T145/Llama-3.1-8B-Instruct-Zeus

ZEUS

Taking inspiration from Dampfinchen/Llama-3.1-8B-Ultra-Instruct and brucethemoose, the goal of this merge is to create an abliterated, conversational AI within 8B parameters that's coherent over long conversations. Using "Ultra-Instruct" as a baseline (which has problems with grammar and coherent conversations), preliminary results seem to show these goals are met. Expect responses in the Markdown format by default.

Merge Details

Merge Method

This model was merged using the DARE TIES merge method using meta-llama/Llama-3.1-8B-Instruct as a base.

Models Merged

The following models were included in the merge:

• akjindal53244/Llama-3.1-Storm-8B
• arcee-ai/Llama-3.1-SuperNova-Lite
• Orenguteng/Llama-3.1-8B-Lexi-Uncensored-V2

Configuration

The following YAML configuration was used to produce this model:

base_model: meta-llama/Llama-3.1-8B-Instruct
dtype: bfloat16
merge_method: dare_ties
parameters:
  int8_mask: 1.0
slices:
- sources:
  - layer_range: [0, 32]
    model: akjindal53244/Llama-3.1-Storm-8B
    parameters:
      density: 0.7
      weight: 0.2
  - layer_range: [0, 32]
    model: arcee-ai/Llama-3.1-SuperNova-Lite
    parameters:
      density: 0.7
      weight: 0.3
  - layer_range: [0, 32]
    model: Orenguteng/Llama-3.1-8B-Lexi-Uncensored-V2
    parameters:
      density: 0.7
      weight: 0.5
  - layer_range: [0, 32]
    model: meta-llama/Llama-3.1-8B-Instruct
tokenizer_source: meta-llama/Llama-3.1-8B-Instruct

Open LLM Leaderboard Evaluation Results

Detailed results can be found here!

Metric	Value
Avg.	29.59
IFEval (0-Shot)	79.41
BBH (3-Shot)	31.39
MATH Lvl 5 (4-Shot)	19.18
GPQA (0-shot)	6.82
MuSR (0-shot)	8.57
MMLU-PRO (5-shot)	32.14

• Falls about 1 point behind "Ultra-Instruct" on IFEval and BBH, but everything else is a significant improvement.