Model Stock: All we need is just a few fine-tuned models
Paper
• 2403.19522 • Published
• 14
Image source: The Angel’s Message
This is a merge of pre-trained language models created using mergekit.
This model was merged using the Model Stock merge method using /Users/sthenno/models/tempesthenno-ppo-enchanted as a base.
The following models were included in the merge:
The following YAML configuration was used to produce this model:
name: tempesthenno-ms-0218
merge_method: model_stock
base_model: /Users/sthenno/models/tempesthenno-ppo-enchanted
tokenizer:
source: base
dtype: float32
out_dtype: bfloat16
parameters:
int8_mask: true
normalize: true
rescale: false
models:
- model: /Users/sthenno/models/tempesthenno-sft-0218-ckpt60
- model: /Users/sthenno/models/tempesthenno-sft-0218-ckpt80
- model: /Users/sthenno/models/tempesthenno-sft-0218-stage2-ckpt40
- model: /Users/sthenno/models/tempesthenno-sft-0218-stage2-ckpt50
- model: /Users/sthenno/models/tempesthenno-sft-0218-stage2-ckpt60
Detailed results can be found here
| Metric | Value |
|---|---|
| Avg. | 42.90 |
| IFEval (0-Shot) | 76.56 |
| BBH (3-Shot) | 50.64 |
| MATH Lvl 5 (4-Shot) | 51.44 |
| GPQA (0-shot) | 17.79 |
| MuSR (0-shot) | 13.21 |
| MMLU-PRO (5-shot) | 47.75 |