Why prune?
Even though Falcon-11B is trained on 5T tokens, it is still undertrained, as can be seen by this graph:
This is why the choice is made to prune 50% of the layers.
Note that ~1B of continued pre-training (~1M rows of 1k tokens) is still required to restore the perplexity of this model in the desired language.
I'm planning on doing that for certain languages, depending on how much compute will be available.
sliced
This is a merge of pre-trained language models created using mergekit.
Merge Details
Merge Method
This model was pruned using the passthrough merge method.
Models Merged
The following models were included in the merge:
Configuration
The following YAML configuration was used to produce this model:
slices:
- sources:
- model: tiiuae/falcon-11B
layer_range: [0, 24]
- sources:
- model: tiiuae/falcon-11B
layer_range: [55, 59]
merge_method: passthrough
dtype: bfloat16
PruneMe has been utilized using the wikimedia/wikipedia Spanish (es) subset by investigating layer similarity with 2000 samples. The layer ranges for pruning were determined based on this analysis to maintain performance while reducing model size.
Direct Use
Research on large language models; as a foundation for further specialization and finetuning for specific usecases (e.g., summarization, text generation, chatbot, etc.)
Out-of-Scope Use
Production use without adequate assessment of risks and mitigation; any use cases which may be considered irresponsible or harmful.
Bias, Risks, and Limitations
Falcon2-5.5B is trained mostly on English, but also German, Spanish, French, Italian, Portuguese, Polish, Dutch, Romanian, Czech, Swedish. It will not generalize appropriately to other languages. Furthermore, as it is trained on a large-scale corpora representative of the web, it will carry the stereotypes and biases commonly encountered online.
Recommendations
We recommend users of Falcon2-5.5B to consider finetuning it for the specific set of tasks of interest, and for guardrails and appropriate precautions to be taken for any production use.