about

Original name : NexesMess/Llama_3.x_70b_Dolnemlimwhitessachi_v1.0

Release name : Dolmen v1.2

Replacing : https://huggingface.co/NexesMess/Llama_3.x_70b_Dolnemhertulimtess_v1.0 (Dolmen v1.0)

changes

OUT :

Tulu goes out, too "messy" and somehow "shackling" the model.
Hermes goes out, despite its quality, due to its tokenizer with a differt EOS token. I had too much problems with that model.

IN :

Hitachi FLDx2 enters, for intelligence and low perplexity (it's my new champion, PPL wikitext Eng 512 2.84 instead of 2.92), joining Tess 3 in those roles.
WhiteRabbitNeo enters, as a stabilizator.

feel

There's definitively progress. The prose is more structured, and the EOS is now working.

credits

Blackroot, for his own observations on his Mirai Series, and the hint about the Hitachi model, which I then tested with delight.

merge

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

Merge Details

Merge Method

This model was merged using the Model Stock merge method using Nexesenex/Llama_3.x_70b_L3.3_Dolphin_128K_v1.02 as a base.

Models Merged

The following models were included in the merge:

Configuration

The following YAML configuration was used to produce this model:

merge_method: model_stock
models:
  - model: Nexesenex/Llama_3.x_70b_L3.3_Dolphin_128K_v1.02
    parameters:
      weight: 1.0
  - model: huihui-ai/Llama-3.1-Nemotron-70B-Instruct-HF-abliterated
    parameters:
      weight: 1.0
  - model: huihui-ai/Tess-R1-Limerick-Llama-3.1-70B-abliterated
    parameters:
      weight: 1.0
  - model: WhiteRabbitNeo/Llama-3.1-WhiteRabbitNeo-2-70B
    parameters:
      weight: 1.0
  - model: migtissera/Tess-3-Llama-3.1-70B
    parameters:
      weight: 1.0
  - model: hitachi-nlp/Llama-3.1-70B-FLDx2 
    parameters:
      weight: 1.0
base_model: Nexesenex/Llama_3.x_70b_L3.3_Dolphin_128K_v1.02
dtype: bfloat16
out_dtype: bfloat16
parameters:
  int8_mask: true
  normalize: true
  rescale: false
  filter_wise: false
  smooth: false
  allow_negative_weights: false
chat_template: auto
tokenizer:
  source: union