Papers › Native Sparse Attention: Hardware-Aligned and Natively Trainable Sparse Attention

Native Sparse Attention: Hardware-Aligned and Natively Trainable Sparse Attention

16 Feb 2025arXiv:2502.11089archive 2025-07-28

Jingyang Yuan, Huazuo Gao, Damai Dai, Junyu Luo, Liang Zhao, Zhengyan Zhang, Zhenda Xie, Y. X. Wei, Lean Wang, Zhiping Xiao, Yuqing Wang, Chong Ruan, Ming Zhang, Wenfeng Liang, Wangding Zeng

Long-context modeling is crucial for next-generation language models, yet the high computational cost of standard attention mechanisms poses significant computational challenges. Sparse attention offers a promising direction for improving efficiency while maintaining model capabilities. We present NSA, a Natively trainable Sparse Attention mechanism that integrates algorithmic innovations with hardware-aligned optimizations to achieve efficient long-context modeling. NSA employs a dynamic hierarchical sparse strategy, combining coarse-grained token compression with fine-grained token selection to preserve both global context awareness and local precision. Our approach advances sparse attention design with two key innovations: (1) We achieve substantial speedups through arithmetic intensity-balanced algorithm design, with implementation optimizations for modern hardware. (2) We enable end-to-end training, reducing pretraining computation without sacrificing model performance. As shown in Figure 1, experiments show the model pretrained with NSA maintains or exceeds Full Attention models across general benchmarks, long-context tasks, and instruction-based reasoning. Meanwhile, NSA achieves substantial speedups over Full Attention on 64k-length sequences across decoding, forward propagation, and backward propagation, validating its efficiency throughout the model lifecycle.

PaperPDFCodeCode Syntology ranMethod as code

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 19 code samples harvested from 2 repositories linked to this paper: 16 ran (of which 2 constructed an object rather than computing a result; 15 with no instrument failure: 0 honoured, 0 violated, 15 with no contract checked; 1 where Syntology's instrument failed) · 3 unverified. Where Syntology's instrument failed: 1 where our contract draft was wrong; failures of Syntology's instrument, not of the code.

By repository: community (archive-listed): 19 samples from 2 repositories, 16 ran (15 with no instrument failure; 1 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.

No sample is linked to this paper from 1 of its repositories that Syntology harvested for other papers:

Levels written between 2026-09-22 and 2026-09-26 (UTC): the dates Syntology's graph wrote each sample's verification level, not when the code executed. 1 of the 16 samples that ran has no recorded date.

britacatalin/bsa mentioned on GitHubpytorchSyntology: 3 samples, 3 ran (0 instrument-failed) · 0 unverified report
microsoft/seerattention mentioned on GitHubpytorchMITSyntology: 16 samples, 13 ran (1 instrument-failed) · 3 unverified report
sandai-org/magiattention mentioned on GitHubpytorchSyntology: no sample linked to this paper (harvested for another paper). Syntology's graph links these samples to Ring Attention with Blockwise Transformers for Near-Infinite Context (arXiv:2310.01889; that paper's own run record: 20 ran (of which 0 constructed an object rather than computing a result; 14 with no instrument failure: 3 honoured, 3 violated, 8 with no contract checked; 6 where Syntology's instrument failed) · 8 unverified); nothing checks that they implement this paper's method. report

Repository list and official/mentioned flags are the archive's, frozen 2025-07-28. 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.

A paper named beside a repository with no sample linked to this paper is shown with that paper's own run record, not this paper's: “ran” means executed on a synthesized input, not that the code is correct, and “where Syntology's instrument failed” counts failures of Syntology's instrument, not of the code. Papers are listed in arXiv-id order, at most three per repository.

Code Syntology ran Syntology

19 samples harvested: 16 ran (of which 2 constructed an object rather than computing a result; 15 with no instrument failure: 0 honoured, 0 violated, 15 with no contract checked; 1 where Syntology's instrument failed) · 3 unverified. “Where Syntology's instrument failed” counts failures of Syntology's instrument, not of the code. Syntology's record for this page has not changed since , the first build that kept a record date for it; when this build read Syntology's graph is in the build record. That date is not when the samples ran.

1ran · our draft was wrong
15ran
3unverified

Licence: 2 of the 19 samples are pointer only, meaning Syntology does not serve that copy's text. For 2 of them, this paper's copy is under a permissive licence and cleared, but the stored copy of the same code that Syntology serves is not; the licence label's tooltip gives Syntology's recorded reason. This page shows no code text for any sample; each one links to its file in the repository.

Harvested from 2 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. A census tag appears when Syntology's census ran implementations with that name on one shared input per bucket: it gives how many were compared, from how many papers, and how many distinct outputs they produced, and names this sample's bucket when this sample was one of them.

britacatalin/bsa community (archive-listed) 3 samples, 3 ran (0 instrument-failed) · 0 unverified
NativeSparseAttention nsa/native_sparse_attention.py ran MIT (permissive) · b93448058eb3a756 · report fixture: returned NativeSparseAttention · level recorded 2026-09-22
StaticGatedAttention nsa/native_sparse_attention.py ran MIT (permissive) · f36889830d661d6f · report fixture: signature branches:Tensor/5/float · returned Tensor · level recorded 2026-09-24
TokenGatedAttention nsa/native_sparse_attention.py ran MIT (permissive) · d9f141ee5201879b · report fixture: returned TokenGatedAttention · level recorded 2026-09-22
microsoft/seerattention community (archive-listed) 16 samples, 13 ran (1 instrument-failed) · 3 unverified
apply_chat_template in 44 papers distillation_decode.py ran MIT (permissive) · 202aa55ac60ff1f6 · report fixture: signature example:dict/0/?, tokenizer:MockTokenizer/0/?, add_prefix:bool/0/bool, prefix:str/0/str · returned dict · level recorded 2026-09-24
apply_rotary_pos_emb in 393 papers
census: 14 implementations from 15 papers, 8 distinct outputs across 7 buckets, one shared input per bucket
seer_attn/modules/common.py ran MIT (permissive) · 8c7b215a2004bfc6 · report fixture: signature q:Tensor/4/float, k:Tensor/4/float, cos:Tensor/3/float, sin:Tensor/3/float, position_ids:NoneType/0/?, unsqueeze_dim:int/0/int · returned tuple · level recorded 2026-09-24
average_groups eval/longbench/average_sparsity.py ran MIT (permissive) · a8c8717e85efa3ea · report fixture: signature file_path:str/0/str · returned dict · level recorded 2026-09-24
build_optimized_chunks distillation_decode.py ran MIT (permissive) · 4334f4eeb6104bee · report fixture: signature dataset:Dataset/0/?, chunk_size:int/0/int · returned Dataset · level recorded 2026-09-24
calculate_averages_by_range eval/longbench/average_score.py ran MIT (permissive) · e9c41059655ed7e3 · report fixture: signature data:dict/0/? · returned dict · level recorded 2026-09-25
generate_base_sparsity_mask eval/efficiency/benchmark_sparse_attn.py ran MIT (permissive) · 8498177d23b25d59 · report fixture: signature max_seqlen_q:int/0/int, max_seqlen_k:int/0/int, round_base:int/0/int, m_block_dim:int/0/int, n_block_dim:int/0/int, sparsity:float/0/float, causal:bool/0/bool, device:str/0/str · returned Tensor · level recorded 2026-09-24
get_tensors eval/efficiency/benchmark_sparse_attn.py ran MIT (permissive) · 09bb2cb546d9e69e · report fixture: signature bs:int/0/int, num_heads:int/0/int, seq_len:int/0/int, head_dim:int/0/int, dtype:dtype/0/? · returned tuple · level recorded 2026-09-24
layer_norm_ref in 12 papers seer_attn/modules/layernorm.py ran MIT (permissive) here; the stored copy is pointer only · 2ce8fcf5bedd19ec · report fixture: signature x:Tensor/2/float, weight:Tensor/1/float, bias:Tensor/1/float, residual:Tensor/2/float, eps:float/0/float, prenorm:bool/0/bool, upcast:bool/0/bool · returned Tensor · level recorded 2026-09-26
load_json in 344 papers eval/longbench/average_score.py ran MIT (permissive) · cc892598a14c82a5 · report fixture: returned dict · level recorded 2026-09-24
rms_norm_ref in 13 papers seer_attn/modules/layernorm.py ran MIT (permissive) here; the stored copy is pointer only · ab802ae3195c6316 · report fixture: signature x:Tensor/2/float, weight:Tensor/1/float, bias:Tensor/1/float, residual:Tensor/2/float, eps:float/0/float, prenorm:bool/0/bool, upcast:bool/0/bool · returned Tensor · level recorded 2026-09-26
rotate_half in 424 papers
census: 24 implementations from 96 papers, 4 distinct outputs across 2 buckets, one shared input per bucket this sample: bucket 1 of 2, arg 1: rank 2, kind float, dtype float32
seer_attn/modules/common.py ran · our draft was wrong fingerprinted MIT (permissive) · e03d53ba9d4f9ae5 · report fixture: signature x:Tensor/2/float · returned Tensor · level recorded, date not recorded
tokenize_fn in 9 papers distillation_decode.py ran MIT (permissive) · 3ef366d8b91a362a · report fixture: signature tokenizer:callable/0/?, examples:dict/0/? · returned dict · level recorded 2026-09-24
tokenize_fn in 9 papers distillation_prefill.py ran MIT (permissive) · f9963e3bc9860eda · report fixture: signature tokenizer:callable/0/?, tranining_max_length:int/0/int, example:dict/0/? · returned dict · level recorded 2026-09-24
apply_rotary_pos_emb_single in 11 papers seer_attn/modules/common.py unverified MIT (permissive) · 89cf2076c1b8ce86 · report
ref_program_fa eval/efficiency/decode_kernel_eval.py unverified MIT (permissive) · ea5becf731d3b58d · report
run_benchmark in 11 papers eval/efficiency/benchmark_sparse_attn.py unverified MIT (permissive) · 046184b390b18314 · report

Method as code Syntology

What Syntology's graph records as this paper's claimed method, lifted into code: each assembly names its bricks (functions and classes), each with its own role and its own verification level. An assembly is not a check that the code implements the method correctly or reproduces the paper's results. Levels are each brick's own verification level in Syntology's graph (level 0: no recorded run), listed brick by brick and never combined into one level; “ran” means the brick executed on a synthesized input, not that its output is correct, and the instrument figure counts failures of Syntology's instrument, not of the code. Where a model read the code against the paper's claim, each assembly's label verdict is one model's judgement, kept in this page's JSON twin; it is not calibrated, so no card prints it.

1 assembly on this paper: 1 harvested, 0 generated.

harvested Assembly for Native Sparse Attention (NSA) (graph method: Native Sparse Attention (NSA)). 3 bricks; individually on a fixture: 3 ran (of which 2 constructed an object rather than computing a result; 3 with no instrument failure: 0 honoured, 0 violated, 3 with no contract checked; 0 where Syntology's instrument failed) · 0 unverified. The assembly has not been run end-to-end.

  1. NativeSparseAttention entry · level 1 · ran · b93448058eb3a756
  2. StaticGatedAttention helper · level 1 · ran · f36889830d661d6f
  3. TokenGatedAttention helper · level 1 · ran · d9f141ee5201879b

Each brick ends with its code_sha256, Syntology's identity for that exact code; click one to copy the get_code call that returns it. For agents, on Syntology's MCP service: assembly(arxiv_id="2502.11089") returns these assemblies with each brick's role and wiring (how to connect).

Results from the paper archive 2025-07-28

No leaderboard rows for this paper in the archive.

Methods

AttentionSoftmax

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