Eyas 17B

Overview

Eyas 17B is a frankenmerge based on the Falcon 3-10B architecture. Built using the mergekit library, Eyas 17B is optimized for a range of natural language processing tasks.

Merge Details

Merge Method

This model was created using the passthrough merge method. This method allows for a seamless integration of model layers to produce a new, high-performance model while maintaining compatibility with the Hugging Face transformers library.

Models Merged

The following models were included in the merge:

tiiuae/Falcon3-10B-Base

Configuration

The following YAML configuration was used to produce Eyas 17B:

slices:
- sources:
  - layer_range: [0, 10]
    model: tiiuae/Falcon3-10B-Base
- sources:
  - layer_range: [5, 15]
    model: tiiuae/Falcon3-10B-Base
- sources:
  - layer_range: [10, 20]
    model: tiiuae/Falcon3-10B-Base
- sources:
  - layer_range: [15, 25]
    model: tiiuae/Falcon3-10B-Base
- sources:
  - layer_range: [20, 30]
    model: tiiuae/Falcon3-10B-Base
- sources:
  - layer_range: [25, 35]
    model: tiiuae/Falcon3-10B-Base
- sources:
  - layer_range: [30, 40]
    model: tiiuae/Falcon3-10B-Base
merge_method: passthrough
dtype: float16