llama-3-Nephilim-v3-8B
This repo contains a merge of pre-trained language models created using mergekit.
GGUF quants are here.
Although none of the components of this merge were trained for roleplay nor intended for it, the model can be used effectively in that role.
Tested with temperature 1 and minP 0.01. This model leans toward being creative, so adjust temperature upward or downward as desired.
There are initial format consistency issues with the merged model, but this can be mitigated in an Instruct prompt. Additionally, promptsteering was employed to vary the text generation output to avoid some of the common failings observed during text generation with Llama 3 8B models. The complete Instruct prompt used during testing is available below.
Built with Meta Llama 3.
Merge Details
Merge Method
This model was merged using the task arithmetic merge method using grimjim/Llama-3-Instruct-8B-SPPO-Iter3-SimPO-merge 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: grimjim/Llama-3-Instruct-8B-SPPO-Iter3-SimPO-merge
dtype: bfloat16
merge_method: task_arithmetic
parameters:
normalize: false
slices:
- sources:
- layer_range: [0, 32]
model: grimjim/Llama-3-Instruct-8B-SPPO-Iter3-SimPO-merge
- layer_range: [0, 32]
model: tokyotech-llm/Llama-3-Swallow-8B-Instruct-v0.1
parameters:
weight: 0.1
Open LLM Leaderboard Evaluation Results
Detailed results can be found here
| Metric | Value |
|---|---|
| Avg. | 20.54 |
| IFEval (0-Shot) | 41.74 |
| BBH (3-Shot) | 28.96 |
| MATH Lvl 5 (4-Shot) | 9.14 |
| GPQA (0-shot) | 6.04 |
| MuSR (0-shot) | 8.33 |
| MMLU-PRO (5-shot) | 29.02 |