KRONOS 8B V4
The goal with this merge is to have a significant step towards why the older TIES merges nuke the IFEval score. Present hypothesis is that merging on the non-instruct model causes this disparity, as IFEval measures instruction-following capacity. It's also possible that either P1 or P3 causes the problem.
Merge Details
Merge Method
This model was merged using the Model Stock merge method using unsloth/Meta-Llama-3.1-8B-Instruct as a base.
Models Merged
The following models were included in the merge:
Configuration
The following YAML configuration was used to produce this model:
base_model: unsloth/Meta-Llama-3.1-8B-Instruct
dtype: bfloat16
merge_method: model_stock
slices:
- sources:
- layer_range: [0, 32]
model: T145/KRONOS-8B-V1-P1
- layer_range: [0, 32]
model: T145/KRONOS-8B-V1-P2
- layer_range: [0, 32]
model: T145/KRONOS-8B-V1-P3
- layer_range: [0, 32]
model: mukaj/Llama-3.1-Hawkish-8B
- layer_range: [0, 32]
model: unsloth/Meta-Llama-3.1-8B-Instruct
tokenizer_source: base
Open LLM Leaderboard Evaluation Results
Detailed results can be found here! Summarized results can be found here!
| Metric | % Value |
|---|---|
| Avg. | 28.60 |
| IFEval (0-Shot) | 78.89 |
| BBH (3-Shot) | 30.14 |
| MATH Lvl 5 (4-Shot) | 18.58 |
| GPQA (0-shot) | 5.26 |
| MuSR (0-shot) | 7.77 |
| MMLU-PRO (5-shot) | 30.95 |