gpt2-instruct-tuned-translator2

<!-- 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 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.

Model description

This model leverages a Stanford Alpaca style instruction tuning dataset, the format is as follows:

###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