SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2401.16405 · 2024

Scaling Sparse Fine-Tuning to Large Language Models

arXiv · PDF · Open in the Atlas

Code that ran

13 of 15 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; 11 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/2401.16405.json · badge: /api/ran/2401.16405.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
ducdauge/sft-llm canonical (classifier label)
classifier basis: census
10 ran · 2 no recorded run
alanansell/peft canonical (classifier label)
classifier basis: classifier_2025
2 ran · 0 no recorded run
copy not recorded community
not in the paper's declared repositories
1 ran · 0 no recorded run
FunctionStatusWhere it lives
b2mb Ran
ran, honoured the contract we drafted
this paper's copy was not recorded; identical code first harvested from huggingface/peft
pointer only · get_code("32380a050ca033d4")
How it ran

The body of this sample is withheld: this paper's copy of the code was not recorded (its link names no repository or source path), so no licence was read for it. 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.

create_prompt_with_llama2_chat_format Ran
ran; our contract draft was wrong, not the code
ducdauge/sft-llm/eval/templates.py
code served (permissive licence) · get_code("e51c9713fadad16f")
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 torch

# Define the arguments for create_prompt_with_llama2_chat_format
ARGS = {
    'messages': [
        {'role': 'system', 'content': 'You are a helpful assistant.'},
        {'role': 'user', 'content': 'Hello, how are you?'},
        {'role': 'assistant', 'content': 'I am doing well, thank you for asking.'},
        {'role': 'user', 'content': 'What is your name?'},
        {'role': 'assistant', 'content': 'I am Claude, an AI assistant.'}
    ],
    'bos': '<s>',
    'eos': '</s>',
    'add_bos': True
}
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 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
create_prompt_with_tulu_chat_format Ran
ran; our contract draft was wrong, not the code
ducdauge/sft-llm/eval/templates.py
pointer only (licence: MIT); the stored copy get_code serves, from thu-keg/crab, is pointer only (licence: NONE) · get_code("697d92a5d51eff3e")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (MIT), but get_code serves the stored copy, first harvested from thu-keg/crab, 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.

dispatch_openai_chat_requesets Ran
ran; our contract draft was wrong, not the code
ducdauge/sft-llm/eval/dispatch_openai_requests.py
pointer only (licence: MIT); the stored copy get_code serves, from thu-keg/crab, is pointer only (licence: NONE) · get_code("dcad9ac62979ca49")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (MIT), but get_code serves the stored copy, first harvested from thu-keg/crab, 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.

dispatch_openai_prompt_requesets Ran
ran; our contract draft was wrong, not the code
ducdauge/sft-llm/eval/dispatch_openai_requests.py
pointer only (licence: MIT); the stored copy get_code serves, from thu-keg/crab, is pointer only (licence: NONE) · get_code("71e4259026911444")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (MIT), but get_code serves the stored copy, first harvested from thu-keg/crab, 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.

encode_with_messages_format Ran
ran; our contract draft was wrong, not the code
ducdauge/sft-llm/finetune/finetune.py
code served (permissive licence) · get_code("1cb0eeb16cf04ee3")
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
from unittest.mock import Mock

torch.manual_seed(0)

class MockTokenizer:
    def __init__(self):
        self.eos_token = '<|end|>'
    
    def __call__(self, text, return_tensors=None, max_length=None, truncation=None):
        # Return a mock tokenization result with input_ids
        # We need to return different lengths based on the text to simulate real tokenization
        text_len = len(text.split())
        token_count = min(text_len + 5, max_length) if max_length else text_len + 5
        
        result = Mock()
        result.input_ids = torch.randint(0, 1000, (1, token_count))
        return result

example = {
    'messages': [
        {
            'role': 'system',
            'content': 'You are a helpful assistant.'
        },
        {
            'role': 'user',
            'content': 'Hello, how are you?'
        },
        {
            'role': 'assistant',
            'content': 'I am doing well, thank you for asking.'
        }
    ]
}

tokenizer = MockTokenizer()
max_seq_length = 128

ARGS = {
    'example': example,
    'tokenizer': tokenizer,
    'max_seq_length': max_seq_length
}
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
encode_with_prompt_completion_format Ran
ran; our contract draft was wrong, not the code
ducdauge/sft-llm/finetune/finetune.py
code served (permissive licence) · get_code("88b8bf1223d522a4")
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
from transformers import AutoTokenizer

torch.manual_seed(0)

# Create a minimal tokenizer stand-in class
class MinimalTokenizer:
    def __init__(self):
        self.eos_token = "</s>"
    
    def __call__(self, text, return_tensors=None, max_length=None, truncation=False):
        # Simulate tokenization by creating token IDs
        # For simplicity, map each character to a unique ID
        token_ids = [ord(c) % 100 for c in text[:max_length]] if max_length else [ord(c) % 100 for c in text]
        
        if not token_ids:
            token_ids = [1]  # Ensure at least one token
        
        input_ids = torch.tensor([token_ids], dtype=torch.long)
        
        if return_tensors == 'pt':
            return type('TokenizerOutput', (), {'input_ids': input_ids})()
        return {'input_ids': input_ids}

# Create example data
example = {
    'prompt': 'What is the capital of France?',
    'completion': ' The capital of France is Paris.'
}

# Create tokenizer
tokenizer = MinimalTokenizer()

# Set max sequence length
max_seq_length = 128

# Define ARGS dictionary
ARGS = {
    'example': example,
    'tokenizer': tokenizer,
    'max_seq_length': max_seq_length
}
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 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
estimate_pass_at_k Ran
ran; our contract draft was wrong, not the code
ducdauge/sft-llm/eval/codex_humaneval/evaluation.py
pointer only (licence: MIT); the stored copy get_code serves, from rucaibox/bamboo, is pointer only (licence: NONE) · get_code("e080db2eaa81fdb3")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (MIT), but get_code serves the stored copy, first harvested from rucaibox/bamboo, 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.

generate_completions Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
ducdauge/sft-llm/eval/utils.py
code served (permissive licence) · get_code("62004a6900e0e332")
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. 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 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
from transformers import PreTrainedModel, PreTrainedTokenizer

torch.manual_seed(0)

# Minimal model that implements generate() and has device attribute
class MinimalModel(nn.Module):
    def __init__(self):
        super().__init__()
        self.linear = nn.Linear(4, 4)
    
    @property
    def device(self):
        return torch.device('cpu')
    
    def generate(self, input_ids, attention_mask=None, stopping_criteria=None, **kwargs):
        batch_size = input_ids.shape[0]
        num_return_sequences = kwargs.get('num_return_sequences', 1)
        # Generate a simple output with some tokens appended
        output_length = input_ids.shape[1] + 5
        outputs = torch.randint(0, 100, (batch_size * num_return_sequences, output_length))
        return outputs

# Minimal tokenizer with required methods
class MinimalTokenizer:
    def __init__(self):
        self.pad_token_id = 0
        self.vocab_size = 100
    
    def __call__(self, text, padding=None, return_tensors=None, add_special_tokens=True):
        # Handle both single string and list of strings
        if isinstance(text, str):
            text = [text]
        
        # Create fake token ids
        batch_size = len(text)
        seq_length = 6
        input_ids = torch.randint(1, 50, (batch_size, seq_length))
        attention_mask = torch.ones(batch_size, seq_length)
        
        class TokenizerOutput:
            def __init__(self, input_ids, attention_mask):
                self.input_ids = input_ids
                self.attention_mask = attention_mask
        
        return TokenizerOutput(input_ids, attention_mask)
    
    def batch_decode(self, token_ids, skip_special_tokens=True):
        # Return strings for each sequence in batch
        return ['This is a decoded text'] * token_ids.shape[0]

model = MinimalModel()
tokenizer = MinimalTokenizer()
prompts = ['Hello world', 'Test prompt']
batch_size = 2
stop_id_sequences = [[1, 2], [3, 4]]
add_special_tokens = True
disable_tqdm = True
num_return_sequences = 1
max_length = 20

ARGS = {
    'model': model,
    'tokenizer': tokenizer,
    'prompts': prompts,
    'batch_size': batch_size,
    'stop_id_sequences': stop_id_sequences,
    'add_special_tokens': add_special_tokens,
    'disable_tqdm': disable_tqdm,
    'max_length': max_length,
    'num_return_sequences': num_return_sequences,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch, tqdm, transformers — 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
get_closest_label Ran
ran; our contract draft was wrong, not the code
alanansell/peft/examples/causal_language_modeling/peft_lora_clm_accelerate_ds_zero3_offload.py
code served (permissive licence) · get_code("e2284f55bbde4f81")
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 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 sys
import numpy as np

def levenshtein_distance(str1, str2):
    if str1 == str2:
        return 0
    num_rows = len(str1) + 1
    num_cols = len(str2) + 1
    dp_matrix = np.empty((num_rows, num_cols))
    dp_matrix[0, :] = range(num_cols)
    dp_matrix[:, 0] = range(num_rows)
    for i in range(1, num_rows):
        for j in range(1, num_cols):
            if str1[i - 1] == str2[j - 1]:
                dp_matrix[i, j] = dp_matrix[i - 1, j - 1]
            else:
                dp_matrix[i, j] = min(dp_matrix[i - 1, j - 1], dp_matrix[i - 1, j], dp_matrix[i, j - 1]) + 1
    return dp_matrix[num_rows - 1, num_cols - 1]

def get_closest_label(eval_pred, classes):
    min_id = sys.maxsize
    min_edit_distance = sys.maxsize
    for i, class_label in enumerate(classes):
        edit_distance = levenshtein_distance(eval_pred.strip(), class_label)
        if edit_distance < min_edit_distance:
            min_id = i
            min_edit_distance = edit_distance
    return classes[min_id]

ARGS = {
    "eval_pred": "  hello  ",
    "classes": ["hello", "hallo", "world", "help"]
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
numpy — 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_next_word_predictions Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
ducdauge/sft-llm/eval/utils.py
code served (permissive licence) · get_code("d134c88af0236263")
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
import torch.nn as nn
from unittest.mock import Mock

torch.manual_seed(0)

# Build a minimal model that satisfies function requirements
class MinimalLanguageModel(nn.Module):
    def __init__(self, vocab_size=100, hidden_size=4):
        super().__init__()
        self.linear = nn.Linear(hidden_size, vocab_size)
        self.device_obj = Mock()
        self.device_obj.type = 'cpu'
    
    @property
    def device(self):
        return self.device_obj
    
    def forward(self, input_ids, attention_mask):
        # input_ids: (batch_size, seq_len)
        # Return logits of shape (batch_size, seq_len, vocab_size)
        batch_size, seq_len = input_ids.shape
        hidden = torch.randn(batch_size, seq_len, 4)
        logits = self.linear(hidden)
        return Mock(logits=logits)

# Build a minimal tokenizer that satisfies function requirements
class MinimalTokenizer:
    def __init__(self, vocab_size=100):
        self.vocab_size = vocab_size
    
    def __call__(self, texts, padding='longest', return_tensors='pt', add_special_tokens=True):
        if isinstance(texts, str):
            texts = [texts]
        max_len = max(len(text.split()) for text in texts) + 2
        batch_size = len(texts)
        input_ids = torch.randint(0, self.vocab_size, (batch_size, max_len))
        attention_mask = torch.ones(batch_size, max_len)
        return Mock(input_ids=input_ids, attention_mask=attention_mask)
    
    def convert_ids_to_tokens(self, ids):
        if isinstance(ids, torch.Tensor):
            ids = ids.tolist()
        if isinstance(ids, list):
            return [f'token_{id}' for id in ids]
        return f'token_{ids}'

model = MinimalLanguageModel(vocab_size=100, hidden_size=4)
tokenizer = MinimalTokenizer(vocab_size=100)
prompts = ["Hello world", "How are you"]
candidate_token_ids = torch.tensor([5, 10, 15, 20])

ARGS = {
    'model': model,
    'tokenizer': tokenizer,
    'prompts': prompts,
    'candidate_token_ids': candidate_token_ids,
    'batch_size': 1,
    'return_token_predictions': False,
    'add_special_tokens': True,
    'disable_tqdm': True,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch, tqdm — 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
levenshtein_distance Ran
ran, honoured the contract we drafted
alanansell/peft/examples/causal_language_modeling/peft_lora_clm_accelerate_ds_zero3_offload.py
code served (permissive licence) · get_code("01f1ac304787218d")
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 float64, 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 numpy as np

# Setup code for levenshtein_distance function
# The function takes two strings and computes the Levenshtein distance between them

ARGS = {
    "str1": "kitten",
    "str2": "sitting"
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type float64, 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
read_problems Ran
ran; our contract draft was wrong, not the code
ducdauge/sft-llm/eval/codex_humaneval/data.py
code served (permissive licence) · get_code("776a712a353130e7")
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 json
import gzip
import os
import tempfile

# Create a temporary directory for our test files
temp_dir = tempfile.mkdtemp()

# Create a minimal JSONL.GZ file with sample data
sample_data = [
    {"task_id": "HumanEval/0", "prompt": "def hello():", "canonical_solution": "return 'hello'", "test": "assert hello() == 'hello'"},
    {"task_id": "HumanEval/1", "prompt": "def add(a, b):", "canonical_solution": "return a + b", "test": "assert add(1, 2) == 3"}
]

# Write the gzipped JSONL file
test_file = os.path.join(temp_dir, "test_problems.jsonl.gz")
with gzip.open(test_file, 'wt') as f:
    for item in sample_data:
        f.write(json.dumps(item) + '\n')

# Define ARGS with the parameter for read_problems
ARGS = {
    'evalset_file': test_file
}
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 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
check_correctness Not yet run
unverified (harvested, no recorded run)
ducdauge/sft-llm/eval/codex_humaneval/execution.py
code served (permissive licence) · get_code("12d4fef547070f7c")
How it ran

no invocation recorded.

score_completions Not yet run
unverified (harvested, no recorded run)
ducdauge/sft-llm/eval/utils.py
code served (permissive licence) · get_code("e86d5bc4e595d32c")
How it ran

no invocation recorded.

Also present in other papers' code

Of 15 samples lifted from this paper's linked repositories, 15 carry a function identity and 0 do not. Of the 15 distinct functions with an identity, 14 are also present in other papers whose code we hold (9 with at least one paper dated earlier, 10 the same month, 14 later) and 1 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
estimate_pass_at_k 20 other papers
8 dated earlier, 1 the same month, 11 later
2107.03374 (2021-07)
Evaluating Large Language Models Trained on Code
2207.14255 (2022-07)
Efficient Training of Language Models to Fill in the Middle
2210.14868 (2022-10)
Multi-lingual Evaluation of Code Generation Models
and 17 more not named here
b2mb 11 other papers
1 dated earlier, 10 later
2303.18223 (2023-03)
A Survey of Large Language Models
2402.02834 (2024-02)
Shortened LLaMA: Depth Pruning for Large Language Models with Comparison of Retraining Methods
2403.08370 (2024-03)
SMART: Submodular Data Mixture Strategy for Instruction Tuning
and 8 more not named here
get_next_word_predictions 10 other papers
1 the same month, 9 later
2401.15006 (2024-01)
Airavata: Introducing Hindi Instruction-tuned LLM
2402.16705 (2024-02)
SelectIT: Selective Instruction Tuning for LLMs via Uncertainty-Aware Self-Reflection
2405.05008 (2024-05)
ADELIE: Aligning Large Language Models on Information Extraction
and 7 more not named here
read_problems 10 other papers
1 dated earlier, 2 the same month, 7 later
2310.20329 (2023-10)
InstructCoder: Instruction Tuning Large Language Models for Code Editing
2401.08565 (2024-01)
Tuning Language Models by Proxy
2401.14196 (2024-01)
DeepSeek-Coder: When the Large Language Model Meets Programming -- The Rise of Code Intelligence
and 7 more not named here
encode_with_prompt_completion_format 9 other papers
2 dated earlier, 3 the same month, 4 later
2311.05657 (2023-11)
Agent Lumos: Unified and Modular Training for Open-Source Language Agents
2312.07540 (2023-12)
diff History for Neural Language Agents
2401.08565 (2024-01)
Tuning Language Models by Proxy
and 6 more not named here
create_prompt_with_tulu_chat_format 8 other papers
1 the same month, 7 later
2401.10768 (2024-01)
Knowledge Verification to Nip Hallucination in the Bud
2403.03870 (2024-03)
Learning to Decode Collaboratively with Multiple Language Models
2405.05008 (2024-05)
ADELIE: Aligning Large Language Models on Information Extraction
and 5 more not named here
dispatch_openai_prompt_requesets 8 other papers
1 dated earlier, 2 the same month, 5 later
2310.06830 (2023-10)
Lemur: Harmonizing Natural Language and Code for Language Agents
2401.10768 (2024-01)
Knowledge Verification to Nip Hallucination in the Bud
2401.15006 (2024-01)
Airavata: Introducing Hindi Instruction-tuned LLM
and 5 more not named here
encode_with_messages_format 8 other papers
3 dated earlier, 2 the same month, 3 later
2310.08491 (2023-10)
Prometheus: Inducing Fine-grained Evaluation Capability in Language Models
2310.11511 (2023-10)
Self-RAG: Learning to Retrieve, Generate, and Critique through Self-Reflection
2311.05657 (2023-11)
Agent Lumos: Unified and Modular Training for Open-Source Language Agents
and 5 more not named here
dispatch_openai_chat_requesets 7 other papers
1 dated earlier, 2 the same month, 4 later
2310.06830 (2023-10)
Lemur: Harmonizing Natural Language and Code for Language Agents
2401.10768 (2024-01)
Knowledge Verification to Nip Hallucination in the Bud
2401.15006 (2024-01)
Airavata: Introducing Hindi Instruction-tuned LLM
and 4 more not named here
create_prompt_with_llama2_chat_format 6 other papers
2 the same month, 4 later
2401.08565 (2024-01)
Tuning Language Models by Proxy
2401.10768 (2024-01)
Knowledge Verification to Nip Hallucination in the Bud
2403.03870 (2024-03)
Learning to Decode Collaboratively with Multiple Language Models
and 3 more not named here
generate_completions 6 other papers
1 dated earlier, 2 the same month, 3 later
2310.08491 (2023-10)
Prometheus: Inducing Fine-grained Evaluation Capability in Language Models
2401.10768 (2024-01)
Knowledge Verification to Nip Hallucination in the Bud
2401.15006 (2024-01)
Airavata: Introducing Hindi Instruction-tuned LLM
and 3 more not named here
get_closest_label 4 other papers
1 dated earlier, 3 later
2303.18223 (2023-03)
A Survey of Large Language Models
2408.15300 (2024-08)
GIFT-SW: Gaussian noise Injected Fine-Tuning of Salient Weights for LLMs
2411.04358 (2024-11)
Robust and Efficient Fine-tuning of LLMs with Bayesian Reparameterization of Low-Rank Adaptation
and 1 more not named here
check_correctness 3 other papers
3 later
2404.02078 (2024-04)
Advancing LLM Reasoning Generalists with Preference Trees
2602.02979 (2026-02)
CPMöbius: Iterative Coach-Player Reasoning for Data-Free Reinforcement Learning
2602.13773 (2026-02)
On Representation Redundancy in Large-Scale Instruction Tuning Data Selection
levenshtein_distance 1 other paper
1 later
2411.16775 (2024-11)
Parameter Efficient Instruction Tuning: An Empirical Study

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

Large Language Models (LLMs) are difficult to fully fine-tune (e.g., with instructions or human feedback) due to their sheer number of parameters. A family of parameter-efficient sparse fine-tuning methods have proven promising in terms of performance but their memory requirements increase proportionally to the size of the LLMs. In this work, we scale sparse fine-tuning to state-of-the-art LLMs like LLaMA 2 7B and 13B. We propose SpIEL, a novel sparse fine-tuning method which, for a desired density level, maintains an array of parameter indices and the deltas of these parameters relative to their pretrained values. It iterates over: (a) updating the active deltas, (b) pruning indices (based on the change of magnitude of their deltas) and (c) regrowth of indices. For regrowth, we explore two criteria based on either the accumulated gradients of a few candidate parameters or their approximate momenta estimated using the efficient SM3 optimizer. We experiment with instruction-tuning of LLMs on standard dataset mixtures, finding that SpIEL is often superior to popular parameter-efficient fine-tuning methods like LoRA (low-rank adaptation) in terms of performance and comparable in terms of run time. We additionally show that SpIEL is compatible with both quantization and efficient optimizers, to facilitate scaling to ever-larger model sizes. We release the code for SpIEL at https://github.com/AlanAnsell/peft and for the instruction-tuning experiments at https://github.com/ducdauge/sft-llm.

For agents

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

get_harvested_code_for_paper("2401.16405")
get_code_for_paper("2401.16405")
have("2401.16405")

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

curl https://syntology.ai/api/ran/2401.16405.json

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

[![Syntology run record](https://syntology.ai/api/ran/2401.16405.svg)](https://syntology.ai/paper/2401.16405)

Connect an agent — have() is free.