sft-xcomet_xl_xxl-chosen-10lp-shuff-full-tiny64

This model is a fine-tuned version of TinyLlama/TinyLlama-1.1B-intermediate-step-1431k-3T on the Unbabel/TowerAligned-v0.1 dataset. It achieves the following results on the evaluation set:

• Loss: 0.7255
• Nll Loss: 0.7255
• Logps/best: -72.0598
• Rewards/chosen: 3.1294
• Rewards/rejected: 2.7165
• Rewards/accuracies: 0.6760
• Rewards/margins: 0.4130
• Logps/rejected: -70.2582
• Logps/chosen: -72.0598
• Logits/rejected: -1.7625
• Logits/chosen: -1.8900

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: 1e-05
• train_batch_size: 1
• eval_batch_size: 4
• seed: 42
• gradient_accumulation_steps: 64
• total_train_batch_size: 64
• optimizer: Adam with betas=(0.9,0.95) 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	Nll Loss	Logps/best	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen
0.7157	0.4254	100	0.7343	0.7343	-72.9338	3.0420	2.6428	0.6900	0.3992	-70.9947	-72.9338	-1.7591	-1.8861
0.7177	0.8508	200	0.7255	0.7255	-72.0598	3.1294	2.7165	0.6760	0.4130	-70.2582	-72.0598	-1.7625	-1.8900

Framework versions

• Transformers 4.43.3
• Pytorch 2.4.0+cu121
• Datasets 2.17.0
• Tokenizers 0.19.1