Reinforcement Learning
Transformers
English
post-training
distillation
agentic-coding
composer-2.5
cursor
kimi-k2
grpo
dapo
diloco
openenv
trl
verl
research
methodology
Instructions to use Codeseys/composer-replication-framework with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use Codeseys/composer-replication-framework with Transformers:
# Load model directly from transformers import AutoModel model = AutoModel.from_pretrained("Codeseys/composer-replication-framework", dtype="auto") - Notebooks
- Google Colab
- Kaggle
File size: 2,612 Bytes
639a760 | 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 | % Composer 2.5 Replication Framework — BibTeX citation file
% https://huggingface.co/Codeseys/composer-replication-framework
%
% Citation order: this work first, then the upstream sources you'd typically
% cite alongside it (Cursor blog, OPSD, SDPO).
@misc{composer-replication-framework-2026,
author = {Codeseys},
title = {Composer 2.5 Replication Framework: Methodology and Integration Architecture for Open Replication of Cursor's Agentic Coding Recipe},
year = {2026},
publisher = {HuggingFace},
howpublished = {\url{https://huggingface.co/Codeseys/composer-replication-framework}},
note = {Pre-experimental v0.0 release. Methodology, integration architecture across TRL/VeRL/OpenEnv, and economic-feasibility result for novel multi-teacher trace-replay channel. Empirical training validation in follow-up paper.}
}
@article{cursor2026composer25,
title = {Introducing {C}omposer 2.5},
author = {{Cursor Team}},
year = {2026},
url = {https://cursor.com/blog/composer-2-5},
note = {Cursor blog. Cited in Section 2 of the framework's methodology paper.}
}
@article{zhao2026opsd,
title = {Self-{D}istilled {R}easoner: {O}n-{P}olicy {S}elf-{D}istillation for {L}arge {L}anguage {M}odels},
author = {Zhao, Siyan and Xie, Zhihui and Liu, Mengchen and Huang, Jing and Pang, Guan and Chen, Feiyu and Grover, Aditya},
year = {2026},
journal = {arXiv preprint arXiv:2601.18734},
url = {https://arxiv.org/abs/2601.18734},
note = {OPSD. MIT-licensed reference implementation at \url{https://github.com/siyan-zhao/OPSD}; the framework lifts \texttt{generalized\_jsd\_loss} from this codebase.}
}
@article{hubotter2026sdpo,
title = {Reinforcement {L}earning via {S}elf-{D}istillation},
author = {H{\"u}botter, Jonas and L{\"u}beck, Frederike and Behric, Lejs and Baumann, Anton and Bagatella, Marco and Marta, Daniel and Hakimi, Ido and Shenfeld, Idan and Buening, Thomas Kleine and Guestrin, Carlos and Krause, Andreas},
year = {2026},
journal = {arXiv preprint arXiv:2601.20802},
url = {https://arxiv.org/abs/2601.20802},
note = {SDPO. ICLR 2026 Scaling Post-training Workshop. Mathematically equivalent to Cursor's ``Targeted RL with Textual Feedback.''}
}
@article{moonshot2026kimi-k25,
title = {{K}imi {K}2.5},
author = {{Moonshot AI}},
year = {2026},
url = {https://huggingface.co/moonshotai/Kimi-K2-Thinking},
note = {Open-source 1T-total / 32B-active MoE base model used by Cursor for Composer 2 / 2.5.}
}
|