metadata

base_model:
  - v000000/MN-12B-Part1
  - v000000/MN-12B-Part2
library_name: transformers
tags:
  - mergekit
  - merge
  - mistral

Mistral-Nemo-12B-Estrella-v1

Untested! Mistral Instruct / ChatML format.

merge

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

Merge Details

Merge Method

This model was with a multi-step merge using the DELLA, DELLA_LINEAR and SLERP merge method.

Models Merged

The following models were included in the merge:

Configuration

The following YAML configuration was used to produce this model:

#Step 1
models:
  - model: Sao10K/MN-12B-Lyra-v1
    parameters:
      weight: 0.15
      density: 0.77
  - model: shuttleai/shuttle-2.5-mini
    parameters:
      weight: 0.20
      density: 0.78
  - model: anthracite-org/magnum-12b-v2
    parameters:
      weight: 0.35
      density: 0.85
  - model: nothingiisreal/MN-12B-Celeste-V1.9
    parameters:
      weight: 0.55
      density: 0.90
merge_method: della
base_model: Sao10K/MN-12B-Lyra-v1
parameters:
  int8_mask: true
  epsilon: 0.05
  lambda: 1
dtype: bfloat16
#Step 2
models:
  - model: BeaverAI/mistral-doryV2-12b
    parameters:
      weight: 0.10
      density: 0.4
  - model: unsloth/Mistral-Nemo-Instruct-2407
    parameters:
      weight: 0.20
      density: 0.4
  - model: UsernameJustAnother/Nemo-12B-Marlin-v5
    parameters:
      weight: 0.25
      density: 0.5
  - model: invisietch/Atlantis-v0.1-12B
    parameters:
      weight: 0.3
      density: 0.5
  - model: NeverSleep/Lumimaid-v0.2-12B
    parameters:
      weight: 0.4
      density: 0.8
merge_method: della_linear
base_model: anthracite-org/magnum-12b-v2
parameters:
  int8_mask: true
  epsilon: 0.05
  lambda: 1
dtype: bfloat16
#Step 3
slices:
  - sources:
      - model: v000000/MN-12B-Part2
        layer_range: [0, 40]
      - model: v000000/MN-12B-Part1
        layer_range: [0, 40]
merge_method: slerp
base_model: v000000/MN-12B-Part1
parameters: #smooth gradient prio part1
  t:
    - filter: self_attn
      value: [0, 0.5, 0.3, 0.7, 0.6, 0.1, 0.6, 0.3, 0.8, 0.5]
    - filter: mlp
      value: [0, 0.5, 0.4, 0.3, 0, 0.3, 0.4, 0.7, 0.2, 0.5]
    - value: 0.5
dtype: bfloat16