Home › Census › Attention

Attention

Syntologyfunction-name censuscensus 2026-09-22battery b986f7e04d79all samples with this name

Attention: 13 implementations from 12 papers ran on one shared input (census 2026-09-22, battery b986f7e04d79); they produced 10 distinct outputs across 4 buckets, one shared input per bucket.

Identical values to six decimals (the recorded digest) on the shared input are agreement on those inputs, not equivalence. Implementations are compared only within one bucket, the positional (rank, kind, dtype) of each array argument; the argument name is not part of the key because the harness draws the shared array from (rank, kind) and casts it to the dtype, whatever the name; each bucket's shared input is fixed by that key, so members of one bucket saw bitwise-identical inputs under their own scalar arguments. A cluster is the set of members whose recorded output digest is identical. Nothing here says which computation a paper's method intended, and nothing reproduces a paper's results.

Not compared, and not in the tables or the counts above:

Bucket 1 of 4: arg 1: rank 3, kind float, dtype float32

7 implementations from 7 papers share this bucket (rank, kind, dtype of each array argument, positional; each member's recorded signature, argument name included, is shown under it); 4 distinct outputs, largest cluster first. Values are the first 8 of the recorded output, flattened.

Cluster (same digest to six decimals)MembersShared output on this bucket's input
2 implementations
2 papers
61b654db79d6
recorded values identical
class
models/backbone.py 8598c6b9
recorded x:3/float/float32
class
models/bidir_modeling_crossattn.py b181227b
recorded x:3/float/float32
[-0.215092, 0.272654, -0.166096, -0.224872, -0.0535068, 0.195691, 0.032025, 0.469848, …]
shape [2, 4, 8] · float32 · Tensor
2 implementations
2 papers
70a37b829270
recorded values identical
class
sonics/models/spectttra.py c3ac8c34
recorded x:3/float/float32
class
models/rfmsr.py cf18d01a
recorded x:3/float/float32; rope=None
[-0.215092, 0.272654, -0.166096, -0.224872, -0.0535068, 0.195691, 0.0320251, 0.469848, …]
shape [2, 4, 8] · float32 · Tensor
2 implementations
2 papers
f9b2f741e78d
recorded values identical
class
golf/model.py 6dc45d11
recorded x:3/float/float32; pos=None, ret_attn=False
class
Codes/multimae/multimae.py d969a5ba
recorded x:3/float/float32
[-0.313875, 0.375622, -0.12825, -0.314344, -0.0956915, 0.406332, 0.125604, 0.486219, …]
shape [2, 4, 8] · float32 · Tensor
1 implementation
1 paper
a331af7abd4a
class
models/early_exit.py f9763b61
recorded x:3/float/float32
[-0, 0.302949, -0.184552, -0, -0.059452, 0.217434, 0.0355834, 0.522054, …]
shape [2, 4, 8] · float32 · Tensor

Bucket 2 of 4: arg 1: rank 3, kind float, dtype float32 · arg 2: rank 3, kind float, dtype float32 · arg 3: rank 3, kind float, dtype float32

4 implementations from 3 papers share this bucket (rank, kind, dtype of each array argument, positional; each member's recorded signature, argument name included, is shown under it); 4 distinct outputs, largest cluster first. Values are the first 8 of the recorded output, flattened.

Cluster (same digest to six decimals)MembersShared output on this bucket's input
1 implementation
1 paper
2626132f9192
class
src/nlp/layers/encoder.py 6f534d51
recorded q:3/float/float32, k:3/float/float32, v:3/float/float32; mask=None
[-0, 1.48999, -1.28467, -0, 0.235625, -1.2001, -0.311517, -0.154612, …]
shape [2, 4, 8] · float32 · Tensor
1 implementation
1 paper
5a2eda2611f0
class
lightglue/lightglue.py a08f05b6
recorded q:3/float/float32, k:3/float/float32, v:3/float/float32; mask=None
[-1.00379, 1.34099, -1.1562, -0.339274, 0.212062, -1.08009, -0.280365, -0.139151, …]
shape [2, 4, 8] · float32 · Tensor
1 implementation
1 paper
e5a7b22e38a6
class
TRANSFORMERS.py 8d12ff40
recorded query:3/float/float32, key:3/float/float32, value:3/float/float32; mask=None
[-0.459559, -0.6186, 0.396802, -0.109149, 0.162055, -0.0981505, -0.0105498, -0.117563, …]
shape [2, 4, 8] · float32 · Tensor
1 implementation
1 paper
f1e879c2058d
class
network/models/d2ls.py e843618e
recorded q:3/float/float32, k:3/float/float32, v:3/float/float32
[0.240397, -0.0802591, -0.205289, 0.0541918, -0.292073, 0.605929, -0.417172, -0.00929932, …]
shape [2, 4, 8] · float32 · Tensor

Bucket 3 of 4: arg 1: rank 3, kind float, dtype float32 · arg 2: rank 3, kind float, dtype float32

1 implementation from 1 paper share this bucket (rank, kind, dtype of each array argument, positional; each member's recorded signature, argument name included, is shown under it); 1 distinct output, largest cluster first. Values are the first 8 of the recorded output, flattened.

Cluster (same digest to six decimals)MembersShared output on this bucket's input
1 implementation
1 paper
61b654db79d6
class
models/hd_former.py d252bbac
recorded fv:3/float/float32, fe:3/float/float32
[-0.215092, 0.272654, -0.166096, -0.224872, -0.0535068, 0.195691, 0.032025, 0.469848, …]
shape [2, 4, 8] · float32 · Tensor

Bucket 4 of 4: arg 1: rank 4, kind float, dtype float32

1 implementation from 1 paper share this bucket (rank, kind, dtype of each array argument, positional; each member's recorded signature, argument name included, is shown under it); 1 distinct output, largest cluster first. Values are the first 8 of the recorded output, flattened.

Cluster (same digest to six decimals)MembersShared output on this bucket's input
1 implementation
1 paper
859fd91fbb1a
class
code/model/decoder.py cc958048
recorded x:4/float/float32
[0.193572, -0.224079, 0.129158, 0.104469, 0.155505, -0.198878, 0.179483, 0.125382, …]
shape [2, 3, 4, 4] · float32 · Tensor

Identical values to six decimals (the recorded digest) on the shared input are agreement on those inputs, not equivalence; where a cluster's members carry recorded values, the largest difference among them is shown under the cluster. Paper titles are the archive's archive 2025-07-28 where the paper is in the archive and the graph's where it was added by Syntology; papers with no page here are shown by their recorded paper id only. A paper count above the implementation count means one implementation (one code sha) is held from several papers' repositories and counts once. Per-sample status, licence and fingerprint records for each paper are on its paper page. JSON twin: /census/attention.json.

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