|
--- |
|
library_name: transformers |
|
license: mit |
|
base_model: raghavbali/gpt2-finetuned-headliner |
|
tags: |
|
- generated_from_trainer |
|
model-index: |
|
- name: gpt2-instruct-tuned-translator2 |
|
results: [] |
|
--- |
|
|
|
<!-- 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. --> |
|
|
|
# GPT2 Instruction Tuned English To German Headline Translation Model |
|
|
|
- This model makes use of a english to german news headline translation dataset derived from [Harvard/abc-news-dataset](https://dataverse.harvard.edu/dataset.xhtml?persistentId=doi:10.7910/DVN/SYBGZL) for the task of instruction tuning |
|
- The dataset was derived using LLaMA3.1 and GPT4o models for generating the translations |
|
- This model is a fine-tuned version of [raghavbali/gpt2-finetuned-headliner](https://huggingface.co/raghavbali/gpt2-finetuned-headliner). |
|
|
|
## Model description |
|
|
|
This model leverages a Stanford Alpaca style instruction tuning dataset, the format is as follows: |
|
```md |
|
###Translate English Text to German:{text} ###Output: {translated_text} |
|
``` |
|
The format is slightly modified to reduce the additional tokens required for the instructions as GPT2 context size is very limited. |
|
The model is trained on small ~5k sample to showcase the impact of instruction tuning on overall alignment of the model towards requested task |
|
|
|
## Intended uses & limitations |
|
|
|
This is only for learning purposes. The model seems to have picked up German vocabulary as well as sentence structures to a good extent but the actual translations are at time grossly incorrect. |
|
The model also attempts at completing the news headlines given as prompt and has a high tendency to hallucinate. |
|
|
|
|
|
## Training procedure |
|
|
|
### Training hyperparameters |
|
|
|
The following hyperparameters were used during training: |
|
- learning_rate: 5e-05 |
|
- train_batch_size: 16 |
|
- eval_batch_size: 16 |
|
- seed: 42 |
|
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08 |
|
- lr_scheduler_type: linear |
|
- lr_scheduler_warmup_steps: 4 |
|
- num_epochs: 1 |
|
|
|
### Training results |
|
|
|
|
|
|
|
### Framework versions |
|
|
|
- Transformers 4.44.2 |
|
- Pytorch 2.4.0+cu121 |
|
- Datasets 2.21.0 |
|
- Tokenizers 0.19.1 |
|
|