7B Merges
Collection
A collection of my 7B parameter merges
•
3 items
•
Updated
🏵️ Anthesis_7B
Anthesis_7B is a merge of the following models:
Formatting/Preset
Alpaca works best, but Mistral gives good outputs as well.
Configuration
models:
- model: ResplendentAI/Paradigm_Shift_7B
# no parameters necessary for base model
- model: rmdhirr/Foxglove_7B
parameters:
density: 0.60
weight: 0.40
merge_method: dare_ties
base_model: ResplendentAI/Paradigm_Shift_7B
parameters:
int8_mask: true
dtype: bfloat16
Open LLM Leaderboard Evaluation Results
Detailed results can be found here
| Metric | Value |
|---|---|
| Avg. | 67.97 |
| AI2 Reasoning Challenge (25-Shot) | 69.03 |
| HellaSwag (10-Shot) | 86.20 |
| MMLU (5-Shot) | 62.06 |
| TruthfulQA (0-shot) | 68.65 |
| Winogrande (5-shot) | 78.93 |
| GSM8k (5-shot) | 42.99 |
- Downloads last month
- 3,016
Merge of
Collection including rmdhirr/Anthesis_7B
Evaluation results
- normalized accuracy on AI2 Reasoning Challenge (25-Shot)test set Open LLM Leaderboard69.030
- normalized accuracy on HellaSwag (10-Shot)validation set Open LLM Leaderboard86.200
- accuracy on MMLU (5-Shot)test set Open LLM Leaderboard62.060
- mc2 on TruthfulQA (0-shot)validation set Open LLM Leaderboard68.650
- accuracy on Winogrande (5-shot)validation set Open LLM Leaderboard78.930
- accuracy on GSM8k (5-shot)test set Open LLM Leaderboard42.990