Unsupervised Post-Training of Foundation Models: A Survey

Yijie Xu; Qianyi Cai; Huizai Yao; Yili Wang; Tianfu Wang; Cehao Yang; Xingbo Yao; Zhiyu Guo

cs.CLAug 25, 2026

Unsupervised Post-Training of Foundation Models: A Survey

Authors: Yijie Xu, Qianyi Cai, Huizai Yao, Yili Wang, Tianfu Wang, Cehao Yang, Xingbo Yao, Zhiyu Guo,

Organizations: HKUST(GZ) · Xiaohongshu Inc. · WeChat, Tencent · HKUST ·

Abstract

Foundation-model post-training usually relies on human labels, preference data, stronger teachers, or executable verifiers. We study Unsupervised Post-Training (UPT): update-bearing adaptation on unlabeled inputs whose learning signal is derived from same-lineage model artifacts rather than an external oracle. We catalog 80 strict UPT methods and organize them by the object that supplies the update signal: a prediction statistic, a sample relation, a self-generated target, or an internal evaluator. Beyond inventory, we show how the choice of internal signal and task structure determines whether post-training improves the model or recursively amplifies error. An orthogonal Input Visibility $\times$ Update Persistence view maps deployment regimes and defines a unified framework for UPT selection and evaluation.

Figures & tables

Panel of Figure 1 — Figure 1: Update-object taxonomy of Unsupervised Post-Training . Strict UPT methods are grouped by the internal update object. Dashed boxes denote boundary-adjacent methods that fail at least one strict UPT check.

Method	NLL	CPT	Train
	Signal	Mechanism	Regime
CPT-LM ( Ke et al., 2023 )	✓	✓	✓
Simple CPT ( Ibrahim et al., 2024 )	✓	✓	✓
LangAdapt CPT ( Elhady et al., 2025 )	✓	✓	✓
Stability-Gap CPT ( Guo et al., 2025 )	✓	✓	✓
ReplayAlign CPT ( Abbes et al., 2026 )	✓	✓	✓

Table 1: Strict UPT methods in Family I (§ 3 ). Signal : NLL, entropy (Ent.), self-confidence (Conf.), or other geometric/rule-based statistic (Geom./Rule). Mechanism : continued pretraining (CPT), test-time training (TTT), entropy/confidence minimization (EM-min), entropy/confidence or other internal-statistic policy-gradient reward (EM-RL), sample-local state update (State). LC : long-context target. ‡ Family I/IV bridge case (§ 6 ).

	Consensus signal						Regime
Method	Majority	Semantic	Self-	Path	Pairwise	Soft	Train	Test	Multimodal
Method	vote	cluster	certainty	consist.	agreement	cluster			Multimodal
EMPO ( Zhang et al., 2025b )		✓					✓
Intuitor ( Zhao et al., 2026b )			✓				✓
CoVo ( Zhang et al., 2025a )				✓			✓
Co-rewarding ( Zhang et al., 2026b )					✓		✓

Table 2: Strict UPT methods in Family II (§ 4 ). Consensus signal : which multi-sample statistic over multiple internal samples drives the gradient. † marks methods where the consensus is the selection mask and a separate intrinsic term shapes advantages or tokens.

Generated target

Selection criterion

Regime

Method

Instr.

Rationale

Curric.

Pref.

Doc.

Cycle

Confid.

Maj.

Debate

MBR

Foresight

Reflect

Train

Test

Self-Tuning † ( Zhang et al., 2025c )

✓

KBAlign ( Zeng et al., 2025 )

✓

CYCLE-INSTRUCT † ( Shen et al., 2025 )

✓

Self-Improve ( Huang et al., 2023 )

✓

Quiet-STaR ( Zelikman et al., 2024 )

✓

Table 3: Strict UPT methods in Family III (§ 5 ). Generated target : instruction/response (Instr.), rationale, curriculum (Curric.), or preference pair (Pref.). Selection criterion : document grounding (Doc.), knowledge-base grounding (KB), cycle consistency, self-consistency (SC), internal confidence (Confid.), majority vote (Maj.), multi-agent debate, minimum-Bayes risk (MBR), stepwise foresight (Foresight), self-reflection (Reflect). † marks a stage- or variant-qualified strict entry.

Method	Update	Issue addressed
Self-Rewarding LM ( Yuan et al., 2024 )	DPO	baseline self-judge
CREAM ( Wang et al., 2025c )	DPO	bias amplification
Meta-Rewarding ( Wu et al., 2024 )	DPO	judge saturation
Temporal SRLM ( Wang et al., 2025b )	DPO	gradient vanishing
CoNL ( Sui and Hooi, 2026 )	PG	critique-as-reward
RLME ( Rentschler and Roberts, 2026 )	PG	NL meta-judge

Table 4: Strict UPT methods in Family IV (§ 6 ). ‡ Family I/IV bridge case (discussed below).

Appendix figures & tables1 asset

Supplementary material from the paper’s appendix.

Appendix

Table 6

LLM Post-Training

Large language model post-training adapts pretrained models into more useful, reliable, and specialized assistants through supervised and preference-based learning.

See all 249 papers

Post-Training in Time Series Foundation Models: A Unifying Framework
Jul 22, 2026Shifeng Xie, Ambroise Odonnat, Zehao Xiao +7Model Compression Time-Series Foundation Models
Noah’s Ark Lab · Université Paris Cité · Inria +2
Training Foundation Models with a Single Algorithm
Oct 29, 2026Yuda Song, Alvin Zhang, Fahim Tajwar +7LLM Training RL for Language Models
Carnegie Mellon University · Johns Hopkins University · Aurora Innovation
LFM: Leveraging Foundation Models for Source-Free Universal Domain Adaptation
Jul 20, 2026Jing Li, Pan Liu, Meng Zhao +7Source-Free Domain Adaptation Unsupervised Domain Adaptation
School of Computer Science and Engineering, Tianjin University of Technology, Tianjin 300384, China · Engineering Research Center of Learning-Based Intelligent System, Ministry of Education of the People’s Republic of China, Tianjin University of Technology, Tianjin 300384, China · School of Artificial Intelligence, Tianjin University, Tianjin 300350, China +1

Jul 22, 2026J/K move · Enter open · S save

Post-Training in Time Series Foundation Models: A Unifying Framework

Shifeng Xie, Ambroise Odonnat, Zehao Xiao, Lei Zan+6

Noah’s Ark Lab · Université Paris Cité · Inria+2

Model Compression Time-Series Foundation Models

Open paper AskPDF arXiv

Time series foundation models (TSFMs) have emerged as general-purpose models for time series analysis, but pretraining alone is often insufficient for reliable downstream deployment. Bridging this gap requires further intervention to handle domain shift, task heterogeneity, limited supervision, and computational constraints, which motivates post-training as a broad class of methods to adapt, augment, compose, calibrate, or specialize pretrained TSFMs for downstream tasks. In this work, we analyze TSFM post-training methods based on their locus of intervention in the prediction pipeline, yielding five categories: parameter adaptation, context augmentation, model composition, output processing and uncertainty control, and compression and specialization. Within each category, we study main representative methods and discuss their current limitations. We further identify future directions toward controlled adaptation, reliable context construction, uncertainty-aware model composition, calibrated output processing, and deployment-aware specialization. Overall, by providing a unifying framework for the emerging TSFM post-training landscape, this work aims to support future research to navigate the design space between a pretrained TSFM and its reliable downstream deployment.

Abstract

LLM Post-Training

Explore similar work

Post-Training in Time Series Foundation Models: A Unifying Framework

Training Foundation Models with a Single Algorithm

LFM: Leveraging Foundation Models for Source-Free Universal Domain Adaptation