distilbert-base-amazon-multi

This model is a fine-tuned version of distilbert-base-multilingual-cased on the mteb/amazon_reviews_multi 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
train_batch_size: 32
eval_batch_size: 320
seed: 42
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
lr_scheduler_type: cosine
lr_scheduler_warmup_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