base_model:
- HuggingFaceTB/cosmo-1b
- Lambent/cosmo-1b-galore-pythontest
- Lambent/cosmo-1b-qlora-pythontest
- Lambent/cosmo-1b-lisa-pythontest
library_name: transformers
tags:
- mergekit
- merge
pythontestmerge
This is a merge of pre-trained language models created using mergekit.
Testing training data validation:
- Model Stock 3/4 Loss: 0.451
My hypothesis that the pretraining was dragging down the stock merge's performance on training data in any way seems inaccurate.
Cosmopedia data validation:
- Model Stock 3/4 Loss: 1.021
On the other hand, it indeed may have pulled it towards forgetfulness. This is a better loss vs catastrophic forgetting than the prior Model Stock or any of the training methods.
I'm going to estimate that using the base model as an anchor point is a strong remedy for catastrophic forgetting when using multiple different training methods on the same dataset. Less sure I can say anything about how it affects adaptation to the new dataset. It's possible that if using this method, you'd want louder/stronger adaptation to start with than you otherwise would.
Merge Details
Merge Method
This model was merged using the Model Stock merge method using HuggingFaceTB/cosmo-1b as a base.
Models Merged
The following models were included in the merge:
- Lambent/cosmo-1b-galore-pythontest
- Lambent/cosmo-1b-qlora-pythontest
- Lambent/cosmo-1b-lisa-pythontest
Configuration
The following YAML configuration was used to produce this model:
models:
- model: Lambent/cosmo-1b-lisa-pythontest
- model: Lambent/cosmo-1b-qlora-pythontest
- model: Lambent/cosmo-1b-galore-pythontest
base_model: HuggingFaceTB/cosmo-1b
merge_method: model_stock
parameters:
filter_wise: false
dtype: float16