kykim0/pythia-1b-tulu-v2-mix-uf-rm

pythia-1b-tulu-v2-mix-nos-rm

This model is a fine-tuned version of kykim0/pythia-1b-tulu-v2-mix-nos on the allenai/ultrafeedback_binarized_cleaned dataset. It achieves the following results on the evaluation set:

• Loss: 0.5227
• Accuracy: 0.7458

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: 1.41e-05
• train_batch_size: 2
• eval_batch_size: 8
• seed: 42
• distributed_type: multi-GPU
• num_devices: 4
• gradient_accumulation_steps: 4
• total_train_batch_size: 32
• total_eval_batch_size: 32
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 1

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy
0.5749	0.0527	100	0.5849	0.6900
0.5873	0.1055	200	0.5581	0.7100
0.5599	0.1582	300	0.5470	0.7212
0.5456	0.2109	400	0.5379	0.7258
0.521	0.2637	500	0.5358	0.7294
0.5361	0.3164	600	0.5363	0.7376
0.5662	0.3691	700	0.5270	0.7412
0.5301	0.4219	800	0.5268	0.7427
0.5661	0.4746	900	0.5301	0.7381
0.5608	0.5274	1000	0.5242	0.7437
0.5223	0.5801	1100	0.5242	0.7422
0.5322	0.6328	1200	0.5249	0.7448
0.4891	0.6856	1300	0.5241	0.7427
0.5111	0.7383	1400	0.5234	0.7437
0.5145	0.7910	1500	0.5225	0.7422
0.4746	0.8438	1600	0.5226	0.7458
0.5551	0.8965	1700	0.5223	0.7448
0.563	0.9492	1800	0.5222	0.7453

Framework versions

• Transformers 4.44.2
• Pytorch 2.1.2+cu121
• Datasets 2.14.6
• Tokenizers 0.19.1