SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2102.07108 · ICML · 2021

CATE: Computation-aware Neural Architecture Encoding with Transformers

Shen Yan, Mi Zhang, Kaiqiang Song, Fei Liu

arXiv · PDF · Open in the Atlas

Code that ran

9 of 10 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (7 of them constructed an object rather than computing a result; 0 honoured the contract we drafted, 0 violated it and 7 have no contract verdict; 2 failures of our instrument; 1 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/2102.07108.json · badge: /api/ran/2102.07108.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
MSU-MLSys-Lab/CATE marked official in Papers with Code
Papers with Code archive
9 ran · 1 no recorded run
FunctionStatusWhere it lives
Encoder Ran
ran on a synthesized input · constructed an object; no result was computed
MSU-MLSys-Lab/CATE/layers/transformer.py
code served (permissive licence) · get_code("ad464810b30d6d25")
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, Encoder: 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 = 4
        self.n_head = 2
        self.d_ff = 8
        self.dropout = 0.1
        self.n_layers = 2

config = Config()

batch_size = 2
seq_len = 3
x = torch.randn(batch_size, seq_len, config.d_model)
mask = torch.ones(batch_size, seq_len)

ARGS = {
    "config": config,
}
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 Encoder — 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 Encoder 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
EncoderBlock Ran
ran on a synthesized input · constructed an object; no result was computed
MSU-MLSys-Lab/CATE/layers/transformer.py
code served (permissive licence) · get_code("2b2a73a9b0d3825e")
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, EncoderBlock: 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 copy import deepcopy as cp
import math
import torch.nn.functional as F

torch.manual_seed(0)

# Helper functions needed by the module
def clones(module, N):
    return nn.ModuleList([cp(module) for _ in range(N)])

def attention(query, key, value, mask=None, dropout=None):
    d_k = query.size(-1)
    scores = torch.matmul(query, key.transpose(-2, -1)) / math.sqrt(d_k)
    if mask is not None:
        scores = scores.masked_fill(mask == 0, -1000000000.0)
    attn = F.softmax(scores, dim=-1)
    if dropout is not None:
        attn = dropout(attn)
    return (torch.matmul(attn, value), attn)

# Define the config class
class Config:
    def __init__(self):
        self.d_model = 8
        self.n_head = 2
        self.d_ff = 16
        self.dropout = 0.1

# Define the building block classes
class PositionwiseFeedForward(nn.Module):
    def __init__(self, config):
        super(PositionwiseFeedForward, self).__init__()
        self.w_1 = nn.Linear(config.d_model, config.d_ff)
        self.w_2 = nn.Linear(config.d_ff, config.d_model)
        self.dropout = nn.Dropout(p=config.dropout)

    def forward(self, x):
        return self.w_2(self.dropout(F.relu(self.w_1(x))))

class MultiHeadAttention(nn.Module):
    def __init__(self, config):
        super(MultiHeadAttention, self).__init__()
        self.d_model = config.d_model
        self.n_head = config.n_head
        self.d_k = config.d_model // config.n_head
        self.linears = clones(nn.Linear(self.d_model, self.d_model), 4)
        self.dropout = nn.Dropout(p=config.dropout)

    def forward(self, query, key, value, mask=None):
        if mask is not None:
            mask = mask.unsqueeze(1)
        batch_size = query.size(0)
        query, key, value = [l(x).view(batch_size, -1, self.n_head, self.d_k).transpose(1, 2) for l, x in zip(self.linears, (query, key, value))]
        x, attn = attention(query, key, value, mask=mask, dropout=self.dropout)
        x = x.transpose(1, 2).contiguous().view(batch_size, -1, self.n_head * self.d_k)
        return (self.linears[3](x), attn)

class FeedForwardBlock(nn.Module):
    def __init__(self, config):
        super(FeedForwardBlock, self).__init__()
        self.norm = nn.LayerNorm(config.d_model)
        self.feed_forward = PositionwiseFeedForward(config)
        self.dropout = nn.Dropout(p=config.dropout)

    def forward(self, x):
        x_ = self.norm(x)
        x_ = self.feed_forward(x_)
        return self.dropout(x_) + x

class SelfAttentionBlock(nn.Module):
    def __init__(self, config):
        super(SelfAttentionBlock, self).__init__()
        self.norm = nn.LayerNorm(config.d_model)
        self.attn = MultiHeadAttention(config)
        self.dropout = nn.Dropout(p=config.dropout)

    def forward(self, x, mask):
        x_ = self.norm(x)
        x_, attn = self.attn(x_, x_, x_, mask)
        return (self.dropout(x_) + x, attn)

class EncoderBlock(nn.Module):
    def __init__(self, config):
        super(EncoderBlock, self).__init__()
        self.self_attn = SelfAttentionBlock(config)
        self.feed_forward = FeedForwardBlock(config)

    def forward(self, x, mask):
        x, attn = self.self_attn(x, mask)
        x = self.feed_forward(x)
        return (x, attn)

# Create config
config = Config()

# Create sample inputs
batch_size = 2
seq_len = 3
d_model = 8

x = torch.randn(batch_size, seq_len, d_model)
mask = torch.ones(batch_size, 1, seq_len, seq_len)

ARGS = {
    "config": config
}
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 EncoderBlock — 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 EncoderBlock 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
FeedForwardBlock Ran
ran on a synthesized input · constructed an object; no result was computed
MSU-MLSys-Lab/CATE/layers/transformer.py
code served (permissive licence) · get_code("9e5a8995a0d0b684")
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, FeedForwardBlock: 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

torch.manual_seed(0)

class Config:
    def __init__(self, d_model, d_ff, dropout):
        self.d_model = d_model
        self.d_ff = d_ff
        self.dropout = dropout

config = Config(d_model=4, d_ff=8, dropout=0.1)

x = torch.randn(2, 3, 4)

ARGS = {
    "config": config
}
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 FeedForwardBlock — 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 FeedForwardBlock 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
MultiHeadAttention Ran
ran on a synthesized input · constructed an object; no result was computed
MSU-MLSys-Lab/CATE/layers/transformer.py
code served (permissive licence) · get_code("ef066f2fb629b698")
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, MultiHeadAttention: 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 copy import deepcopy as cp
import math
import torch.nn.functional as F

torch.manual_seed(0)

def attention(query, key, value, mask=None, dropout=None):
    d_k = query.size(-1)
    scores = torch.matmul(query, key.transpose(-2, -1)) / math.sqrt(d_k)
    if mask is not None:
        scores = scores.masked_fill(mask == 0, -1000000000.0)
    attn = F.softmax(scores, dim=-1)
    if dropout is not None:
        attn = dropout(attn)
    return (torch.matmul(attn, value), attn)

def clones(module, N):
    return nn.ModuleList([cp(module) for _ in range(N)])

class MultiHeadAttention(nn.Module):

    def __init__(self, config):
        super(MultiHeadAttention, self).__init__()
        self.d_model = config.d_model
        self.n_head = config.n_head
        self.d_k = config.d_model // config.n_head
        self.linears = clones(nn.Linear(self.d_model, self.d_model), 4)
        self.dropout = nn.Dropout(p=config.dropout)

    def forward(self, query, key, value, mask=None):
        if mask is not None:
            mask = mask.unsqueeze(1)
        batch_size = query.size(0)
        query, key, value = [l(x).view(batch_size, -1, self.n_head, self.d_k).transpose(1, 2) for l, x in zip(self.linears, (query, key, value))]
        x, attn = attention(query, key, value, mask=mask, dropout=self.dropout)
        x = x.transpose(1, 2).contiguous().view(batch_size, -1, self.n_head * self.d_k)
        return (self.linears[3](x), attn)

class Config:
    def __init__(self, d_model, n_head, dropout):
        self.d_model = d_model
        self.n_head = n_head
        self.dropout = dropout

config = Config(d_model=4, n_head=2, dropout=0.1)

ARGS = {
    "config": config
}
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 MultiHeadAttention — 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 MultiHeadAttention 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
PositionalEmbedding Ran
ran on a synthesized input · constructed an object; no result was computed
MSU-MLSys-Lab/CATE/layers/transformer.py
code served (permissive licence) · get_code("21d20b888bdba6a4")
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, PositionalEmbedding: 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, max_len, d_model):
        self.max_len = max_len
        self.d_model = d_model

config = Config(max_len=8, d_model=4)

ARGS = {
    "config": config
}
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 PositionalEmbedding — 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 PositionalEmbedding 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
PositionwiseFeedForward Ran
ran on a synthesized input · constructed an object; no result was computed
MSU-MLSys-Lab/CATE/layers/transformer.py
code served (permissive licence) · get_code("73d36f4b75757093")
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, PositionwiseFeedForward: 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

torch.manual_seed(0)

class Config:
    def __init__(self):
        self.d_model = 4
        self.d_ff = 8
        self.dropout = 0.1

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 — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type PositionwiseFeedForward — 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 PositionwiseFeedForward 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
SelfAttentionBlock Ran
ran on a synthesized input · constructed an object; no result was computed
MSU-MLSys-Lab/CATE/layers/transformer.py
code served (permissive licence) · get_code("52117ec9c56aa17a")
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, SelfAttentionBlock: 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 copy import deepcopy as cp
import math
import torch.nn.functional as F

torch.manual_seed(0)

# Define helper functions needed by the module
def clones(module, N):
    return nn.ModuleList([cp(module) for _ in range(N)])

def attention(query, key, value, mask=None, dropout=None):
    d_k = query.size(-1)
    scores = torch.matmul(query, key.transpose(-2, -1)) / math.sqrt(d_k)
    if mask is not None:
        scores = scores.masked_fill(mask == 0, -1000000000.0)
    attn = F.softmax(scores, dim=-1)
    if dropout is not None:
        attn = dropout(attn)
    return (torch.matmul(attn, value), attn)

# Define MultiHeadAttention as required
class MultiHeadAttention(nn.Module):
    def __init__(self, config):
        super(MultiHeadAttention, self).__init__()
        self.d_model = config.d_model
        self.n_head = config.n_head
        self.d_k = config.d_model // config.n_head
        self.linears = clones(nn.Linear(self.d_model, self.d_model), 4)
        self.dropout = nn.Dropout(p=config.dropout)

    def forward(self, query, key, value, mask=None):
        if mask is not None:
            mask = mask.unsqueeze(1)
        batch_size = query.size(0)
        query, key, value = [l(x).view(batch_size, -1, self.n_head, self.d_k).transpose(1, 2) for l, x in zip(self.linears, (query, key, value))]
        x, attn = attention(query, key, value, mask=mask, dropout=self.dropout)
        x = x.transpose(1, 2).contiguous().view(batch_size, -1, self.n_head * self.d_k)
        return (self.linears[3](x), attn)

# Define SelfAttentionBlock
class SelfAttentionBlock(nn.Module):
    def __init__(self, config):
        super(SelfAttentionBlock, self).__init__()
        self.norm = nn.LayerNorm(config.d_model)
        self.attn = MultiHeadAttention(config)
        self.dropout = nn.Dropout(p=config.dropout)

    def forward(self, x, mask):
        x_ = self.norm(x)
        x_, attn = self.attn(x_, x_, x_, mask)
        return (self.dropout(x_) + x, attn)

# Create a config class with required attributes
class Config:
    def __init__(self):
        self.d_model = 4
        self.n_head = 2
        self.dropout = 0.1

# Create config instance
config = Config()

# Create input tensors
# batch_size=2, seq_len=3, d_model=4
x = torch.randn(2, 3, 4)
mask = torch.ones(2, 1, 3, 3)

ARGS = {
    "config": config
}
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 SelfAttentionBlock — 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 SelfAttentionBlock 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; our contract draft was wrong, not the code
MSU-MLSys-Lab/CATE/layers/transformer.py
code served (permissive licence) · get_code("bbc7c22f9d8732dc")
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

torch.manual_seed(0)

# Construct tiny tensors for attention mechanism
# batch_size=2, seq_len=3, d_k=4
batch_size = 2
seq_len = 3
d_k = 4

query = torch.randn(batch_size, seq_len, d_k)
key = torch.randn(batch_size, seq_len, d_k)
value = torch.randn(batch_size, seq_len, d_k)

# Create a mask (optional parameter)
# mask should be broadcastable to scores shape (batch_size, seq_len, seq_len)
# Use 1s where we want to attend, 0s where we want to mask
mask = torch.ones(batch_size, seq_len, seq_len)

# Create a dropout layer (optional parameter)
dropout = nn.Dropout(p=0.1)

ARGS = {
    "query": query,
    "key": key,
    "value": value,
    "mask": mask,
    "dropout": dropout,
}
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 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
clones Ran
ran; our contract draft was wrong, not the code
MSU-MLSys-Lab/CATE/layers/transformer.py
code served (permissive licence) · get_code("d013dfb22c365f67")
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 ModuleList. 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 copy import deepcopy as cp

torch.manual_seed(0)

# The clones function takes:
# - module: an nn.Module to be cloned N times
# - N: an integer specifying how many clones to create

# Create a simple nn.Module for testing
class SimpleModule(nn.Module):
    def __init__(self):
        super().__init__()
        self.linear = nn.Linear(4, 3)
    
    def forward(self, x):
        return self.linear(x)

module = SimpleModule()
N = 5

ARGS = {
    "module": module,
    "N": N,
}
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 ModuleList — a ModuleList output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a ModuleList output the recorded ndim 1 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
Transformer Not yet run
unverified (harvested, no recorded run)
MSU-MLSys-Lab/CATE/layers/transformer.py
code served (permissive licence) · get_code("014afd02662c3b66")
How it ran

no invocation recorded.

Also present in other papers' code

Of 10 samples lifted from this paper's linked repositories, 10 carry a function identity and 0 do not. Of the 10 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 10 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

Recent works (White et al., 2020a;Yan et al., 2020) demonstrate the importance of architecture encodings in Neural Architecture Search (NAS). These encodings encode either structure or computation information of the neural architectures. Compared to structure-aware encodings, computation-aware encodings map architectures with similar accuracies to the same region, which improves the downstream architecture search performance (Zhang et al., 2019; White et al., 2020a). In this work, we introduce a Computation-Aware Transformer-based Encoding method called CATE. Different from existing computation-aware encodings based on fixed transformation (e.g. path encoding), CATE employs a pairwise pre-training scheme to learn computation-aware encodings using Transformers with cross-attention. Such learned encodings contain dense and contextualized computation information of neural architectures. We compare CATE with eleven encodings under three major encoding-dependent NAS subroutines in both small and large search spaces. Our experiments show that CATE is beneficial to the downstream search, especially in the large search space. Moreover, the outside search space experiment demonstrates its superior generalization ability beyond the search space on which it was trained. Our code is available at: https://github.com/MSU-MLSys-Lab/CATE.

For agents

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

get_harvested_code_for_paper("2102.07108")
get_code_for_paper("2102.07108")
have("2102.07108")

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

curl https://syntology.ai/api/ran/2102.07108.json

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

[![Syntology run record](https://syntology.ai/api/ran/2102.07108.svg)](https://syntology.ai/paper/2102.07108)

Connect an agent — have() is free.