File size: 3,538 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 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 | # CITATION.cff — Citation File Format
# https://citation-file-format.github.io/
# Used by HF, GitHub, Zenodo to render a "Cite this repository" UI.
cff-version: 1.2.0
message: "If you use this framework or its derivative artifacts, please cite as below."
type: software
title: "Composer 2.5 Replication Framework: Methodology and Integration Architecture for Open Replication of Cursor's Agentic Coding Recipe"
abstract: >
An open-source methodology and integration architecture for replicating
Cursor's Composer 2.5 recipe on a HuggingFace base model, plus a novel
multi-teacher trace-replay distillation reward channel that complements
the published SDPO/OPSD method (which Cursor's "Targeted RL with Textual
Feedback" uses). Pre-experimental v0.0 release: methodology paper, audited
recipe mapping, integration architecture across TRL/VeRL/OpenEnv,
empirical economic-feasibility result for the novel channel ($0.98/trace),
and a working code skeleton with 38 passing unit tests.
authors:
- family-names: "Codeseys"
given-names: ""
affiliation: "Independent researcher"
# Replace with real ORCID if available:
# orcid: "https://orcid.org/0000-0000-0000-0000"
repository-code: "https://huggingface.co/Codeseys/composer-replication-framework"
url: "https://huggingface.co/Codeseys/composer-replication-framework"
date-released: "2026-05-25"
version: "0.0.0"
license: "MIT"
keywords:
- reinforcement-learning
- post-training
- distillation
- agentic-coding
- composer-2.5
- cursor
- kimi-k2
- grpo
- dapo
- sdpo
- opsd
- trl
- verl
- openenv
- llm
# Primary upstream works this framework depends on / cites
references:
- type: article
title: "Introducing Composer 2.5"
authors:
- name: "Cursor Team"
year: 2026
url: "https://cursor.com/blog/composer-2-5"
- type: article
title: "Self-Distilled Reasoner: On-Policy Self-Distillation for Large Language Models"
authors:
- family-names: "Zhao"
given-names: "Siyan"
- family-names: "Xie"
given-names: "Zhihui"
- family-names: "Liu"
given-names: "Mengchen"
- family-names: "Huang"
given-names: "Jing"
- family-names: "Pang"
given-names: "Guan"
- family-names: "Chen"
given-names: "Feiyu"
- family-names: "Grover"
given-names: "Aditya"
year: 2026
url: "https://arxiv.org/abs/2601.18734"
notes: "OPSD — single-LLM self-distillation; provides the reference loss implementation lifted by this framework."
- type: article
title: "Reinforcement Learning via Self-Distillation"
authors:
- family-names: "Hübotter"
given-names: "Jonas"
- family-names: "Lübeck"
given-names: "Frederike"
- family-names: "Behric"
given-names: "Lejs"
- family-names: "Baumann"
given-names: "Anton"
- family-names: "Bagatella"
given-names: "Marco"
- family-names: "Marta"
given-names: "Daniel"
- family-names: "Hakimi"
given-names: "Ido"
- family-names: "Shenfeld"
given-names: "Idan"
- family-names: "Buening"
given-names: "Thomas Kleine"
- family-names: "Guestrin"
given-names: "Carlos"
- family-names: "Krause"
given-names: "Andreas"
year: 2026
url: "https://arxiv.org/abs/2601.20802"
notes: "SDPO — formalizes the same mechanism as Cursor's Targeted RL with Textual Feedback. ICLR 2026 Scaling Post-training Workshop."
|