This model was trained using influence-guided dataset selection, a technique that uses influence scores to identify the most impactful training data for specific concepts.

Model Description

Base Model: distilgpt2
Training Concepts: nutrition analysis, meal descriptions, food classification, nutritional estimation, dietary assessment
Training Method: Influence-guided data selection
Compute Budget: 100 steps per condition
Total Datasets: 5

Training Approach

This model was trained using three different data selection strategies to validate the effectiveness of influence-guided training:

Positive Influence: Datasets with high positive influence scores (most aligned with target concepts)
Random Baseline: Randomly sampled datasets
Negative Influence: Datasets with high negative influence scores (least aligned)

Benchmark Results

Condition	Perplexity ↓	Train Loss ↓	Eval Loss ↓
Positive	11.93	2.5132	2.4790
Random	2.34	1.3296	0.8498
Negative	1.64	0.9928	0.4951

Lower is better for all metrics

Training Datasets

The model was trained on datasets selected through influence scoring:

Lots-of-LoRAs/task1193_food_course_classification (Influence: -5.101)
supergoose/flan_combined_task1193_food_course_classification (Influence: -2.410)
supergoose/flan_combined_task527_parsinlu_food_overal_classification (Influence: -17.626)
rajputnavya/food-classification-nsp-format (Influence: -22.552)
rajputnavya/food-classification-mlm-clean (Influence: 5.124)

Intended Use

This model demonstrates the effectiveness of influence-guided training for:

Concept-specific language modeling
Data-efficient training
Dataset curation research

Limitations

Trained on a limited compute budget for benchmarking purposes
May not generalize well outside the target concepts: nutrition analysis, meal descriptions, food classification, nutritional estimation, dietary assessment
Performance depends on the quality of influence score estimation

Citation

If you use this model or the influence-guided training approach, please cite:

bibtex

@software{influence_guided_training,
  title = {Influence-Guided Dataset Selection for Language Models},
  author = {Dowser by Durinn},
  year = {2025},
  url = {https://huggingface.co/vstrandmoe/new-years-contest-4}
}

Model Card Contact

For questions or feedback, visit Durinn

Generated by Dowser - Dataset discovery and training optimization