Na_M2_1000steps_1e8rate_05beta_cSFTDPO

This model is a fine-tuned version of tsavage68/Na_M2_1000steps_1e7_SFT on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.3285
• Rewards/chosen: 0.2983
• Rewards/rejected: -0.6864
• Rewards/accuracies: 1.0
• Rewards/margins: 0.9847
• Logps/rejected: -81.2962
• Logps/chosen: -47.5358
• Logits/rejected: -2.5349
• Logits/chosen: -2.5474

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-08
• 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: 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
0.6915	0.2667	50	0.6994	0.0060	0.0118	0.5400	-0.0058	-79.8998	-48.1204	-2.5353	-2.5479
0.6635	0.5333	100	0.6459	0.0371	-0.0697	0.7100	0.1068	-80.0629	-48.0583	-2.5354	-2.5480
0.5585	0.8	150	0.5484	0.1041	-0.2242	0.9400	0.3283	-80.3718	-47.9241	-2.5344	-2.5470
0.5041	1.0667	200	0.4568	0.1548	-0.4106	1.0	0.5654	-80.7446	-47.8228	-2.5349	-2.5475
0.4012	1.3333	250	0.3983	0.2253	-0.5152	1.0	0.7405	-80.9538	-47.6818	-2.5354	-2.5479
0.3304	1.6	300	0.3692	0.2306	-0.6109	1.0	0.8415	-81.1452	-47.6712	-2.5346	-2.5472
0.3396	1.8667	350	0.3524	0.2373	-0.6582	1.0	0.8955	-81.2397	-47.6578	-2.5349	-2.5474
0.3311	2.1333	400	0.3304	0.2656	-0.7177	1.0	0.9834	-81.3589	-47.6011	-2.5350	-2.5475
0.3099	2.4	450	0.3378	0.2807	-0.6665	1.0	0.9472	-81.2563	-47.5710	-2.5361	-2.5486
0.3384	2.6667	500	0.3271	0.2743	-0.7151	1.0	0.9894	-81.3535	-47.5838	-2.5349	-2.5474
0.3381	2.9333	550	0.3284	0.2854	-0.7005	1.0	0.9859	-81.3243	-47.5616	-2.5347	-2.5472
0.3328	3.2	600	0.3217	0.2963	-0.7183	1.0	1.0146	-81.3600	-47.5398	-2.5349	-2.5474
0.3162	3.4667	650	0.3252	0.3046	-0.6916	1.0	0.9962	-81.3066	-47.5232	-2.5358	-2.5483
0.2907	3.7333	700	0.3331	0.3002	-0.6711	1.0	0.9713	-81.2656	-47.5319	-2.5350	-2.5475
0.3052	4.0	750	0.3279	0.2998	-0.6877	1.0	0.9875	-81.2988	-47.5328	-2.5350	-2.5474
0.3264	4.2667	800	0.3285	0.2983	-0.6864	1.0	0.9847	-81.2962	-47.5358	-2.5349	-2.5474
0.3196	4.5333	850	0.3285	0.2983	-0.6864	1.0	0.9847	-81.2962	-47.5358	-2.5349	-2.5474
0.2962	4.8	900	0.3285	0.2983	-0.6864	1.0	0.9847	-81.2962	-47.5358	-2.5349	-2.5474
0.3115	5.0667	950	0.3285	0.2983	-0.6864	1.0	0.9847	-81.2962	-47.5358	-2.5349	-2.5474
0.3285	5.3333	1000	0.3285	0.2983	-0.6864	1.0	0.9847	-81.2962	-47.5358	-2.5349	-2.5474

Framework versions

• Transformers 4.44.2
• Pytorch 2.4.0+cu121
• Datasets 2.21.0
• Tokenizers 0.19.1