SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2402.16347 · 2024

CodeS: Towards Building Open-source Language Models for Text-to-SQL

arXiv · PDF · Open in the Atlas

Code that ran

12 of 16 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (0 of them constructed an object rather than computing a result; 2 honoured the contract we drafted, 0 violated it and 0 have no contract verdict; 10 failures of our instrument; 4 no recorded run), measured 2026-11-22; repository links from the Papers with Code archive attributed under CC BY-SA, no affiliation; 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/2402.16347.json · badge: /api/ran/2402.16347.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
ruckbreasoning/codes canonical (classifier label)
classifier basis: census
12 ran · 4 no recorded run
FunctionStatusWhere it lives
compare_sql Ran
ran; our contract draft was wrong, not the code
ruckbreasoning/codes/evaluate_ex.py
code served (permissive licence) · get_code("9c09b83ecd4e3646")
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 sqlite3
import tempfile
import os

# Create a temporary SQLite database with sample data
temp_dir = tempfile.gettempdir()
db_path = os.path.join(temp_dir, "test_compare_sql.db")

# Create and populate a test database
conn = sqlite3.connect(db_path, check_same_thread=False)
cursor = conn.cursor()

# Create a simple test table
cursor.execute('''
    CREATE TABLE IF NOT EXISTS users (
        id INTEGER PRIMARY KEY,
        name TEXT,
        age INTEGER
    )
''')

# Insert sample data
cursor.execute("DELETE FROM users")  # Clear any existing data
cursor.execute("INSERT INTO users (id, name, age) VALUES (1, 'Alice', 30)")
cursor.execute("INSERT INTO users (id, name, age) VALUES (2, 'Bob', 25)")
cursor.execute("INSERT INTO users (id, name, age) VALUES (3, 'Charlie', 35)")
conn.commit()
conn.close()

# Define the arguments
ARGS = {
    "predicted_sql": "SELECT id, name, age FROM users WHERE age > 24 ORDER BY id",
    "ground_truth": "SELECT id, name FROM users WHERE age > 24 UNION SELECT id, name FROM users WHERE age > 24 ORDER BY id",
    "db_path": db_path
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
func_timeout — 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
extract_large_numbers Ran
ran; our contract draft was wrong, not the code
ruckbreasoning/codes/prepare_sft_datasets.py
code served (permissive licence) · get_code("fde45273dd27b383")
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 str. 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 re

def extract_large_numbers(text):
    number_information = []
    patterns = {'thousand': 10 ** 3, 'million': 10 ** 6, 'billion': 10 ** 9, 'trillion': 10 ** 12}
    for word, multiplier in patterns.items():
        matches = re.findall('(\\d+\\.?\\d*)\\s*{}'.format(word), text, flags=re.IGNORECASE)
        for match in matches:
            number = float(match) * multiplier
            number_information.append(match + ' ' + word + ' = ' + str(int(number)))
    for phrase, number in {'thousands of': 10 ** 3, 'millions of': 10 ** 6, 'billions of': 10 ** 9, 'trillions of': 10 ** 12}.items():
        if phrase in text:
            number_information.append(phrase + ' = ' + str(int(number)))
    large_number_evidence = ''
    for info in number_information:
        large_number_evidence += info + '; '
    return large_number_evidence.strip()

text_sample = "The company earned 5.5 million dollars and has billions of users worldwide. We saw 2 thousand transactions and trillions of calculations performed."

ARGS = {
    'text': text_sample
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type str — a str output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a str 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
get_schema Ran
ran; our contract draft was wrong, not the code
ruckbreasoning/codes/schema_item_filter.py
code served (permissive licence) · get_code("56e8906d4b412b25")
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 dict. 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

# Construct a valid argument for the get_schema function
# The function expects tables_and_columns: a list of tuples (table_name, column_name)

tables_and_columns = [
    ('users', 'id'),
    ('users', 'name'),
    ('users', 'email'),
    ('orders', 'order_id'),
    ('orders', 'user_id'),
    ('orders', 'amount'),
    ('products', 'product_id'),
    ('products', 'product_name'),
]

ARGS = {
    'tables_and_columns': tables_and_columns
}
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 dict — a dict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a dict 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
get_sequence_length Ran
ran, honoured the contract we drafted
ruckbreasoning/codes/schema_item_filter.py
code served (permissive licence) · get_code("8af31230a92da0c1")
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 int, ndim 0, dtype family int, shape () as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape 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 transformers import AutoTokenizer

# Create a simple tokenizer stand-in that mimics transformers tokenizer behavior
class SimpleTokenizer:
    def __call__(self, text_list, is_split_into_words=False):
        # Simulate tokenization by returning input_ids
        # Each word gets tokenized into 1-2 token ids
        token_count = 0
        for word in text_list:
            if word == '|' or word == ':' or word == ',':
                token_count += 1
            else:
                token_count += 2  # Simulate words being split into multiple tokens
        
        return {
            'input_ids': list(range(token_count)),
            'attention_mask': [1] * token_count
        }

# Set up arguments
text = "What is the total revenue?"

tables_and_columns = [
    ('users', 'user_id'),
    ('users', 'name'),
    ('orders', 'order_id'),
    ('orders', 'amount'),
]

tokenizer = SimpleTokenizer()

ARGS = {
    'text': text,
    'tables_and_columns': tables_and_columns,
    'tokenizer': tokenizer,
}
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 int, ndim 0, dtype family int, shape () — shape () as a later run recorded it
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: none by the harness; any seed is in the setup block
masked_softmax Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
ruckbreasoning/codes/modeling_gpt_bigcode.py
code served (permissive licence) · get_code("49334501abd461a4")
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 Tensor, ndim 2, 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.functional as F

torch.manual_seed(0)

# Create small tensors for testing
batch_size = 2
seq_len = 4

# x: input tensor for softmax computation
x = torch.randn(batch_size, seq_len)

# mask: boolean tensor indicating which positions to keep (True = keep, False = mask)
mask = torch.tensor([
    [True, True, False, False],
    [True, True, True, False]
])

# mask_value: the value to fill in masked positions (typically large negative number for softmax)
mask_value = torch.tensor(float('-inf'))

ARGS = {
    "x": x,
    "mask": mask,
    "mask_value": mask_value
}
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 Tensor, ndim 2, dtype family float — no output shape is stated: observe_signature.py could not observe it (import:OSError)
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: none by the harness; any seed is in the setup block
prepare_input_ids_and_attention_mask Ran
ran; our contract draft was wrong, not the code
ruckbreasoning/codes/text2sql_few_shot.py
code served (permissive licence) · get_code("93d18eb1da1b928c")
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 dict. 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)

class MockTokenizer:
    def __init__(self, name_or_path='bert-base-uncased', bos_token_id=101):
        self.name_or_path = name_or_path
        self.bos_token_id = bos_token_id
    
    def __call__(self, text, truncation=False):
        # Return a mock tokenization result with input_ids
        return {'input_ids': [1, 2, 3, 4, 5, 6, 7, 8, 9, 10]}

tokenizer = MockTokenizer(name_or_path='bert-base-uncased', bos_token_id=101)
input_seq = "This is a test sequence for tokenization."
max_input_length = 15
device = 'cpu'

ARGS = {
    'tokenizer': tokenizer,
    'input_seq': input_seq,
    'max_input_length': max_input_length,
    'device': device
}
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 dict — a dict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a dict 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
prepare_inputs_and_labels Ran
ran; our contract draft was wrong, not the code
ruckbreasoning/codes/schema_item_filter.py
code served (permissive licence) · get_code("379783e2b5131276")
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. The code mentions cuda, and the harness aliases cuda to the CPU for every sample; whether this run needed that is 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)

class MockTokenizer:
    def __call__(self, words, return_tensors=None, is_split_into_words=False, padding=None, max_length=None, truncation=False):
        # Simulate tokenization of word list
        # Return a mock object with input_ids, attention_mask, and word_ids method
        batch_size = 1
        seq_length = min(len(words) * 2 + 5, max_length) if max_length else len(words) * 2 + 5
        
        input_ids = torch.randint(100, 1000, (batch_size, seq_length))
        attention_mask = torch.ones(batch_size, seq_length)
        
        # Store word_ids mapping for word_ids() call
        # Map tokens back to word indices
        word_ids_list = []
        token_idx = 0
        for word_idx, word in enumerate(words):
            # Each word typically maps to 1-3 tokens
            num_tokens = 2 if word_idx % 3 == 0 else 1
            for _ in range(num_tokens):
                if token_idx < seq_length:
                    word_ids_list.append(word_idx)
                    token_idx += 1
        
        # Pad with None
        while len(word_ids_list) < seq_length:
            word_ids_list.append(None)
        
        class MockTokenizedOutput:
            def __init__(self, input_ids, attention_mask, word_ids_list):
                self.input_ids = input_ids
                self.attention_mask = attention_mask
                self._word_ids_list = word_ids_list
            
            def __getitem__(self, key):
                if key == 'input_ids':
                    return self.input_ids
                elif key == 'attention_mask':
                    return self.attention_mask
                return None
            
            def word_ids(self, batch_index=0):
                return self._word_ids_list
        
        return MockTokenizedOutput(input_ids, attention_mask, word_ids_list)

sample = {
    'text': 'What are the employees?',
    'schema': {
        'schema_items': [
            {
                'table_name': 'employees',
                'column_names': ['id', 'name', 'salary']
            },
            {
                'table_name': 'departments',
                'column_names': ['dept_id', 'dept_name']
            }
        ]
    }
}

tokenizer = MockTokenizer()

ARGS = {
    'sample': sample,
    'tokenizer': tokenizer
}
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
remove_similar_comments Ran
ran; our contract draft was wrong, not the code
ruckbreasoning/codes/prepare_sft_datasets.py
code served (permissive licence) · get_code("c1c7de5d97c5b309")
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 list, ndim 1, 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

ARGS = {
    "names": ["user_id", "account_name", "created_at"],
    "comments": ["userid", "account name", "The time when the record was created"]
}
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 list, ndim 1, dtype family float — a list output has no array shape of its own; none is stated
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: none by the harness; any seed is in the setup block
resume_model_and_optimizer Ran
ran, honoured the contract we drafted
ruckbreasoning/codes/train_causal_lm.py
code served (permissive licence) · get_code("4a150066109c1a86")
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 int, ndim 0, dtype family int, shape () as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape 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 unittest.mock import Mock

torch.manual_seed(0)

class SimpleModel(nn.Module):
    def __init__(self):
        super().__init__()
        self.linear = nn.Linear(4, 3)
    
    def load_checkpoint(self, load_dir, tag, load_optimizer_states=False):
        """Mock checkpoint loading method"""
        checkpoint_state_dict = {
            'last_global_step': 42,
            'model_state': self.state_dict(),
        }
        if load_optimizer_states:
            checkpoint_state_dict['optimizer_state'] = {}
        
        return None, checkpoint_state_dict

model = SimpleModel()
load_dir = "/tmp/checkpoint"
tag = "latest"

ARGS = {
    'model': model,
    'load_dir': load_dir,
    'tag': tag,
}
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 int, ndim 0, dtype family int, shape () — shape () as a later run recorded it
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: none by the harness; any seed is in the setup block
text2sql_func Ran
ran; our contract draft was wrong, not the code
ruckbreasoning/codes/text2sql_zero_shot.py
code served (permissive licence) · get_code("c9ee130e5a1966a9")
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 list, ndim 1, 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

torch.manual_seed(0)

# Create a minimal model that supports generate() method
class MockTextToSQLModel(nn.Module):
    def __init__(self):
        super().__init__()
        self.embed = nn.Embedding(100, 16)
        self.linear = nn.Linear(16, 100)
    
    def generate(self, input_ids, attention_mask=None, max_new_tokens=10, num_beams=4, num_return_sequences=4):
        # Return generated token ids with shape (batch_size * num_return_sequences, seq_len + max_new_tokens)
        batch_size = input_ids.shape[0]
        seq_len = input_ids.shape[1]
        output_seq_len = seq_len + max_new_tokens
        total_sequences = batch_size * num_return_sequences
        # Generate random token ids between 0 and 99
        generate_ids = torch.randint(0, 100, (total_sequences, output_seq_len))
        return generate_ids

# Create a minimal tokenizer with batch_decode method
class MockTokenizer:
    def __init__(self):
        self.vocab = {i: f"token_{i}" for i in range(100)}
    
    def batch_decode(self, token_ids, skip_special_tokens=False, clean_up_tokenization_spaces=False):
        # Convert token ids to strings
        decoded = []
        for tokens in token_ids:
            decoded_str = " ".join([self.vocab.get(t.item(), f"token_{t.item()}") for t in tokens])
            decoded.append(decoded_str)
        return decoded

# Create model, tokenizer, and inputs
model = MockTextToSQLModel()
tokenizer = MockTokenizer()

# Create input tensors with batch_size=2, seq_len=5
batch_size = 2
seq_len = 5
input_ids = torch.randint(0, 100, (batch_size, seq_len))
attention_mask = torch.ones((batch_size, seq_len), dtype=torch.long)

inputs = {
    'input_ids': input_ids,
    'attention_mask': attention_mask
}

max_new_tokens = 8

ARGS = {
    'model': model,
    'inputs': inputs,
    'tokenizer': tokenizer,
    'max_new_tokens': max_new_tokens
}
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 list, ndim 1, dtype family float — a list output has no array shape of its own; none is stated
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: none by the harness; any seed is in the setup block
upcast_masked_softmax Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
ruckbreasoning/codes/modeling_gpt_bigcode.py
code served (permissive licence) · get_code("38cab2f301010b6d")
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 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.functional

torch.manual_seed(0)

# Create small tensors
batch_size = 2
seq_len = 4
hidden_dim = 3

# x: input tensor (batch_size, seq_len, hidden_dim)
x = torch.randn(batch_size, seq_len, hidden_dim, dtype=torch.float32)

# mask: boolean mask with same shape as x
mask = torch.ones(batch_size, seq_len, hidden_dim, dtype=torch.bool)
# Set some values to False to test masking
mask[0, 0, 0] = False
mask[1, 2, 1] = False

# mask_value: scalar tensor to use where mask is False
mask_value = torch.tensor(float('-inf'), dtype=torch.float32)

# scale: float scaling factor
scale = 0.5

# softmax_dtype: the dtype to use for softmax computation
softmax_dtype = torch.float32

ARGS = {
    "x": x,
    "mask": mask,
    "mask_value": mask_value,
    "scale": scale,
    "softmax_dtype": softmax_dtype,
}
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 Tensor, ndim 3, dtype family float — no output shape is stated: observe_signature.py could not observe it (import:OSError)
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: none by the harness; any seed is in the setup block
upcast_softmax Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
ruckbreasoning/codes/modeling_gpt_bigcode.py
code served (permissive licence) · get_code("49527b68e554c71a")
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 Tensor, ndim 2, 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.functional as F

torch.manual_seed(0)

# Create a small tensor for testing
# Using shape (2, 4) - batch size 2, feature dim 4
x = torch.randn(2, 4, dtype=torch.float32)

# Define scale parameter
scale = 2.0

# Define softmax_dtype parameter
softmax_dtype = torch.float32

ARGS = {
    "x": x,
    "scale": scale,
    "softmax_dtype": softmax_dtype
}
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 Tensor, ndim 2, dtype family float — no output shape is stated: observe_signature.py could not observe it (import:OSError)
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: none by the harness; any seed is in the setup block
execute_sql Not yet run
unverified (harvested, no recorded run)
ruckbreasoning/codes/evaluate_ex.py
code served (permissive licence) · get_code("e4ab0d16733f055d")
How it ran

no invocation recorded.

extract_skeleton Not yet run
unverified (harvested, no recorded run)
ruckbreasoning/codes/text2sql_few_shot.py
code served (permissive licence) · get_code("5d9edf7a65f4c0f4")
How it ran

no invocation recorded.

prepare Not yet run
unverified (harvested, no recorded run)
ruckbreasoning/codes/tokenize_pt_corpus.py
code served (permissive licence) · get_code("e459a1d6427bf16a")
How it ran

no invocation recorded.

prepare_batch_inputs_and_labels Not yet run
unverified (harvested, no recorded run)
ruckbreasoning/codes/train_schema_item_filter.py
code served (permissive licence) · get_code("0f37491b931ec512")
How it ran

no invocation recorded.

Also present in other papers' code

Of 16 samples lifted from this paper's linked repositories, 16 carry a function identity and 0 do not. Of the 16 distinct functions with an identity, 3 are also present in other papers whose code we hold (0 with at least one paper dated earlier, 0 the same month, 3 later) and 13 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.

FunctionAlso present inEarliest-datedOthers named
masked_softmax 1 other paper
1 later
2404.10933 (2024-04)
LLMem: Estimating GPU Memory Usage for Fine-Tuning Pre-Trained LLMs
upcast_masked_softmax 1 other paper
1 later
2404.10933 (2024-04)
LLMem: Estimating GPU Memory Usage for Fine-Tuning Pre-Trained LLMs
upcast_softmax 1 other paper
1 later
2404.10933 (2024-04)
LLMem: Estimating GPU Memory Usage for Fine-Tuning Pre-Trained LLMs

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

Language models have shown promising performance on the task of translating natural language questions into SQL queries (Text-to-SQL). However, most of the state-of-the-art (SOTA) approaches rely on powerful yet closed-source large language models (LLMs), such as ChatGPT and GPT-4, which may have the limitations of unclear model architectures, data privacy risks, and expensive inference overheads. To address the limitations, we introduce CodeS, a series of pre-trained language models with parameters ranging from 1B to 15B, specifically designed for the text-to-SQL task. CodeS is a fully open-source language model, which achieves superior accuracy with much smaller parameter sizes. This paper studies the research challenges in building CodeS. To enhance the SQL generation abilities of CodeS, we adopt an incremental pre-training approach using a specifically curated SQL-centric corpus. Based on this, we address the challenges of schema linking and rapid domain adaptation through strategic prompt construction and a bi-directional data augmentation technique. We conduct comprehensive evaluations on multiple datasets, including the widely used Spider benchmark, the newly released BIRD benchmark, robustness-diagnostic benchmarks such as Spider-DK, Spider-Syn, Spider-Realistic, and Dr.Spider, as well as two real-world datasets created for financial and academic applications. The experimental results show that our CodeS achieves new SOTA accuracy and robustness on nearly all challenging text-to-SQL benchmarks.

For agents

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

get_harvested_code_for_paper("2402.16347")
get_code_for_paper("2402.16347")
have("2402.16347")

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

curl https://syntology.ai/api/ran/2402.16347.json

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

[![Syntology run record](https://syntology.ai/api/ran/2402.16347.svg)](https://syntology.ai/paper/2402.16347)

Connect an agent — have() is free.