Pith. sign in

REVIEW 7 cited by

Jump to Conclusions: Short-Cutting Transformers With Linear Transformations

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 2303.09435 v2 pith:DVVDNV45 submitted 2023-03-16 cs.CL

classification cs.CL
keywords representationslayershiddenmethodfinalbertearlygpt-2
verification ladder T0 review T1 audit T2 compute T3 formal

Signed reviews

No signed human review yet.

0 comments
read the original abstract

Transformer-based language models create hidden representations of their inputs at every layer, but only use final-layer representations for prediction. This obscures the internal decision-making process of the model and the utility of its intermediate representations. One way to elucidate this is to cast the hidden representations as final representations, bypassing the transformer computation in-between. In this work, we suggest a simple method for such casting, using linear transformations. This approximation far exceeds the prevailing practice of inspecting hidden representations from all layers, in the space of the final layer. Moreover, in the context of language modeling, our method produces more accurate predictions from hidden layers, across various model scales, architectures, and data distributions. This allows "peeking" into intermediate representations, showing that GPT-2 and BERT often predict the final output already in early layers. We then demonstrate the practicality of our method to recent early exit strategies, showing that when aiming, for example, at retention of 95% accuracy, our approach saves additional 7.9% layers for GPT-2 and 5.4% layers for BERT. Last, we extend our method to linearly approximate sub-modules, finding that attention is most tolerant to this change. Our code and learned mappings are publicly available at https://github.com/sashayd/mat.

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 7 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Off-Axis, On Purpose: Where a Transformer Computes Concepts and Why it Does So

    cs.CL 2026-08 conditional novelty 7.0 of 10

    Intermediate transformer states sit far from the output axis on purpose: that position insulates attention's cross-token mixing, and the frame can be prescribed in advance without loss.

  2. How Transformers Reject Wrong Answers: Rotational Dynamics of Factual Constraint Processing

    cs.CL 2026-02 reject novelty 6.0 of 10

    Correct and incorrect single-token continuations of factual queries are separated by rotation of displacement vectors in transformer hidden states, with larger models also suppressing the correct token when forced to ...

  3. AdaDecode: Accelerating LLM Decoding with Adaptive Layer Parallelism

    cs.CL 2025-06 conditional novelty 6.0 of 10

    AdaDecode speeds up LLM generation by predicting tokens at early layers when confidence is high, running the skipped layers in parallel, and verifying the output exactly matches standard decoding.

  4. Discovering Chunks in Neural Embeddings for Interpretability

    cs.LG 2025-02 conditional novelty 5.0 of 10

    Recurring 'chunks' in neural embeddings can be extracted, predict input patterns, and be perturbed to steer a model's outputs.

  5. One Jump Is All You Need: Short-Cutting Transformers for Early Exit Prediction with One Jump to Fit All Exit Levels

    cs.LG 2025-04 conditional novelty 4.0 of 10

    Selecting one pre-trained low-rank shortcut jump and reusing it at every transformer block level gives over 30x parameter savings while largely matching per-level shortcuts in early-exit prediction.

  6. Look Ahead Text Understanding and LLM Stitching

    cs.CL 2024-12 conditional novelty 4.0 of 10

    LASI predicts the section label of a sentence before it appears, and stitching GPT to BERT with a loss or attention module beats plain BERT by about one accuracy point on PubMed abstracts.

  7. Linear Feedback Control Systems for Iterative Prompt Optimization in Large Language Models

    cs.LG 2025-01 reject novelty 3.0 of 10

    A conceptual framework that maps iterative prompt refinement onto PID feedback control, without empirical validation.

Pith tools