|
--- |
|
license: mit |
|
datasets: |
|
- siacus/dv_subject |
|
base_model: |
|
- meta-llama/Llama-2-7b-chat-hf |
|
new_version: siacus/llama-2-7b-small-dv |
|
--- |
|
The data used to train the model are on Huggingface under [siacus/dv_subject](https://huggingface.co/datasets/siacus/dv_subject) |
|
|
|
The small-dv version of the fine-tuned model works on a training-set of 5,000 randomly sampled data. |
|
|
|
The large version works on the whole 76.1K training records. |
|
|
|
The test set is of size 32.6K rows. |
|
|
|
F16 version from merged weights created with llama.cpp on a CUDA GPU and the 4bit quantized version created on a Mac M2 Ultra Metal architecture. If you want to use the 4bit quantized version on CUDA, please quantize it directly from the F16 version. |
|
|
|
For more information about this model refer the main repository for the supplementary material of the manuscript |
|
[Rethinking Scale: The Efficacy of Fine-Tuned Open-Source LLMs in Large-Scale Reproducible Social Science Research](https://arxiv.org/abs/2411.00890). |
|
|
