IE_L3_1000steps_1e5rate_03beta_SFT

This model is a fine-tuned version of tsavage68/IE_L3_1000steps_1e6rate_SFT on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.1802
• Rewards/chosen: -1.8241
• Rewards/rejected: -17.1487
• Rewards/accuracies: 0.7400
• Rewards/margins: 15.3246
• Logps/rejected: -132.7896
• Logps/chosen: -88.8782
• Logits/rejected: -0.8401
• Logits/chosen: -0.7195

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-05
• 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
0.1906	0.4	50	0.1802	-1.7929	-17.0298	0.7400	15.2369	-132.3931	-88.7740	-0.8398	-0.7194
0.1386	0.8	100	0.1802	-1.7771	-17.0460	0.7400	15.2689	-132.4471	-88.7214	-0.8398	-0.7192
0.1386	1.2	150	0.1802	-1.7982	-17.0858	0.7400	15.2876	-132.5800	-88.7917	-0.8401	-0.7193
0.1733	1.6	200	0.1802	-1.7978	-17.0381	0.7400	15.2403	-132.4209	-88.7903	-0.8396	-0.7190
0.2253	2.0	250	0.1802	-1.7877	-17.0275	0.7400	15.2398	-132.3854	-88.7567	-0.8395	-0.7189
0.1386	2.4	300	0.1802	-1.8012	-17.0499	0.7400	15.2487	-132.4602	-88.8018	-0.8399	-0.7195
0.1213	2.8	350	0.1802	-1.7983	-17.0687	0.7400	15.2705	-132.5230	-88.7921	-0.8395	-0.7189
0.1906	3.2	400	0.1802	-1.7995	-17.0794	0.7400	15.2799	-132.5586	-88.7960	-0.8403	-0.7193
0.1906	3.6	450	0.1802	-1.8034	-17.0941	0.7400	15.2908	-132.6077	-88.8090	-0.8399	-0.7193
0.2079	4.0	500	0.1802	-1.8158	-17.1281	0.7400	15.3123	-132.7209	-88.8505	-0.8397	-0.7185
0.156	4.4	550	0.1802	-1.8012	-17.1383	0.7400	15.3371	-132.7549	-88.8016	-0.8406	-0.7196
0.1213	4.8	600	0.1802	-1.7944	-17.0830	0.7400	15.2886	-132.5706	-88.7792	-0.8403	-0.7195
0.1906	5.2	650	0.1802	-1.7935	-17.1490	0.7400	15.3555	-132.7905	-88.7761	-0.8407	-0.7197
0.2426	5.6	700	0.1802	-1.7991	-17.1635	0.7400	15.3644	-132.8388	-88.7946	-0.8399	-0.7188
0.2599	6.0	750	0.1802	-1.7918	-17.1508	0.7400	15.3590	-132.7967	-88.7704	-0.8392	-0.7182
0.1213	6.4	800	0.1802	-1.8045	-17.1834	0.7400	15.3789	-132.9053	-88.8128	-0.8395	-0.7183
0.2426	6.8	850	0.1802	-1.8050	-17.1755	0.7400	15.3706	-132.8791	-88.8143	-0.8416	-0.7202
0.1733	7.2	900	0.1802	-1.7886	-17.1414	0.7400	15.3528	-132.7653	-88.7597	-0.8403	-0.7193
0.1386	7.6	950	0.1802	-1.8171	-17.1472	0.7400	15.3300	-132.7844	-88.8548	-0.8401	-0.7195
0.156	8.0	1000	0.1802	-1.8241	-17.1487	0.7400	15.3246	-132.7896	-88.8782	-0.8401	-0.7195

Framework versions

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