Instructions to use yorko/scibert_scivocab_uncased_long_4096 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use yorko/scibert_scivocab_uncased_long_4096 with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("fill-mask", model="yorko/scibert_scivocab_uncased_long_4096")# Load model directly from transformers import AutoTokenizer, AutoModelForMaskedLM tokenizer = AutoTokenizer.from_pretrained("yorko/scibert_scivocab_uncased_long_4096") model = AutoModelForMaskedLM.from_pretrained("yorko/scibert_scivocab_uncased_long_4096", device_map="auto") - Inference
- Notebooks
- Google Colab
- Kaggle
Speed
#1
by Borach - opened
What do you mean with 2.5 slower?
On a sequence of 512 it is 2.5 slower than scibert? Or a sequence of 4096 is 2.5 slower than scibert on 512?
Hi! sorry for missing this, I don't have numbers at hand anymore, but AFAIR that was the comparison for a sequence of 4096 for SciBERT long and 512 for SciBERT, so not that bad for the longformer version.
yorko changed discussion status to closed