mistralit2_1000_STEPS_1e7_rate_0.1_beta_DPO

This model is a fine-tuned version of mistralai/Mistral-7B-Instruct-v0.2 on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.5956
• Rewards/chosen: -1.5853
• Rewards/rejected: -2.2163
• Rewards/accuracies: 0.6308
• Rewards/margins: 0.6310
• Logps/rejected: -50.7358
• Logps/chosen: -39.2390
• Logits/rejected: -2.7784
• Logits/chosen: -2.7790

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-07
• 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
0.6877	0.1	50	0.6853	-0.0549	-0.0714	0.5846	0.0165	-29.2866	-23.9352	-2.8632	-2.8635
0.6422	0.2	100	0.6445	-0.6597	-0.8190	0.5758	0.1593	-36.7627	-29.9831	-2.8277	-2.8281
0.5395	0.29	150	0.6276	-1.2920	-1.6093	0.6000	0.3173	-44.6654	-36.3059	-2.8059	-2.8065
0.5656	0.39	200	0.6108	-1.0994	-1.4528	0.6000	0.3533	-43.1002	-34.3802	-2.8123	-2.8129
0.6317	0.49	250	0.5945	-0.9164	-1.3078	0.6176	0.3914	-41.6506	-32.5496	-2.8115	-2.8120
0.5648	0.59	300	0.6008	-1.5057	-1.9938	0.6198	0.4882	-48.5106	-38.4425	-2.8042	-2.8048
0.5332	0.68	350	0.6081	-1.6363	-2.1476	0.6154	0.5113	-50.0487	-39.7490	-2.7837	-2.7843
0.5852	0.78	400	0.5973	-1.4834	-2.0381	0.6330	0.5547	-48.9534	-38.2196	-2.7903	-2.7909
0.6266	0.88	450	0.5981	-1.4246	-1.9676	0.6374	0.5430	-48.2480	-37.6317	-2.7910	-2.7916
0.5184	0.98	500	0.5916	-1.2806	-1.8253	0.6286	0.5447	-46.8255	-36.1919	-2.7922	-2.7928
0.4736	1.07	550	0.5909	-1.3628	-1.9386	0.6330	0.5758	-47.9585	-37.0137	-2.7867	-2.7874
0.4708	1.17	600	0.5950	-1.4680	-2.0567	0.6330	0.5887	-49.1391	-38.0658	-2.7836	-2.7842
0.5232	1.27	650	0.5965	-1.5438	-2.1546	0.6308	0.6108	-50.1188	-38.8241	-2.7804	-2.7811
0.455	1.37	700	0.5976	-1.5823	-2.2032	0.6308	0.6209	-50.6042	-39.2085	-2.7793	-2.7799
0.4032	1.46	750	0.5958	-1.5721	-2.1999	0.6352	0.6278	-50.5717	-39.1070	-2.7788	-2.7795
0.4487	1.56	800	0.5957	-1.5857	-2.2157	0.6308	0.6300	-50.7295	-39.2429	-2.7785	-2.7791
0.5015	1.66	850	0.5972	-1.5836	-2.2125	0.6308	0.6289	-50.6972	-39.2220	-2.7785	-2.7791
0.419	1.76	900	0.5966	-1.5861	-2.2157	0.6308	0.6296	-50.7298	-39.2470	-2.7783	-2.7790
0.4581	1.86	950	0.5956	-1.5849	-2.2158	0.6308	0.6309	-50.7304	-39.2349	-2.7784	-2.7790
0.381	1.95	1000	0.5956	-1.5853	-2.2163	0.6308	0.6310	-50.7358	-39.2390	-2.7784	-2.7790

Framework versions

• Transformers 4.38.2
• Pytorch 2.0.0+cu117
• Datasets 2.18.0
• Tokenizers 0.15.2