Quantization made by Richard Erkhov.
Bloom-1b7-glue-mrpc-Cont-IT-Step3 - bnb 4bits
Original model description:
license: bigscience-bloom-rail-1.0
base_model: alonzogarbanzo/Bloom-1b7-ropes-Cont-IT-Step2
tags:
- generated_from_trainer
model-index:
- name: Bloom-1b7-glue-mrpc-Cont-IT-Step3
results: []
Bloom-1b7-glue-mrpc-Cont-IT-Step3
This model is a fine-tuned version of
alonzogarbanzo/Bloom-1b7-ropes-Cont-IT-Step2 on an unknown 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: 5e-05
- train_batch_size: 1
- eval_batch_size: 8
- seed: 42
- gradient_accumulation_steps: 4
- total_train_batch_size: 4
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- num_epochs: 10
- mixed_precision_training: Native AMP
Training results
Final results: {'loss': 0.0741, 'grad_norm': 3.153198719024658, 'learning_rate': 3.0000000000000004e-07, 'epoch': 10.0}
Average results: {'train_runtime': 362.8198, 'train_samples_per_second': 5.512, 'train_steps_per_second': 1.378, 'train_loss': 0.5167060216665268, 'epoch': 10.0}
Framework versions
- Transformers 4.38.1
- Pytorch 2.2.0+cu121
- Datasets 2.17.0
- Tokenizers 0.15.2
