fnet-base-finetuned-mnli

This model is a fine-tuned version of google/fnet-base on the GLUE MNLI dataset. It achieves the following results on the evaluation set:

• Loss: 0.6443
• Accuracy: 0.7675

The model was fine-tuned to compare google/fnet-base as introduced in this paper against bert-base-cased.

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

This model is trained using the run_glue script. The following command was used:

#!/usr/bin/bash

python ../run_glue.py \\n  --model_name_or_path google/fnet-base \\n  --task_name mnli \\n  --do_train \\n  --do_eval \\n  --max_seq_length 512 \\n  --per_device_train_batch_size 16 \\n  --learning_rate 2e-5 \\n  --num_train_epochs 3 \\n  --output_dir fnet-base-finetuned-mnli \\n  --push_to_hub \\n  --hub_strategy all_checkpoints \\n  --logging_strategy epoch \\n  --save_strategy epoch \\n  --evaluation_strategy epoch \\n```

### Training hyperparameters

The following hyperparameters were used during training:
- learning_rate: 2e-05
- train_batch_size: 16
- eval_batch_size: 8
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- num_epochs: 3.0

### Training results

| Training Loss | Epoch | Step  | Validation Loss | Accuracy |
|:-------------:|:-----:|:-----:|:---------------:|:--------:|
| 0.7143        | 1.0   | 24544 | 0.6169          | 0.7504   |
| 0.5407        | 2.0   | 49088 | 0.6218          | 0.7627   |
| 0.4178        | 3.0   | 73632 | 0.6564          | 0.7658   |


### Framework versions

- Transformers 4.11.0.dev0
- Pytorch 1.9.0
- Datasets 1.12.1
- Tokenizers 0.10.3