Tarek07/Thalassic-Alpha-LLaMa-70B

After having some fair success with merging some of my favorite Llama models, particularly on Tarek07/Progenitor-V1.1-LLaMa-70B, which was based on the excellent Steelskull/L3.3-MS-Nevoria-70b merge, just with a couple extra ingredients and different merge methods. I then decided to try my hand at working the new deepseek-ai/DeepSeek-R1-Distill-Llama-70B in. I have decided to try mix it in 3 different ways. For this first model I decided to use Deepseek as the base, using the same formula as my Progenitor V1.1 model. From early testing it seems very promising.

merge

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

Merge Details

Merge Method

This model was merged using the Linear DELLA merge method using deepseek-ai/DeepSeek-R1-Distill-Llama-70B as a base.

Models Merged

The following models were included in the merge:

• EVA-UNIT-01/EVA-LLaMA-3.33-70B-v0.1
• SicariusSicariiStuff/Negative_LLAMA_70B
• Sao10K/70B-L3.3-Cirrus-x1
• Sao10K/L3.1-70B-Hanami-x1
• TheDrummer/Anubis-70B-v1

Configuration

The following YAML configuration was used to produce this model:

models:
  - model: Sao10K/L3.1-70B-Hanami-x1
    parameters:
      weight: 0.20
      density: 0.7
  - model: Sao10K/70B-L3.3-Cirrus-x1
    parameters:
      weight: 0.20
      density: 0.7
  - model: SicariusSicariiStuff/Negative_LLAMA_70B
    parameters:
      weight: 0.20
      density: 0.7
  - model: TheDrummer/Anubis-70B-v1
    parameters:
      weight: 0.20
      density: 0.7
  - model: EVA-UNIT-01/EVA-LLaMA-3.33-70B-v0.1
    parameters:
      weight: 0.20
      density: 0.7
merge_method: della_linear
base_model: deepseek-ai/DeepSeek-R1-Distill-Llama-70B
parameters:
  epsilon: 0.2
  lambda: 1.1
dtype: bfloat16
tokenizer_source: base