Parameter Efficient Fine-tuning of Transformer-Based Language Models Using Dataset Pruning
Bibliographic record
Abstract
The widespread use of transformer-based language models is in part owed to their ease of adaptation to various tasks. Fine-tuning is a method of adapting pre-trained language models to a downstream task. The resource requirements for fine-tuning, although still less than pre-training, has been increasing due to the significant growth in the number of parameters of language models. Parameter efficient fine-tuning methods limit the set of model parameters that are updated during fine-tuning, leading to reductions in both memory usage and fine-tuning time. Dataset pruning is another method of efficient fine-tuning which removes training data points, thus reducing training time, while maintaining the evaluation performance of the fine-tuned model. In this work, we apply dataset pruning on top of parameter efficient fine-tuning to further reduce the hardware requirements of the fine-tuning. Our approach benefits from lower memory usage of parameter efficient methods while addressing their long fine-tuning time with dataset pruning. On average, our proposed method uses 22% of the fine-tuning dataset while updating only 0.5% of model parameters. As a result, while achieving an evaluation performance similar to full fine-tuning, our method reduces the peak memory usage of the fine-tuning by 40% and its wall clock time by 83%.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.001 | 0.008 |
| Meta-epidemiology (narrow) | 0.001 | 0.001 |
| Meta-epidemiology (broad) | 0.002 | 0.001 |
| Bibliometrics | 0.001 | 0.001 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.001 | 0.003 |
| Open science | 0.003 | 0.002 |
| Research integrity | 0.001 | 0.002 |
| Insufficient payload (model declined to judge) | 0.003 | 0.002 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".