Full-bandwidth transformer

Hugging Face Daily Papers Papers

Summary

A new transformer variant, the full-bandwidth transformer, feeds back top-layer hidden states through a gated linear unit to improve reasoning and efficiency without altering the core architecture. Trained up to 400B tokens, it matches standard transformers trained with 1.5x more data while producing shorter reasoning traces.

Autoregressive transformers compute along two axes: horizontally across generated tokens, and vertically through model depth. Dense attention gives each token broad horizontal access to the past, but the vertical feedback channel between decoding steps remains narrow: only the sampled token returns to the bottom of the stack, while the top-layer hidden state is discarded. We introduce the full-bandwidth transformer, which widens this channel with latent feedback: at each decoding step, the previous top-layer hidden state is fused with the sampled token embedding through a gated linear unit and fed back as the next input. Latent feedback lets non-verbalized computation re-enter the stack with a renewed depth budget, while preserving the standard transformer architecture, KV cache, and language-modeling objective. To train full-bandwidth transformers without losing parallel teacher forcing, we use a scheduled multi-pass objective that introduces latent feedback late in pretraining and mixes a small fraction of deeper feedback passes for stability. We train 1B-parameter full-bandwidth transformers up to 400B tokens and find that latent feedback improves validation loss, 5-shot language-model evaluation, math and coding generation, and instruction-tuned performance. With negligible per-token decoding overhead, full-bandwidth transformers match or approach standard transformers trained with roughly 1.5times more tokens, and manage to produce shorter reasoning traces at equal or better accuracy.
Original Article
View Cached Full Text

Cached at: 08/14/26, 03:26 AM

Paper page - Full-bandwidth transformer

Source: https://huggingface.co/papers/2608.08888

Abstract

Full-bandwidth transformers use latent feedback of top-layer hidden states to improve reasoning and efficiency without altering the core architecture.

Autoregressive transformerscompute along two axes: horizontally across generated tokens, and vertically through model depth.Dense attentiongives each token broad horizontal access to the past, but the vertical feedback channel between decoding steps remains narrow: only the sampled token returns to the bottom of the stack, while the top-layer hidden state is discarded. We introduce thefull-bandwidth transformer, which widens this channel withlatent feedback: at each decoding step, the previous top-layer hidden state is fused with the sampled token embedding through agated linear unitand fed back as the next input.Latent feedbacklets non-verbalized computation re-enter the stack with a renewed depth budget, while preserving the standard transformer architecture,KV cache, and language-modeling objective. To trainfull-bandwidth transformers without losing parallelteacher forcing, we use ascheduled multi-pass objectivethat introduceslatent feedbacklate in pretraining and mixes a small fraction of deeper feedback passes for stability. We train 1B-parameterfull-bandwidth transformers up to 400B tokens and find thatlatent feedbackimproves validation loss, 5-shot language-model evaluation, math and coding generation, and instruction-tuned performance. With negligible per-token decoding overhead,full-bandwidth transformers match or approach standard transformers trained with roughly 1.5times more tokens, and manage to produce shorter reasoning traces at equal or better accuracy.

View arXiv pageView PDFAdd to collection

Get this paper in your agent:

hf papers read 2608\.08888

Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash

Models citing this paper0

No model linking this paper

Cite arxiv.org/abs/2608.08888 in a model README.md to link it from this page.

Datasets citing this paper0

No dataset linking this paper

Cite arxiv.org/abs/2608.08888 in a dataset README.md to link it from this page.

Spaces citing this paper0

No Space linking this paper

Cite arxiv.org/abs/2608.08888 in a Space README.md to link it from this page.

Collections including this paper0

No Collection including this paper

Add this paper to acollectionto link it from this page.

Similar Articles

Variable-Width Transformers

Hugging Face Daily Papers

Proposes a nonuniform width allocation transformer (hourglass shape) that outperforms uniform baselines in language modeling, reducing FLOPs and KV cache size.

@ZhihuFrontier: Half a year ago, a Zhihu contributor predicted that the next Transformer would absorb loops, recurrent state, sparse ro…

X AI KOLs Timeline

A Zhihu contributor's half-year-old prediction that the next Transformer would absorb loops, recurrent state, sparse routing, and latent reasoning is gaining relevance as Loop Engineering advances. The article explores how future Transformer architectures may evolve into hybrid models blending linear-complexity layers for background context with attention for precise reasoning, plus finer-grained sparsity and native System 2 reasoning.