cpo-xcomet-xl_xxl-inc7b-10p-shuff-5e-7-full-tiny2

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: 2.5858
• Nll Loss: 0.9632
• Logps/best: -93.9459
• Rewards/chosen: -9.3946
• Rewards/rejected: -8.9636
• Rewards/accuracies: 0.4740
• Rewards/margins: -0.4310
• Logps/rejected: -89.6356
• Logps/chosen: -93.9459
• Logits/rejected: -1.8013
• Logits/chosen: -1.9355

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: 1
• eval_batch_size: 4
• seed: 42
• gradient_accumulation_steps: 16
• total_train_batch_size: 16
• optimizer: Adam with betas=(0.9,0.95) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 3

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
3.1908	0.5317	500	2.7136	1.0203	-99.1796	-9.9180	-9.3876	0.4600	-0.5304	-93.8759	-99.1796	-1.8188	-1.9550
2.7347	1.0635	1000	2.6365	0.9846	-95.9023	-9.5902	-9.1174	0.4720	-0.4728	-91.1739	-95.9023	-1.8087	-1.9438
2.5644	1.5952	1500	2.6035	0.9703	-94.5918	-9.4592	-9.0135	0.4680	-0.4456	-90.1355	-94.5918	-1.8043	-1.9388
2.6495	2.1270	2000	2.5883	0.9646	-94.0702	-9.4070	-8.9746	0.4720	-0.4324	-89.7462	-94.0702	-1.8018	-1.9361
2.4747	2.6587	2500	2.5858	0.9632	-93.9459	-9.3946	-8.9636	0.4740	-0.4310	-89.6356	-93.9459	-1.8013	-1.9355

Framework versions

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