Pith. sign in

REVIEW 13 cited by

Well-Read Students Learn Better: On the Importance of Pre-training Compact Models

Not yet reviewed by Pith; the record is open.

This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.

SPECIMEN: schema-true, not a live event

T0 review · schema-true

One-sentence machine reading of the paper's core claim.

pith:XXXXXXXX · record.json · timestamp

arxiv 1908.08962 v2 pith:RWZNSQPC submitted 2019-08-23 cs.CL

classification cs.CL
keywords modelspre-trainedpre-trainingbeencompactdistillationdataexplore
verification ladder T0 review T1 audit T2 compute T3 formal
0 comments
read the original abstract

Recent developments in natural language representations have been accompanied by large and expensive models that leverage vast amounts of general-domain text through self-supervised pre-training. Due to the cost of applying such models to down-stream tasks, several model compression techniques on pre-trained language representations have been proposed (Sun et al., 2019; Sanh, 2019). However, surprisingly, the simple baseline of just pre-training and fine-tuning compact models has been overlooked. In this paper, we first show that pre-training remains important in the context of smaller architectures, and fine-tuning pre-trained compact models can be competitive to more elaborate methods proposed in concurrent work. Starting with pre-trained compact models, we then explore transferring task knowledge from large fine-tuned models through standard knowledge distillation. The resulting simple, yet effective and general algorithm, Pre-trained Distillation, brings further improvements. Through extensive experiments, we more generally explore the interaction between pre-training and distillation under two variables that have been under-studied: model size and properties of unlabeled task data. One surprising observation is that they have a compound effect even when sequentially applied on the same data. To accelerate future research, we will make our 24 pre-trained miniature BERT models publicly available.

Discussion (0). Sign in to comment.

Forward citations

Cited by 13 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. OpenAlex reports about 429 citations worldwide. Full citation record

  1. Transporting Task Vectors across Different Architectures without Training

    cs.LG 2026-02 unverdicted novelty 7.0 of 10

    Theseus transports task vectors across models of different widths by functionally matching activations via orthogonal Procrustes alignment, yielding a closed-form solution that improves over baselines without training.

  2. Basis Transformers for Multi-Task Tabular Regression

    cs.LG 2025-06 conditional novelty 7.0 of 10

    Basis transformers beat fine-tuned LLMs on 34 multi-task tabular regression datasets while using five times fewer parameters and no data preprocessing.

  3. Learn from A Rationalist: Distilling Intermediate Interpretable Rationales

    cs.LG 2026-01 conditional novelty 6.0 of 10

    Distilling a teacher rationale-extraction model's feature selections and predictions into smaller students improves student accuracy by up to ~14 points on CIFAR-10 while keeping the same rationale sparsity.

  4. Mitigating Attention Localization in Small Scale: Self-Attention Refinement via One-step Belief Propagation

    cs.CL 2025-09 conditional novelty 6.0 of 10

    A one-step belief propagation refinement with a repulsive Potts prior increases attention entropy and improves downstream accuracy of small Transformers, with GTD as a diagnostic for multi-hop attention.

  5. GenRecal: Generation after Recalibration from Large to Small Vision-Language Models

    cs.CL 2025-06 conditional novelty 6.0 of 10

    A learnable Recalibrator bridges different tokenizers so that small VLMs can distill knowledge from any large VLM, improving their benchmark scores.

  6. Clustering and Median Aggregation Improve Differentially Private Inference

    cs.LG 2025-06 conditional novelty 6.0 of 10

    Clustering seed texts and privately aggregating median token logits improves representativeness and reduces reported privacy cost for DP synthetic text generation.

  7. Team DACTYL at PAN 2026: Bayesian Data Mixing and Empirical X-risk Minimization for AI-text Detection

    cs.CL 2026-07 conditional novelty 5.0 of 10

    A Bayesian-uncertainty text filter, partial-AUROC training, and MCGrad calibration produce the second-ranked AI-text detector (0.974 mean score) on the PAN 2026 leaderboard.

  8. The Impact of Background Speech on Interruption Detection in Collaborative Groups

    cs.CL 2025-07 conditional novelty 5.0 of 10

    Interruption detection in small-group dialogue remains accurate in simulated multi-group background noise when acoustic and textual features are combined, while an overlap-based heuristic collapses.

  9. ReqBrain: Task-Specific Instruction Tuning of LLMs for AI-Assisted Requirements Generation

    cs.SE 2025-05 conditional novelty 5.0 of 10

    ReqBrain, a LoRA-fine-tuned Zephyr-7b-beta model, produces software requirements that human evaluators could not reliably tell apart from human-authored ones, with automatic metrics favoring it over untuned ChatGPT-4o.

  10. X-ARES: A Comprehensive Framework for Assessing Audio Encoder Performance

    cs.SD 2025-05 conditional novelty 5.0 of 10

    X-ARES evaluates 13 audio encoders on 22 speech, sound, and music tasks using linear probing and nearest-neighbor classifiers, revealing strong domain-dependent performance differences.

  11. Compressed Models are NOT Trust-equivalent to Their Large Counterparts

    cs.CL 2025-08 conditional novelty 4.0 of 10

    Compressed BERT models share at most 67% of their top decision features with BERT-base and show different calibration profiles even at similar accuracy, so accuracy parity does not ensure trust-equivalence.

  12. Semantic Convergence: Investigating Shared Representations Across Scaled LLMs

    cs.CL 2025-07 conditional novelty 4.0 of 10

    Gemma-2-2B and Gemma-2-9B align most strongly on SAE-derived features in middle layers, with preliminary evidence for shared multi-token concept subspaces.

  13. QFFN-BERT: An Empirical Study of Depth, Performance, and Data Efficiency in Hybrid Quantum-Classical Transformers

    cs.CL 2025-07 reject novelty 4.0 of 10

    Replacing the feed-forward networks of a tiny BERT with four-qubit quantum circuits yielded 81.19% versus 79.59% on SST-2, but the result rests on a single seed and a CLS-only circuit application.

Pith tools