-
Notifications
You must be signed in to change notification settings - Fork 34
Description
Blurr is superfast for training huggingface models. Tried training a 1B param XNL-ROBERTA without it on free kaggle notebook and it took forever.
So I went looking over the web. Came about Blurr Framework! Goldmine for sure! But needs a slight update before you work with it
The steps below will help you excape the "BaseModelWrapper not working" error that took me a couple of hours today to solve.
STEP 1:
!pip install -q huggingface_hub["fastai"] -qq
! git clone https://github.com/ohmeow/blurr.git
STEP 2:
%cd /kaggle/working/blurr
!pip install -e ".[dev]" -qq
STEP 3
!pip uninstall -q numpy -y #downgrade
!pip install "numpy<2"
!pip uninstall -q datasets -y #downgrade
!pip install -q 'datasets<2.14.0' #datasets==2.16.0 #downgrade
!pip install -q seqeval
STEP4:
import torch
import transformers
from fastai.text.all import *
from blurr.text.data.all import *
from blurr.text.modeling.all import *
Good luck!