miscii-14b-1225

Image source: Rrhar'il | Phigros

Prompting & Usage

See miscii-14b-1028 for more details.

Training Details

Coming soon

Merge Details

This is a merge of pre-trained language models created using mergekit.

Merge Method

This model was merged using the TIES merge method using miscii-14b-1028 as a base.

Models Merged

The following models were included in the merge:

• sthenno/exp-002
• sthenno/miscii-1218

Configuration

The following YAML configuration was used to produce this model:

tokenizer_source: "base"
chat_template: "chatml"

merge_method: ties
dtype: bfloat16

parameters:
  normalize: true

base_model: sthenno-com/miscii-14b-1028

models:
  - model: sthenno-com/miscii-14b-1028
    parameters:
      weight: 1
      density: 0.5
  - model: sthenno/miscii-1218
    parameters:
      weight: 1
      density: 0.5
  - model: sthenno/exp-002
    parameters:
      weight: 0.9
      density: 0.5
  - model: sthenno/miscii-1218
    parameters:
      weight: 0.6
      density: 0.5

Open LLM Leaderboard Evaluation Results

Congratulations to the miscii series models for surpassing 40 points for the first time! As of December 25, 2024, this should be the best-performing 14B model in the tests, right?

Metric	Value
Avg.	40.08
IFEval (0-Shot)	78.78
BBH (3-Shot)	50.91
MATH Lvl 5 (4-Shot)	31.57
GPQA (0-shot)	17.00
MuSR (0-shot)	14.77
MMLU-PRO (5-shot)	47.46

Refined since 2025-02-15

Metric	Value
Avg.	42.35
IFEval (0-Shot)	78.78
BBH (3-Shot)	50.91
MATH Lvl 5 (4-Shot)	45.17
GPQA (0-shot)	17.00
MuSR (0-shot)	14.77
MMLU-PRO (5-shot)	47.46