Papers › Fast and Stable Triangular Inversion for Delta-Rule Linear Transformers

Fast and Stable Triangular Inversion for Delta-Rule Linear Transformers

20 May 2026arXiv:2605.21325added by Syntology

Aleksandros Sobczyk, Gioele Gottardo, Christos K. Matzoros, Mirko De Vita, Filip Skogh, Anastasios Zouzias, Jiawei Zhuang

Title, abstract, authors and date from arXiv's metadata (CC0); this paper is not in the Papers with Code archive (frozen 2025-07-28).

Linear attention has emerged as a cornerstone for efficient long-context architectures, as evidenced by its integration into state-of-the-art open-source models including Qwen3.5/3.6, Kimi Linear, and RWKV-7. Models that incorporate linear attention layers with the so-called Delta-Rule involve the inversion of triangular matrices as a core sub-routine. This operation often forms a performance bottleneck, and, due to its high-sensitivity to numerical errors, it can significantly deteriorate end-to-end model accuracy if it is not carefully implemented. This work provides a systematic analysis of both direct and iterative triangular inversion algorithms, targeting methods that are rich in matrix products, and, therefore, have the potential to efficiently utilize modern hardware. To that end, our analysis covers a broad spectrum of mathematical and practical aspects, with a heavy focus on numerical stability, computational complexity, and, ultimately, hardware efficiency and practical considerations. We provide a rigorous experimental evaluation to verify these properties in practical scenarios, and in low-precision floating-point representations, highlighting the strengths and limitations of each method. Performance benchmarks on NPUs reveal up to 4.3× speed-up against the state-of-the-art implementations of SGLang for triangular matrix inversion, leading to significant performance improvements on the entire layer level, while maintaining full end-to-end model accuracy.

PaperPDFCode Syntology ran

In Syntology View this paper on Syntology, its page in Syntology's graph. That page adds which of its functions are also present in other papers' code Syntology holds, with dates, and its dated run record as JSON and a README badge. It also lists the repositories linked to the paper, the abstract and the calls for agents.

Open this paper in Syntology's Atlas, the map of the papers in Syntology's graph and their citations.

For agents, on Syntology's MCP service (how to connect):

Code

Syntology 28 code samples harvested from 3 repositories linked to this paper: 15 ran (of which 0 constructed an object rather than computing a result; 4 with no instrument failure: 1 honoured, 0 violated, 3 with no contract checked; 11 where Syntology's instrument failed) · 13 unverified. Where Syntology's instrument failed: 7 where our contract draft was wrong; 4 where the contract check could not run; failures of Syntology's instrument, not of the code.

By repository: found in paper text by Syntology: 28 samples from 3 repositories, 15 ran (4 with no instrument failure; 11 where Syntology's instrument failed). The run record, sample by sample. “Ran” means executed on a synthesized input, not that the code is correct or reproduces the paper.

Levels written on 2026-09-24 (UTC): the dates Syntology's graph wrote each sample's verification level, not when the code executed.

huawei-csl/pto-kernels found in paper text by SyntologySyntology: 9 samples, 6 ran (5 instrument-failed) · 3 unverified report
huawei-csl/gdn-tri-inverse found in paper text by SyntologySyntology: 10 samples, 3 ran (2 instrument-failed) · 7 unverified report
huawei-csl/megagdn-pto found in paper text by SyntologySyntology: 9 samples, 6 ran (4 instrument-failed) · 3 unverified report

Reachability, where shown, is from one Syntology probe window (2026-09-16 to 2026-09-18); repositories not probed show nothing. GitHub stars are not tracked.

Code Syntology ran Syntology

28 samples harvested: 15 ran (of which 0 constructed an object rather than computing a result; 4 with no instrument failure: 1 honoured, 0 violated, 3 with no contract checked; 11 where Syntology's instrument failed) · 13 unverified. “Where Syntology's instrument failed” counts failures of Syntology's instrument, not of the code. Syntology's record for this page last changed ; when this build read Syntology's graph is in the build record. That date is not when the samples ran.

1ran · honoured contract
7ran · our draft was wrong
4ran; the contract check could not run
3ran
13unverified

Licence: 28 of the 28 samples are pointer only, meaning Syntology does not serve that copy's text. This page shows no code text for any sample; each one links to its file in the repository.

Harvested from 3 repositories linked to this paper, official or community; the list below is grouped by repository, each group headed by its name and what it is to the paper. “Ran” means the sample executed on a synthesized input. It does not mean the output is correct, and nothing here reproduces the paper's results. A line beginning “fixture:” under a sample that ran gives the arguments Syntology's fixture called the sample with, each as name:type/rank/element kind (? where none was recorded), and the type of what the fixture's run returned, exactly as the harness recorded them; the arguments and the returned type were recorded by two separate runs of the same fixture. No output value is shown, and a returned type says nothing about whether the output was right. “Level recorded” gives the UTC date when Syntology's graph wrote the sample's verification level, not when the code executed; where the level was written before Syntology's graph kept that date, the row says “level recorded, date not recorded”. “Honoured” and “violated” refer to a contract Syntology drafted from the code itself. “Our draft was wrong” and “the contract check could not run” are failures of Syntology's instrument, not of the code: the sample ran on Syntology's synthesized input, and then the drafted contract did not describe it, or the check itself raised an error (RAISES), lacked a dependency (DEP_MISSING) or ran out of time (TIMEOUT); hover the label for which. A metamorphic tier, where shown, is evidence of the output's shape, not that the sample implements the method.

Each sample ends with its code_sha256. A sample's code_sha256 is Syntology's identity for that exact code. It is printed in full (16 hexadecimal characters for every sample in this build), and get_code(code_sha256="…") takes it exactly as printed. An agent calls it on Syntology's MCP service (how to connect); click an identity to copy that call.

Repository labels, per sample. official repository: The archive marks this repository official for the paper. named in the paper: The archive records that the paper mentions this repository; it is not marked official. community (archive-listed): In the archive's code links for this paper, not marked official and not recorded as mentioned in the paper. found in paper text by Syntology: Syntology found this repository in the paper's own text; whether it is the authors' implementation is not asserted. community: Not in the archive's code links for this paper; a community repository Syntology harvested. Samples are grouped by repository: a repository marked official first, then by how directly the repository is tied to the paper, then in the archive's order; never by how many ran. Licence labels name the repository's licence as recorded at harvest. “Pointer only” means Syntology does not serve that copy's text, for one of six reasons: no licence file was found; the licence was not identified; the licence is recorded as permissive but that copy's record is not marked cleared; the licence is outside the permissive list Syntology serves text under (MIT, Apache-2.0, BSD and similar); this paper's copy was not located, so its licence is not recorded; or this paper's copy is cleared, but the stored copy of the same code that Syntology serves is not. Some licences outside that list permit redistribution, such as WTFPL, and GPL-3.0 under its conditions; they are simply not on the list. Hover a licence label for the reason. File links open the file on GitHub at the default branch, which may have changed since the harvest.

A function or class name links to its page when the name occurs in the harvested code of at least 5 papers; the name is a string, not an identity.

huawei-csl/pto-kernels found in paper text by Syntology 9 samples, 6 ran (5 instrument-failed) · 3 unverified
compile_kernel .skills/testing-pto-kernels/reference/pto_demo_utils.py ran · our draft was wrong BSD-3-Clause-Clear · pointer only · 49cd9155190e49b2 · report fixture: signature compile_sh:PosixPath/0/?, kernel:str/0/str · returned PosixPath · level recorded 2026-09-24
npu_from_cpu .skills/testing-pto-kernels/reference/dynamic_multi_core/a5/run_kernel_ctypes.py ran · our draft was wrong BSD-3-Clause-Clear · pointer only · 29854c7c42411c2b · report fixture: signature arr:ndarray/2/float, device:str/0/str · returned Tensor · level recorded 2026-09-24
npu_tensor .skills/testing-pto-kernels/reference/dynamic_multi_core/a5/run_kernel_pybind.py ran · our draft was wrong BSD-3-Clause-Clear · pointer only · 43739c7f7a9f008f · report fixture: signature arr:ndarray/2/float, device:str/0/str · returned Tensor · level recorded 2026-09-24
summarize in 36 papers .skills/testing-pto-kernels/reference/dynamic_multi_core/a2a3/benchmark.py ran; the contract check could not run (RAISES) fingerprinted BSD-3-Clause-Clear · pointer only · 7140e33bf7bef21f · report fixture: signature samples_us:list/1/float · returned tuple · level recorded 2026-09-24
tensor_ptr .skills/testing-pto-kernels/reference/pto_demo_utils.py ran · our draft was wrong BSD-3-Clause-Clear · pointer only · 6c19edb11fb6754c · report fixture: signature t:Tensor/2/float · returned c_void_p · level recorded 2026-09-24
workspace .skills/testing-pto-kernels/reference/dynamic_multi_core/a2a3/run_mix_ctypes.py ran · honoured contract BSD-3-Clause-Clear · pointer only · bf1d857bd30bc516 · report fixture: signature block_dim:int/0/int, device:str/0/str · returned Tensor · level recorded 2026-09-24
add_reference_to_path .skills/testing-pto-kernels/reference/pto_demo_utils.py unverified BSD-3-Clause-Clear · pointer only · b26eb7a5b59d8e1f · report
build_kernel .skills/testing-pto-kernels/reference/dynamic_multi_core/a2a3/run_kernel_pybind.py unverified BSD-3-Clause-Clear · pointer only · bcc14bcd0eec3750 · report
event_bench .skills/testing-pto-kernels/reference/dynamic_multi_core/a2a3/benchmark.py unverified BSD-3-Clause-Clear · pointer only · b0bfd6df1362ea99 · report
huawei-csl/gdn-tri-inverse found in paper text by Syntology 10 samples, 3 ran (2 instrument-failed) · 7 unverified
chunk_gated_delta_rule_native src/gdn_tri_inverse/triton_chunk_gdn/fla/chunk.py ran BSD-3-Clause-Clear · pointer only · 8c58a327ad934418 · report fixture: signature query:Tensor/4/float, key:Tensor/4/float, value:Tensor/4/float, g:Tensor/3/float, beta:Tensor/3/float, chunk_size:int/0/int, initial_state:Tensor/4/float, output_final_state:bool/0/bool, use_qk_l2norm_in_kernel:bool/0/bool, tri_inv_fn:callable/0/? · returned tuple · level recorded 2026-09-24
inv_tril_inplace src/gdn_tri_inverse/triton_chunk_gdn/fla/chunk.py ran · our draft was wrong fingerprinted BSD-3-Clause-Clear · pointer only · 65c98279ed930932 · report fixture: signature A:Tensor/3/float · returned Tensor · level recorded 2026-09-24
tri_inv_qwen3_next_default src/gdn_tri_inverse/linalg.py ran · our draft was wrong fingerprinted BSD-3-Clause-Clear · pointer only · 27aa399aa0968a73 · report fixture: signature A:Tensor/3/float · returned Tensor · level recorded 2026-09-24
fast_inv_tril src/gdn_tri_inverse/triton_chunk_gdn/fla/chunk.py unverified BSD-3-Clause-Clear · pointer only · 3872a6d7e9a842a6 · report
recurrent_gated_delta_rule_ref src/gdn_tri_inverse/core.py unverified BSD-3-Clause-Clear · pointer only · 94a3d31751316112 · report
ref_chunk_cumsum src/gdn_tri_inverse/tilelang_chunk_gdn/opt_gdn_chunk_cumsum.py unverified BSD-3-Clause-Clear · pointer only · 1902ecb472db2fcc · report
ref_chunk_h src/gdn_tri_inverse/tilelang_chunk_gdn/opt_gdn_chunk_h.py unverified BSD-3-Clause-Clear · pointer only · 2996cf978e93d4a1 · report
ref_chunk_o src/gdn_tri_inverse/tilelang_chunk_gdn/opt_gdn_chunk_o.py unverified BSD-3-Clause-Clear · pointer only · 7136a894cedf2d88 · report
ref_seq_gdn src/gdn_tri_inverse/tilelang_chunk_gdn/opt_gdn_full.py unverified BSD-3-Clause-Clear · pointer only · f3d36efe796d3c18 · report
ref_wy_fast src/gdn_tri_inverse/tilelang_chunk_gdn/opt_gdn_wy_fast.py unverified BSD-3-Clause-Clear · pointer only · 18bceabcfb10e5ac · report
huawei-csl/megagdn-pto found in paper text by Syntology 9 samples, 6 ran (4 instrument-failed) · 3 unverified
bind_triton vllm_patch/chunk_kda_pto.py ran; the contract check could not run (DEP_MISSING) no licence file found · pointer only · 3be586d513227377 · report fixture: signature _triton_impl:callable/0/? · returned function · level recorded 2026-09-24
chunk_kda_pto vllm_patch/chunk_kda_pto.py ran no licence file found · pointer only · 88c21f1f458ccf5c · report fixture: signature q:Tensor/3/float, k:Tensor/3/float, v:Tensor/3/float, g:Tensor/3/float, beta:Tensor/3/float, scale:NoneType/0/?, initial_state:NoneType/0/?, output_final_state:bool/0/bool, use_qk_l2norm_in_kernel:bool/0/bool, cu_seqlens:Tensor/1/int, _triton_impl:callable/0/? · returned tuple · level recorded 2026-09-24
get_tp_context docker/overrides/model_runner_v1.py ran · our draft was wrong no licence file found · pointer only · 179b4ecb6419827a · report fixture: signature drafter:callable/0/? · returned nullcontext · level recorded 2026-09-24
total_chunks megagdn_pto/kernel_libs.py ran no licence file found · pointer only · e4e5891b8b4488b1 · report fixture: signature batch_size:int/0/int, seq_len:int/0/int, chunk_size:int/0/int, cu_seqlens:Tensor/1/int · returned int · level recorded 2026-09-24
transpose_beta megagdn_pto/kernel_libs.py ran; the contract check could not run (RAISES) no licence file found · pointer only · 33a4536ecf79ed25 · report fixture: signature beta:Tensor/3/float · returned Tensor · level recorded 2026-09-24
transpose_gates megagdn_pto/kernel_libs.py ran; the contract check could not run (RAISES) no licence file found · pointer only · e0779707b00e56f4 · report fixture: signature g_sum:Tensor/3/float · returned Tensor · level recorded 2026-09-24
compile_chunk_kernel megagdn_pto/compile.py unverified no licence file found · pointer only · 1a65bd82810630c4 · report
compile_tri_inverse megagdn_pto/compile.py unverified no licence file found · pointer only · 8991055d8785b3a2 · report
staged_separate_launch benchmarks/kernel/bench_pto_pipeline_latency.py unverified no licence file found · pointer only · 04b68677ccfb33a0 · report

Results from the paper

The Papers with Code archive ends with its 2025-07-28 snapshot. This paper's arXiv identifier, 2605.21325, was issued in May 2026, after that date, so the archive has no leaderboard rows for it.

Placed on leaderboards by Syntology Syntology

Syntology's extractor, the model Claude Sonnet 4.5, judged this paper's tables to report on 4 leaderboards (a model's judgement, not a result) and has not placed the paper on any of them: Multiple Choice Question Answering (MCQA) · MMLU (Abstract Algebra) (refused by rules: the value is not on the scale of the leaderboard column's archive values; the quoted value appears more than once in the row); Multiple Choice Question Answering (MCQA) · MMLU (Astronomy) (refused by a rule: the value is not on the scale of the leaderboard column's archive values); Multiple Choice Question Answering (MCQA) · MMLU (College Biology) (refused by a rule: the value is not on the scale of the leaderboard column's archive values); Multiple Choice Question Answering (MCQA) · MMLU (College Mathematics) (refused by a rule: the value is not on the scale of the leaderboard column's archive values). What is not shown.

Syntology has checked 8,886 of the 18,637 papers on this site that are newer than the archive (for 3,717 of them no archive leaderboard matched the paper's tables, so there was nothing further to check); 733 were read and have nothing to place (arXiv has no HTML version of the paper, or that version has no tables), 41 could not be read (the extractor's reply could not be parsed), and results from the other 8,977 appear after they are checked.

Report a problem or propose a change · a person checks every report against the paper or source before anything changes; decisions are listed on /corrections