merge
This is a merge of pre-trained language models created using mergekit.
Merge Details
Merge Method
This model was merged using the SLERP merge method.
Models Merged
The following models were included in the merge:
Configuration
The following YAML configuration was used to produce this model:
slices:
- sources:
- model: bamec66557/MISCHIEVOUS-12B-Mix_0.4v
layer_range: [0, 8]
- model: bamec66557/MISCHIEVOUS-12B-Mix_III_IV_V
layer_range: [0, 8]
parameters:
t:
- value: 0.72
- sources:
- model: bamec66557/MISCHIEVOUS-12B-Mix_0.4v
layer_range: [8, 16]
- model: bamec66557/MISCHIEVOUS-12B-Mix_III_IV_V
layer_range: [8, 16]
parameters:
t:
- value: [0.75, 0.85, 0.75]
- sources:
- model: bamec66557/MISCHIEVOUS-12B-Mix_0.4v
layer_range: [16, 24]
- model: bamec66557/MISCHIEVOUS-12B-Mix_III_IV_V
layer_range: [16, 24]
parameters:
t:
- value: [0.85, 1.0, 0.85]
- filter: feed_forward
value: [0.9, 1.0, 1.1]
- sources:
- model: bamec66557/MISCHIEVOUS-12B-Mix_0.4v
layer_range: [24, 32]
- model: bamec66557/MISCHIEVOUS-12B-Mix_III_IV_V
layer_range: [24, 32]
parameters:
t:
- value: [0.95, 1.0, 0.95]
- sources:
- model: bamec66557/MISCHIEVOUS-12B-Mix_0.4v
layer_range: [32, 40]
- model: bamec66557/MISCHIEVOUS-12B-Mix_III_IV_V
layer_range: [32, 40]
parameters:
t:
- value: 1.0
- filter: self_attn
value: [0.92, 1.0, 1.08]
merge_method: slerp
base_model: bamec66557/MISCHIEVOUS-12B-Mix_0.4v
regularization:
- method: weight_clipping
clip_range: [-0.04, 0.04]
postprocessing:
- operation: gaussian_smoothing
sigma: 0.9
- operation: normalize
- operation: quantize
target_dtype: int8
dtype: bfloat16
Open LLM Leaderboard Evaluation Results
Detailed results can be found here
| Metric | Value |
|---|---|
| Avg. | 23.84 |
| IFEval (0-Shot) | 43.66 |
| BBH (3-Shot) | 34.73 |
| MATH Lvl 5 (4-Shot) | 12.31 |
| GPQA (0-shot) | 10.40 |
| MuSR (0-shot) | 12.34 |
| MMLU-PRO (5-shot) | 29.58 |