[Built with Axolotl](https://github.com/Open Access-AI-Collective/axolotl)

medusa-1.0-ELYZA-japanese-Llama-2-7b-instruct

This model is a fine-tuned version of elyza/ELYZA-japanese-Llama-2-7b-instruct on the None dataset. It achieves the following results on the evaluation set:

  • Loss: 1.8838

Model description

This is a Medusa-1 created using Medusa.

Intended uses & limitations

Training and evaluation data

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

  • learning_rate: 0.0005
  • train_batch_size: 4
  • eval_batch_size: 4
  • seed: 42
  • gradient_accumulation_steps: 4
  • total_train_batch_size: 16
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: cosine
  • lr_scheduler_warmup_steps: 40
  • num_epochs: 2

Training results

Training Loss Epoch Step Validation Loss
2.4268 0.06 40 2.4129
2.204 0.11 80 2.2328
2.1341 0.17 120 2.1939
2.1774 0.23 160 2.1762
2.1331 0.28 200 2.1652
2.1485 0.34 240 2.1537
2.1608 0.4 280 2.1485
2.0998 0.45 320 2.1340
2.1478 0.51 360 2.1233
2.1374 0.57 400 2.1165
2.0657 0.62 440 2.0984
2.1227 0.68 480 2.0834
2.0573 0.74 520 2.0739
2.0501 0.79 560 2.0602
2.0664 0.85 600 2.0431
2.0231 0.91 640 2.0277
2.0263 0.96 680 2.0119
1.8635 1.02 720 1.9990
1.8844 1.07 760 1.9905
1.8585 1.13 800 1.9796
1.8018 1.19 840 1.9728
1.8134 1.24 880 1.9595
1.7795 1.3 920 1.9498
1.7603 1.36 960 1.9371
1.8302 1.41 1000 1.9262
1.7909 1.47 1040 1.9172
1.7787 1.53 1080 1.9083
1.7476 1.58 1120 1.9010
1.7897 1.64 1160 1.8945
1.7795 1.7 1200 1.8895
1.7329 1.75 1240 1.8864
1.7221 1.81 1280 1.8846
1.7624 1.87 1320 1.8840
1.793 1.92 1360 1.8838
1.7839 1.98 1400 1.8838

Framework versions

  • Transformers 4.34.1
  • Pytorch 2.1.2
  • Datasets 2.16.1
  • Tokenizers 0.14.1
Downloads last month
225
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for noguchis/medusa-1.0-ELYZA-japanese-Llama-2-7b-instruct

Quantized
(2)
this model