OpenELM-1_1B-DPO-full-random-pair

This model was trained from scratch on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 164.4180
• Rewards/chosen: -560.0
• Rewards/rejected: -482.0
• Rewards/accuracies: 0.4277
• Rewards/margins: -76.0
• Logps/rejected: -48640.0
• Logps/chosen: -56320.0
• Logits/rejected: 3.1562
• Logits/chosen: 2.7031

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: 5e-05
• train_batch_size: 8
• eval_batch_size: 16
• seed: 42
• distributed_type: multi-GPU
• num_devices: 4
• gradient_accumulation_steps: 2
• total_train_batch_size: 64
• total_eval_batch_size: 64
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 3

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen
0.6914	0.1047	100	0.6966	-0.2227	-0.2197	0.4570	-0.0034	-310.0	-340.0	-9.875	-10.25
0.6914	0.2094	200	4.2809	-11.1875	-10.375	0.4316	-0.8555	-1328.0	-1440.0	-8.6875	-9.1875
0.6914	0.3141	300	95.7161	-324.0	-280.0	0.4258	-44.0	-28416.0	-32768.0	-2.5	-2.5156
0.6914	0.4188	400	99.5534	-338.0	-292.0	0.4277	-46.0	-29440.0	-34048.0	-2.625	-2.6406
0.6914	0.5236	500	103.5082	-352.0	-304.0	0.4277	-48.0	-30592.0	-35328.0	-1.9688	-2.0469
0.6914	0.6283	600	107.6879	-366.0	-316.0	0.4316	-50.0	-31872.0	-36864.0	-1.1328	-1.2656
0.6914	0.7330	700	111.8930	-380.0	-328.0	0.4297	-52.0	-33024.0	-38400.0	-0.5117	-0.7031
0.6914	0.8377	800	116.2988	-394.0	-340.0	0.4355	-54.0	-34304.0	-39680.0	1.3906	1.0781
0.6914	0.9424	900	120.7803	-410.0	-354.0	0.4316	-55.75	-35584.0	-41216.0	1.9062	1.5391
0.6914	1.0471	1000	125.1435	-424.0	-366.0	0.4355	-57.75	-36864.0	-42752.0	4.4688	3.9062
0.6914	1.1518	1100	129.3826	-440.0	-380.0	0.4316	-59.75	-38144.0	-44288.0	4.0312	3.5156
0.6914	1.2565	1200	133.6557	-454.0	-392.0	0.4297	-62.0	-39424.0	-45568.0	3.8438	3.2188
0.6914	1.3613	1300	137.5098	-466.0	-404.0	0.4355	-63.5	-40704.0	-46848.0	2.1094	1.7891
0.6914	1.4660	1400	141.6271	-482.0	-416.0	0.4355	-65.5	-41728.0	-48384.0	3.2344	2.7656
0.6914	1.5707	1500	145.0692	-492.0	-426.0	0.4336	-67.0	-42752.0	-49664.0	3.4844	2.9844
0.6914	1.6754	1600	148.4839	-504.0	-436.0	0.4297	-68.5	-43776.0	-50688.0	3.3281	2.8594
0.6914	1.7801	1700	151.1965	-512.0	-444.0	0.4316	-69.5	-44544.0	-51712.0	3.7031	3.2188
0.6914	1.8848	1800	154.0215	-524.0	-452.0	0.4336	-71.0	-45568.0	-52736.0	4.2188	3.6875
0.6914	1.9895	1900	156.4897	-532.0	-460.0	0.4316	-72.5	-46080.0	-53504.0	3.3125	2.875
0.6914	2.0942	2000	158.3665	-540.0	-466.0	0.4336	-73.0	-46848.0	-54016.0	3.1875	2.75
0.6914	2.1990	2100	160.3225	-544.0	-470.0	0.4297	-74.0	-47360.0	-54784.0	3.3438	2.8906
0.6914	2.3037	2200	161.6044	-548.0	-474.0	0.4316	-74.5	-47616.0	-55296.0	2.7344	2.3594
0.6914	2.4084	2300	162.5378	-552.0	-478.0	0.4316	-75.0	-48128.0	-55552.0	2.8281	2.4062
0.6914	2.5131	2400	163.3184	-556.0	-480.0	0.4336	-75.5	-48128.0	-55808.0	3.0	2.5469
0.6914	2.6178	2500	163.9196	-556.0	-482.0	0.4316	-75.5	-48384.0	-56064.0	3.1875	2.75
0.6914	2.7225	2600	164.2697	-556.0	-482.0	0.4297	-76.0	-48640.0	-56064.0	3.1719	2.7344
0.6914	2.8272	2700	164.3540	-560.0	-482.0	0.4297	-76.0	-48640.0	-56064.0	3.1562	2.7188
0.6914	2.9319	2800	164.4180	-560.0	-482.0	0.4277	-76.0	-48640.0	-56320.0	3.1562	2.7031

Framework versions

• Transformers 4.44.2
• Pytorch 2.3.0
• Datasets 2.21.0
• Tokenizers 0.19.1