<!-- This model card has been generated automatically according to the information the Trainer had access to. You should probably proofread and complete it, then remove this comment. -->
distilgpt2-nepali-patrakar-qa
This model is a fine-tuned version of Sakonii/distilgpt2-nepali on an unknown dataset. It achieves the following results on the evaluation set:
- Loss: 3.9077
Model description
Refer to original distilgpt2
Intended uses & limitations
This marginally fine-tuned model can be used for Nepali text generation and possibly question answering and intends to be fine-tuned on Nepali language focused generative downstream task. The language model being trained on a data with texts grouped to a block size of 512, it handles text sequence up to 512 tokens.
Training procedure
The model is trained with the same configuration as the original distilgpt2; but with 512 tokens per instance, 72 instances per batch, and around 34.14K training steps (excluding the pre-training with CLM Objective).
Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 5e-05
- train_batch_size: 72
- eval_batch_size: 72
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- num_epochs: 5
Training results
| Training Loss | Epoch | Step | Validation Loss |
|---|---|---|---|
| 4.1278 | 1.0 | 6829 | 4.0184 |
| 3.9461 | 2.0 | 13658 | 3.9630 |
| 3.8268 | 3.0 | 20487 | 3.9319 |
| 3.7978 | 4.0 | 27316 | 3.9140 |
| 3.7949 | 5.0 | 34145 | 3.9077 |
Framework versions
- Transformers 4.32.1
- Pytorch 2.0.0
- Datasets 2.1.0
- Tokenizers 0.13.3
