sft-dynamoguard

This model is a fine-tuned version of dynamofl/dynamo-1.6B-v0.4-mosaic-dynamoDPO-iter0-2978 on the dynamofl/toxicity-70k-april26 dataset. It achieves the following results on the evaluation set:

• Loss: 0.1604

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: 2e-05
• train_batch_size: 8
• eval_batch_size: 4
• seed: 42
• distributed_type: multi-GPU
• num_devices: 6
• gradient_accumulation_steps: 16
• total_train_batch_size: 768
• total_eval_batch_size: 24
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 20

Training results

Training Loss	Epoch	Step	Validation Loss
0.705	0.9711	21	0.5762
0.2553	1.9884	43	0.2231
0.2009	2.9595	64	0.1939
0.1822	3.9769	86	0.1838
0.1778	4.9942	108	0.1780
0.1735	5.9653	129	0.1740
0.1686	6.9827	151	0.1709
0.1662	8.0	173	0.1685
0.1632	8.9711	194	0.1666
0.1608	9.9884	216	0.1650
0.1588	10.9595	237	0.1637
0.1597	11.9769	259	0.1627
0.1571	12.9942	281	0.1620
0.1568	13.9653	302	0.1614
0.1553	14.9827	324	0.1610
0.1536	16.0	346	0.1607
0.1544	16.9711	367	0.1605
0.1553	17.9884	389	0.1605
0.1554	18.9595	410	0.1604
0.156	19.4220	420	0.1604

Framework versions

• Transformers 4.40.1
• Pytorch 2.3.0+cu121
• Datasets 2.19.0
• Tokenizers 0.19.1