
|
sv_reduction: 1.0 # strict value: "1.0 / number of task vector" if number of task vector is >=2 (percentage of singular components to keep for each TV) |
|
sv_scaling: 1.0 # float (hyperparameter): suggested nearby "number of task vector / 2.0" if number of task vector is >=2 else 1.0 (reduction to the stretching factor of the singular values) |
|
num_iterations: 4 # number of iterations for the probabilistic SVD algorithm (2 is a good start and fastest option but could be inaccurate, 32 is the most accurate but slowest option. I don't recommend going above 32 use SVD probabilistic at False instead) |
|
dtype: float16 |
|
|
|
``` |
