SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2407.09121 · 2024

Refuse Whenever You Feel Unsafe: Improving Safety in LLMs via Decoupled Refusal Training

arXiv · PDF · Open in the Atlas

Code that ran

3 of 3 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; 0 honoured the contract we drafted, 0 violated it and 0 have no contract verdict; 3 failures of our instrument; 0 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/2407.09121.json · badge: /api/ran/2407.09121.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
robustnlp/derta canonical (classifier label)
classifier basis: classifier_2025
3 ran · 0 no recorded run
FunctionStatusWhere it lives
load_prompts Ran
ran; our contract draft was wrong, not the code
robustnlp/derta/evaluation.py
code served (permissive licence) · get_code("ce51d685abf1ad76")
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) 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 json
import tempfile
import os

# Create a temporary JSON file for testing
temp_dir = tempfile.mkdtemp()
mmlu_file = os.path.join(temp_dir, 'mmlu_test.json')
gsm8k_file = os.path.join(temp_dir, 'gsm8k_test.json')

# Write sample MMLU data
mmlu_data = [
    {'question': 'What is 2+2?', 'answer': 'A'},
    {'question': 'What is the capital of France?', 'answer': 'B'}
]
with open(mmlu_file, 'w', encoding='utf-8') as f:
    for item in mmlu_data:
        f.write(json.dumps(item) + '\n')

# Write sample GSM8K data
gsm8k_data = [
    {'question': 'If John has 5 apples and gives 2 away, how many does he have?', 'answer': '3'},
    {'question': 'Mary buys 3 books at $10 each. How much does she spend?', 'answer': '$30'}
]
with open(gsm8k_file, 'w', encoding='utf-8') as f:
    for item in gsm8k_data:
        f.write(json.dumps(item) + '\n')

# Define a simple create_prompt function
def create_prompt(data):
    """Simple prompt creator that returns the question as-is or formatted"""
    if isinstance(data, list) and len(data) > 0:
        if isinstance(data[0], dict) and 'question' in data[0]:
            return [item['question'] for item in data]
    return data

# Define ARGS
ARGS = {
    'create_prompt': create_prompt,
    'eval_data': mmlu_file
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
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 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
read_from_json Ran
ran; our contract draft was wrong, not the code
robustnlp/derta/evaluation.py
code served (permissive licence) · get_code("ecc4e13936789c38")
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) 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 json
import tempfile
import os

# Create a temporary JSON file with sample data
temp_dir = tempfile.mkdtemp()
temp_file = os.path.join(temp_dir, "test_data.jsonl")

# Write sample JSONL data
sample_data = [
    {"id": 1, "text": "hello world", "label": 0},
    {"id": 2, "text": "foo bar", "label": 1},
    {"id": 3, "text": "test data", "label": 0}
]

with open(temp_file, 'w', encoding='utf-8') as f:
    for item in sample_data:
        f.write(json.dumps(item) + '\n')

ARGS = {
    "filename": temp_file
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
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 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
read_input Ran
ran; our contract draft was wrong, not the code
robustnlp/derta/evaluation.py
code served (permissive licence) · get_code("2a686c7dfc4f6740")
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) 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 tempfile
import os

# Create a temporary file with some test data
temp_file = tempfile.NamedTemporaryFile(mode='w', delete=False, encoding='utf-8', suffix='.txt')
temp_file.write("line 1\n")
temp_file.write("line 2\n")
temp_file.write("line 3\n")
temp_file.close()

ARGS = {
    'path': temp_file.name
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
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 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

Also present in other papers' code

Of 3 samples lifted from this paper's linked repositories, 3 carry a function identity and 0 do not. Of the 3 distinct functions with an identity, 1 are also present in other papers whose code we hold (1 with at least one paper dated earlier, 0 the same month, 0 later) and 2 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
read_input 4 other papers
4 dated earlier
2308.12674 (2023-08)
Improving Translation Faithfulness of Large Language Models via Augmenting Instructions
2311.09071 (2023-11)
How Vocabulary Sharing Facilitates Multilingualism in LLaMA?
2401.08350 (2024-01)
Salute the Classic: Revisiting Challenges of Machine Translation in the Age of Large Language Models
and 1 more not named here

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

This study addresses a critical gap in safety tuning practices for Large Language Models (LLMs) by identifying and tackling a refusal position bias within safety tuning data, which compromises the models' ability to appropriately refuse generating unsafe content. We introduce a novel approach, Decoupled Refusal Training (DeRTa), designed to empower LLMs to refuse compliance to harmful prompts at any response position, significantly enhancing their safety capabilities. DeRTa incorporates two novel components: (1) Maximum Likelihood Estimation (MLE) with Harmful Response Prefix, which trains models to recognize and avoid unsafe content by appending a segment of harmful response to the beginning of a safe response, and (2) Reinforced Transition Optimization (RTO), which equips models with the ability to transition from potential harm to safety refusal consistently throughout the harmful response sequence. Our empirical evaluation, conducted using LLaMA3 and Mistral model families across six attack scenarios, demonstrates that our method not only improves model safety without compromising performance but also surpasses baseline methods in defending against attacks.

For agents

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

get_harvested_code_for_paper("2407.09121")
get_code_for_paper("2407.09121")
have("2407.09121")

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

curl https://syntology.ai/api/ran/2407.09121.json

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

[![Syntology run record](https://syntology.ai/api/ran/2407.09121.svg)](https://syntology.ai/paper/2407.09121)

Connect an agent — have() is free.