SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2403.14112 · ACL · 2024

Benchmarking Chinese Commonsense Reasoning of LLMs: From Chinese-Specifics to Reasoning-Memorization Correlations

Wei Li, Jiang Wu, Hang Yan, Conghui He, Jiaxing Sun, Songyang Zhang, Chenya Gu, Weiquan Huang, Liu Xiang, Ma Long, Zou Shiming

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 (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; 4 failures of our instrument; 2 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/2403.14112.json · badge: /api/ran/2403.14112.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
01-ai/Yi canonical (classifier label)
classifier basis: census
6 ran · 2 no recorded run
FunctionStatusWhere it lives
calculate_flops Ran
ran, honoured the contract we drafted
01-ai/Yi/finetune/utils/perf.py
code served (permissive licence) · get_code("069f56cb00324e10")
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 float, ndim 0, dtype family float, 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

class HFConfig:
    def __init__(self, num_hidden_layers, hidden_size, vocab_size):
        self.num_hidden_layers = num_hidden_layers
        self.hidden_size = hidden_size
        self.vocab_size = vocab_size

ARGS = {
    'checkpoint_activations_factor': 1.0,
    'batch_size': 2,
    'seq_length': 8,
    'hf_config': HFConfig(
        num_hidden_layers=4,
        hidden_size=16,
        vocab_size=1000
    )
}
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 float, ndim 0, dtype family float, 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
generate Ran
ran; our contract draft was wrong, not the code
01-ai/Yi/finetune/sft/prompt_eval.py
pointer only (licence: Apache-2.0); the stored copy get_code serves, from lyan62/FoodieQA, is pointer only (licence: NONE) · get_code("c383e2541e15d801")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (Apache-2.0), but get_code serves the stored copy, first harvested from lyan62/FoodieQA, and GitHub detected no licence in that repository when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

get_eval_ds_config Ran
ran; our contract draft was wrong, not the code
01-ai/Yi/finetune/utils/ds_utils.py
pointer only (licence: Apache-2.0); the stored copy get_code serves, from lyan62/FoodieQA, is pointer only (licence: NONE) · get_code("1182381b66bc654d")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (Apache-2.0), but get_code serves the stored copy, first harvested from lyan62/FoodieQA, and GitHub detected no licence in that repository when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

get_hf_configs Ran
ran, honoured the contract we drafted
01-ai/Yi/finetune/utils/perf.py
code served (permissive licence) · get_code("df20c2f4083d17af")
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, ndim 1, dtype family int; 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 numpy as np

class MockHFConfig:
    """Mock HuggingFace config object with necessary attributes."""
    def __init__(self, num_layers, hidden_size, vocab_size):
        self.num_hidden_layers = num_layers
        self.hidden_size = hidden_size
        self.vocab_size = vocab_size

# Create a mock config with valid values
mock_config = MockHFConfig(
    num_layers=12,
    hidden_size=768,
    vocab_size=50257
)

ARGS = {
    'hf_config': mock_config
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple, ndim 1, dtype family int — a tuple 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
parse_text Ran
ran; our contract draft was wrong, not the code
01-ai/Yi/demo/web_demo.py
pointer only (licence: Apache-2.0); the stored copy get_code serves, from lyan62/FoodieQA, is pointer only (licence: NONE) · get_code("ca6cfdfc224a6e07")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (Apache-2.0), but get_code serves the stored copy, first harvested from lyan62/FoodieQA, and GitHub detected no licence in that repository when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

to_device Ran
ran; our contract draft was wrong, not the code
01-ai/Yi/finetune/utils/utils.py
code served (permissive licence) · get_code("1cbe0f5b4bb933ec")
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
import torch.nn as nn

torch.manual_seed(0)

# The function iterates over batch.items() and calls .to(device) on each value
# It handles exceptions gracefully, so we need a dict with mixed types:
# - some values that have .to() method (tensors)
# - some values that don't (to test the exception handling)

batch = {
    "tensor_field": torch.randn(2, 4),
    "another_tensor": torch.randn(3, 5),
    "string_field": "this is a string",
    "int_field": 42,
    "list_field": [1, 2, 3],
}

device = "cpu"

ARGS = {
    "batch": batch,
    "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
get_tokenizer Not yet run
unverified (harvested, no recorded run)
01-ai/Yi/finetune/utils/utils.py
code served (permissive licence) · get_code("8b21444b16ee2b78")
How it ran

no invocation recorded.

load_hf_tokenizer Not yet run
unverified (harvested, no recorded run)
01-ai/Yi/finetune/utils/utils.py
code served (permissive licence) · get_code("3c7a6b46926fca7f")
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, 5 are also present in other papers whose code we hold (4 with at least one paper dated earlier, 0 the same month, 5 later) and 3 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
parse_text 9 other papers
7 dated earlier, 2 later
2203.13474 (2022-03)
CodeGen: An Open Large Language Model for Code with Multi-Turn Program Synthesis
2305.12524 (2023-05)
TheoremQA: A Theorem-driven Question Answering Dataset ♠
2305.14314 (2023-05)
QLORA: Efficient Finetuning of Quantized LLMs
and 6 more not named here
calculate_flops 6 other papers
2 dated earlier, 4 later
2310.10505 (2023-10)
ReMax: A Simple, Effective, and Efficient Reinforcement Learning Method for Aligning Large Language Models
2402.05808 (2024-02)
Training Large Language Models for Reasoning through Reverse Curriculum Reinforcement Learning
2405.16455 (2024-05)
On the Algorithmic Bias of Aligning Large Language Models with RLHF: Preference Collapse and Matching Regularization
and 3 more not named here
get_hf_configs 6 other papers
2 dated earlier, 4 later
2310.10505 (2023-10)
ReMax: A Simple, Effective, and Efficient Reinforcement Learning Method for Aligning Large Language Models
2402.05808 (2024-02)
Training Large Language Models for Reasoning through Reverse Curriculum Reinforcement Learning
2405.16455 (2024-05)
On the Algorithmic Bias of Aligning Large Language Models with RLHF: Preference Collapse and Matching Regularization
and 3 more not named here
get_eval_ds_config 4 other papers
1 dated earlier, 2 later, 1 same year, order not determinable
2024.naacl-long.326 (2024 (year only))
LSTDial: Enhancing Dialogue Generation via Long-and Short-Term Measurement Feedback
2402.18150 (2024-02)
Unsupervised Information Refinement Training of Large Language Models for Retrieval-Augmented Generation
2406.08434 (2024-06)
TASTE: Teaching Large Language Models to Translate through Self-Reflection
and 1 more not named here
generate 1 other paper
1 later
2406.11030 (2024-06)
FoodieQA: A Multimodal Dataset for Fine-Grained Understanding of Chinese Food Culture

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

We introduce CHARM, the first benchmark for comprehensively and in-depth evaluating the commonsense reasoning ability of large language models (LLMs) in Chinese, which covers both globally known and Chinese-specific commonsense. We evaluated 7 English and 12 Chinese-oriented LLMs on CHARM, employing 5 representative prompt strategies for improving LLMs' reasoning ability, such as Chain-of-Thought. Our findings indicated that the LLM's language orientation and the task's domain influence the effectiveness of the prompt strategy, which enriches previous research findings. We built closely-interconnected reasoning and memorization tasks, and found that some LLMs struggle with memorizing Chinese commonsense, affecting their reasoning ability, while others show differences in reasoning despite similar memorization performance. We also evaluated the LLMs' memorizationindependent reasoning abilities and analyzed the typical errors. Our study precisely identified the LLMs' strengths and weaknesses, providing the clear direction for optimization. It can also serve as a reference for studies in other fields. We will release CHARM at https://github.com/opendatalab/CHARM.

For agents

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

get_harvested_code_for_paper("2403.14112")
get_code_for_paper("2403.14112")
have("2403.14112")

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

curl https://syntology.ai/api/ran/2403.14112.json

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

[![Syntology run record](https://syntology.ai/api/ran/2403.14112.svg)](https://syntology.ai/paper/2403.14112)

Connect an agent — have() is free.