SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2601.02569 · 2026

LoRA-Drop: Temporal LoRA Decoding for Efficient LLM Inference

Il-Min Kim, Hossein Rajabzadeh, Hyock Kwon, Maryam Dialameh, Chul Park

arXiv · PDF · Open in the Atlas

Code that ran

6 of 8 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (5 of them constructed an object rather than computing a result; 0 honoured the contract we drafted, 0 violated it and 5 have no contract verdict; 1 failure of our instrument; 2 no recorded run), measured 2026-11-22; bodies inline only under a permissive licence as GitHub detects it. Not a reproduction of the paper's results.

"Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results. This record, dated, as JSON: /api/ran/2601.02569.json · badge: /api/ran/2601.02569.svg

Each repository's role is a label: a classifier's output recorded on the link, or, where no role is recorded, what the Papers with Code archive marked. The line under it names its basis. Neither says who wrote the code.

RepositoryRoleRan
hosseinbv/LoRA-Drop community
Papers with Code archive
6 ran · 2 no recorded run
FunctionStatusWhere it lives
AnchorOutputCache Ran
ran on a synthesized input · constructed an object; no result was computed
hosseinbv/LoRA-Drop/src/llamafactory/my_models/qwen2_lora_drop/modeling_qwen2_lora_drop.py
code served (permissive licence) · get_code("b35152d0331daf84")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned an instance of its own class, AnchorOutputCache: the run constructed the object and did not call it, so no result was computed. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch

torch.manual_seed(0)

ARGS = {
    "num_layers": 3,
    "device": None,
    "dtype": None,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type AnchorOutputCache — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a AnchorOutputCache output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
LoRADropAdapter Ran
ran on a synthesized input · constructed an object; no result was computed
hosseinbv/LoRA-Drop/src/llamafactory/my_models/qwen2_lora_drop/modeling_qwen2_lora_drop.py
code served (permissive licence) · get_code("7a15cc47294a5a73")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned an instance of its own class, LoRADropAdapter: the run constructed the object and did not call it, so no result was computed. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
from torch import nn
import math

torch.manual_seed(0)

ARGS = {
    "hidden_size": 8,
    "rank": 4,
    "alpha": 16.0,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type LoRADropAdapter — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a LoRADropAdapter output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
Qwen2_Lora_DropConfig Ran
ran on a synthesized input · constructed an object; no result was computed
hosseinbv/LoRA-Drop/src/llamafactory/my_models/qwen2_lora_drop/modeling_qwen2_lora_drop.py
code served (permissive licence) · get_code("a94662289330b9c8")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned an instance of its own class, Qwen2_Lora_DropConfig: the run constructed the object and did not call it, so no result was computed. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn
from transformers.configuration_utils import PretrainedConfig

torch.manual_seed(0)

# Construct arguments for Qwen2_Lora_DropConfig
ARGS = {
    "vocab_size": 151936,
    "hidden_size": 4096,
    "intermediate_size": 22016,
    "num_hidden_layers": 32,
    "num_attention_heads": 32,
    "num_key_value_heads": 32,
    "hidden_act": "silu",
    "max_position_embeddings": 32768,
    "initializer_range": 0.02,
    "rms_norm_eps": 1e-6,
    "use_cache": True,
    "tie_word_embeddings": False,
    "rope_theta": 10000.0,
    "rope_scaling": {
        "rope_type": "default",
        "factor": 1.0,
    },
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch, transformers — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Qwen2_Lora_DropConfig — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a Qwen2_Lora_DropConfig output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
Qwen2_Lora_DropMLP Ran
ran on a synthesized input · constructed an object; no result was computed
hosseinbv/LoRA-Drop/src/llamafactory/my_models/qwen2_lora_drop/modeling_qwen2_lora_drop.py
code served (permissive licence) · get_code("a6aa393721f0f595")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned an instance of its own class, Qwen2_Lora_DropMLP: the run constructed the object and did not call it, so no result was computed. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn
from transformers.activations import ACT2FN

torch.manual_seed(0)

class Config:
    def __init__(self):
        self.hidden_size = 4
        self.intermediate_size = 8
        self.hidden_act = "silu"

config = Config()

ARGS = {
    "config": config
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch, transformers — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Qwen2_Lora_DropMLP — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a Qwen2_Lora_DropMLP output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
Qwen2_Lora_DropRMSNorm Ran
ran on a synthesized input · constructed an object; no result was computed
hosseinbv/LoRA-Drop/src/llamafactory/my_models/qwen2_lora_drop/modeling_qwen2_lora_drop.py
code served (permissive licence) · get_code("9e976091a320bd6c")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned an instance of its own class, Qwen2_Lora_DropRMSNorm: the run constructed the object and did not call it, so no result was computed. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
from torch import nn

torch.manual_seed(0)

# Construct arguments for Qwen2_Lora_DropRMSNorm.__init__
hidden_size = 4
eps = 1e-6

ARGS = {
    'hidden_size': hidden_size,
    'eps': eps
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch, transformers — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Qwen2_Lora_DropRMSNorm — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a Qwen2_Lora_DropRMSNorm output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
eager_attention_forward Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
hosseinbv/LoRA-Drop/src/llamafactory/my_models/qwen2_lora_drop/modeling_qwen2_lora_drop.py
code served (permissive licence) · get_code("7019040ccf80940c")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned tuple. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn
from torch.nn import functional as F

torch.manual_seed(0)

# Create a minimal module that satisfies the function's requirements
class MinimalAttentionModule(nn.Module):
    def __init__(self, num_key_value_groups=2):
        super().__init__()
        self.num_key_value_groups = num_key_value_groups
        self.training = False
    
    def forward(self, x):
        return x

# Set up dimensions with distinct values
batch_size = 2
num_attention_heads = 4
num_key_value_heads = 2
seq_len = 3
head_dim = 5

# Create the module
module = MinimalAttentionModule(num_key_value_groups=2)

# Create tensors with appropriate shapes
# query: (batch, num_attention_heads, seqlen, head_dim)
query = torch.randn(batch_size, num_attention_heads, seq_len, head_dim)

# key: (batch, num_key_value_heads, seqlen, head_dim)
key = torch.randn(batch_size, num_key_value_heads, seq_len, head_dim)

# value: (batch, num_key_value_heads, seqlen, head_dim)
value = torch.randn(batch_size, num_key_value_heads, seq_len, head_dim)

# attention_mask: (batch, 1, seqlen, seqlen) - causal mask added to attention weights
attention_mask = torch.zeros(batch_size, 1, seq_len, seq_len, dtype=torch.float32)
# Make it causal (lower triangular)
attention_mask = torch.tril(torch.ones(seq_len, seq_len)) - 1
attention_mask = attention_mask * float('-inf')
attention_mask = attention_mask.unsqueeze(0).unsqueeze(0)
attention_mask = attention_mask.expand(batch_size, 1, -1, -1)

# scaling factor for attention
scaling = 1.0 / (head_dim ** 0.5)

# dropout probability
dropout = 0.1

ARGS = {
    "module": module,
    "query": query,
    "key": key,
    "value": value,
    "attention_mask": attention_mask,
    "scaling": scaling,
    "dropout": dropout,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch, transformers — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple — a tuple output has no array shape of its own; none is stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
Qwen2_Lora_DropAttention Not yet run
unverified (harvested, no recorded run)
hosseinbv/LoRA-Drop/src/llamafactory/my_models/qwen2_lora_drop/modeling_qwen2_lora_drop.py
code served (permissive licence) · get_code("70ab38f3d574a076")
How it ran

no invocation recorded.

Qwen2_Lora_DropDecoderLayer Not yet run
unverified (harvested, no recorded run)
hosseinbv/LoRA-Drop/src/llamafactory/my_models/qwen2_lora_drop/modeling_qwen2_lora_drop.py
code served (permissive licence) · get_code("d1bacaf27ebf7334")
How it ran

no invocation recorded.

Also present in other papers' code

Of 8 samples lifted from this paper's linked repositories, 8 carry a function identity and 0 do not. Of the 8 distinct functions with an identity, 0 are also present in other papers whose code we hold (0 with at least one paper dated earlier, 0 the same month, 0 later) and 8 are only here among the 52,977 papers whose code Syntology holds (as of 2026-09-26).

Only here means: among the 52,977 papers whose code Syntology holds (51,887 with lifted samples, 1,090 through their assemblies only), as of 2026-09-26, no other paper's lifted samples or current assembly bricks contain an identical function body after normalisation.

Presence, not direction: an identical function body in two papers' repositories does not say who wrote it, which came from which, or that either paper's method rests on it. Matching is exact after normalisation, so a renamed or edited function is not matched. Dates are when papers appeared, not when code was written. Only .py files the harvester lifted are covered; notebooks are not read. A new-style arXiv id YYMM.NNNN dates a paper to the month 20YY-MM; otherwise Paper.year dates it to the year; otherwise it is undated. Dates are when the paper appeared, not when the code was written. Read 2026-11-22.

Repositories linked to this paper

Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.

Abstract

Autoregressive large language models (LLMs) are bottlenecked by sequential decoding, where each new token typically requires executing all transformer layers. Existing dynamic-depth and layer-skipping methods reduce this cost, but often rely on auxiliary routing mechanisms or incur accuracy degradation when bypassed layers are left uncompensated. We present LoRA-Drop, a plug-and-play inference framework that accelerates decoding by applying a temporal compute schedule to a fixed subset of intermediate layers: on most decoding steps, selected layers reuse the previous-token hidden state and apply a low-rank LoRA correction, while periodic refresh steps execute the full model to prevent drift. LoRA-Drop requires no routing network, is compatible with standard KV caching, and can reduce KV-cache footprint by skipping KV updates in droppable layers during LoRA steps and refreshing periodically. Across LLaMA2-7B, LLaMA3-8B, Qwen2.5-7B, and Qwen2.5-14B, LoRA-Drop achieves up to 2.6× faster decoding and 45-55% KV-cache reduction while staying within 0.5 percentage points (pp) of baseline accuracy. Evaluations on reasoning (GSM8K, MATH, BBH), code generation (HumanEval, MBPP), and longcontext/multilingual benchmarks (LongBench, XNLI, XCOPA) identify a consistent safe zone of scheduling configurations that preserves quality while delivering substantial efficiency gains, providing a simple path toward adaptive-capacity inference in LLMs. Codes are available at https://github.com/hosseinbv/LoRA-Drop.git.

For agents

The same record, over MCP at https://syntology.ai/mcp:

get_harvested_code_for_paper("2601.02569")
get_code_for_paper("2601.02569")
have("2601.02569")

The run record, dated, one paper per request, free:

curl https://syntology.ai/api/ran/2601.02569.json

A badge for a README (the split and the date, never a ratio):

[![Syntology run record](https://syntology.ai/api/ran/2601.02569.svg)](https://syntology.ai/paper/2601.02569)

Connect an agent — have() is free.