IE_M2_1000steps_1e7rate_03beta_SFT

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: -0.4432
• Rewards/rejected: -6.7623
• Rewards/accuracies: 0.4600
• Rewards/margins: 6.3191
• Logps/rejected: -63.5627
• Logps/chosen: -43.6829
• Logits/rejected: -2.8851
• Logits/chosen: -2.8225

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
0.4682	0.4	50	0.3782	-0.1103	-2.3818	0.4600	2.2716	-48.9613	-42.5731	-2.9040	-2.8424
0.3812	0.8	100	0.3743	-0.3057	-5.2338	0.4600	4.9281	-58.4679	-43.2247	-2.8913	-2.8290
0.3119	1.2	150	0.3743	-0.4620	-6.2918	0.4600	5.8298	-61.9944	-43.7454	-2.8899	-2.8276
0.3639	1.6	200	0.3743	-0.4045	-6.1963	0.4600	5.7918	-61.6762	-43.5540	-2.8874	-2.8248
0.4332	2.0	250	0.3743	-0.4216	-6.3719	0.4600	5.9503	-62.2614	-43.6108	-2.8860	-2.8234
0.3986	2.4	300	0.3743	-0.4257	-6.4310	0.4600	6.0053	-62.4585	-43.6244	-2.8858	-2.8233
0.3986	2.8	350	0.3743	-0.4206	-6.4901	0.4600	6.0695	-62.6555	-43.6075	-2.8857	-2.8232
0.4505	3.2	400	0.3743	-0.4331	-6.5613	0.4600	6.1281	-62.8927	-43.6493	-2.8859	-2.8233
0.4505	3.6	450	0.3743	-0.4385	-6.6329	0.4600	6.1945	-63.1316	-43.6671	-2.8854	-2.8229
0.4332	4.0	500	0.3743	-0.4451	-6.6895	0.4600	6.2444	-63.3203	-43.6893	-2.8853	-2.8227
0.3292	4.4	550	0.3743	-0.4424	-6.7191	0.4600	6.2766	-63.4188	-43.6803	-2.8853	-2.8227
0.3639	4.8	600	0.3743	-0.4424	-6.7393	0.4600	6.2969	-63.4861	-43.6801	-2.8854	-2.8228
0.4505	5.2	650	0.3743	-0.4464	-6.7495	0.4600	6.3031	-63.5201	-43.6934	-2.8852	-2.8225
0.4505	5.6	700	0.3743	-0.4436	-6.7510	0.4600	6.3074	-63.5251	-43.6842	-2.8853	-2.8227
0.3639	6.0	750	0.3743	-0.4452	-6.7582	0.4600	6.3130	-63.5491	-43.6895	-2.8852	-2.8225
0.2426	6.4	800	0.3743	-0.4492	-6.7644	0.4600	6.3152	-63.5699	-43.7027	-2.8854	-2.8227
0.5025	6.8	850	0.3743	-0.4443	-6.7593	0.4600	6.3150	-63.5528	-43.6864	-2.8850	-2.8224
0.3119	7.2	900	0.3743	-0.4434	-6.7628	0.4600	6.3194	-63.5646	-43.6836	-2.8853	-2.8226
0.3466	7.6	950	0.3743	-0.4431	-6.7625	0.4600	6.3194	-63.5635	-43.6825	-2.8851	-2.8225
0.3812	8.0	1000	0.3743	-0.4432	-6.7623	0.4600	6.3191	-63.5627	-43.6829	-2.8851	-2.8225

Framework versions

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