jialicheng/unlearn_speech_commands_hubert-large_neggrad_8_42

superb_ks_42

This model is a fine-tuned version of facebook/hubert-large-ll60k on the superb dataset. It achieves the following results on the evaluation set:

• Loss: -9.0111
• Accuracy: 0.6209
• Test Accuracy: 0.6209
• Df Accuracy: 0.1395
• Unlearn Overall Accuracy: 0.7407
• Unlearn Time: 929.3382

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 5e-05
• train_batch_size: 32
• eval_batch_size: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 10.0

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy	Overall Accuracy	Unlearn Overall Accuracy	Time
No log	1.0	128	-9.0111	0.1395	0.7407	0.7407	-1
No log	2.0	256	-20.9958	0.1395	0.7407	0.7407	-1
No log	3.0	384	-35.4087	0.1395	0.7407	0.7407	-1
No log	4.0	512	-51.3901	0.1395	0.7407	0.7407	-1
No log	5.0	640	-67.7050	0.1395	0.7407	0.7407	-1
No log	6.0	768	-83.0873	0.1395	0.7407	0.7407	-1
No log	7.0	896	-96.3518	0.1395	0.7407	0.7407	-1
-113.6834	8.0	1024	-106.5309	0.1395	0.7407	0.7407	-1
-113.6834	9.0	1152	-112.9372	0.1395	0.7407	0.7407	-1
-113.6834	10.0	1280	-115.1336	0.1395	0.7407	0.7407	-1

Framework versions

• Transformers 4.39.3
• Pytorch 2.2.2+cu118
• Datasets 2.18.0
• Tokenizers 0.15.2