ProdeusUnity/Stellar-Odyssey-12b-v0.0

Stellar Odyssey 12b v0.0

Join my dream, it's just the right time, whoa... Leave it all behind... Get ready now... Riise up into my world~

Listen to the song on youtube: https://www.youtube.com/watch?v=npyiiInMA0w

This is my second attempt at a model merge, This time, these models were used

• mistralai/Mistral-Nemo-Base-2407
• Sao10K/MN-12B-Lyra-v4
• nothingiisreal/MN-12B-Starcannon-v2
• Gryphe/Pantheon-RP-1.5-12b-Nemo

License for this model is: Apache 2.0 (due to the base model, Mistral Nemo Base 2407)

Intended Use case: Roleplay

Instruction Format: ChatML

Thank you to AuriAetherwiing for helping me merge the models.

Data?

This is a hard question to answer, I didn't add any data to the model itself, rather it's a merge of other models, so the data used for them applies to this model too, though it won't be the same.

Merge Details

This model was merged using the della_linear merge method using mistralai/Mistral-Nemo-Base-2407 as a base.

Models Merged

The following models were included in the merge:

• Sao10K/MN-12B-Lyra-v4
• Gryphe/Pantheon-RP-1.5-12b-Nemo
• nothingiisreal/MN-12B-Starcannon-v2

Configuration

The following YAML configuration was used to produce this model:

models:
  - model: C:\Users\\Downloads\Mergekit-Fixed\mergekit\Sao10K_MN-12B-Lyra-v4
    parameters:
      weight: 0.3
      density: 0.25
  - model: C:\Users\\Downloads\Mergekit-Fixed\mergekit\nothingiisreal_MN-12B-Starcannon-v2
    parameters:
      weight: 0.1
      density: 0.4
  - model: C:\Users\\Downloads\Mergekit-Fixed\mergekit\Gryphe_Pantheon-RP-1.5-12b-Nemo
    parameters:
      weight: 0.4
      density: 0.5
merge_method: della_linear
base_model: C:\Users\\Downloads\Mergekit-Fixed\mergekit\mistralai_Mistral-Nemo-Base-2407
parameters:
  epsilon: 0.05
  lambda: 1
merge_method: della_linear
dtype: bfloat16

Notes

Della_Linear: Refer to https://arxiv.org/abs/2406.11617 and https://arxiv.org/abs/2212.04089, as it is quite long to explain what Della_Linear is BFloat16: Brain Floating Point 16, a way to run models faster on Nvidia GPUs Density: Fraction of weights in differences from the base model to retain Epsilon: Maximum change in drop probability based on magnitude. Drop probabilities assigned will range