ClaudioItaly/Fimbulvetr-40

Changing the value of kv_count from 34 to 40 indicates an increase in the number of key-value pairs in the model. These key-value pairs are mainly used to represent attention information within neural networks, particularly in Transformer-type models such as LLaMA.

merge

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

Merge Details

Merge Method

This model was merged using the passthrough merge method using Sao10K/Fimbulvetr-11B-v2 as a base.

Models Merged

The following models were included in the merge:

Configuration

The following YAML configuration was used to produce this model:

base_model:  Sao10K/Fimbulvetr-11B-v2 
merge_method: passthrough
dtype: float16
parameters:
  normalize: true

slices:
  - sources:
      - model: Sao10K/Fimbulvetr-11B-v2
        layer_range: [0, 48]  # Assumi che il modello abbia 48 layer
    densify:
      - linear
      - "rope:alpha=8192/4096"  # Estende il contesto a 8192

tokens:
  - source: Sao10K/Fimbulvetr-11B-v2
    mode: stretch