SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2602.03473 · ICML · 2026

Scaling Continual Learning to 300+ Tasks with Bi-Level Routing Mixture-of-Experts

Yizhou Yu, Meng Lou, Yunxiang Fu

arXiv · PDF · Open in the Atlas

Code that ran

4 of 5 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (3 of them constructed an object rather than computing a result; 0 honoured the contract we drafted, 0 violated it and 4 have no contract verdict; 0 failures of our instrument; 1 no recorded run), measured 2026-11-21; 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/2602.03473.json · badge: /api/ran/2602.03473.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
LMMMEng/CaRE community
Papers with Code archive
4 ran · 1 no recorded run
FunctionStatusWhere it lives
Adapter Ran
ran on a synthesized input · constructed an object; no result was computed
LMMMEng/CaRE/backbone/vit_brmoe.py
code served (permissive licence) · get_code("7b7f101ebda35088")
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, Adapter: 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
import math

torch.manual_seed(0)

class Config:
    def __init__(self):
        self.d_model = 8
        self.attn_bn = 4

config = Config()

ARGS = {
    'config': config,
    'd_model': None,
    'bottleneck': None,
    'dropout': 0.1,
    'adapter_layernorm_option': 'in',
    'non_linear_func': nn.ReLU
}
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 Adapter — 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 Adapter 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
Attention Ran
ran on a synthesized input
LMMMEng/CaRE/backbone/vit_brmoe.py
code served (permissive licence) · get_code("52a9b5305ff1f5ef")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, class construct-then-call prompt, model not recorded) and returned Tensor, ndim 3, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim 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

# INIT_ARGS for Attention.__init__
INIT_ARGS = {
    'dim': 12,
    'num_heads': 3,
    'qkv_bias': True,
    'attn_drop': 0.1,
    'proj_drop': 0.1
}

# ARGS for Attention.forward
ARGS = {
    'x': torch.randn(2, 5, 12)  # (batch_size=2, seq_len=5, dim=12)
}
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 3, dtype family float — no output shape is stated: observe_signature.py observed this class with its constructor sizes canonicalised where it could, which the node does not record, so its shape is not stated as this setup block's
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
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: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
Block Ran
ran on a synthesized input · constructed an object; no result was computed
LMMMEng/CaRE/backbone/vit_brmoe.py
code served (permissive licence) · get_code("e5a863df7a76f7e2")
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, Block: 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 timm.models.vision_transformer import DropPath

torch.manual_seed(0)

# Stand-in config class with attributes used in Block.forward()
class Config:
    def __init__(self):
        self.ffn_option = 'sequential'

# Stand-in adapt module that matches the signature used in Block.forward()
class AdaptModule(nn.Module):
    def __init__(self, dim):
        super().__init__()
        self.dim = dim
    
    def forward(self, x, add_residual=True, task_id=0, with_task_id=False):
        # Returns (adapted_x, cls_token)
        cls = torch.randn(x.shape[0], 1, self.dim)
        return x, cls

# Build the config
config = Config()

# Build the adapt module
adapt_module = AdaptModule(dim=4)

# Define all arguments for Block.__init__
ARGS = {
    'dim': 4,
    'num_heads': 2,
    'mlp_ratio': 4,
    'qkv_bias': False,
    'drop': 0.0,
    'attn_drop': 0.0,
    'drop_path': 0.0,
    'act_layer': nn.GELU,
    'norm_layer': nn.LayerNorm,
    'config': config,
    'layer_id': 0,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
timm, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Block — 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 Block 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
MoEAdapter Ran
ran on a synthesized input · constructed an object; no result was computed
LMMMEng/CaRE/backbone/vit_brmoe.py
code served (permissive licence) · get_code("682e645892e64729")
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, MoEAdapter: 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
import copy
import math

torch.manual_seed(0)

class MockConfig:
    def __init__(self):
        self.d_model = 64
        self.attn_bn = 16
        self.expert_dim = 8
        self.adapter_dim = 32
        self.topk_routers = 2
        self.topk_experts = 3
        self.init_cls = 10
        self._device = 'cpu'
    
    def get(self, key, default=None):
        return getattr(self, key, default)

config = MockConfig()

ARGS = {
    'config': config,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
einops, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type MoEAdapter — 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 MoEAdapter 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
VisionTransformer Not yet run
unverified (harvested, no recorded run)
LMMMEng/CaRE/backbone/vit_brmoe.py
code served (permissive licence) · get_code("afa43a669cda3d4f")
How it ran

no invocation recorded.

Also present in other papers' code

Of 5 samples lifted from this paper's linked repositories, 5 carry a function identity and 0 do not. Of the 5 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 5 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 2026-11-21.

Repositories linked to this paper

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

Abstract

Continual learning, especially class-incremental learning (CIL), on the basis of a pre-trained model (PTM) has garnered substantial research interest in recent years. However, how to effectively learn both discriminative and comprehensive feature representations while maintaining stability and plasticity over very long task sequences remains an open problem. We propose CaRE, a scalable Continual Learner with efficient Bi-Level Routing Mixture-of-Experts (BR-MoE). The core idea of BR-MoE is a bi-level routing mechanism: a router selection stage that dynamically activates relevant task-specific routers, followed by an expert routing phase that dynamically activates and aggregates experts, aiming to inject discriminative and comprehensive representations into every intermediate network layer. On the other hand, we introduce a challenging dataset, OmniBenchmark-1K, for CIL performance evaluation on very long task sequences with hundreds of tasks. Extensive experiments show that CaRE demonstrates leading performance across a variety of datasets and task settings, including commonly used CIL datasets with classical CIL settings (e.g., 5-20 tasks). To the best of our knowledge, CaRE is the first continual learner that scales to very long task sequences (ranging from 100 to over 300 non-overlapping tasks), while outperforming all baselines by a large margin on such task sequences. We hope that this work will inspire further research into continual learning over extremely long task sequences. Code and dataset are publicly released at https://github.com/LMMMEng/CaRE.

For agents

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

get_harvested_code_for_paper("2602.03473")
get_code_for_paper("2602.03473")
have("2602.03473")

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

curl https://syntology.ai/api/ran/2602.03473.json

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

[![Syntology run record](https://syntology.ai/api/ran/2602.03473.svg)](https://syntology.ai/paper/2602.03473)

Connect an agent — have() is free.