Modelpublic

arnabdhar/distilbert-base-amazon-multi

sourceHugging Faceapache-2.0updated 3y agoView on Hugging Face

0likes34downloads

Model Card

distilbert-base-amazon-multi

This model is a fine-tuned version of distilbert-base-multilingual-cased on the mteb/amazonreviewsmulti dataset. It achieves the following results on the evaluation set:

—Loss: 0.9292
—Accuracy: 0.6055
—Matthews Correlation: 0.5072

Training procedure

This model was fine tuned on Google Colab using a single NVIDIA V100 GPU with 16GB of VRAM. It took around 13 hours to finish the finetuning of 10_000 steps.

Training hyperparameters

The following hyperparameters were used during training:

—learning_rate: 5e-05
—trainbatchsize: 32
—evalbatchsize: 320
—seed: 42
—optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
—lrschedulertype: cosine
—lrschedulerwarmup_ratio: 0.1
—training_steps: 100000

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy	Matthews Correlation
1.0008	0.26	10000	1.0027	0.5616	0.4520
0.9545	0.51	20000	0.9705	0.5810	0.4788
0.9216	0.77	30000	0.9415	0.5883	0.4868
0.8765	1.03	40000	0.9495	0.5891	0.4871
0.8837	1.28	50000	0.9254	0.5992	0.4997
0.8753	1.54	60000	0.9199	0.6014	0.5029
0.8572	1.8	70000	0.9108	0.6090	0.5117
0.7851	2.05	80000	0.9276	0.6052	0.5066
0.7918	2.31	90000	0.9292	0.6055	0.5072
0.793	2.57	100000	0.9288	0.6064	0.5084

Framework versions

—Transformers 4.35.2
—Pytorch 2.1.0+cu118
—Datasets 2.15.0
—Tokenizers 0.15.0