PMoL: Parameter Efficient MoE for Preference Mixing of LLM Alignment Article Swipe

PDF

Deyu Liu , Bing Xu , Yinzhuo Chen , Baiping Xu , Wenpeng Lü , Muyun Yang , Tiejun Zhao ·

YOU? · · 2024 · Open Access · · DOI: https://doi.org/10.48550/arxiv.2411.01245

Reinforcement Learning from Human Feedback (RLHF) has been proven to be an effective method for preference alignment of large language models (LLMs) and is widely used in the post-training process of LLMs. However, RLHF struggles with handling multiple competing preferences. This leads to a decrease in the alignment of LLMs with human preferences. To address this issue, we propose Preference Mixture of LoRAs (PMoL) from the perspective of model architecture, which can adapt to any number of preferences to mix. PMoL combines Mixture of Experts (MoE) and Low Rank Adaptor (LoRA). This architecture is innovatively applied to the research of preference alignment and has achieved significant performance improvement. The expert group soft loss is used to enable MoE with the ability to mix preferences. Through comprehensive evaluation by the reward model and GPT-4o, the experiment results show that PMoL has superior preference mixing capabilities compared to baseline methods. PMoL achieves better preference alignment with lower training costs.

Related Topics

Concepts

Mixing (physics) Preference Mathematics Computer science Physics Statistics Quantum mechanics

Metadata

Type: preprint
Language: en
Landing Page: http://arxiv.org/abs/2411.01245
PDF: https://arxiv.org/pdf/2411.01245
OA Status: green
Related Works: 10
OpenAlex ID: https://openalex.org/W4404351351

All OpenAlex metadata

Raw OpenAlex JSON

OpenAlex ID: https://openalex.org/W4404351351

Canonical identifier for this work in OpenAlex
DOI: https://doi.org/10.48550/arxiv.2411.01245

Digital Object Identifier
Title: PMoL: Parameter Efficient MoE for Preference Mixing of LLM Alignment

Work title
Type: preprint

OpenAlex work type
Language: en

Primary language
Publication year: 2024

Year of publication
Publication date: 2024-11-02

Full publication date if available
Authors: Deyu Liu, Bing Xu, Yinzhuo Chen, Baiping Xu, Wenpeng Lü, Muyun Yang, Tiejun Zhao

List of authors in order
Landing page: https://arxiv.org/abs/2411.01245

Publisher landing page
PDF URL: https://arxiv.org/pdf/2411.01245

Direct link to full text PDF
Open access: Yes

Whether a free full text is available
OA status: green

Open access status per OpenAlex
OA URL: https://arxiv.org/pdf/2411.01245

Direct OA link when available
Concepts: Mixing (physics), Preference, Mathematics, Computer science, Physics, Statistics, Quantum mechanics

Top concepts (fields/topics) attached by OpenAlex
Cited by: 0

Total citation count in OpenAlex
Related works (count): 10

Other works algorithmically related by OpenAlex

Full payload

id	https://openalex.org/W4404351351
doi	https://doi.org/10.48550/arxiv.2411.01245
ids.doi	https://doi.org/10.48550/arxiv.2411.01245
ids.openalex	https://openalex.org/W4404351351
fwci
type	preprint
title	PMoL: Parameter Efficient MoE for Preference Mixing of LLM Alignment
biblio.issue
biblio.volume
biblio.last_page
biblio.first_page
topics[0].id	https://openalex.org/T10181
topics[0].field.id	https://openalex.org/fields/17
topics[0].field.display_name	Computer Science
topics[0].score	0.9793000221252441
topics[0].domain.id	https://openalex.org/domains/3
topics[0].domain.display_name	Physical Sciences
topics[0].subfield.id	https://openalex.org/subfields/1702
topics[0].subfield.display_name	Artificial Intelligence
topics[0].display_name	Natural Language Processing Techniques
topics[1].id	https://openalex.org/T11063
topics[1].field.id	https://openalex.org/fields/17
topics[1].field.display_name	Computer Science
topics[1].score	0.953499972820282
topics[1].domain.id	https://openalex.org/domains/3
topics[1].domain.display_name	Physical Sciences
topics[1].subfield.id	https://openalex.org/subfields/1703
topics[1].subfield.display_name	Computational Theory and Mathematics
topics[1].display_name	Rough Sets and Fuzzy Logic
topics[2].id	https://openalex.org/T13999
topics[2].field.id	https://openalex.org/fields/17
topics[2].field.display_name	Computer Science
topics[2].score	0.9232000112533569
topics[2].domain.id	https://openalex.org/domains/3
topics[2].domain.display_name	Physical Sciences
topics[2].subfield.id	https://openalex.org/subfields/1710
topics[2].subfield.display_name	Information Systems
topics[2].display_name	Digital Rights Management and Security
is_xpac	False
apc_list
apc_paid
concepts[0].id	https://openalex.org/C138777275
concepts[0].level	2
concepts[0].score	0.7729140520095825
concepts[0].wikidata	https://www.wikidata.org/wiki/Q6884054
concepts[0].display_name	Mixing (physics)
concepts[1].id	https://openalex.org/C2781249084
concepts[1].level	2
concepts[1].score	0.6197450160980225
concepts[1].wikidata	https://www.wikidata.org/wiki/Q908656
concepts[1].display_name	Preference
concepts[2].id	https://openalex.org/C33923547
concepts[2].level	0
concepts[2].score	0.3845593333244324
concepts[2].wikidata	https://www.wikidata.org/wiki/Q395
concepts[2].display_name	Mathematics
concepts[3].id	https://openalex.org/C41008148
concepts[3].level	0
concepts[3].score	0.32169491052627563
concepts[3].wikidata	https://www.wikidata.org/wiki/Q21198
concepts[3].display_name	Computer science
concepts[4].id	https://openalex.org/C121332964
concepts[4].level	0
concepts[4].score	0.2775244116783142
concepts[4].wikidata	https://www.wikidata.org/wiki/Q413
concepts[4].display_name	Physics
concepts[5].id	https://openalex.org/C105795698
concepts[5].level	1
concepts[5].score	0.2504141330718994
concepts[5].wikidata	https://www.wikidata.org/wiki/Q12483
concepts[5].display_name	Statistics
concepts[6].id	https://openalex.org/C62520636
concepts[6].level	1
concepts[6].score	0.0
concepts[6].wikidata	https://www.wikidata.org/wiki/Q944
concepts[6].display_name	Quantum mechanics
keywords[0].id	https://openalex.org/keywords/mixing
keywords[0].score	0.7729140520095825
keywords[0].display_name	Mixing (physics)
keywords[1].id	https://openalex.org/keywords/preference
keywords[1].score	0.6197450160980225
keywords[1].display_name	Preference
keywords[2].id	https://openalex.org/keywords/mathematics
keywords[2].score	0.3845593333244324
keywords[2].display_name	Mathematics
keywords[3].id	https://openalex.org/keywords/computer-science
keywords[3].score	0.32169491052627563
keywords[3].display_name	Computer science
keywords[4].id	https://openalex.org/keywords/physics
keywords[4].score	0.2775244116783142
keywords[4].display_name	Physics
keywords[5].id	https://openalex.org/keywords/statistics
keywords[5].score	0.2504141330718994
keywords[5].display_name	Statistics
language	en
locations[0].id	pmh:oai:arXiv.org:2411.01245
locations[0].is_oa	True
locations[0].source.id	https://openalex.org/S4306400194
locations[0].source.issn
locations[0].source.type	repository
locations[0].source.is_oa	True
locations[0].source.issn_l
locations[0].source.is_core	False
locations[0].source.is_in_doaj	False
locations[0].source.display_name	arXiv (Cornell University)
locations[0].source.host_organization	https://openalex.org/I205783295
locations[0].source.host_organization_name	Cornell University
locations[0].source.host_organization_lineage	https://openalex.org/I205783295
locations[0].license
locations[0].pdf_url	https://arxiv.org/pdf/2411.01245
locations[0].version	submittedVersion
locations[0].raw_type	text
locations[0].license_id
locations[0].is_accepted	False
locations[0].is_published	False
locations[0].raw_source_name
locations[0].landing_page_url	http://arxiv.org/abs/2411.01245
locations[1].id	doi:10.48550/arxiv.2411.01245
locations[1].is_oa	True
locations[1].source.id	https://openalex.org/S4306400194
locations[1].source.issn
locations[1].source.type	repository
locations[1].source.is_oa	True
locations[1].source.issn_l
locations[1].source.is_core	False
locations[1].source.is_in_doaj	False
locations[1].source.display_name	arXiv (Cornell University)
locations[1].source.host_organization	https://openalex.org/I205783295
locations[1].source.host_organization_name	Cornell University
locations[1].source.host_organization_lineage	https://openalex.org/I205783295
locations[1].license
locations[1].pdf_url
locations[1].version
locations[1].raw_type	article
locations[1].license_id
locations[1].is_accepted	False
locations[1].is_published
locations[1].raw_source_name
locations[1].landing_page_url	https://doi.org/10.48550/arxiv.2411.01245
indexed_in	arxiv, datacite
authorships[0].author.id	https://openalex.org/A5013126069
authorships[0].author.orcid	https://orcid.org/0000-0002-2578-9525
authorships[0].author.display_name	Deyu Liu
authorships[0].author_position	first
authorships[0].raw_author_name	Liu, Dongxu
authorships[0].is_corresponding	False
authorships[1].author.id	https://openalex.org/A5101926912
authorships[1].author.orcid	https://orcid.org/0000-0002-2094-7002
authorships[1].author.display_name	Bing Xu
authorships[1].author_position	middle
authorships[1].raw_author_name	Xu, Bing
authorships[1].is_corresponding	False
authorships[2].author.id	https://openalex.org/A5079842923
authorships[2].author.orcid
authorships[2].author.display_name	Yinzhuo Chen
authorships[2].author_position	middle
authorships[2].raw_author_name	Chen, Yinzhuo
authorships[2].is_corresponding	False
authorships[3].author.id	https://openalex.org/A5021958428
authorships[3].author.orcid	https://orcid.org/0000-0003-2891-5635
authorships[3].author.display_name	Baiping Xu
authorships[3].author_position	middle
authorships[3].raw_author_name	Xu, Bufan
authorships[3].is_corresponding	False
authorships[4].author.id	https://openalex.org/A5076564877
authorships[4].author.orcid	https://orcid.org/0000-0002-1840-3540
authorships[4].author.display_name	Wenpeng Lü
authorships[4].author_position	middle
authorships[4].raw_author_name	Lu, Wenpeng
authorships[4].is_corresponding	False
authorships[5].author.id	https://openalex.org/A5108053327
authorships[5].author.orcid	https://orcid.org/0000-0002-5940-0266
authorships[5].author.display_name	Muyun Yang
authorships[5].author_position	middle
authorships[5].raw_author_name	Yang, Muyun
authorships[5].is_corresponding	False
authorships[6].author.id	https://openalex.org/A5038560914
authorships[6].author.orcid	https://orcid.org/0000-0002-1754-5416
authorships[6].author.display_name	Tiejun Zhao
authorships[6].author_position	last
authorships[6].raw_author_name	Zhao, Tiejun
authorships[6].is_corresponding	False
has_content.pdf	False
has_content.grobid_xml	False
is_paratext	False
open_access.is_oa	True
open_access.oa_url	https://arxiv.org/pdf/2411.01245
open_access.oa_status	green
open_access.any_repository_has_fulltext	False
created_date	2025-10-10T00:00:00
display_name	PMoL: Parameter Efficient MoE for Preference Mixing of LLM Alignment
has_fulltext	False
is_retracted	False
updated_date	2025-11-06T06:51:31.235846
primary_topic.id	https://openalex.org/T10181
primary_topic.field.id	https://openalex.org/fields/17
primary_topic.field.display_name	Computer Science
primary_topic.score	0.9793000221252441
primary_topic.domain.id	https://openalex.org/domains/3
primary_topic.domain.display_name	Physical Sciences
primary_topic.subfield.id	https://openalex.org/subfields/1702
primary_topic.subfield.display_name	Artificial Intelligence
primary_topic.display_name	Natural Language Processing Techniques
related_works	https://openalex.org/W2899084033, https://openalex.org/W2748952813, https://openalex.org/W4391375266, https://openalex.org/W1979597421, https://openalex.org/W2007980826, https://openalex.org/W2061531152, https://openalex.org/W3002753104, https://openalex.org/W2077600819, https://openalex.org/W2142036596, https://openalex.org/W2072657027
cited_by_count	0
locations_count	2
best_oa_location.id	pmh:oai:arXiv.org:2411.01245
best_oa_location.is_oa	True
best_oa_location.source.id	https://openalex.org/S4306400194
best_oa_location.source.issn
best_oa_location.source.type	repository
best_oa_location.source.is_oa	True
best_oa_location.source.issn_l
best_oa_location.source.is_core	False
best_oa_location.source.is_in_doaj	False
best_oa_location.source.display_name	arXiv (Cornell University)
best_oa_location.source.host_organization	https://openalex.org/I205783295
best_oa_location.source.host_organization_name	Cornell University
best_oa_location.source.host_organization_lineage	https://openalex.org/I205783295
best_oa_location.license
best_oa_location.pdf_url	https://arxiv.org/pdf/2411.01245
best_oa_location.version	submittedVersion
best_oa_location.raw_type	text
best_oa_location.license_id
best_oa_location.is_accepted	False
best_oa_location.is_published	False
best_oa_location.raw_source_name
best_oa_location.landing_page_url	http://arxiv.org/abs/2411.01245
primary_location.id	pmh:oai:arXiv.org:2411.01245
primary_location.is_oa	True
primary_location.source.id	https://openalex.org/S4306400194
primary_location.source.issn
primary_location.source.type	repository
primary_location.source.is_oa	True
primary_location.source.issn_l
primary_location.source.is_core	False
primary_location.source.is_in_doaj	False
primary_location.source.display_name	arXiv (Cornell University)
primary_location.source.host_organization	https://openalex.org/I205783295
primary_location.source.host_organization_name	Cornell University
primary_location.source.host_organization_lineage	https://openalex.org/I205783295
primary_location.license
primary_location.pdf_url	https://arxiv.org/pdf/2411.01245
primary_location.version	submittedVersion
primary_location.raw_type	text
primary_location.license_id
primary_location.is_accepted	False
primary_location.is_published	False
primary_location.raw_source_name
primary_location.landing_page_url	http://arxiv.org/abs/2411.01245
publication_date	2024-11-02
publication_year	2024
referenced_works_count	0
abstract_inverted_index.a	43
abstract_inverted_index.To	53
abstract_inverted_index.an	11
abstract_inverted_index.be	10
abstract_inverted_index.by	127
abstract_inverted_index.in	26, 45
abstract_inverted_index.is	23, 93, 113
abstract_inverted_index.of	17, 30, 48, 61, 67, 76, 83, 99
abstract_inverted_index.to	9, 42, 73, 78, 96, 115, 121, 145
abstract_inverted_index.we	57
abstract_inverted_index.Low	87
abstract_inverted_index.MoE	117
abstract_inverted_index.The	108
abstract_inverted_index.and	22, 86, 102, 131
abstract_inverted_index.any	74
abstract_inverted_index.can	71
abstract_inverted_index.for	14
abstract_inverted_index.has	6, 103, 139
abstract_inverted_index.mix	122
abstract_inverted_index.the	27, 46, 65, 97, 119, 128, 133
abstract_inverted_index.LLMs	49
abstract_inverted_index.PMoL	80, 138, 148
abstract_inverted_index.RLHF	33
abstract_inverted_index.Rank	88
abstract_inverted_index.This	40, 91
abstract_inverted_index.been	7
abstract_inverted_index.from	2, 64
abstract_inverted_index.loss	112
abstract_inverted_index.mix.	79
abstract_inverted_index.show	136
abstract_inverted_index.soft	111
abstract_inverted_index.that	137
abstract_inverted_index.this	55
abstract_inverted_index.used	25, 114
abstract_inverted_index.with	35, 50, 118, 153
abstract_inverted_index.(MoE)	85
abstract_inverted_index.Human	3
abstract_inverted_index.LLMs.	31
abstract_inverted_index.LoRAs	62
abstract_inverted_index.adapt	72
abstract_inverted_index.group	110
abstract_inverted_index.human	51
abstract_inverted_index.large	18
abstract_inverted_index.leads	41
abstract_inverted_index.lower	154
abstract_inverted_index.model	68, 130
abstract_inverted_index.which	70
abstract_inverted_index.(LLMs)	21
abstract_inverted_index.(PMoL)	63
abstract_inverted_index.(RLHF)	5
abstract_inverted_index.better	150
abstract_inverted_index.costs.	156
abstract_inverted_index.enable	116
abstract_inverted_index.expert	109
abstract_inverted_index.issue,	56
abstract_inverted_index.method	13
abstract_inverted_index.mixing	142
abstract_inverted_index.models	20
abstract_inverted_index.number	75
abstract_inverted_index.proven	8
abstract_inverted_index.reward	129
abstract_inverted_index.widely	24
abstract_inverted_index.(LoRA).	90
abstract_inverted_index.Adaptor	89
abstract_inverted_index.Experts	84
abstract_inverted_index.GPT-4o,	132
abstract_inverted_index.Mixture	60, 82
abstract_inverted_index.Through	124
abstract_inverted_index.ability	120
abstract_inverted_index.address	54
abstract_inverted_index.applied	95
abstract_inverted_index.process	29
abstract_inverted_index.propose	58
abstract_inverted_index.results	135
abstract_inverted_index.Feedback	4
abstract_inverted_index.However,	32
abstract_inverted_index.Learning	1
abstract_inverted_index.achieved	104
abstract_inverted_index.achieves	149
abstract_inverted_index.baseline	146
abstract_inverted_index.combines	81
abstract_inverted_index.compared	144
abstract_inverted_index.decrease	44
abstract_inverted_index.handling	36
abstract_inverted_index.language	19
abstract_inverted_index.methods.	147
abstract_inverted_index.multiple	37
abstract_inverted_index.research	98
abstract_inverted_index.superior	140
abstract_inverted_index.training	155
abstract_inverted_index.alignment	16, 47, 101, 152
abstract_inverted_index.competing	38
abstract_inverted_index.effective	12
abstract_inverted_index.struggles	34
abstract_inverted_index.Preference	59
abstract_inverted_index.evaluation	126
abstract_inverted_index.experiment	134
abstract_inverted_index.preference	15, 100, 141, 151
abstract_inverted_index.performance	106
abstract_inverted_index.perspective	66
abstract_inverted_index.preferences	77
abstract_inverted_index.significant	105
abstract_inverted_index.architecture	92
abstract_inverted_index.capabilities	143
abstract_inverted_index.improvement.	107
abstract_inverted_index.innovatively	94
abstract_inverted_index.preferences.	39, 52, 123
abstract_inverted_index.Reinforcement	0
abstract_inverted_index.architecture,	69
abstract_inverted_index.comprehensive	125
abstract_inverted_index.post-training	28
cited_by_percentile_year
countries_distinct_count	0
institutions_distinct_count	7
citation_normalized_percentile