chat_1000_STEPS_05beta_CDPOSFT
This model is a fine-tuned version of tsavage68/chat_600STEPS_1e8rate_SFT on an unknown dataset. It achieves the following results on the evaluation set:
- Loss: 1.9591
- Rewards/chosen: -2.3309
- Rewards/rejected: -2.7467
- Rewards/accuracies: 0.4703
- Rewards/margins: 0.4158
- Logps/rejected: -24.2954
- Logps/chosen: -21.4165
- Logits/rejected: 0.7996
- Logits/chosen: 0.8002
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: 4
- eval_batch_size: 1
- seed: 42
- gradient_accumulation_steps: 2
- total_train_batch_size: 8
- 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 |
---|---|---|---|---|---|---|---|---|---|---|---|
0.8761 | 0.0977 | 50 | 0.9959 | -1.8123 | -2.0646 | 0.4703 | 0.2523 | -22.9312 | -20.3793 | -0.5909 | -0.5908 |
1.3595 | 0.1953 | 100 | 1.6440 | -4.1891 | -4.1878 | 0.4000 | -0.0014 | -27.1775 | -25.1329 | 0.0807 | 0.0814 |
2.308 | 0.2930 | 150 | 1.9206 | -2.8317 | -2.4482 | 0.3560 | -0.3835 | -23.6984 | -22.4180 | -0.3052 | -0.3058 |
2.1704 | 0.3906 | 200 | 2.0223 | -1.4807 | -1.0616 | 0.3692 | -0.4191 | -20.9251 | -19.7160 | 0.1414 | 0.1417 |
1.9751 | 0.4883 | 250 | 2.0674 | -2.7530 | -2.4864 | 0.3824 | -0.2666 | -23.7748 | -22.2607 | 0.3122 | 0.3129 |
2.1664 | 0.5859 | 300 | 2.1396 | -3.3148 | -2.9537 | 0.3736 | -0.3610 | -24.7095 | -23.3841 | 0.7138 | 0.7139 |
2.3861 | 0.6836 | 350 | 2.4133 | -3.5559 | -3.0649 | 0.3868 | -0.4910 | -24.9319 | -23.8665 | 0.8230 | 0.8231 |
1.6234 | 0.7812 | 400 | 1.9885 | -2.7362 | -2.6712 | 0.4198 | -0.0650 | -24.1443 | -22.2270 | 0.0176 | 0.0179 |
2.1754 | 0.8789 | 450 | 2.0755 | -3.9759 | -3.8057 | 0.4044 | -0.1701 | -26.4135 | -24.7063 | -1.1567 | -1.1564 |
2.1709 | 0.9766 | 500 | 2.0516 | -2.8482 | -2.7256 | 0.4132 | -0.1227 | -24.2531 | -22.4511 | 0.0334 | 0.0339 |
0.4438 | 1.0742 | 550 | 1.9671 | -2.7066 | -2.6893 | 0.4154 | -0.0173 | -24.1807 | -22.1679 | 0.5746 | 0.5752 |
0.4123 | 1.1719 | 600 | 2.1253 | -2.7676 | -2.8681 | 0.4396 | 0.1005 | -24.5381 | -22.2898 | 0.6889 | 0.6894 |
0.4884 | 1.2695 | 650 | 2.0208 | -2.9445 | -3.1160 | 0.4484 | 0.1715 | -25.0340 | -22.6437 | 0.6377 | 0.6383 |
0.8103 | 1.3672 | 700 | 1.9313 | -2.2016 | -2.4993 | 0.4549 | 0.2977 | -23.8006 | -21.1578 | 0.7416 | 0.7422 |
0.6385 | 1.4648 | 750 | 1.9420 | -2.2243 | -2.5777 | 0.4593 | 0.3533 | -23.9573 | -21.2033 | 0.7943 | 0.7949 |
0.708 | 1.5625 | 800 | 1.9438 | -2.1910 | -2.5802 | 0.4527 | 0.3892 | -23.9624 | -21.1367 | 0.8128 | 0.8134 |
0.5451 | 1.6602 | 850 | 1.9599 | -2.3369 | -2.7540 | 0.4637 | 0.4171 | -24.3099 | -21.4285 | 0.8004 | 0.8010 |
0.5874 | 1.7578 | 900 | 1.9604 | -2.3346 | -2.7464 | 0.4615 | 0.4117 | -24.2947 | -21.4239 | 0.7988 | 0.7994 |
0.4059 | 1.8555 | 950 | 1.9599 | -2.3347 | -2.7487 | 0.4615 | 0.4141 | -24.2995 | -21.4240 | 0.7998 | 0.8003 |
0.2957 | 1.9531 | 1000 | 1.9591 | -2.3309 | -2.7467 | 0.4703 | 0.4158 | -24.2954 | -21.4165 | 0.7996 | 0.8002 |
Framework versions
- Transformers 4.40.1
- Pytorch 2.0.0+cu117
- Datasets 2.19.0
- Tokenizers 0.19.1
- Downloads last month
- 6
This model does not have enough activity to be deployed to Inference API (serverless) yet. Increase its social
visibility and check back later, or deploy to Inference Endpoints (dedicated)
instead.
Model tree for tsavage68/chat_1000_STEPS_05beta_1e5rate_CDPOSFT
Base model
meta-llama/Llama-2-7b-chat-hf
Finetuned
tsavage68/chat_600STEPS_1e8rate_SFT