Add Falcon support + benchmark results

[tomaarsen]

Hello!

Pull Request overview

• Add support for all Falcon models.

Details

As simple as

from attention_sinks import AutoModel

model = AutoModel.from_pretrained("tiiuae/falcon-7b", device_map="auto")

Benchmarks

python benchmark/perplexity.py --model_name_or_path tiiuae/falcon-7b --experiment attention_sinks --output_dir benchmark/outputs_falcon_7b
python benchmark/perplexity.py --model_name_or_path tiiuae/falcon-7b --experiment transformers --output_dir benchmark/outputs_falcon_7b  --num_tokens 5000
python benchmark/perplexity.py --model_name_or_path tiiuae/falcon-7b --experiment windowed --output_dir benchmark/outputs_falcon_7b

python benchmark/plot_perplexity.py --features perplexity vram --title "Log perplexity & VRAM usage of Falcon-7B as a function of input lengths" --output_dir benchmark/outputs_falcon_7b --log_perplexity_limit 4

• Tom Aarsen