Papers › LoLCATs: On Low-Rank Linearizing of Large Language Models

LoLCATs: On Low-Rank Linearizing of Large Language Models

14 Oct 2024arXiv:2410.10254archive 2025-07-28

Michael Zhang, Simran Arora, Rahul Chalamala, Alan Wu, Benjamin Spector, Aaryan Singhal, Krithik Ramesh, Christopher Ré

Recent works show we can linearize large language models (LLMs) -- swapping the quadratic attentions of popular Transformer-based LLMs with subquadratic analogs, such as linear attention -- avoiding the expensive pretraining costs. However, linearizing LLMs often significantly degrades model quality, still requires training over billions of tokens, and remains limited to smaller 1.3B to 7B LLMs. We thus propose Low-rank Linear Conversion via Attention Transfer (LoLCATs), a simple two-step method that improves LLM linearizing quality with orders of magnitudes less memory and compute. We base these steps on two findings. First, we can replace an LLM's softmax attentions with closely-approximating linear attentions, simply by training the linear attentions to match their softmax counterparts with an output MSE loss ("attention transfer"). Then, this enables adjusting for approximation errors and recovering LLM quality simply with low-rank adaptation (LoRA). LoLCATs significantly improves linearizing quality, training efficiency, and scalability. We significantly reduce the linearizing quality gap and produce state-of-the-art subquadratic LLMs from Llama 3 8B and Mistral 7B v0.1, leading to 20+ points of improvement on 5-shot MMLU. Furthermore, LoLCATs does so with only 0.2% of past methods' model parameters and 0.4% of their training tokens. Finally, we apply LoLCATs to create the first linearized 70B and 405B LLMs (50x larger than prior work). When compared with prior approaches under the same compute budgets, LoLCATs significantly improves linearizing quality, closing the gap between linearized and original Llama 3.1 70B and 405B LLMs by 77.8% and 78.1% on 5-shot MMLU.

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 32 code samples harvested from 1 repository linked to this paper: 23 ran (of which 9 constructed an object rather than computing a result; 13 with no instrument failure: 0 honoured, 0 violated, 13 with no contract checked; 10 where Syntology's instrument failed) · 9 unverified. Where Syntology's instrument failed: 7 where our contract draft was wrong; 3 where the contract check could not run; failures of Syntology's instrument, not of the code.

By repository: official repository: 32 samples from 1 repository, 23 ran (13 with no instrument failure; 10 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 between 2026-09-22 and 2026-09-24 (UTC): the dates Syntology's graph wrote each sample's verification level, not when the code executed. 1 of the 23 samples that ran has no recorded date.

hazyresearch/lolcats officialmentioned in papermentioned on GitHubpytorchSyntology: 32 samples, 23 ran (10 instrument-failed) · 9 unverified 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.

Code Syntology ran Syntology

32 samples harvested: 23 ran (of which 9 constructed an object rather than computing a result; 13 with no instrument failure: 0 honoured, 0 violated, 13 with no contract checked; 10 where Syntology's instrument failed) · 9 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.

7ran · our draft was wrong
3ran; the contract check could not run
13ran
9unverified

Licence: 0 of the 32 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 hazyresearch/lolcats. “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”. For 4 samples that ran, the graph records a signature this page does not show: for a class the fixture built without calling, the harness called the object with inputs it made up itself, so the recorded signature is the harness's probe, not the fixture's arguments. “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.

hazyresearch/lolcats official repository 32 samples, 23 ran (10 instrument-failed) · 9 unverified
DynamicNTKScalingRotaryEmbedding src/model/linear_attention/linear_window_attention_sw_linear.py ran Apache-2.0 (permissive) · 363df15c92df5f35 · report fixture: returned DynamicNTKScalingRotaryEmbedding · level recorded 2026-09-22
Exp src/model/linear_attention/linear_window_attention_sw_linear.py ran fingerprinted Apache-2.0 (permissive) · cdf03be65a92f940 · report fixture: signature x:Tensor/4/float · returned Tensor · level recorded 2026-09-24
ExpHalfspace src/model/linear_attention/linear_window_attention_sw_linear.py ran fingerprinted Apache-2.0 (permissive) · 72f5147b8da94cef · report fixture: returned ExpHalfspace · level recorded 2026-09-22
FeatureMap src/model/linear_attention/linear_window_attention_sw_linear.py ran fingerprinted Apache-2.0 (permissive) · d60bb1001847bf81 · report fixture: returned FeatureMap · level recorded 2026-09-22
FeatureMapAct src/model/linear_attention/linear_window_attention_sw_linear.py ran fingerprinted Apache-2.0 (permissive) · 8508acb10b82cae5 · report fixture: returned FeatureMapAct · level recorded 2026-09-22
FeatureMapAdapter src/model/linear_attention/linear_window_attention_sw_linear.py ran Apache-2.0 (permissive) · 11df4b1bdd5a0634 · report fixture: returned FeatureMapAdapter · level recorded 2026-09-22
FeatureMapMLP src/model/linear_attention/linear_window_attention_sw_linear.py ran Apache-2.0 (permissive) · 624f03bff20eeff4 · report fixture: returned FeatureMapMLP · level recorded 2026-09-22
LinearScalingRotaryEmbedding src/model/linear_attention/linear_window_attention_sw_linear.py ran Apache-2.0 (permissive) · 8e7478585c4a072f · report fixture: returned LinearScalingRotaryEmbedding · level recorded 2026-09-22
PosELU src/model/linear_attention/linear_window_attention_sw_linear.py ran fingerprinted Apache-2.0 (permissive) · 222e712a93c7c816 · report fixture: returned PosELU · level recorded 2026-09-22
ReLU in 8 papers src/model/linear_attention/linear_window_attention_sw_linear.py ran fingerprinted Apache-2.0 (permissive) · f8a915d0998d5a10 · report fixture: signature x:Tensor/4/float · returned Tensor · level recorded 2026-09-24
RotaryEmbedding in 27 papers src/model/linear_attention/linear_window_attention_sw_linear.py ran Apache-2.0 (permissive) · c2be9851321e656a · report fixture: returned RotaryEmbedding · level recorded 2026-09-22
SoftmaxDim src/model/linear_attention/linear_window_attention_sw_linear.py ran fingerprinted Apache-2.0 (permissive) · 45348551e263b890 · report fixture: signature x:Tensor/4/float · returned Tensor · level recorded 2026-09-24
SoftmaxDimHalfspace src/model/linear_attention/linear_window_attention_sw_linear.py ran fingerprinted Apache-2.0 (permissive) · 79ecf2b5021ec165 · report fixture: signature x:Tensor/4/float · returned Tensor · 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
src/model/linear_attention/linear_window_attention_sw_linear.py ran; the contract check could not run (RAISES) Apache-2.0 (permissive) · 8f7f3f7e4221d5c0 · report fixture: signature q:Tensor/3/float, k:Tensor/3/float, cos:Tensor/2/float, sin:Tensor/2/float, position_ids:Tensor/2/int, unsqueeze_dim:int/0/int · returned tuple · level recorded 2026-09-22
get_rotary_embeddings src/model/linear_attention/linear_window_attention_sw_linear.py ran · our draft was wrong Apache-2.0 (permissive) · 3d075a97ebad15d7 · report fixture: signature rope_scaling_type:str/0/str, head_dim:int/0/int, max_position_embeddings:int/0/int, rope_theta:float/0/float, rope_scaling_factor:float/0/float, device:NoneType/0/? · returned LinearScalingRotaryEmbedding · level recorded 2026-09-22
init_feature_map src/model/linear_attention/linear_window_attention_sw_linear.py ran · our draft was wrong Apache-2.0 (permissive) · 4fcb658ea76a7904 · report fixture: signature name:str/0/str, mlp:callable/0/?, head_dim_idx:int/0/int, eps:float/0/float, fullspace:bool/0/bool · returned FeatureMap · level recorded 2026-09-22
init_feature_map_act src/model/linear_attention/linear_window_attention_sw_linear.py ran · our draft was wrong Apache-2.0 (permissive) · d9fed240ed04f9af · report fixture: signature name:str/0/str, fullspace:bool/0/bool, eps:float/0/float · returned SoftmaxDim · level recorded 2026-09-22
init_learned_kernel src/model/linear_attention/linear_window_attention_sw_linear.py ran · our draft was wrong Apache-2.0 (permissive) · 7ea7286d8db1055b · report fixture: signature name:str/0/str, num_heads:int/0/int, head_dim:int/0/int, feature_dim:int/0/int, dtype:dtype/0/?, device:device/0/?, skip_connection:bool/0/bool, bias:bool/0/bool, zero_init:bool/0/bool, normal_init:bool/0/bool · returned FeatureMapMLP · level recorded 2026-09-22
quadratic_attention src/model/linear_attention/linear_window_attention_sw_linear.py ran; the contract check could not run (RAISES) fingerprinted Apache-2.0 (permissive) · 95f5da87ec68a1ed · report fixture: signature q:Tensor/4/float, k:Tensor/4/float, v:Tensor/4/float, causal:bool/0/bool, fp32_attention:bool/0/bool, eps:float/0/float · returned tuple · level recorded 2026-09-22
repeat_kv in 244 papers
census: 15 implementations from 60 papers, 2 distinct outputs on one shared input this sample: bucket 1 of 1, arg 1: rank 4, kind float, dtype float32
src/model/linear_attention/linear_window_attention_sw_linear.py ran; the contract check could not run (RAISES) fingerprinted Apache-2.0 (permissive) · ab05d5e4c4236572 · report fixture: signature hidden_states:Tensor/4/float, n_rep:int/0/int · returned Tensor · level recorded 2026-09-22
sliding_window_softmax_attention src/model/linear_attention/linear_window_attention_sw_linear.py ran · our draft was wrong fingerprinted Apache-2.0 (permissive) · 8e336d9e009bb875 · report fixture: signature q:Tensor/4/float, k:Tensor/4/float, v:Tensor/4/float, window_size:int/0/int, window_factor:float/0/float, mask_value:float/0/float · returned tuple · level recorded 2026-09-22
softmax_attention src/model/linear_attention/linear_attention.py ran · our draft was wrong Apache-2.0 (permissive) · 95392cd040b0c2e7 · report fixture: signature q:Tensor/4/float, k:Tensor/4/float, v:Tensor/4/float, causal:bool/0/bool, fp32_attention:bool/0/bool · returned tuple · level recorded, date not recorded
softmax_attention src/model/linear_attention/linear_window_attention_sw_linear.py ran · our draft was wrong Apache-2.0 (permissive) · f122b39d7b617e75 · report fixture: signature q:Tensor/4/float, k:Tensor/4/float, v:Tensor/4/float, causal:bool/0/bool, fp32_attention:bool/0/bool · returned tuple · level recorded 2026-09-22
LolcatsLinearAttention src/model/linear_attention/linear_window_attention_sw_linear.py unverified Apache-2.0 (permissive) · 5ca6886743273ec2 · report
LolcatsLinearSlidingWindowAttention src/model/linear_attention/linear_window_attention_sw_linear.py unverified Apache-2.0 (permissive) · 22e03adc01e6f4af · report
causal_dot_product src/model/linear_attention/linear_attention.py unverified Apache-2.0 (permissive) · c61cbc5f0923bbf5 · report
evaluate_model in 54 papers distill_llama.py unverified Apache-2.0 (permissive) · 99199bb3bba7f5ed · report
flash_attention_2 src/model/linear_attention/linear_window_attention_sw_linear.py unverified Apache-2.0 (permissive) · 7adc1fdae7b7a47d · report
hybrid_attention_linear src/model/linear_attention/linear_window_attention_sw_linear.py unverified Apache-2.0 (permissive) · 6cc2057faa9846d2 · report
linear_attention in 7 papers src/model/linear_attention/linear_attention.py unverified Apache-2.0 (permissive) · d8595b54ec29491c · report
linear_attention in 7 papers src/model/linear_attention/linear_window_attention_sw_linear.py unverified Apache-2.0 (permissive) · 61e674e010a5aa26 · report
under_window_linear_attention src/model/linear_attention/linear_window_attention_sw_linear.py unverified Apache-2.0 (permissive) · 20430ccdc09ccdc0 · 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 LoLCATs (graph method: LoLCATs). 30 bricks; individually on a fixture: 23 ran (of which 9 constructed an object rather than computing a result; 13 with no instrument failure: 0 honoured, 0 violated, 13 with no contract checked; 10 where Syntology's instrument failed) · 7 unverified. The assembly has not been run end-to-end.

  1. LolcatsLinearSlidingWindowAttention entry · level 0 · unverified · 22e03adc01e6f4af
  2. LolcatsLinearAttention helper · level 0 · unverified · 5ca6886743273ec2
  3. flash_attention_2 helper · level 0 · unverified · 7adc1fdae7b7a47d
  4. hybrid_attention_linear model · level 0 · unverified · 6cc2057faa9846d2
  5. apply_rotary_pos_emb helper · level 1 · ran; the contract check could not run (RAISES) · 8f7f3f7e4221d5c0
  6. get_rotary_embeddings helper · level 1 · ran · our draft was wrong · 3d075a97ebad15d7
  7. init_feature_map helper · level 1 · ran · our draft was wrong · 4fcb658ea76a7904
  8. init_learned_kernel helper · level 1 · ran · our draft was wrong · 7ea7286d8db1055b
  9. linear_attention helper · level 0 · unverified · 61e674e010a5aa26
  10. quadratic_attention helper · level 1 · ran; the contract check could not run (RAISES) · 95f5da87ec68a1ed
  11. repeat_kv helper · level 1 · ran; the contract check could not run (RAISES) · ab05d5e4c4236572
  12. softmax_attention helper · level 1 · ran · our draft was wrong · f122b39d7b617e75
  13. sliding_window_softmax_attention model · level 1 · ran · our draft was wrong · 8e336d9e009bb875
  14. under_window_linear_attention model · level 0 · unverified · 20430ccdc09ccdc0
  15. rotate_half helper · level 1 · ran · our draft was wrong · e03d53ba9d4f9ae5
  16. DynamicNTKScalingRotaryEmbedding helper · level 1 · ran · 363df15c92df5f35
  17. LinearScalingRotaryEmbedding helper · level 1 · ran · 8e7478585c4a072f
  18. RotaryEmbedding helper · level 1 · ran · c2be9851321e656a
  19. FeatureMap helper · level 1 · ran · d60bb1001847bf81
  20. FeatureMapAdapter helper · level 1 · ran · 11df4b1bdd5a0634
  21. FeatureMapMLP helper · level 1 · ran · 624f03bff20eeff4
  22. causal_dot_product helper · level 0 · unverified · c61cbc5f0923bbf5
  23. init_feature_map_act helper · level 1 · ran · our draft was wrong · d9fed240ed04f9af
  24. Exp helper · level 1 · ran · cdf03be65a92f940
  25. ExpHalfspace helper · level 1 · ran · 72f5147b8da94cef
  26. PosELU helper · level 1 · ran · 222e712a93c7c816
  27. ReLU helper · level 1 · ran · f8a915d0998d5a10
  28. SoftmaxDim helper · level 1 · ran · 45348551e263b890
  29. SoftmaxDimHalfspace helper · level 1 · ran · 79ecf2b5021ec165
  30. FeatureMapAct helper · level 1 · ran · 8508acb10b82cae5

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="2410.10254") returns these assemblies with each brick's role and wiring (how to connect).

Tasks

MMLU

Results from the paper archive 2025-07-28

No leaderboard rows for this paper in the archive.

Methods

AttentionBASELLaMASoftmax

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