This model is a fine-tuned version of lole25/phi-2-sft-ultrachat-lora on the HuggingFaceH4/ultrafeedback_binarized dataset. It achieves the following results on the evaluation set:

Loss: 0.0093
Rewards/chosen: -0.0154
Rewards/rejected: -0.0218
Rewards/accuracies: 0.3500
Rewards/margins: 0.0064
Logps/rejected: -96.3794
Logps/chosen: -93.2678
Logits/rejected: 0.7520
Logits/chosen: 0.7332

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
num_devices: 4
gradient_accumulation_steps: 4
total_train_batch_size: 64
total_eval_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: 2

Training results

Training Loss	Epoch	Step	Logits/chosen	Logits/rejected	Logps/chosen	Logps/rejected	Validation Loss	Rewards/accuracies	Rewards/chosen	Rewards/margins	Rewards/rejected
0.01	1.04	100	0.8011	0.8188	-91.7671	-94.2623	0.0100	0.25	-0.0004	0.0003	-0.0007
0.0098	0.42	200	0.0098	-0.0018	-0.0032	0.3060	0.0015	-94.5191	-91.9032	0.8107	0.7928
0.0095	0.63	300	0.0096	-0.0058	-0.0088	0.3060	0.0030	-95.0819	-92.3092	0.7982	0.7800
0.0091	0.84	400	0.0094	-0.0110	-0.0157	0.3340	0.0047	-95.7642	-92.8250	0.7753	0.7565
0.0094	1.05	500	0.0093	-0.0132	-0.0192	0.3400	0.0060	-96.1150	-93.0463	0.7679	0.7492
0.0093	1.26	600	0.0093	-0.0144	-0.0207	0.3440	0.0063	-96.2631	-93.1677	0.7578	0.7383
0.009	1.47	700	0.0093	-0.0152	-0.0212	0.3480	0.0060	-96.3198	-93.2491	0.7545	0.7355
0.009	1.67	800	0.0093	-0.0155	-0.0218	0.3420	0.0063	-96.3791	-93.2749	0.7523	0.7328
0.0091	1.88	900	0.0093	-0.0156	-0.0218	0.3480	0.0063	-96.3809	-93.2841	0.7515	0.7320

Framework versions

PEFT 0.7.1
Transformers 4.36.2
Pytorch 2.1.2+cu118
Datasets 2.14.6
Tokenizers 0.15.2

lole25/phi-2-gpo-ultrachat-lora-2

phi-2-gpo-ultrachat-lora-2

Model description

Intended uses & limitations

Training and evaluation data

Training procedure

Training hyperparameters

Training results

Framework versions

lole25/
phi-2-gpo-ultrachat-lora-2