IE_M2_1000steps_1e7rate_05beta_cSFTDPO

This model is a fine-tuned version of tsavage68/IE_M2_1000steps_1e7rate_SFT on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.3743
• Rewards/chosen: -0.4914
• Rewards/rejected: -7.5827
• Rewards/accuracies: 0.4600
• Rewards/margins: 7.0912
• Logps/rejected: -56.1871
• Logps/chosen: -43.1884
• Logits/rejected: -2.8937
• Logits/chosen: -2.8314

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
0.4538	0.4	50	0.3745	-0.1879	-3.8972	0.4600	3.7093	-48.8163	-42.5814	-2.9040	-2.8425
0.3812	0.8	100	0.3743	-0.3333	-6.0415	0.4600	5.7082	-53.1048	-42.8721	-2.8978	-2.8359
0.3119	1.2	150	0.3743	-0.4581	-7.1508	0.4600	6.6927	-55.3234	-43.1217	-2.8957	-2.8335
0.3639	1.6	200	0.3743	-0.4596	-7.1592	0.4600	6.6996	-55.3402	-43.1247	-2.8956	-2.8335
0.4332	2.0	250	0.3743	-0.4698	-7.3147	0.4600	6.8449	-55.6512	-43.1451	-2.8944	-2.8322
0.3986	2.4	300	0.3743	-0.4645	-7.3358	0.4600	6.8713	-55.6934	-43.1344	-2.8945	-2.8322
0.3986	2.8	350	0.3743	-0.4758	-7.3590	0.4600	6.8832	-55.7398	-43.1571	-2.8947	-2.8325
0.4505	3.2	400	0.3743	-0.4808	-7.3913	0.4600	6.9105	-55.8044	-43.1671	-2.8944	-2.8321
0.4505	3.6	450	0.3743	-0.4859	-7.4793	0.4600	6.9934	-55.9805	-43.1774	-2.8942	-2.8319
0.4332	4.0	500	0.3743	-0.4895	-7.5333	0.4600	7.0438	-56.0884	-43.1845	-2.8937	-2.8314
0.3292	4.4	550	0.3743	-0.4880	-7.5663	0.4600	7.0782	-56.1543	-43.1815	-2.8938	-2.8316
0.3639	4.8	600	0.3743	-0.4870	-7.5730	0.4600	7.0860	-56.1677	-43.1795	-2.8936	-2.8313
0.4505	5.2	650	0.3743	-0.4897	-7.5693	0.4600	7.0796	-56.1604	-43.1849	-2.8935	-2.8312
0.4505	5.6	700	0.3743	-0.4895	-7.5788	0.4600	7.0893	-56.1795	-43.1845	-2.8938	-2.8316
0.3639	6.0	750	0.3743	-0.4877	-7.5842	0.4600	7.0965	-56.1901	-43.1808	-2.8935	-2.8312
0.2426	6.4	800	0.3743	-0.4987	-7.5876	0.4600	7.0889	-56.1971	-43.2030	-2.8938	-2.8316
0.5025	6.8	850	0.3743	-0.4942	-7.5824	0.4600	7.0882	-56.1866	-43.1939	-2.8937	-2.8314
0.3119	7.2	900	0.3743	-0.4890	-7.5862	0.4600	7.0972	-56.1942	-43.1835	-2.8936	-2.8314
0.3466	7.6	950	0.3743	-0.4914	-7.5827	0.4600	7.0912	-56.1871	-43.1884	-2.8937	-2.8314
0.3812	8.0	1000	0.3743	-0.4914	-7.5827	0.4600	7.0912	-56.1871	-43.1884	-2.8937	-2.8314

Framework versions

• Transformers 4.44.2
• Pytorch 2.0.0+cu117
• Datasets 3.0.0
• Tokenizers 0.19.1