zephyr-7b-dpo-full-ultrabin-reward-scale-01-random
This model is a fine-tuned version of alignment-handbook/zephyr-7b-sft-full on an unknown dataset. It achieves the following results on the evaluation set:
- Loss: 0.5400
- Rewards/chosen: -2.0899
- Rewards/rejected: -3.6849
- Rewards/accuracies: 0.7539
- Rewards/margins: 1.5951
- Logps/rejected: -631.1537
- Logps/chosen: -471.6162
- Logits/rejected: 4.0473
- Logits/chosen: 2.8493
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-07
- train_batch_size: 8
- eval_batch_size: 8
- seed: 55
- distributed_type: multi-GPU
- num_devices: 8
- gradient_accumulation_steps: 2
- total_train_batch_size: 128
- 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: 1
Training results
Training Loss | Epoch | Step | Validation Loss | Rewards/chosen | Rewards/rejected | Rewards/accuracies | Rewards/margins | Logps/rejected | Logps/chosen | Logits/rejected | Logits/chosen |
---|---|---|---|---|---|---|---|---|---|---|---|
0.6757 | 0.1046 | 50 | 0.6523 | 0.0185 | -0.0845 | 0.6914 | 0.1030 | -271.1101 | -260.7814 | -2.5756 | -2.6127 |
0.5878 | 0.2092 | 100 | 0.5948 | -1.0094 | -1.7751 | 0.7305 | 0.7657 | -440.1679 | -363.5702 | 0.4396 | 0.1394 |
0.5614 | 0.3138 | 150 | 0.5571 | -1.1386 | -2.1151 | 0.7344 | 0.9765 | -474.1727 | -376.4938 | 1.4578 | 0.8611 |
0.5467 | 0.4184 | 200 | 0.5479 | -1.3072 | -2.5365 | 0.75 | 1.2293 | -516.3164 | -393.3519 | 1.9356 | 0.9992 |
0.5401 | 0.5230 | 250 | 0.5436 | -1.2459 | -2.5137 | 0.7539 | 1.2678 | -514.0286 | -387.2157 | 2.2974 | 1.4064 |
0.5223 | 0.6276 | 300 | 0.5449 | -1.7241 | -3.1547 | 0.7461 | 1.4306 | -578.1356 | -435.0401 | 3.0431 | 2.1942 |
0.5133 | 0.7322 | 350 | 0.5299 | -1.6714 | -3.0561 | 0.7578 | 1.3847 | -568.2688 | -429.7706 | 2.9402 | 1.9451 |
0.5234 | 0.8368 | 400 | 0.5391 | -2.0819 | -3.6352 | 0.7539 | 1.5533 | -626.1800 | -470.8206 | 4.0879 | 2.9776 |
0.5253 | 0.9414 | 450 | 0.5400 | -2.0899 | -3.6849 | 0.7539 | 1.5951 | -631.1537 | -471.6162 | 4.0473 | 2.8493 |
Framework versions
- Transformers 4.44.0.dev0
- Pytorch 2.1.2
- Datasets 2.20.0
- Tokenizers 0.19.1
- Downloads last month
- 0
Model tree for sfulay/zephyr-7b-dpo-full-ultrabin-reward-scale-01-random
Base model
mistralai/Mistral-7B-v0.1
Finetuned
alignment-handbook/zephyr-7b-sft-full