Pith. sign in

REVIEW 16 cited by

Primer: Searching for Efficient Transformers for Language Modeling

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 2109.08668 v2 pith:4PVEPCN5 submitted 2021-09-17 cs.LG cs.AIcs.CLcs.NE

Primer: Searching for Efficient Transformers for Language Modeling

classification cs.LG cs.AIcs.CLcs.NE
keywords primertrainingtransformerlanguagecomputecostmodelingmodels
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved
0 comments
read the original abstract

Large Transformer models have been central to recent advances in natural language processing. The training and inference costs of these models, however, have grown rapidly and become prohibitively expensive. Here we aim to reduce the costs of Transformers by searching for a more efficient variant. Compared to previous approaches, our search is performed at a lower level, over the primitives that define a Transformer TensorFlow program. We identify an architecture, named Primer, that has a smaller training cost than the original Transformer and other variants for auto-regressive language modeling. Primer's improvements can be mostly attributed to two simple modifications: squaring ReLU activations and adding a depthwise convolution layer after each Q, K, and V projection in self-attention. Experiments show Primer's gains over Transformer increase as compute scale grows and follow a power law with respect to quality at optimal model sizes. We also verify empirically that Primer can be dropped into different codebases to significantly speed up training without additional tuning. For example, at a 500M parameter size, Primer improves the original T5 architecture on C4 auto-regressive language modeling, reducing the training cost by 4X. Furthermore, the reduced training cost means Primer needs much less compute to reach a target one-shot performance. For instance, in a 1.9B parameter configuration similar to GPT-3 XL, Primer uses 1/3 of the training compute to achieve the same one-shot performance as Transformer. We open source our models and several comparisons in T5 to help with reproducibility.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 16 Pith papers

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

  1. Can Transformers Really Do It All? On the Compatibility of Inductive Biases Across Tasks

    cs.LG 2026-07 conditional novelty 7.0

    Learned replacement non-linearities show transformers are rarely optimal for algorithmic tasks, with benefits that are task-specific, while language/code gains are smaller and more transferable.

  2. Bug or Feature$^2$: Weight Drift, Activation Sparsity and Spikes

    cs.LG 2026-05 accept novelty 7.0

    The paper proves negative weight drift at initialization under MSE or cross-entropy with asymmetric activations, links it to up to 90% sparsity in GPT-nano, maps the sparsity-accuracy cliff across 79 configurations, a...

  3. Bug or Feature$^2$: Weight Drift, Activation Sparsity and Spikes

    cs.LG 2026-05 accept novelty 7.0

    Standard losses induce negative weight drift with positive-biased activations, producing up to 90% sparsity in GPT-nano and an accuracy cliff above ~70% sparsity; clipped ReLU² and GELU² improve the tradeoff.

  4. From Competition to Collaboration: Designing Sustainable Mechanisms Between LLMs and Online Forums

    cs.AI 2026-02 unverdicted novelty 7.0

    A new sequential interaction framework lets LLMs propose questions to forums, with simulations on real Stack Exchange data showing players can reach roughly half the utility of an ideal full-information scenario despi...

  5. Fast Inference from Transformers via Speculative Decoding

    cs.LG 2022-11 accept novelty 7.0

    Speculative decoding accelerates exact sampling from large autoregressive models by 2-3x on T5-XXL by running smaller approximation models in parallel to propose token sequences that the large model then verifies in b...

  6. Flamingo: a Visual Language Model for Few-Shot Learning

    cs.CV 2022-04 unverdicted novelty 7.0

    Flamingo models reach new state-of-the-art few-shot results on image and video tasks by bridging frozen vision and language models with cross-attention layers trained on interleaved web-scale data.

  7. On the global convergence of gradient descent for wide shallow models with bounded nonlinearities

    math.OC 2026-05 unverdicted novelty 6.0

    Gradient descent on wide shallow models with bounded nonlinearities converges globally in the mean-field limit as non-global critical points are unstable under the dynamics.

  8. Three-Phase Transformer

    cs.CL 2026-04 unverdicted novelty 6.0

    Three-Phase Transformer partitions hidden states into N cyclic channels with phase-respecting RMSNorm and Givens rotations plus an orthogonal Gabriel's horn DC injection, delivering 7.2% lower perplexity and 1.93x fas...

  9. Physics of Language Models: Part 4.1, Architecture Design and the Magic of Canon Layers

    cs.CL 2025-12 conditional novelty 6.0

    Canon layers—residual 1-d causal convolutions over adjacent tokens—boost synthetic reasoning depth 2-4x, lift NoPE to RoPE level, and bring GLA up to Mamba2/GDN, with qualitative real-world confirmation.

  10. Resting Neurons, Active Insights: Robustifying Activation Sparsity in LLMs via Spontaneity

    cs.LG 2025-12 unverdicted novelty 6.0

    SPON adds learnable persistent activation anchors trained via distribution matching to restore LLM accuracy under high activation sparsity by preventing representational distribution shifts.

  11. ST-MoE: Designing Stable and Transferable Sparse Expert Models

    cs.CL 2022-02 unverdicted novelty 6.0

    ST-MoE introduces stability techniques for sparse expert models, allowing a 269B-parameter model to achieve state-of-the-art transfer learning results across reasoning, summarization, and QA tasks at the compute cost ...

  12. Mapping the Schedule x Bit-Width Boundary in Sub-100M Quantisation-Aware Training

    cs.LG 2026-05 unverdicted novelty 5.0

    Factorial experiments with over 1300 runs falsify the hypothesis that INT6 QAT needs a different LR schedule from higher precision and identify a 50M-parameter boundary for INT4 schedule sensitivity.

  13. Domyn-Small: A European 10B Reasoning Language Model

    cs.CL 2026-05 conditional novelty 5.0

    Domyn-Small is a 10B reasoning LLM that claims to deliver roughly one-third the inference tokens of Qwen3.5-9B at competitive accuracy, though results are marked as preliminary.

  14. ELAS: Efficient Pre-Training of Low-Rank Large Language Models via 2:4 Activation Sparsity

    cs.LG 2026-05 unverdicted novelty 5.0

    ELAS pre-trains low-rank LLMs by applying 2:4 activation sparsity after squared ReLU to cut memory and accelerate training with minimal performance loss.

  15. NVIDIA Nemotron 3: Efficient and Open Intelligence

    cs.CL 2025-12 unverdicted novelty 5.0

    NVIDIA releases the Nemotron 3 model family with hybrid Mamba-Transformer architecture, LatentMoE, NVFP4 training, MTP layers, and multi-environment RL post-training for reasoning and agentic tasks.

  16. Resting Neurons, Active Insights: Robustifying Activation Sparsity in LLMs via Spontaneity

    cs.LG 2025-12 unverdicted novelty 5.0

    SPON adds a small set of trainable input-independent activation vectors as representational anchors, trained by distribution matching, to stabilize sparse activation in LLMs and recover performance lost to hidden-stat...