# Nova: Generative Language Models for Assembly Code with Hierarchical Attention and Contrastive Learning > Nova is a generative LLM for x86-64 assembly, trained from DeepSeek-Coder checkpoints on stripped, normalized gcc -O0 to -O3 assembly built from AnghaBench and The Stack. It adds hierarchical attention that builds per-instruction summaries, and contrastive objectives that keep a function's source and O0–O3 assembly close while ordering them by optimization level. Fine-tuned for decompilation, Nova-1.3B and Nova-6.7B reach 25.17% and 34.36% average Pass@1 on HumanEval-Decompile, 14.84 and 12.41 points above same-size LLM4Decompile; separately fine-tuned models report the highest average Recall@1 for O0-to-O3 binary code similarity on six real-world projects. - Authors: Nan Jiang; Chengxiao Wang; Kevin Liu; Xiangzhe Xu; Lin Tan; Xiangyu Zhang; Petr Babkin - Venue: ICLR 2025 - Status: Conference paper - Publication / posting date: 2025 (exact date not verified) - Date source: Publication year stated by the official ICLR proceedings; no finer publication date is asserted. The proceedings page's citation date of 2025-05-01 is not adopted, and the arXiv dates are not substituted. - First arXiv submission: November 22, 2023 - arXiv: 2311.13721 - Reviewed arXiv version: v7 - Sources reviewed: 2026-10-18 - Added to the catalog: 2026-10-19 - Original paper: - PDF: - Code: - BibTeX: - Catalog page: - Record JSON: - Chinese version: [简体中文](https://awesomerea.org/zh/papers/paper-nova/index.md) ## Scope and limitations Decompilation is evaluated function by function on the 164 HumanEval-Decompile C problems, each compiled for x86-64 at O0–O3, by running 20 samples per function against the problems' test cases; Pass@k on these tests is neither whole-program nor equivalence evidence. Baselines other than LLM4Decompile and Nova were prompted with three-shot examples rather than fine-tuned. Similarity results cover O0 queries against O3 candidates in pools of 50–500 functions from Binutils, Curl, ImageMagick, SQLite, OpenSSL and PuTTY, after fine-tuning on BinaryCorp-3M. Decompilation fine-tuning used a 2.16M-pair sample, the README says only 50K files came from The Stack, and the paper limits Nova to x86-64 assembly and 1.3B and 6.7B models. Inputs require the paper's assembly normalization. ## Why it fits this collection Adds an assembly-specialized foundation model evaluated on both decompilation and binary code similarity, complementing the decompilation-specific LLM papers in this section. ## Publication and review notes Authors follow the proceedings, which list pages 95905–95926. arXiv has seven versions: v1 of 2023-11-22 was titled 'Nova+: Generative Language Models for Binaries' (written Nova$^+$ in the arXiv metadata) and listed six authors, without Petr Babkin; the checked later versions (v3, v5, v7) use the final title, and v7 of 2025-11-11 states ICLR 2025 publication. The repository README's abstract gives a relative decompilation gain of up to 146.54%, while the paper reports absolute percentage points. The reviewed 2026-01-04 repository commit carries a BSD 3-Clause LICENSE, and the Hugging Face cards for nova-1.3b, nova-6.7b and their -bcr decompilation variants are tagged bsd-3-clause-clear; the models start from DeepSeek-Coder checkpoints. requirements.txt pins torch 2.3.0, transformers 4.40.2, vllm 0.5.0 and deepspeed 0.14.4 for a Python 3.10 environment, and training examples use four GPUs. ## Primary sources 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. ## About this entry Reviewed against the paper and its primary sources. Cite the original authors using the linked paper or BibTeX. Original editorial content: MIT. Linked projects, papers and assets retain their own rights and licenses.