metadata
library_name: transformers
base_model: /scratch/gpfs/jg9904/saved_models/Mistral-7B-Instruct-v0.3
tags:
- alignment-handbook
- generated_from_trainer
datasets:
- >-
/scratch/gpfs/jg9904/unintentional-unalignment/data_files/data-mistral-7b-instruct-sppo-iter1/50_new
model-index:
- name: mistral-dpo-lr-5.0e-7-beta-0.01
results: []
mistral-dpo-lr-5.0e-7-beta-0.01
This model is a fine-tuned version of /scratch/gpfs/jg9904/saved_models/Mistral-7B-Instruct-v0.3 on the /scratch/gpfs/jg9904/unintentional-unalignment/data_files/data-mistral-7b-instruct-sppo-iter1/50_new dataset. It achieves the following results on the evaluation set:
- Loss: 0.4740
- Rewards/chosen: -0.4546
- Rewards/rejected: -1.1932
- Rewards/accuracies: 0.8036
- Rewards/margins: 0.7386
- Logps/rejected: -459.2464
- Logps/chosen: -346.3625
- Logits/rejected All: -2.7774
- Logits/chosen All: -2.7702
- Logits/rejected Sum: 8023.3535
- Logits/chosen Sum: 8554.5498
- Logits/rejected Avg: 21.6078
- Logits/chosen Avg: 21.0986
- Gradient/inner Product: 463470592.0
- Gradient/nabla Chosen Logps: 28288.0
- Gradient/nabla Rejected Logps: 37632.0
- Gradient/correlation: 0.4004
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: 4
- seed: 42
- distributed_type: multi-GPU
- num_devices: 8
- total_train_batch_size: 64
- total_eval_batch_size: 32
- 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 All | Logits/chosen All | Logits/rejected Sum | Logits/chosen Sum | Logits/rejected Avg | Logits/chosen Avg | Gradient/inner Product | Gradient/nabla Chosen Logps | Gradient/nabla Rejected Logps | Gradient/correlation |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| No log | 0 | 0 | 0.6931 | 0.0 | 0.0 | 0.0 | 0.0 | -339.9275 | -300.9012 | -2.8672 | -2.8605 | 7351.9551 | 7878.5537 | 19.8359 | 19.5574 | 86507520.0 | 16384.0 | 17152.0 | 0.2451 |
| 0.667 | 0.6803 | 100 | 0.4740 | -0.4546 | -1.1932 | 0.8036 | 0.7386 | -459.2464 | -346.3625 | -2.7774 | -2.7702 | 8023.3535 | 8554.5498 | 21.6078 | 21.0986 | 463470592.0 | 28288.0 | 37632.0 | 0.4004 |
Framework versions
- Transformers 4.45.0
- Pytorch 2.5.1+cu124
- Datasets 2.14.6
- Tokenizers 0.20.4