Falcon2-5.5B-Dutch / README.md
ssmits's picture
Update README.md
847e1c6 verified
|
raw
history blame
1.19 kB
metadata
license: apache-2.0
tags:
  - merge
  - mergekit
  - lazymergekit
  - tiiuae/falcon-11B
language:
  - nl

Falcon-5.5B-Dutch

Falcon-5.5B-Dutch is a pruned version of Falcon-11B using mergekit:

🧩 Configuration

slices:
  - sources:
      - model: tiiuae/falcon-11B
        layer_range: [0, 25]
  - sources:
      - model: tiiuae/falcon-11B
        layer_range: [56, 59]
            
merge_method: passthrough
dtype: bfloat16

PruneMe has been utilized using the AgentWaller/dutch-oasst1 dataset by investigating layer similarity with 4000 samples. The layer ranges for pruning were determined based on this analysis to maintain performance while reducing model size.

image/png

Note: This is a base language model and has not been optimized for conversational or chat applications. Further fine-tuning may be required to adapt it for specific use cases.