zephyr-7b-dpo-full-ultrabin-low-bleu
This model is a fine-tuned version of alignment-handbook/zephyr-7b-sft-full on the HuggingFaceH4/ultrafeedback_binarized dataset. It achieves the following results on the evaluation set:
- Loss: 0.5708
- Rewards/chosen: -0.2797
- Rewards/rejected: -0.8021
- Rewards/accuracies: 0.7227
- Rewards/margins: 0.5224
- Logps/rejected: -342.8771
- Logps/chosen: -290.6019
- Logits/rejected: -0.9196
- Logits/chosen: -1.1155
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 |
---|---|---|---|---|---|---|---|---|---|---|---|
0.5746 | 0.3484 | 50 | 0.6150 | -0.0560 | -0.3281 | 0.7188 | 0.2721 | -295.4749 | -268.2298 | -2.4155 | -2.4561 |
0.478 | 0.6969 | 100 | 0.5756 | -0.3770 | -0.9031 | 0.7227 | 0.5261 | -352.9680 | -300.3292 | -0.8534 | -1.0395 |
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-low-bleu
Base model
mistralai/Mistral-7B-v0.1
Finetuned
alignment-handbook/zephyr-7b-sft-full