SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2312.00407 · EMNLP · 2023

CoLLiE: Collaborative Training of Large Language Models in an Efficient Way

Yu Sun, Shuo Zhang, Xipeng Qiu, Qipeng Guo, Kai Lv, Yuqing Yang, Xiaoran Liu, Hang Yan, Tengxiao Liu, Keyu Chen, Jiawei Hong, Honglin Guo, and 3 more

arXiv · PDF · Open in the Atlas

Code that ran

4 of 7 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; 2 failures of our instrument; 3 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/2312.00407.json · badge: /api/ran/2312.00407.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
OpenLMLab/collie canonical (classifier label)
classifier basis: census
4 ran · 3 no recorded run
FunctionStatusWhere it lives
inputs_to_kv_cache_for_model Ran
ran, honoured the contract we drafted
OpenLMLab/collie/collie/models/utils.py
code served (permissive licence) · get_code("7accb4cc8f6b2131")
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 6, 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)

# Construct inputs dictionary with kv_cache entries
# Each layer has past_key_values_layer{i}_key and past_key_values_layer{i}_value
inputs = {
    'past_key_values_layer0_key': torch.randn(2, 4, 8, 4),  # (batch, seq_len, num_heads, head_dim)
    'past_key_values_layer0_value': torch.randn(2, 4, 8, 4),
    'past_key_values_layer1_key': torch.randn(2, 4, 8, 4),
    'past_key_values_layer1_value': torch.randn(2, 4, 8, 4),
    'past_key_values_layer2_key': torch.randn(2, 4, 8, 4),
    'past_key_values_layer2_value': torch.randn(2, 4, 8, 4),
}

num_hidden_layers = 3

ARGS = {
    'num_hidden_layers': num_hidden_layers,
    'inputs': inputs,
}
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, ndim 6, dtype family float — 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
kv_cache_to_inputs_for_model Ran
ran; our contract draft was wrong, not the code
OpenLMLab/collie/collie/models/utils.py
code served (permissive licence) · get_code("4bdb94ab969119dc")
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)

# Construct past_key_values as a nested tuple structure
# Each layer has a tuple of (key_tensor, value_tensor)
# Using distinct sizes: batch=2, seq_len=3, num_heads=4, head_dim=5

past_key_values = (
    (
        torch.randn(2, 4, 3, 5),  # layer 0 key: (batch, num_heads, seq_len, head_dim)
        torch.randn(2, 4, 3, 5),  # layer 0 value
    ),
    (
        torch.randn(2, 4, 3, 5),  # layer 1 key
        torch.randn(2, 4, 3, 5),  # layer 1 value
    ),
    (
        torch.randn(2, 4, 3, 5),  # layer 2 key
        torch.randn(2, 4, 3, 5),  # layer 2 value
    ),
)

ARGS = {
    "past_key_values": past_key_values,
}
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
load_config Ran
ran; our contract draft was wrong, not the code
OpenLMLab/collie/collie/config.py
code served (permissive licence) · get_code("3240d072cd82bbff")
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 tempfile
import json
import yaml
import os

# Create a temporary YAML file
temp_yaml_dir = tempfile.mkdtemp()
yaml_path = os.path.join(temp_yaml_dir, "config.yaml")
yaml_content = {
    "model": "test_model",
    "batch_size": 32,
    "learning_rate": 0.001
}
with open(yaml_path, 'w') as f:
    yaml.dump(yaml_content, f)

# Create a temporary JSON file
json_path = os.path.join(temp_yaml_dir, "config.json")
json_content = {
    "model": "test_model",
    "batch_size": 16,
    "learning_rate": 0.0005
}
with open(json_path, 'w') as f:
    json.dump(json_content, f)

ARGS = {
    "path": yaml_path
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
yaml — 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_callback Ran
ran, honoured the contract we drafted
OpenLMLab/collie/collie/callbacks/callback_manager.py
code served (permissive licence) · get_code("25e3c49687867f6c")
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
from typing import Sequence

torch.manual_seed(0)

# Define the Callback base class that the function expects
class Callback:
    def __init__(self, name="callback"):
        self.name = name

# Create some concrete callback instances
callback1 = Callback("callback_1")
callback2 = Callback("callback_2")

# Test case 1: None input
ARGS = {
    "callbacks": None
}

# Alternative test cases to verify the function works:
# Single callback
# ARGS = {
#     "callbacks": callback1
# }

# List of callbacks
# ARGS = {
#     "callbacks": [callback1, callback2]
# }

# Tuple of callbacks
# ARGS = {
#     "callbacks": (callback1, callback2)
# }
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
config_command_parser Not yet run
unverified (harvested, no recorded run)
OpenLMLab/collie/collie_cli/config.py
code served (permissive licence) · get_code("91010674e84e0e76")
How it ran

no invocation recorded.

flash_attention Not yet run
unverified (harvested, no recorded run)
OpenLMLab/collie/collie/models/utils.py
code served (permissive licence) · get_code("f854c14e7ec1871d")
How it ran

no invocation recorded.

generate_deepspeed_config Not yet run
unverified (harvested, no recorded run)
OpenLMLab/collie/collie_cli/config.py
code served (permissive licence) · get_code("605ce2d0520693bc")
How it ran

no invocation recorded.

Also present in other papers' code

Of 7 samples lifted from this paper's linked repositories, 7 carry a function identity and 0 do not. Of the 7 distinct functions with an identity, 0 are also present in other papers whose code we hold (0 with at least one paper dated earlier, 0 the same month, 0 later) and 7 are only here among the 52,977 papers whose code Syntology holds (as of 2026-09-26).

Only here means: among the 52,977 papers whose code Syntology holds (51,887 with lifted samples, 1,090 through their assemblies only), as of 2026-09-26, no other paper's lifted samples or current assembly bricks contain an identical function body after normalisation.

Presence, not direction: an identical function body in two papers' repositories does not say who wrote it, which came from which, or that either paper's method rests on it. Matching is exact after normalisation, so a renamed or edited function is not matched. Dates are when papers appeared, not when code was written. Only .py files the harvester lifted are covered; notebooks are not read. A new-style arXiv id YYMM.NNNN dates a paper to the month 20YY-MM; otherwise Paper.year dates it to the year; otherwise it is undated. Dates are when the paper appeared, not when the code was written. Read 2026-11-22.

Repositories linked to this paper

Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.

Abstract

Large language models (LLMs) are increasingly pivotal in a wide range of natural language processing tasks. Access to pre-trained models, courtesy of the open-source community, has made it possible to adapt these models to specific applications for enhanced performance. However, the substantial resources required for training these models necessitate efficient solutions. This paper introduces CoL-LiE, an efficient library that facilitates collaborative training of large language models using 3D parallelism, parameter-efficient finetuning (PEFT) methods, and optimizers such as Lion, Adan, Sophia, and LOMO. With its modular design and comprehensive functionality, CoLLiE offers a balanced blend of efficiency, ease of use, and customization. CoLLiE has proven superior training efficiency in comparison with prevalent solutions in pre-training and fine-tuning scenarios. Furthermore, we provide an empirical evaluation of the correlation between model size and GPU memory consumption under different optimization methods, as well as an analysis of the throughput. Lastly, we carry out a comprehensive comparison of various optimizers and PEFT methods within the instruction-tuning context. CoLLiE is available at https://github.com/OpenLMLab/collie.

For agents

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

get_harvested_code_for_paper("2312.00407")
get_code_for_paper("2312.00407")
have("2312.00407")

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

curl https://syntology.ai/api/ran/2312.00407.json

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

[![Syntology run record](https://syntology.ai/api/ran/2312.00407.svg)](https://syntology.ai/paper/2312.00407)

Connect an agent — have() is free.