This model is a fine-tuned version of 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

DUAL-GPO-2/phi-2-gpo-renew2-i0

phi-2-gpo-renew2-i0

Model description

Intended uses & limitations

Training and evaluation data

Training procedure

Training hyperparameters

Training results

Framework versions

DUAL-GPO-2/
phi-2-gpo-renew2-i0