DatPySci/pythia-1b-self-kto-iter0

pythia-1b-kto-iter0-epoch1

This model is a fine-tuned version of DatPySci/pythia-1b-sft-full on the DatPySci/SPIN_iter0 dataset. It achieves the following results on the evaluation set:

• Loss: 0.3355
• Rewards/real: -0.0044
• Rewards/generated: -1.1080
• Rewards/accuracies: 0.9600
• Rewards/margins: 1.1036
• Logps/generated: -571.6245
• Logps/real: -468.8171
• Logits/generated: 0.2235
• Logits/real: -0.2848

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-07
• train_batch_size: 4
• eval_batch_size: 4
• seed: 42
• distributed_type: multi-GPU
• num_devices: 2
• gradient_accumulation_steps: 8
• total_train_batch_size: 64
• total_eval_batch_size: 8
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 1

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/real	Rewards/generated	Rewards/accuracies	Rewards/margins	Logps/generated	Logps/real	Logits/generated	Logits/real
0.4016	0.38	300	0.3914	0.0270	-0.8143	0.9320	0.8413	-568.6872	-468.5030	0.2573	-0.2517
0.3451	0.77	600	0.3355	-0.0044	-1.1080	0.9600	1.1036	-571.6245	-468.8171	0.2235	-0.2848

Framework versions

• Transformers 4.38.1
• Pytorch 2.2.1
• Datasets 2.17.1
• Tokenizers 0.15.2