Home › Census › repeat_kv

repeat_kv

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

repeat_kv: 15 implementations from 60 papers ran on one shared input (census 2026-09-22, battery b986f7e04d79); they produced 2 distinct outputs.

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:

Every examined implementation of this name that took an array argument ran on the shared input.
No implementation of this name ran on its own fixture arguments only: every one that ran took an array argument.
Every compared output was digested.

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

15 implementations from 60 papers share this bucket (rank, kind, dtype of each array argument, positional; each member's recorded signature, argument name included, is shown under it); 2 distinct outputs, largest cluster first. Values are the first 8 of the recorded output, flattened.

Cluster (same digest to six decimals)	Members	Shared output on this bucket's input
12 implementations 56 papers 78125b9a9282 recorded values identical	DreamLLM: Synergistic Multimodal Comprehension and Creation A Paradigm Shift in Machine Translation: Boosting... Hierarchical Context Merging: Better Long Context... Semantic Density: Uncertainty Quantification for Large... Beyond Next Token Prediction: Patch-Level Training for... and 29 more papers holding this code (all in the JSON twin) one code sha held from 34 papers' repositories modeling_xalma.py 30d7eec4 recorded hidden_states:4/float/float32; n_rep=2 Time-MoE: Billion-Scale Time Series Foundation Models... KVzip: Query-Agnostic KV Cache Compression with Context... Draft-based Approximate Inference for LLMs Leave No Context Behind: Efficient Infinite Context... Natural language guidance of high-fidelity... and 5 more papers holding this code (all in the JSON twin) one code sha held from 10 papers' repositories time_moe/models/modeling_time_moe.py 3c76e528 recorded hidden_states:4/float/float32; n_rep=2 Transformer-XL: Attentive Language Models Beyond a... DashAttention: Differentiable and Adaptive Sparse... added by Syntology one code sha held from 2 papers' repositories src/gemma.py 4e762613 recorded hidden_states:4/float/float32; n_rep=2 Mixture of Lookup Experts Scaling Particle Collision Data Analysis one code sha held from 2 papers' repositories modeling_mole.py ab40781b recorded hidden_states:4/float/float32; n_rep=2 Beyond Steering Vector: Flow-based Activation Steering... added by Syntology src/flas/model.py 0f07b44f recorded hidden_states:4/float/float32; n_rep=2 SVD-LLM V2: Optimizing Singular Value Truncation for... SVDLLM.py 25ef0712 recorded hidden_states:4/float/float32; n_rep=2 LLaMA: Open and Efficient Foundation Language Models model.py 31333a36 recorded hidden_states:4/float/float32; n_rep=2 Attention Satisfies: A Constraint-Satisfaction Lens on... model_lib/attention_tools.py 3281dccd recorded hidden_states:4/float/float32; n_rep=2 ST-Lite: Training-Free KV Cache Compression with... added by Syntology eval/attention_helpers.py 404c0af5 recorded hidden_states:4/float/float32; n_rep=2 TroL: Traversal of Layers for Large Language and Vision Models trol/arch_internlm2/modeling_trol.py 6045450d recorded hidden_states:4/float/float32; n_rep=2 UniLMv2: Pseudo-Masked Language Models for Unified... Diff-Transformer/multihead_attention.py 8b9e23cf recorded x:4/float/float32; n_rep=2 LoLCATs: On Low-Rank Linearizing of Large Language Models src/model/linear_attention/linear_window_attention_sw_linear.py ab05d5e4 recorded hidden_states:4/float/float32; n_rep=2	[0.723105, 0.653079, 1.44609, 1.09854, 1.1373, -0.199927, -0.271487, 0.831979, …] shape [2, 6, 4, 4] · float32 · Tensor
3 implementations 4 papers 2f77db535cd3 recorded values identical	Bayesian Attention Mechanism: A Probabilistic Framework... Training Compute-Optimal Large Language Models one code sha held from 2 papers' repositories models/bam.py 6d2a08dc recorded x:4/float/float32; n_rep=2 Periodic RoPE for Infinite Context LLMs added by Syntology model/model_miniwin.py 07d6ef50 recorded x:4/float/float32; n_rep=2 MobileLLM: Optimizing Sub-billion Parameter Language... model/model_minimind.py 226c2c73 recorded x:4/float/float32; n_rep=2	[0.723105, 0.653079, 1.44609, 1.09854, 0.723105, 0.653079, 1.44609, 1.09854, …] shape [2, 3, 8, 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/repeat-kv.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