sft-xcomet_xl_xxl-chosen-10lp-shuff-full-tiny3

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: 0.7148
• Nll Loss: 0.7148
• Logps/best: -71.0023
• Rewards/chosen: 3.2352
• Rewards/rejected: 2.8073
• Rewards/accuracies: 0.6780
• Rewards/margins: 0.4279
• Logps/rejected: -69.3502
• Logps/chosen: -71.0023
• Logits/rejected: -1.7526
• Logits/chosen: -1.8804

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: 32
• total_train_batch_size: 32
• 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
0.7347	0.2127	100	0.7451	0.7451	-73.9864	2.9368	2.5650	0.6820	0.3718	-71.7727	-73.9864	-1.7676	-1.8958
0.7192	0.4254	200	0.7245	0.7245	-71.9551	3.1399	2.7227	0.6760	0.4172	-70.1954	-71.9551	-1.7508	-1.8778
0.7184	0.6381	300	0.7170	0.7170	-71.2174	3.2137	2.7824	0.6800	0.4312	-69.5984	-71.2174	-1.7526	-1.8800
0.6793	0.8508	400	0.7148	0.7148	-71.0023	3.2352	2.8073	0.6780	0.4279	-69.3502	-71.0023	-1.7526	-1.8804

Framework versions

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