File size: 10,933 Bytes
1227045
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
00e61d8
1227045
 
 
 
 
00e61d8
1227045
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
\begin{thebibliography}{999}

\bibitem[Yao et~al.(2023)Yao, Aminabadi, Ruwase, Rajbhandari, Wu, Awan, Rasley, Zhang, Li, Holmes, Zhou, Wyatt, Smith, Kurilenko, Qin, Tanaka, Che, Song, and He]{yao2023deepspeedchat}
Yao, Z.; Aminabadi, R.Y.; Ruwase, O.; Rajbhandari, S.; Wu, X.; Awan, A.A.; Rasley, J.; Zhang, M.; Li, C.; Holmes, C.;  et~al.
\newblock {DeepSpeed-Chat}: Easy, Fast and Affordable {RLHF} Training of {ChatGPT}-like Models at All Scales.
\newblock {\em arXiv preprint arXiv:2308.01320} {\bf 2023},  \href{http://arxiv.org/abs/2308.01320}{{\normalfont [arXiv:cs.LG/2308.01320]}}.
\newblock {\url{https://doi.org/10.48550/arXiv.2308.01320}}.

\bibitem[Sheng et~al.(2025)Sheng, Zhang, Ye, Wu, Zhang, Zhang, Peng, Lin, and Wu]{sheng2024hybridflow}
Sheng, G.; Zhang, C.; Ye, Z.; Wu, X.; Zhang, W.; Zhang, R.; Peng, Y.; Lin, H.; Wu, C.
\newblock {HybridFlow}: A Flexible and Efficient {RLHF} Framework.
\newblock In Proceedings of the Twentieth European Conference on Computer Systems. ACM,  2025, pp. 1279--1297.
\newblock {\url{https://doi.org/10.1145/3689031.3696075}}.

\bibitem[Zhong et~al.(2025)Zhong, Zhang, Wu, Liu, Chen, Wan, Hu, Xia, Ming, Zhu, and Jin]{zhong2024rlhfuse}
Zhong, Y.; Zhang, Z.; Wu, B.; Liu, S.; Chen, Y.; Wan, C.; Hu, H.; Xia, L.; Ming, R.; Zhu, Y.;  et~al.
\newblock Optimizing {RLHF} Training for Large Language Models with Stage Fusion.
\newblock In Proceedings of the 22nd USENIX Symposium on Networked Systems Design and Implementation (NSDI 25). USENIX Association,  2025, pp. 489--503.

\bibitem[Mei et~al.(2025)Mei, Fu, Li, Wang, Zhang, and Wu]{mei2024realhf}
Mei, Z.; Fu, W.; Li, K.; Wang, G.; Zhang, H.; Wu, Y.
\newblock {ReaL}: Efficient {RLHF} Training of Large Language Models with Parameter Reallocation.
\newblock In Proceedings of the Machine Learning and Systems,  2025, Vol.~7.

\bibitem[Kwon et~al.(2023)Kwon, Li, Zhuang, Sheng, Zheng, Yu, Gonzalez, Zhang, and Stoica]{kwon2023pagedattention}
Kwon, W.; Li, Z.; Zhuang, S.; Sheng, Y.; Zheng, L.; Yu, C.H.; Gonzalez, J.E.; Zhang, H.; Stoica, I.
\newblock Efficient Memory Management for Large Language Model Serving with {PagedAttention}.
\newblock In Proceedings of the 29th Symposium on Operating Systems Principles,  2023, pp. 611--626.
\newblock {\url{https://doi.org/10.1145/3600006.3613165}}.

\bibitem[{vLLM Project}()]{vllm2026sleepmode}
{vLLM Project}.
\newblock {vLLM} Sleep Mode.
\newblock vLLM 0.18.0 documentation.
\newblock Accessed September 14, 2026.

\bibitem[Santacroce et~al.(2023)Santacroce, Lu, Yu, Li, and Shen]{santacroce2023hydra}
Santacroce, M.; Lu, Y.; Yu, H.; Li, Y.; Shen, Y.
\newblock Efficient {RLHF}: Reducing the Memory Usage of {PPO}.
\newblock {\em arXiv preprint arXiv:2309.00754} {\bf 2023},  \href{http://arxiv.org/abs/2309.00754}{{\normalfont [arXiv:cs.LG/2309.00754]}}.
\newblock {\url{https://doi.org/10.48550/arXiv.2309.00754}}.

\bibitem[Zhou et~al.(2024)Zhou, Yang, Tang, Xiang, Guan, and Liu]{zhou2024rlhfmemory}
Zhou, J.; Yang, H.; Tang, S.J.; Xiang, M.; Guan, H.; Liu, T.
\newblock Understanding and Alleviating Memory Consumption in {RLHF} for {LLMs}.
\newblock In Proceedings of the Machine Learning for Systems Workshop at NeurIPS 2024,  2024.

\bibitem[Gao et~al.(2020)Gao, Liu, Zhang, Li, Zhu, Lin, and Yang]{gao2020dnnmem}
Gao, Y.; Liu, Y.; Zhang, H.; Li, Z.; Zhu, Y.; Lin, H.; Yang, M.
\newblock Estimating {GPU} Memory Consumption of Deep Learning Models.
\newblock In Proceedings of the 28th ACM Joint Meeting on European Software Engineering Conference and Symposium on the Foundations of Software Engineering. ACM,  2020, pp. 1342--1352.
\newblock {\url{https://doi.org/10.1145/3368089.3417050}}.

\bibitem[Kim et~al.(2024)Kim, Wang, Chaturvedi, Gupta, Kim, Kwon, and Ha]{kim2024llmem}
Kim, T.; Wang, Y.; Chaturvedi, V.; Gupta, L.; Kim, S.; Kwon, Y.; Ha, S.
\newblock {LLMem}: Estimating {GPU} Memory Usage for Fine-Tuning Pre-Trained {LLMs}.
\newblock In Proceedings of the Thirty-Third International Joint Conference on Artificial Intelligence. International Joint Conferences on Artificial Intelligence Organization,  2024, pp. 6324--6332.
\newblock {\url{https://doi.org/10.24963/ijcai.2024/699}}.

\bibitem[{DeepSpeed Contributors}(2026)]{deepspeed2026memory}
{DeepSpeed Contributors}.
\newblock Memory Requirements.
\newblock DeepSpeed documentation,  2026.
\newblock Accessed October 16, 2026.

\bibitem[Lysenst{\o}en(2026)]{lysenstoen2026sloguard}
Lysenst{\o}en, C.
\newblock {SLO-Guard}: Crash-Aware, Budget-Consistent Autotuning for {SLO}-Constrained {LLM} Serving.
\newblock {\em arXiv preprint arXiv:2604.17627} {\bf 2026},  \href{http://arxiv.org/abs/2604.17627}{{\normalfont [arXiv:cs.LG/2604.17627]}}.
\newblock {\url{https://doi.org/10.48550/arXiv.2604.17627}}.

\bibitem[Fu et~al.(2025)Fu, Gao, Shen, Zhu, Mei, He, Xu, Wei, Mei, Wang, Yang, Yuan, and Wu]{fu2025areal}
Fu, W.; Gao, J.; Shen, X.; Zhu, C.; Mei, Z.; He, C.; Xu, S.; Wei, G.; Mei, J.; Wang, J.;  et~al.
\newblock {AReaL}: A Large-Scale Asynchronous Reinforcement Learning System for Language Reasoning.
\newblock In Proceedings of the Advances in Neural Information Processing Systems. Curran Associates, Inc.,  2025, Vol.~38, pp. 36256--36282.
\newblock {\url{https://doi.org/10.52202/085713-1218}}.

\bibitem[Zeng et~al.(2027)Zeng, Wang, and Cheng]{zeng2027stateattention}
Zeng, Y.; Wang, X.; Cheng, Y.
\newblock Similarity-guided state attention for visual reinforcement learning.
\newblock {\em Neural Networks} {\bf 2027}, {\em 205},~109560.
\newblock Published online September 1, 2026, {\url{https://doi.org/10.1016/j.neunet.2026.109560}}.

\bibitem[Cobbe et~al.(2021)Cobbe, Kosaraju, Bavarian, Chen, Jun, Kaiser, Plappert, Tworek, Hilton, Nakano, Hesse, and Schulman]{cobbe2021gsm8k}
Cobbe, K.; Kosaraju, V.; Bavarian, M.; Chen, M.; Jun, H.; Kaiser, L.; Plappert, M.; Tworek, J.; Hilton, J.; Nakano, R.;  et~al.
\newblock Training Verifiers to Solve Math Word Problems.
\newblock {\em arXiv preprint arXiv:2110.14168} {\bf 2021},  \href{http://arxiv.org/abs/2110.14168}{{\normalfont [2110.14168]}}.
\newblock {\url{https://doi.org/10.48550/arXiv.2110.14168}}.

\bibitem[Hendrycks et~al.(2021)Hendrycks, Burns, Kadavath, Arora, Basart, Tang, Song, and Steinhardt]{hendrycks2021math}
Hendrycks, D.; Burns, C.; Kadavath, S.; Arora, A.; Basart, S.; Tang, E.; Song, D.; Steinhardt, J.
\newblock Measuring Mathematical Problem Solving With the {MATH} Dataset.
\newblock In Proceedings of the Neural Information Processing Systems Track on Datasets and Benchmarks,  2021, Vol.~1.

\bibitem[Li et~al.(2022)Li, Choi, Chung, Kushman, Schrittwieser, Leblond, Eccles, Keeling, Gimeno, Dal~Lago, Hubert, Choy, de~Masson~d'Autume, Babuschkin, Chen, Huang, Welbl, Gowal, Cherepanov, Molloy, Mankowitz, Sutherland~Robson, Kohli, de~Freitas, Kavukcuoglu, and Vinyals]{li2022alphacode}
Li, Y.; Choi, D.; Chung, J.; Kushman, N.; Schrittwieser, J.; Leblond, R.; Eccles, T.; Keeling, J.; Gimeno, F.; Dal~Lago, A.;  et~al.
\newblock Competition-level code generation with {AlphaCode}.
\newblock {\em Science} {\bf 2022}, {\em 378},~1092--1097.
\newblock {\url{https://doi.org/10.1126/science.abq1158}}.

\bibitem[{IT4Innovations National Supercomputing Center}(2026)]{it4i2026karolina}
{IT4Innovations National Supercomputing Center}.
\newblock {Karolina}: System Description,  2026.
\newblock Accessed September 14, 2026.

\bibitem[Shao et~al.(2024)Shao, Wang, Zhu, Xu, Song, Bi, Zhang, Zhang, Li, Wu, and Guo]{shao2024deepseekmath}
Shao, Z.; Wang, P.; Zhu, Q.; Xu, R.; Song, J.; Bi, X.; Zhang, H.; Zhang, M.; Li, Y.K.; Wu, Y.;  et~al.
\newblock {DeepSeekMath}: Pushing the Limits of Mathematical Reasoning in Open Language Models.
\newblock {\em arXiv preprint arXiv:2402.03300} {\bf 2024},  \href{http://arxiv.org/abs/2402.03300}{{\normalfont [arXiv:cs.CL/2402.03300]}}.
\newblock {\url{https://doi.org/10.48550/arXiv.2402.03300}}.

\bibitem[Hu et~al.(2022)Hu, Shen, Wallis, Allen-Zhu, Li, Wang, Wang, and Chen]{hu2022lora}
Hu, E.J.; Shen, Y.; Wallis, P.; Allen-Zhu, Z.; Li, Y.; Wang, S.; Wang, L.; Chen, W.
\newblock {LoRA}: Low-Rank Adaptation of Large Language Models.
\newblock In Proceedings of the International Conference on Learning Representations,  2022.

\bibitem[{Qwen Team}(2024)]{qwen2024qwen253b}
{Qwen Team}.
\newblock {Qwen2.5-3B-Instruct} Model Card.
\newblock Hugging Face model card,  2024.
\newblock Accessed September 14, 2026.

\bibitem[{Microsoft}(2025)]{microsoft2025phi4mini}
{Microsoft}.
\newblock {Phi-4-mini-instruct} Model Card.
\newblock Hugging Face model card,  2025.
\newblock Accessed September 14, 2026.

\bibitem[{IBM Granite Team}(2025)]{ibm2025granite33}
{IBM Granite Team}.
\newblock {Granite-3.3-2B-Instruct} Model Card.
\newblock Hugging Face model card,  2025.
\newblock Accessed September 14, 2026.

\bibitem[{Qwen Team}(2024)]{qwen2024qwen257b}
{Qwen Team}.
\newblock {Qwen2.5-7B-Instruct} Model Card.
\newblock Hugging Face model card,  2024.
\newblock Accessed September 16, 2026.

\bibitem[{Mistral AI}(2024)]{mistral2024mistral7bv03}
{Mistral AI}.
\newblock {Mistral-7B-Instruct-v0.3} Model Card.
\newblock Hugging Face model card,  2024.
\newblock Accessed September 16, 2026.

\bibitem[{LUMI Consortium}(2026)]{lumi2026}
{LUMI Consortium}.
\newblock {GPU} Nodes -- {LUMI-G}, {LUMI} Documentation,  2026.
\newblock Accessed October 21, 2026.

\bibitem[Henderson et~al.(2018)Henderson, Islam, Bachman, Pineau, Precup, and Meger]{henderson2018matters}
Henderson, P.; Islam, R.; Bachman, P.; Pineau, J.; Precup, D.; Meger, D.
\newblock Deep Reinforcement Learning that Matters.
\newblock In Proceedings of the Thirty-Second AAAI Conference on Artificial Intelligence,  2018.
\newblock {\url{https://doi.org/10.1609/aaai.v32i1.11694}}.

\bibitem[Agarwal et~al.(2021)Agarwal, Schwarzer, Castro, Courville, and Bellemare]{agarwal2021precipice}
Agarwal, R.; Schwarzer, M.; Castro, P.S.; Courville, A.; Bellemare, M.G.
\newblock Deep Reinforcement Learning at the Edge of the Statistical Precipice.
\newblock In Proceedings of the Advances in Neural Information Processing Systems,  2021, Vol.~34, pp. 29304--29320.

\bibitem[Kobza and {\v S}ediv{\'y}(2026{\natexlab{a}})]{kobza2026benchmark}
Kobza, O.; {\v S}ediv{\'y}, J.
\newblock {RLVRAMBench}: Benchmark Evidence Release.
\newblock Hugging Face dataset,  2026.
\newblock Version electronics-submission-2026-10-23-r2; released October 23, 2026, {\url{https://doi.org/10.57967/hf/10780}}.

\bibitem[Kobza and {\v S}ediv{\'y}(2026{\natexlab{b}})]{kobza2026reproduction}
Kobza, O.; {\v S}ediv{\'y}, J.
\newblock {RLVRAMBench}: Resource Transfer Benchmark and Reproduction Code.
\newblock GitHub repository,  2026.
\newblock Tag electronics-submission-2026-10-23-r2; released October 23, 2026.

\bibitem[{OpenAI}(2026)]{openai2026codex}
{OpenAI}.
\newblock {Codex}.
\newblock \url{https://openai.com/codex/},  2026.
\newblock Coding assistance and text refinement; accessed September 13, 2026.

\end{thebibliography}