metadata
license: apache-2.0
base_model: mosaicml/mpt-7b
tags:
- generated_from_trainer
datasets:
- mozilla-foundation/common_voice_11_0
model-index:
- name: laft_kin_mpt
results: []
laft_kin_mpt
This model is a fine-tuned version of mosaicml/mpt-7b on the mozilla-foundation/common_voice_11_0 rw dataset. It achieves the following results on the evaluation set:
- Loss: 2.5256
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: 0.0003
- train_batch_size: 1
- eval_batch_size: 1
- seed: 42
- distributed_type: multi-GPU
- optimizer: Adam with betas=(0.9,0.95) and epsilon=1e-05
- lr_scheduler_type: cosine
- lr_scheduler_warmup_steps: 2000
- num_epochs: 6.0
Training results
| Training Loss | Epoch | Step | Validation Loss |
|---|---|---|---|
| 2.7031 | 1.0 | 21495 | 2.4057 |
| 2.3594 | 2.0 | 42990 | 2.3244 |
| 2.2812 | 3.0 | 64485 | 2.2475 |
| 2.1406 | 4.0 | 85980 | 2.1605 |
| 1.4062 | 5.0 | 107475 | 2.1943 |
| 1.0938 | 6.0 | 128970 | 2.5256 |
Framework versions
- Transformers 4.44.0
- Pytorch 2.4.0+cu121
- Datasets 2.20.0
- Tokenizers 0.19.1