--- license: mit library_name: peft tags: - alignment-handbook - generated_from_trainer - trl - dpo base_model: microsoft/phi-2 datasets: - HuggingFaceH4/ultrafeedback_binarized model-index: - name: phi-2-gpo-renew2-i0 results: [] --- # phi-2-gpo-renew2-i0 This model is a fine-tuned version of [lole25/phi-2-sft-lora-ultrachat](https://huggingface.co/lole25/phi-2-sft-lora-ultrachat) on the HuggingFaceH4/ultrafeedback_binarized dataset. It achieves the following results on the evaluation set: - Loss: 0.0391 - Rewards/chosen: -0.0132 - Rewards/rejected: -0.0427 - Rewards/accuracies: 0.6330 - Rewards/margins: 0.0295 - Logps/rejected: -252.3540 - Logps/chosen: -280.1870 - Logits/rejected: 1.0400 - Logits/chosen: 0.9376 ## 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-06 - train_batch_size: 4 - eval_batch_size: 4 - seed: 42 - distributed_type: multi-GPU - gradient_accumulation_steps: 4 - total_train_batch_size: 16 - optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08 - lr_scheduler_type: cosine - lr_scheduler_warmup_ratio: 0.1 - num_epochs: 1 ### Training results | Training Loss | Epoch | Step | Logits/chosen | Logits/rejected | Logps/chosen | Logps/rejected | Validation Loss | Rewards/accuracies | Rewards/chosen | Rewards/margins | Rewards/rejected | |:-------------:|:-----:|:----:|:-------------:|:---------------:|:------------:|:--------------:|:---------------:|:------------------:|:--------------:|:---------------:|:----------------:| | 0.0659 | 0.03 | 100 | 0.9711 | 1.0635 | -277.5683 | -243.8923 | 0.0536 | 0.4745 | -0.0002 | 0.0005 | -0.0008 | | 0.0597 | 0.05 | 200 | 0.9688 | 1.0617 | -277.1979 | -243.9651 | 0.0518 | 0.5880 | 0.0035 | 0.0050 | -0.0015 | | 0.0564 | 0.08 | 300 | 0.9499 | 1.0440 | -276.5096 | -244.6272 | 0.0475 | 0.6175 | 0.0104 | 0.0185 | -0.0081 | | 0.0402 | 0.1 | 400 | 0.8995 | 0.9932 | -277.3771 | -246.9109 | 0.0438 | 0.6325 | 0.0017 | 0.0326 | -0.0309 | | 0.0421 | 0.13 | 500 | 0.8362 | 0.9295 | -281.6956 | -251.9139 | 0.0411 | 0.6195 | -0.0415 | 0.0395 | -0.0810 | | 0.0439 | 0.16 | 600 | 0.8607 | 0.9520 | -284.5547 | -255.5005 | 0.0395 | 0.6175 | -0.0701 | 0.0468 | -0.1168 | | 0.0363 | 0.18 | 700 | 0.8949 | 0.9895 | -281.1619 | -251.8926 | 0.0390 | 0.6310 | -0.0362 | 0.0446 | -0.0808 | | 0.0402 | 0.21 | 800 | 0.9001 | 0.9937 | -282.6901 | -253.8720 | 0.0382 | 0.6220 | -0.0514 | 0.0491 | -0.1006 | | 0.0381 | 0.24 | 900 | 0.9534 | 1.0465 | -283.0851 | -254.8047 | 0.0376 | 0.6315 | -0.0554 | 0.0545 | -0.1099 | | 0.0421 | 0.26 | 1000 | 0.9448 | 1.0399 | -281.6268 | -253.1114 | 0.0374 | 0.6270 | -0.0408 | 0.0522 | -0.0930 | | 0.0393 | 0.29 | 1100 | 0.9609 | 1.0557 | -283.3031 | -254.3491 | 0.0370 | 0.6285 | -0.0576 | 0.0478 | -0.1053 | | 0.0533 | 0.31 | 1200 | 0.9417 | 1.0368 | -283.6022 | -255.3544 | 0.0369 | 0.6210 | -0.0606 | 0.0548 | -0.1154 | | 0.0392 | 0.34 | 1300 | 0.9660 | 1.0634 | -279.6129 | -250.9576 | 0.0367 | 0.6120 | -0.0207 | 0.0508 | -0.0714 | | 0.0432 | 0.37 | 1400 | 0.9482 | 1.0463 | -279.0112 | -250.1082 | 0.0367 | 0.6260 | -0.0146 | 0.0483 | -0.0629 | | 0.0304 | 0.39 | 1500 | 0.9496 | 1.0471 | -282.7773 | -254.4339 | 0.0359 | 0.6360 | -0.0523 | 0.0539 | -0.1062 | | 0.0436 | 0.42 | 1600 | 0.9585 | 1.0586 | -280.7699 | -252.2616 | 0.0359 | 0.6340 | -0.0322 | 0.0522 | -0.0845 | | 0.0405 | 0.44 | 1700 | 0.9322 | 1.0312 | -282.8529 | -254.8697 | 0.0355 | 0.6335 | -0.0531 | 0.0575 | -0.1105 | | 0.0352 | 0.47 | 1800 | 0.9539 | 1.0533 | -281.2394 | -253.3721 | 0.0354 | 0.6220 | -0.0369 | 0.0586 | -0.0956 | | 0.0392 | 0.5 | 1900 | 0.9508 | 1.0498 | -280.3594 | -252.4193 | 0.0355 | 0.6210 | -0.0281 | 0.0579 | -0.0860 | | 0.0368 | 0.52 | 2000 | 0.9577 | 1.0563 | -279.8615 | -251.5159 | 0.0354 | 0.6300 | -0.0231 | 0.0539 | -0.0770 | | 0.0326 | 0.55 | 2100 | 0.9760 | 1.0751 | -281.1432 | -252.9630 | 0.0352 | 0.6300 | -0.0360 | 0.0555 | -0.0915 | | 0.0368 | 0.58 | 2200 | 0.9640 | 1.0642 | -281.4595 | -253.4691 | 0.0352 | 0.6345 | -0.0391 | 0.0574 | -0.0965 | | 0.0315 | 0.6 | 2300 | 0.9676 | 1.0685 | -280.0628 | -251.8242 | 0.0351 | 0.6330 | -0.0252 | 0.0549 | -0.0801 | | 0.0341 | 0.63 | 2400 | 0.9405 | 1.0420 | -279.9447 | -251.8426 | 0.0352 | 0.6320 | -0.0240 | 0.0563 | -0.0803 | | 0.0488 | 0.65 | 2500 | 0.9378 | 1.0394 | -280.7594 | -252.9968 | 0.0350 | 0.6340 | -0.0321 | 0.0597 | -0.0918 | | 0.0279 | 0.68 | 2600 | 0.9350 | 1.0361 | -281.3765 | -253.7721 | 0.0349 | 0.6315 | -0.0383 | 0.0613 | -0.0996 | | 0.0427 | 0.71 | 2700 | 0.9319 | 1.0336 | -280.6644 | -252.9290 | 0.0348 | 0.6310 | -0.0312 | 0.0600 | -0.0911 | | 0.0331 | 0.73 | 2800 | 0.9335 | 1.0354 | -280.4611 | -252.5369 | 0.0349 | 0.6290 | -0.0291 | 0.0581 | -0.0872 | | 0.0415 | 0.76 | 2900 | 0.9228 | 1.0248 | -280.5276 | -252.6469 | 0.0349 | 0.6315 | -0.0298 | 0.0585 | -0.0883 | | 0.0404 | 0.79 | 3000 | 0.9277 | 1.0305 | -280.2291 | -252.4009 | 0.0349 | 0.6295 | -0.0268 | 0.0590 | -0.0859 | | 0.0362 | 0.81 | 3100 | 0.9270 | 1.0296 | -280.1861 | -252.3079 | 0.0348 | 0.6305 | -0.0264 | 0.0585 | -0.0849 | | 0.0412 | 0.84 | 3200 | 0.9313 | 1.0338 | -280.2876 | -252.4237 | 0.0348 | 0.6260 | -0.0274 | 0.0587 | -0.0861 | | 0.0485 | 0.86 | 3300 | 0.9336 | 1.0359 | -279.9648 | -252.0546 | 0.0347 | 0.6270 | -0.0242 | 0.0582 | -0.0824 | | 0.0376 | 0.89 | 3400 | 0.9354 | 1.0377 | -280.1902 | -252.3589 | 0.0346 | 0.6310 | -0.0264 | 0.0590 | -0.0854 | | 0.0352 | 0.92 | 3500 | 0.9392 | 1.0418 | -280.2037 | -252.3726 | 0.0346 | 0.6260 | -0.0266 | 0.0590 | -0.0856 | | 0.0379 | 0.94 | 3600 | 0.9390 | 1.0414 | -280.1781 | -252.3377 | 0.0347 | 0.6315 | -0.0263 | 0.0589 | -0.0852 | | 0.0361 | 0.97 | 3700 | 0.9377 | 1.0399 | -280.2047 | -252.3741 | 0.0346 | 0.6310 | -0.0266 | 0.0590 | -0.0856 | | 0.0298 | 0.99 | 3800 | 0.9387 | 1.0412 | -280.1767 | -252.3201 | 0.0347 | 0.6275 | -0.0263 | 0.0587 | -0.0850 | ### Framework versions - PEFT 0.7.1 - Transformers 4.36.2 - Pytorch 2.1.2 - Datasets 2.14.6 - Tokenizers 0.15.2