This model is a fine-tuned version of tsavage68/mistralit2_1000_STEPS_5e7_SFT on an unknown dataset. It achieves the following results on the evaluation set:

Loss: 1.4255
Rewards/chosen: -5.6433
Rewards/rejected: -5.2981
Rewards/accuracies: 0.3758
Rewards/margins: -0.3452
Logps/rejected: -79.5378
Logps/chosen: -80.1048
Logits/rejected: -6.4077
Logits/chosen: -6.4074

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: 4
eval_batch_size: 1
seed: 42
gradient_accumulation_steps: 2
total_train_batch_size: 8
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
lr_scheduler_type: cosine
lr_scheduler_warmup_steps: 100
training_steps: 1000

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen
1.162	0.0977	50	1.0241	-1.1697	-0.9124	0.3604	-0.2573	-35.6813	-35.3689	-2.8702	-2.8700
1.8581	0.1953	100	1.6441	-6.0913	-6.0028	0.4242	-0.0885	-86.5851	-84.5852	-3.1044	-3.1044
1.601	0.2930	150	1.3641	-4.2549	-4.0426	0.4022	-0.2122	-66.9833	-66.2202	-3.2868	-3.2868
2.1307	0.3906	200	1.3342	-4.1303	-3.9079	0.4132	-0.2225	-65.6354	-64.9751	-4.3387	-4.3387
1.3343	0.4883	250	1.3463	-4.0915	-3.8831	0.4198	-0.2084	-65.3879	-64.5864	-4.3330	-4.3330
1.3841	0.5859	300	1.3320	-4.0505	-3.8262	0.4154	-0.2244	-64.8184	-64.1769	-4.5895	-4.5895
1.7277	0.6836	350	1.3279	-4.0165	-3.7899	0.4220	-0.2266	-64.4557	-63.8364	-4.6158	-4.6158
1.335	0.7812	400	1.3359	-4.0896	-3.8538	0.3978	-0.2358	-65.0948	-64.5675	-5.0034	-5.0034
1.5947	0.8789	450	1.3465	-4.1603	-3.9312	0.4000	-0.2291	-65.8693	-65.2751	-5.3332	-5.3332
1.7213	0.9766	500	1.3347	-4.1493	-3.9087	0.3912	-0.2407	-65.6435	-65.1649	-4.9376	-4.9376
1.3263	1.0742	550	1.3675	-4.4116	-4.1427	0.3912	-0.2689	-67.9843	-67.7880	-5.2901	-5.2901
1.5059	1.1719	600	1.4638	-5.1683	-4.8431	0.3868	-0.3253	-74.9878	-75.3551	-6.8721	-6.8717
1.2282	1.2695	650	1.4415	-5.0469	-4.7332	0.3846	-0.3137	-73.8891	-74.1404	-5.9456	-5.9454
1.3857	1.3672	700	1.3755	-4.5698	-4.2789	0.3912	-0.2909	-69.3454	-69.3694	-5.9554	-5.9552
1.3042	1.4648	750	1.3729	-4.5565	-4.2730	0.3956	-0.2835	-69.2869	-69.2369	-6.0268	-6.0267
1.2375	1.5625	800	1.3925	-4.9273	-4.6201	0.3934	-0.3072	-72.7581	-72.9446	-6.2783	-6.2781
1.3352	1.6602	850	1.4065	-5.2972	-4.9675	0.3736	-0.3297	-76.2315	-76.6436	-6.2527	-6.2524
1.1524	1.7578	900	1.4293	-5.6724	-5.3274	0.3736	-0.3451	-79.8304	-80.3960	-6.5300	-6.5297
1.2406	1.8555	950	1.4259	-5.6442	-5.2991	0.3736	-0.3451	-79.5474	-80.1136	-6.4213	-6.4210
1.7265	1.9531	1000	1.4255	-5.6433	-5.2981	0.3758	-0.3452	-79.5378	-80.1048	-6.4077	-6.4074

Framework versions

Transformers 4.40.1
Pytorch 2.0.0+cu117
Datasets 2.19.0
Tokenizers 0.19.1

tsavage68/Mistral2_1000_STEPS_01beta_1e5_CDPOSFT

Mistral2_1000_STEPS_01beta_CDPOSFT

Model description

Intended uses & limitations

Training and evaluation data

Training procedure

Training hyperparameters

Training results

Framework versions

tsavage68/
Mistral2_1000_STEPS_01beta_1e5_CDPOSFT