LaMa-Merged-slerp / README.md
SamaOkasha's picture
Upload folder using huggingface_hub
c8811f4 verified
|
raw
history blame
No virus
1.06 kB
---
license: apache-2.0
tags:
- merge
- mergekit
- lazymergekit
- allenai/led-base-16384
- allenai/led-base-16384
---
# LaMa-Merged-slerp
LaMa-Merged-slerp is a merge of the following models using [mergekit](https://github.com/cg123/mergekit):
* [allenai/led-base-16384](https://huggingface.co/allenai/led-base-16384)
* [allenai/led-base-16384](https://huggingface.co/allenai/led-base-16384)
## 🧩 Configuration
```yaml
slices:
- sources:
- model: allenai/led-base-16384
layer_range: [0, 12] # Taking the initial layers from LED Base model
- model: allenai/led-base-16384
layer_range: [12, 24] # Taking the later layers from LED Base model
merge_method: slerp
base_model: allenai/led-base-16384 # Using LED Base model as the base model
parameters:
t:
- filter: self_attn
value: [0, 0.5, 0.3, 0.7, 1] # Interpolation values for self-attention layers
- filter: mlp
value: [1, 0.5, 0.7, 0.3, 0] # Interpolation values for MLP layers
- value: 0.5 # Default interpolation value
dtype: bfloat16
```