Bit-Llama2-jp-123M
This model was built by referring to the config in TinyLlama/TinyLlama-1.1B-Chat-v1.0, 123M The model is a pre-trained Bit-Llama2 of Parameters with only 1 epoch on a Japanese dataset. The dataset used is range3/wiki40b-ja.
- Loss: 3.7091
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: 0.0005
- train_batch_size: 156
- eval_batch_size: 156
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: cosine
- lr_scheduler_warmup_steps: 500
- num_epochs: 1
Training results
| Training Loss | Epoch | Step | Validation Loss |
|---|---|---|---|
| 19.3793 | 0.04 | 1000 | 5.3113 |
| 5.0921 | 0.08 | 2000 | 4.9641 |
| 4.8154 | 0.12 | 3000 | 4.7104 |
| 4.6664 | 0.16 | 4000 | 4.5876 |
| 4.5545 | 0.2 | 5000 | 4.5258 |
| 4.4743 | 0.24 | 6000 | 4.4283 |
| 4.4061 | 0.28 | 7000 | 4.3539 |
| 4.3117 | 0.32 | 8000 | 4.2735 |
| 4.2433 | 0.36 | 9000 | 4.2243 |
| 4.2037 | 0.4 | 10000 | 4.1739 |
| 4.1576 | 0.44 | 11000 | 4.1266 |
| 4.0925 | 0.48 | 12000 | 4.0624 |
| 4.0615 | 0.52 | 13000 | 4.0433 |
| 4.0151 | 0.56 | 14000 | 3.9993 |
| 3.9721 | 0.6 | 15000 | 3.9721 |
| 3.941 | 0.64 | 16000 | 3.9185 |
| 3.9 | 0.68 | 17000 | 3.8841 |
| 3.8719 | 0.72 | 18000 | 3.8539 |
| 3.8376 | 0.76 | 19000 | 3.8189 |
| 3.8131 | 0.8 | 20000 | 3.7946 |
| 3.7801 | 0.84 | 21000 | 3.7739 |
| 3.7604 | 0.88 | 22000 | 3.7515 |
| 3.7413 | 0.92 | 23000 | 3.7365 |
| 3.7265 | 0.96 | 24000 | 3.7231 |
| 3.7152 | 1.0 | 25000 | 3.7091 |
Framework versions
- Transformers 4.36.2
- Pytorch 2.2.0+cu121
- Datasets 2.18.0
- Tokenizers 0.15.2