# GBLA: reading long recommendation histories efficiently

A technical and patent-practice reading of gated bidirectional linear attention, its hybrid encoder and the limits of the reported speed and retrieval results.

Source: https://www.iplexlaw.co.kr/en/blog/gbla-generative-retrieval

HOME / NEWS & INSIGHTS NEWS & INSIGHTS GBLA: reading long recommendation histories efficiently A technical and patent-practice reading of gated bidirectional linear attention, its hybrid encoder and the limits of the reported speed and retrieval results. Artificial Intelligence 2026.10.27 published IPLEX 12 min read The research problem Gated Bidirectional Linear Attention for Generative Retrieval, by Artem Matveev, Vladislav Tytskiy, Sergei Makeev and Sergei Liamaev of Yandex, studies efficient use of long interaction histories. This analysis follows arXiv version 2, dated 8 June 2026, of the paper accepted at SIGIR 2026. Generative retrieval produces semantic item identifiers for candidate retrieval; it is not ordinary generation of a textual answer. Bidirectional history, autoregressive output The encoder can compare both directions within the history already available at prediction time, while the decoder generates item-ID tokens autoregressively. Bidirectionality within that observed history does not mean access to future target labels. In the reported industrial test at history length 2,048, bidirectional self-attention achieved Recall@1000 of 0.8667 versus 0.8353 for a causal encoder, a difference of 3.14 percentage points. What makes attention linear Ordinary attention has a sequence-length cost proportional to L²dh. Kernel-based bidirectional linear attention rearranges the calculation as φ(Q)(φ(K)ᵀV), normalised by φ(Q)(φ(K)ᵀ1), using a positive feature map such as ELU+1. The core calculation scales as Ldh² and is linear in history length L when feature width dh is fixed. GBLA adds three mechanisms: local causal Conv1D mixing before the Q/K/V projections, a scalar key gate normalised across sequence positions, and an output gate combining RMSNorm with SiLU. The key gate weights contributions in both the feature aggregate and its denominator. It is soft importance weighting, not binary deletion or a recurrent forgetting factor. The convolution kernel size in the reported setup is four. Original Figure 1: local mixing and key and output gates Why the encoder remains hybrid The model concatenates multi-hash embeddings and projects them into the encoder, followed by a decoder with cross-attention. Its nine-layer hybrid encoder repeats [self-attention, GBLA, GBLA], retaining three self-attention layers and replacing six. Thus the full hybrid is not a wholly linear-time model. At length 2,048, hybrid Recall@1000 was 0.8668, compared with 0.8586 for all-GBLA layers, a 0.82-point gap. Original Tables 1 and 2: bidirectionality and hybrid placement Industrial retrieval results and ablations The private Yandex experiment used seven days of data, about 400 million training examples and next-day evaluation, with a nine-layer encoder of width 1,024 and 16 heads and a one-layer causal decoder. Effective batch size was matched, but peak learning rates differed: 7×10⁻⁴ for self-attention and 5×10⁻⁴ for the hybrid. Recall@1000 measures candidate coverage for a later ranker, not clicks or revenue. Self-attention/hybrid Recall@1000 was 0.8667/0.8668 at 2,048, 0.8784/0.8790 at 4,096 and 0.8854/0.8842 at 8,192. At the longest of these, Recall@10 was 0.3013/0.2948. The method therefore did not improve every metric. At 2,048, removing the key gate, convolution or gated RMSNorm reduced Recall@1000 from 0.8668 to 0.8661, 0.8652 or 0.8641; plain linear attention reached 0.8620. Its Recall@10 was 0.2675 versus 0.2780 for GBLA. These point estimates do not themselves prove statistical significance. Original Table 3: retrieval quality by history length Original Table 4: GBLA component ablations Layer speed and full training are different measurements On H100 GPUs, the single-layer benchmark used batch size eight and averages over ten runs against FlashAttention-v3. At length 2,048, GBLA took 0.618 ms against 0.585 ms and was slower. Speedups were 1.50× at 4,096, 2.46× at 8,192 and 8.22× at 32,768, where times were 8.217 versus 67.542 ms. The 8.2× headline applies to that layer benchmark, not an entire service. End-to-end training at length 8,192 and batch eight fell from 5,400 to 4,200 ms, about 1.29×. At length 2,048 and batch 32, the hybrid took 3,100 versus 3,000 ms. Plain linear attention was faster still at 3,800 ms for 8,192, with its quality trade-off. Batch sizes differ across sequence lengths, so these rows are not controlled measurements of length alone. Original Figure 2: single attention-layer latency Original Tables 5 and 6: layer and end-to-end training times Public benchmarks and remaining limits The Amazon Beauty, Toys and Sports experiments followed the GRID/Tiger setup with 5-core filtering, the last interaction for testing and the penultimate for validation. Flan-T5 embeddings and RK-Means semantic IDs were used; histories were at most 128 items, and results averaged five seeds. Tiger/GBLA Recall@10 was 0.0641/0.0611, 0.0584/0.0579 and 0.0345/0.0329 respectively. The reported Recall and NDCG values were slightly lower for GBLA in these tests. Similar magnitude is not proof of equivalence, and the paper did not benchmark efficiency on these short histories. Industrial data are private, and reported improvements must be understood within the stated hardware, architecture and evaluation conditions. Original Table 7: retrieval results on public Amazon datasets A patent-practice reading The useful technical question is how the problem of long histories is connected to a particular arrangement of local mixing, key weighting, output gating and retained self-attention. A patent analysis should distinguish individual components from their combination, identify what the claims would actually require and connect alleged effects to supporting experiments. The paper is not evidence that any proposed claim is novel or patentable. Prior-art comparison, claim scope and disclosure sufficiency require separate assessment. For implementation, evaluate candidate recall, downstream ranking, memory and end-to-end latency together. The original diagrams and tables are retained below with local-language explanations; the linked paper and CC BY 4.0 attribution identify their source. The original figure labels and forms are preserved; the text explains the source material. Related service: AI patents Paper and figures: Matveev et al., GBLA v2, 2026 (CC BY 4.0) · CC BY 4.0 Read the Korean source This article reflects the information available when it was published. Contact us to discuss your circumstances. Discuss this topic ↗ All articles ON THIS PAGE The research problem Bidirectional history, autoregressive output What makes attention linear Why the encoder remains hybrid Industrial retrieval results and ablations Layer speed and full training are different measurements Public benchmarks and remaining limits A patent-practice reading TALK TO IPLEX Discuss your IP questions We consider your technology and business needs together. ↗ Contact us Newer Saxophone pad repairs: process wording and inventive step ↗ Older Design invalidation: prove prior public disclosure before comparing similarity ↗ Related insights Artificial Intelligence 2026.10.29 SwiLA: organizing memory around context How Switching Linear Attention combines linear regressors within a fixed recurrent state, what its recall experiments establish, and where capacity and computational limits remain. ↗ Read article Artificial Intelligence 2026.10.28 Microsoft’s speech synthesis patent: combining sounds, words and sentence structure US12094447B2 explains how phoneme or character information can be combined with word embeddings and grammatical structure to generate speech. ↗ Read article Artificial Intelligence 2026.10.22 DFlare: widening the information path to a speculative drafter Layer-specific feature fusion, separate KV projections and a progressive training loss increase the capacity of a block-diffusion drafter, with measured gains and important limits. ↗ Read article

- https://www.iplexlaw.co.kr/en
- https://www.iplexlaw.co.kr/en/blog/category/ai
- https://www.iplexlaw.co.kr/en/ai-patents
- https://arxiv.org/abs/2606.07317v2
- https://creativecommons.org/licenses/by/4.0/
- https://www.iplexlaw.co.kr/ko/blog/gbla-generative-retrieval
- https://www.iplexlaw.co.kr/en/contact
- https://www.iplexlaw.co.kr/en/blog
- https://www.iplexlaw.co.kr/en/blog/gbla-generative-retrieval#article-1885-en-0
- https://www.iplexlaw.co.kr/en/blog/gbla-generative-retrieval#article-1885-en-1
- https://www.iplexlaw.co.kr/en/blog/gbla-generative-retrieval#article-1885-en-2
- https://www.iplexlaw.co.kr/en/blog/gbla-generative-retrieval#article-1885-en-3
- https://www.iplexlaw.co.kr/en/blog/gbla-generative-retrieval#article-1885-en-4
- https://www.iplexlaw.co.kr/en/blog/gbla-generative-retrieval#article-1885-en-5
- https://www.iplexlaw.co.kr/en/blog/gbla-generative-retrieval#article-1885-en-6
- https://www.iplexlaw.co.kr/en/blog/gbla-generative-retrieval#article-1885-en-7
- https://www.iplexlaw.co.kr/en/contact
- https://www.iplexlaw.co.kr/en/blog/brunch-case-1886
- https://www.iplexlaw.co.kr/en/blog/brunch-case-1884
- https://www.iplexlaw.co.kr/en/blog/swila-switching-linear-attention
- https://www.iplexlaw.co.kr/en/blog/swila-switching-linear-attention
- https://www.iplexlaw.co.kr/en/blog/microsoft-multilevel-tts
- https://www.iplexlaw.co.kr/en/blog/microsoft-multilevel-tts
- https://www.iplexlaw.co.kr/en/blog/dflare-block-diffusion-drafting
- https://www.iplexlaw.co.kr/en/blog/dflare-block-diffusion-drafting