6.7b-ri-reproduce-combined-4-gpu-0-val
This model is a fine-tuned version of facebook/opt-6.7b on the ChaiML/dalio_combined_v1 dataset.
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: 9e-07
- train_batch_size: 1
- eval_batch_size: 8
- seed: 100
- distributed_type: multi-GPU
- num_devices: 4
- total_train_batch_size: 4
- total_eval_batch_size: 32
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: constant
- num_epochs: 15.0
Training results
Framework versions
- Transformers 4.25.0.dev0
- Pytorch 1.12.1+cu113
- Datasets 2.3.2
- Tokenizers 0.12.1