Merge Details

Merge Method

This model was merged using the DARE TIES merge method using NousResearch/Hermes-3-Llama-3.1-8B as a base.

Models Merged

The following models were included in the merge:

• cognitivecomputations/dolphin-2.9.4-llama3.1-8b
• arcee-ai/Llama-3.1-SuperNova-Lite
• deepseek-ai/DeepSeek-R1-Distill-Llama-8B

Configuration

The following YAML configuration was used to produce this model:

yaml
Copy code
# DARE-TIES merge recipe for 'anarva-8b' (beast model)
# Merges 4 premium Llama-3.1-8B models: Hermes-3, DeepSeek-R1-Distill-Llama, Dolphin-2.9.4, and SuperNova.
merge_method: dare_ties
base_model: NousResearch/Hermes-3-Llama-3.1-8B
dtype: bfloat16

models:
  - model: NousResearch/Hermes-3-Llama-3.1-8B
    parameters:
      density: 0.60
      weight: 1.0
  - model: deepseek-ai/DeepSeek-R1-Distill-Llama-8B
    parameters:
      density: 0.45
      weight: 0.40
  - model: cognitivecomputations/dolphin-2.9.4-llama3.1-8b
    parameters:
      density: 0.45
      weight: 0.35
  - model: arcee-ai/Llama-3.1-SuperNova-Lite
    parameters:
      density: 0.45
      weight: 0.30

parameters:
  normalize: true
  int8_mask: true

tokenizer:
  source: "union"
  tokens:
    "<|im_start|>":
      source: "cognitivecomputations/dolphin-2.9.4-llama3.1-8b"
    "<|im_end|>":
      source: "cognitivecomputations/dolphin-2.9.4-llama3.1-8b"
    "<think>":
      source: "deepseek-ai/DeepSeek-R1-Distill-Llama-8B"
      force: true