MedQA_L3_1000steps_1e7rate_05beta_CSFTDPO

This model is a fine-tuned version of tsavage68/MedQA_L3_1000steps_1e6rate_SFT on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.5679
• Rewards/chosen: 0.9256
• Rewards/rejected: 0.5812
• Rewards/accuracies: 0.7407
• Rewards/margins: 0.3444
• Logps/rejected: -32.6925
• Logps/chosen: -29.4774
• Logits/rejected: -0.7357
• Logits/chosen: -0.7349

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: 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
0.6857	0.0489	50	0.6947	-0.0249	-0.0232	0.4879	-0.0018	-33.9011	-31.3784	-0.7318	-0.7312
0.6799	0.0977	100	0.6734	0.3881	0.3450	0.6681	0.0432	-33.1649	-30.5522	-0.7330	-0.7323
0.6275	0.1466	150	0.6484	0.5732	0.4639	0.6813	0.1093	-32.9271	-30.1822	-0.7310	-0.7303
0.5934	0.1954	200	0.6321	0.1707	0.0172	0.6989	0.1535	-33.8203	-30.9871	-0.7310	-0.7303
0.6358	0.2443	250	0.6181	0.4355	0.2501	0.7253	0.1854	-33.3546	-30.4574	-0.7315	-0.7308
0.5727	0.2931	300	0.6007	0.5633	0.3322	0.7429	0.2311	-33.1904	-30.2020	-0.7321	-0.7314
0.5786	0.3420	350	0.5923	0.7025	0.4439	0.7407	0.2586	-32.9670	-29.9235	-0.7343	-0.7335
0.545	0.3908	400	0.5830	0.9347	0.6493	0.7385	0.2854	-32.5562	-29.4591	-0.7336	-0.7328
0.5497	0.4397	450	0.5795	0.9735	0.6722	0.7385	0.3014	-32.5105	-29.3814	-0.7346	-0.7338
0.5857	0.4885	500	0.5781	1.0925	0.7817	0.7407	0.3108	-32.2914	-29.1435	-0.7356	-0.7348
0.5168	0.5374	550	0.5714	1.0244	0.6925	0.7385	0.3319	-32.4698	-29.2796	-0.7358	-0.7350
0.567	0.5862	600	0.5699	0.9715	0.6353	0.7407	0.3362	-32.5842	-29.3855	-0.7356	-0.7349
0.5375	0.6351	650	0.5689	0.9102	0.5695	0.7429	0.3407	-32.7158	-29.5081	-0.7357	-0.7349
0.5541	0.6839	700	0.5698	0.9277	0.5885	0.7385	0.3391	-32.6778	-29.4732	-0.7359	-0.7351
0.5824	0.7328	750	0.5693	0.9133	0.5709	0.7516	0.3424	-32.7129	-29.5019	-0.7358	-0.7350
0.5769	0.7816	800	0.5684	0.9103	0.5658	0.7429	0.3444	-32.7232	-29.5080	-0.7354	-0.7346
0.6223	0.8305	850	0.5678	0.9317	0.5868	0.7473	0.3449	-32.6812	-29.4651	-0.7360	-0.7352
0.5968	0.8793	900	0.5687	0.9231	0.5807	0.7385	0.3424	-32.6935	-29.4824	-0.7361	-0.7353
0.5673	0.9282	950	0.5678	0.9259	0.5813	0.7407	0.3446	-32.6921	-29.4767	-0.7357	-0.7349
0.4742	0.9770	1000	0.5679	0.9256	0.5812	0.7407	0.3444	-32.6925	-29.4774	-0.7357	-0.7349

Framework versions

• Transformers 4.41.1
• Pytorch 2.0.0+cu117
• Datasets 2.19.1
• Tokenizers 0.19.1