ChrisToukmaji/focus_kin_llama_focus_trained

Paper and Citation

Paper: Prompt, Translate, Fine-Tune, Re-Initialize, or Instruction-Tune? Adapting LLMs for In-Context Learning in Low-Resource Languages

@misc{toukmaji2025prompttranslatefinetunereinitialize,
      title={Prompt, Translate, Fine-Tune, Re-Initialize, or Instruction-Tune? Adapting LLMs for In-Context Learning in Low-Resource Languages}, 
      author={Christopher Toukmaji and Jeffrey Flanigan},
      year={2025},
      eprint={2506.19187},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2506.19187}, 
}

focus_kin_llama_focus_trained

This model is a fine-tuned version of final_models/focus_kin_llama_after_focus_reinit on the mozilla-foundation/common_voice_11_0 rw dataset. It achieves the following results on the evaluation set:

• Loss: 5.0483

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: 0.0003
• train_batch_size: 1
• eval_batch_size: 1
• seed: 42
• distributed_type: multi-GPU
• optimizer: Adam with betas=(0.9,0.95) and epsilon=1e-05
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 2000
• num_epochs: 6.0

Training results

Training Loss	Epoch	Step	Validation Loss
4.8538	1.0	11772	5.1956
5.096	2.0	23544	4.9890
4.6339	3.0	35316	4.8349
3.7299	4.0	47088	4.6903
2.188	5.0	58860	4.8203
1.6435	6.0	70632	5.0483

Framework versions

• Transformers 4.44.0
• Pytorch 2.5.1+cu124
• Datasets 3.2.0
• Tokenizers 0.19.1