This model is a fine-tuned version of tsavage68/UTI_M2_1000steps_1e7rate_SFT on an unknown dataset. It achieves the following results on the evaluation set:

Loss: 2.8757
Rewards/chosen: -6.0873
Rewards/rejected: -4.3997
Rewards/accuracies: 0.0700
Rewards/margins: -1.6876
Logps/rejected: -18.1734
Logps/chosen: -16.7171
Logits/rejected: -2.6754
Logits/chosen: -2.6755

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: 2
eval_batch_size: 1
seed: 42
gradient_accumulation_steps: 2
total_train_batch_size: 4
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
lr_scheduler_type: cosine
lr_scheduler_warmup_steps: 100
training_steps: 750

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen
2.0624	0.3333	25	2.5226	-1.8449	-0.4090	0.0700	-1.4358	-10.1920	-8.2322	-2.5575	-2.5577
3.5044	0.6667	50	4.5890	-5.8971	-3.0050	0.0800	-2.8921	-15.3840	-16.3367	-1.8026	-1.8027
3.7907	1.0	75	2.9383	-6.0760	-3.7525	0.0300	-2.3235	-16.8790	-16.6946	-3.0195	-3.0194
2.0779	1.3333	100	3.1415	-5.1412	-2.7057	0.0400	-2.4355	-14.7854	-14.8249	-2.3865	-2.3865
2.1307	1.6667	125	4.6202	-7.0602	-3.8899	0.0700	-3.1703	-17.1538	-18.6629	-2.7484	-2.7482
2.7273	2.0	150	3.0326	-5.1418	-2.7562	0.0300	-2.3856	-14.8864	-14.8261	-2.0909	-2.0909
1.5353	2.3333	175	3.1462	-5.0136	-2.5758	0.0400	-2.4378	-14.5256	-14.5696	0.3331	0.3331
4.0743	2.6667	200	2.9971	-4.9664	-2.5983	0.0300	-2.3681	-14.5706	-14.4753	-0.0159	-0.0159
0.9986	3.0	225	3.0856	-4.9856	-2.5747	0.0300	-2.4109	-14.5234	-14.5137	0.3048	0.3048
1.6525	3.3333	250	3.1592	-5.0282	-2.5848	0.0400	-2.4433	-14.5436	-14.5988	0.4805	0.4805
1.9899	3.6667	275	3.1493	-5.0549	-2.6157	0.0400	-2.4392	-14.6055	-14.6523	0.4901	0.4901
1.4214	4.0	300	3.0891	-5.0107	-2.5984	0.0300	-2.4123	-14.5708	-14.5639	0.2933	0.2933
2.4027	4.3333	325	3.1322	-5.0026	-2.5711	0.0400	-2.4315	-14.5162	-14.5476	0.2898	0.2898
3.0815	4.6667	350	3.1138	-5.0100	-2.5863	0.0300	-2.4236	-14.5466	-14.5624	0.2376	0.2375
3.9086	5.0	375	3.0901	-5.0259	-2.6131	0.0300	-2.4128	-14.6001	-14.5942	0.2565	0.2565
1.9593	5.3333	400	3.1018	-5.0272	-2.6090	0.0300	-2.4182	-14.5919	-14.5968	0.2684	0.2684
2.4838	5.6667	425	3.0355	-4.9905	-2.6034	0.0300	-2.3871	-14.5807	-14.5234	0.1890	0.1890
3.7035	6.0	450	3.0602	-5.0100	-2.6110	0.0300	-2.3990	-14.5959	-14.5624	0.2694	0.2694
1.9548	6.3333	475	3.1280	-4.9982	-2.5683	0.0300	-2.4299	-14.5106	-14.5389	0.2830	0.2830
2.401	6.6667	500	3.0907	-4.9860	-2.5730	0.0300	-2.4129	-14.5200	-14.5144	0.2371	0.2371
3.021	7.0	525	3.0511	-4.8229	-2.4291	0.0300	-2.3938	-14.2321	-14.1882	-0.3978	-0.3978
3.0732	7.3333	550	3.0204	-4.5795	-2.2045	0.0300	-2.3750	-13.7831	-13.7015	-1.2656	-1.2657
1.1973	7.6667	575	3.0684	-4.5390	-2.1704	0.0300	-2.3686	-13.7148	-13.6205	-1.1867	-1.1870
2.8529	8.0	600	3.0303	-4.5060	-2.1298	0.0300	-2.3762	-13.6335	-13.5544	-1.7173	-1.7173
2.8072	8.3333	625	3.0587	-4.5594	-2.1832	0.0300	-2.3762	-13.7404	-13.6613	-1.2299	-1.2300
3.3903	8.6667	650	2.8325	-5.9445	-4.0748	0.0600	-1.8697	-17.5236	-16.4315	-2.8838	-2.8840
2.174	9.0	675	3.1512	-7.3312	-5.4478	0.0700	-1.8834	-20.2696	-19.2049	-2.8619	-2.8620
1.6297	9.3333	700	3.0978	-6.3597	-4.4688	0.0800	-1.8909	-18.3115	-17.2618	-2.6659	-2.6658
2.3962	9.6667	725	2.8899	-6.0876	-4.3898	0.0700	-1.6978	-18.1535	-16.7177	-2.6732	-2.6734
1.6021	10.0	750	2.8757	-6.0873	-4.3997	0.0700	-1.6876	-18.1734	-16.7171	-2.6754	-2.6755

Framework versions

Transformers 4.41.2
Pytorch 2.0.0+cu117
Datasets 2.19.2
Tokenizers 0.19.1

tsavage68/UTI2_M2_750steps_1e5rate_05beta_CSFTDPO

UTI2_M2_750steps_1e5rate_05beta_CSFTDPO

Model description

Intended uses & limitations

Training and evaluation data

Training procedure

Training hyperparameters

Training results

Framework versions

tsavage68/
UTI2_M2_750steps_1e5rate_05beta_CSFTDPO