This model is a fine-tuned version of tsavage68/chat_600STEPS_1e8rate_SFT on an unknown dataset. It achieves the following results on the evaluation set:

Loss: 0.9140
Rewards/chosen: -2.4513
Rewards/rejected: -2.9029
Rewards/accuracies: 0.4901
Rewards/margins: 0.4516
Logps/rejected: -47.8308
Logps/chosen: -41.2673
Logits/rejected: -0.3435
Logits/chosen: -0.3434

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
0.7203	0.0977	50	0.7044	-0.4487	-0.4880	0.4527	0.0393	-23.6818	-21.2416	-0.5577	-0.5575
0.8621	0.1953	100	0.8300	-1.1435	-1.1120	0.3802	-0.0314	-29.9224	-28.1895	-0.4308	-0.4305
0.8777	0.2930	150	0.8612	-1.0764	-1.0264	0.3648	-0.0499	-29.0661	-27.5182	-0.7212	-0.7205
0.8866	0.3906	200	0.8505	-0.4936	-0.4251	0.3912	-0.0684	-23.0532	-21.6902	0.8383	0.8386
0.8943	0.4883	250	0.9037	-1.0321	-0.9550	0.3736	-0.0771	-28.3517	-27.0753	2.2593	2.2595
0.8969	0.5859	300	0.8893	-0.9749	-0.8979	0.3714	-0.0771	-27.7806	-26.5040	1.0067	1.0072
0.9063	0.6836	350	0.9050	-1.2039	-1.1361	0.4044	-0.0679	-30.1626	-28.7939	0.7986	0.7986
0.7892	0.7812	400	0.8418	-1.0320	-1.0841	0.4330	0.0521	-29.6434	-27.0748	0.2635	0.2646
0.7866	0.8789	450	0.8482	-1.6282	-1.6894	0.4242	0.0612	-35.6958	-33.0367	0.4838	0.4849
0.8563	0.9766	500	0.8541	-1.6971	-1.7380	0.4132	0.0408	-36.1819	-33.7261	0.6780	0.6787
0.3046	1.0742	550	0.8749	-1.7613	-1.9274	0.4440	0.1661	-38.0758	-34.3676	0.6885	0.6888
0.3175	1.1719	600	0.9081	-2.1900	-2.4423	0.4615	0.2523	-43.2246	-38.6545	0.2508	0.2509
0.2851	1.2695	650	0.9462	-2.4571	-2.7801	0.4505	0.3231	-46.6032	-41.3252	-0.1026	-0.1026
0.4224	1.3672	700	0.9056	-2.0324	-2.3887	0.4637	0.3562	-42.6887	-37.0789	-0.1502	-0.1499
0.1386	1.4648	750	0.9267	-2.5222	-2.9198	0.4879	0.3975	-47.9997	-41.9771	-0.2790	-0.2789
0.2923	1.5625	800	0.9171	-2.4324	-2.8660	0.4813	0.4336	-47.4622	-41.0790	-0.3229	-0.3228
0.3088	1.6602	850	0.9140	-2.4368	-2.8842	0.4879	0.4474	-47.6445	-41.1227	-0.3429	-0.3428
0.2886	1.7578	900	0.9138	-2.4524	-2.9047	0.4879	0.4524	-47.8490	-41.2781	-0.3427	-0.3427
0.2677	1.8555	950	0.9151	-2.4517	-2.9016	0.4879	0.4499	-47.8176	-41.2713	-0.3438	-0.3437
0.2167	1.9531	1000	0.9140	-2.4513	-2.9029	0.4901	0.4516	-47.8308	-41.2673	-0.3435	-0.3434

Framework versions

Transformers 4.40.1
Pytorch 2.0.0+cu117
Datasets 2.19.0
Tokenizers 0.19.1

tsavage68/chat_1000_STEPS_01beta_1e5rate_CDPOSFT

chat_1000_STEPS_01beta_CDPOSFT

Model description

Intended uses & limitations

Training and evaluation data

Training procedure

Training hyperparameters

Training results

Framework versions

tsavage68/
chat_1000_STEPS_01beta_1e5rate_CDPOSFT