This model was merged using the OTTER-Merge (OT + Top-eigenspace Elastic Regularization) merge method using meta-llama/Meta-Llama-3-8B-Instruct as a base.

Models Merged

The following models were included in the merge:

tokyotech-llm/Llama-3-Swallow-8B-Instruct-v0.1

Configuration

The following YAML configuration was used to produce this model:

yaml

dtype: bfloat16
tokenizer:
  source: union
merge_method: otter_merge
models:
- model: tokyotech-llm/Llama-3-Swallow-8B-Instruct-v0.1
  parameters:
    weight: 0.5
- model: meta-llama/Meta-Llama-3-8B-Instruct
  parameters:
    weight: 0.5
base_model: meta-llama/Meta-Llama-3-8B-Instruct
parameters:
  weights:
  - 0.5
  - 0.5
  align: true
  eps: 0.05
  iters: 50
  rank_r: 64
  k_preserve: 32
  beta: 0.5
  ot_cap: 1024
  group_size: 0
write_readme: README.md

gsjang/ja-llama-3-swallow-8b-instruct-v0.1-x-meta-llama-3-8b-instruct-otter_merge

ja-llama-3-swallow-8b-instruct-v0.1-x-meta-llama-3-8b-instruct-otter_merge

Merge Details

Merge Method

Models Merged

Configuration

gsjang/
ja-llama-3-swallow-8b-instruct-v0.1-x-meta-llama-3-8b-instruct-otter_merge