A Developer’s Guide to Compressing Pre-Trained Transformer Neural Networks Across Different Domains
Bibliographic record
Abstract
Transformer neural networks have become the benchmark across a variety of domains in part thanks to their competence in comprehending the importance and dependencies between large input segments. Pre-trained transformer models typically contain millions of parameters that have been trained with comprehensive datasets improving their generalization, allowing them to be fine-tuned quicker and with less data. Therefore, starting with a pre-trained model provides a substantial advantage over developing a network from the ground up. However, the race to exceed the accuracy of previous model generations leads pre-trained model to increase in size and the required computational resources for inference, as well as training. This is where compression techniques are vital to maximize the impact and performance of the models, while minimizing the cost to deploy. This work covers six of the most successful compression techniques including pruning, knowledge distillation, quantization, adaptive inference (early exit), low-rank factorization and mixture of experts. The study presents a comprehensive overview of their methods, strengths, and weaknesses meant to serve as a guide for machine learning practitioners. The key areas explored are model speedup, model size, accuracy retention, and training requirements.
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 | 0.012 |
| Meta-epidemiology (narrow) | 0.002 | 0.002 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.001 | 0.002 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.002 | 0.002 |
| Open science | 0.003 | 0.002 |
| Research integrity | 0.001 | 0.004 |
| Insufficient payload (model declined to judge) | 0.112 | 0.072 |
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".