Pith. sign in

REVIEW 4 major objections 6 minor 2 cited by

GRACE's claim: repository-level code completion improves when the whole repository is modeled as a layered code graph, retrieved by structure and text, and fused with the incomplete code before the LLM sees it — beating the strongest graph

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · deepseek-v4-flash

2026-08-05 04:41 UTC pith:7IF3KHKQ

load-bearing objection A plausible graph-based RAG system for repository-level code completion, but the current draft doesn't specify its central learned structural components, so the headline gains can't be assessed as written. the 4 major comments →

arxiv 2509.05980 v1 pith:7IF3KHKQ submitted 2025-09-07 cs.SE

GRACE: Graph-Guided Repository-Aware Code Completion through Hierarchical Code Fusion

Xingliang Wang , Baoyi Wang , Chen Zhi , Junxiao Han , Xinkui Zhao , Jianwei Yin , Shuiguang Deng This is my paper
classification cs.SE
keywords code completionrepository-level code completionretrieval-augmented generationcode graphgraph neural networkrerankinggraph fusionlarge language models
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

GRACE claims that repository-level code completion is limited not mainly by model capability but by what RAG systems feed the model: text-only retrieval that ignores how code relates, and naïve concatenation that flattens those relations away. The fix the paper proposes is to model the whole repository as a layered, multi-relation graph — file structure and cross-file dependencies, call and inheritance graphs, ASTs, control and data flow — and keep that structure visible at every stage: retrieval scores combine textual and topological similarity, a graph-aware reranker reorders candidates, and a fusion stage attaches retrieved subgraphs to the incomplete code by cross-attention before serialization into the prompt. If the paper is right, structural context is a lever for code completion that operates alongside parameter scaling and becomes more valuable as models grow. The evidence is first place in 35 of 48 metric-language-dataset cells across two benchmarks and three backbone LLMs, and, with DeepSeek-V3, average gains of +8.19% exact match and +7.51% edit similarity over the strongest graph-based RAG baseline.

Core claim

The paper's discovery, on its own terms: a repository should be treated as a graph database, not a text archive, and every pipeline stage should preserve that graph. GRACE builds a unified graph spanning repository, module, and function levels — folder structure, cross-file dependencies, call and inheritance graphs, ASTs, control/data flow — joined by cross-level edges. Retrieval runs semantic and structural paths in parallel, merges and reranks candidates, then fuses the top-k subgraphs with the query graph by cross-attention. The empirical claim: with DeepSeek-V3, GRACE beats the strongest graph-based RAG baseline by 8.19% EM and 7.51% ES and leads 35 of 48 metric-language-dataset cells.

What carries the argument

The central object is the multi-level, multi-semantic code graph — a heterogeneous graph G = (V, E, T_v, T_e) with repository-level (folders, files), module-level (classes, functions, types), and function-level (AST, control-flow, data-flow) entities unified by cross-level edges such as file→function definition links and function→AST roots. Two mechanisms carry the argument. The hybrid retriever's dual-path encoding represents each node as h_v = [v_c; v_s], concatenating a semantic code embedding with a Laplacian positional encoding from eigenvectors of the normalized Laplacian, making structural similarity searchable. The fusion stage computes cross-attention A = softmax(H_q H_rᵀ / √d) betw

Load-bearing premise

The structural-retrieval and reranking gains presuppose that the graph neural components actually perform learned structural matching, but the paper never specifies how they are trained — no loss, no fitting procedure, no graph-level readout — so if those parameters are untrained or arbitrary, the structure-based improvement has no demonstrated mechanism.

What would settle it

Randomize (or freeze) the parameters of the structural encoder and graph-aware reranker while keeping the semantic retrieval path intact, and re-run the CrossCodeEval Python experiments: if EM and F1 do not fall by the ablation-attributed margins, the learned structural components are not what produces the reported gains. A second check: re-run GRACE and baselines on the original, unmodified RepoEval splits; if the margin over baselines shrinks sharply, part of the headline result depends on the unreleased RepoEval-Updated refresh.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

Share X Bluesky LinkedIn Reddit HN

If this is right

  • Replacing the hybrid retriever with BM25 costs 4–6 F1 points, so structural retrieval carries signal that text similarity alone misses.
  • Removing the fusion stage is the single largest drop (up to –6.2 F1 and –5.4 EM), so wiring retrieved subgraphs into the query graph matters more than retrieval quality alone.
  • The gain widens with backbone size: from Qwen2.5-Coder-14B to DeepSeek-V3 the margin over the best non-graph competitor grows from 2.3 to 5.4 EM points on Python/CrossCodeEval, and scaling Qwen from 0.6B to 32B raises F1 by 56.3%.
  • Retrieval depth saturates at k = 3; adding more retrieved subgraphs beyond that mostly adds redundant edges rather than new cues.
  • AST-only graphs land 2–3% behind the full system, showing call, data-flow, and control-flow edges each contribute complementary structure.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Because the paper gives no training procedure for the structural encoder or reranker, a replication that trains them explicitly, or replaces them with untrained versions, would isolate whether the reported structural gains come from learned matching or from the graph representation itself.
  • The gain widening with backbone scale suggests a testable substitution: whether GRACE at 7B parameters matches vanilla RAG at 32B, which would make structure a cheaper alternative to scale in constrained deployments.
  • The graph construction is language-agnostic in principle (AST, call, inheritance, and data-flow edges exist in most languages), so porting GRACE to C, Go, or Rust is a direct test of whether the mechanism is structural rather than benchmark-specific.
  • Since the fused graph is serialized before the LLM sees it, the serialization format is itself a control variable: re-ordering or re-labeling the same edges should measurably change completion accuracy, quantifying how much of the gain is presentation versus content.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

4 major / 6 minor

Summary. The paper proposes GRACE, a repository-level code completion framework that constructs a multi-level, multi-semantic code graph, retrieves relevant context through a hybrid graph retriever combining semantic and structural similarity, reranks candidates, and fuses retrieved subgraphs with the query graph before serializing into an LLM prompt. The central claim is that GRACE significantly outperforms state-of-the-art RAG and graph-RAG baselines on RepoEval-Updated and CrossCodeEval across three backbone LLMs, with headline gains of +8.19% EM and +7.51% ES over RepoFuse on DeepSeek-V3 and top placement in 35/48 metric–language–dataset combinations.

Significance. If the reported results hold, the contribution is potentially valuable: the multi-level graph construction and graph-to-text serialization are clearly motivated, the experimental design includes external baselines and multiple LLMs, the code is released, and the complexity analysis is a useful addition. The evaluation is also not circular: alpha is tuned on validation data and comparisons are against independent methods. However, the central mechanism—the structural encoder and the graph-aware reranker—is under-specified to the point of being unreproducible as written, and the main results tables contain internal inconsistencies. These issues currently prevent the paper from supporting its claims.

major comments (4)
  1. [§2.2.1–2.2.3, Algorithm 1 (Eqs. (2)–(4))] The structural retrieval path and reranker are not defined at the level required to support the central claim. Eq. (2) defines node-level h_v, but StructSim in §2.2.2 uses graph embeddings h_Gq/h_Gc without any graph-level readout or pooling. The GNN encoders in Eqs. (3)–(4) appear only in §2.3.2 and are not described (layers, message passing, training objective, loss, optimizer, or supervision data). The parameters W_alpha and b_alpha in §2.2.3 are called learnable but no training procedure is given. The abstract additionally promises a 'graph attention network-based re-ranker' that never appears in §2.2.3 or Algorithm 1. Without these specifications, the structural retrieval mechanism and the RQ2 ablation (82.76→78.97 under BM25) cannot be interpreted: the drop could be due to the semantic embedding difference alone. The released code does not cure an under-specified method description
  2. [Table 3 and §3.3 (RQ2)] Table 3's caption states that metrics are macro-averaged over Code and Identifier tasks, but every value equals the Code-only column of Table 2 (e.g., CrossCodeEval-Python F1 82.76, EM 31.38; Java F1 83.38, EM 26.72; RepoEval-Python F1 79.96, EM 29.52; RepoEval-Java F1 79.93, EM 23.85). Macro-averaging the Python F1 with the Identifier F1 of 74.63 would give 78.70, not 82.76. Also, RQ2 states that replacing HGR with BM25 cuts F1 by 4–6% and cites 82.76→78.97, which is a 3.79-point drop. The ablation analysis needs to be re-reported with the correct aggregation and consistent arithmetic.
  3. [Table 2] Several cells appear duplicated across supposedly different conditions. For example, the RepoEval-Updated Java GRACE row contains the Qwen block '23.2 59.85 80.96 78.31 38.47 62.83 68.69 70.08', which is nearly identical to the Python GRACE row's Qwen block '29.20 57.45 80.96 78.31 38.47 62.83 68.69 70.08', including identical Recall/F1 values across different languages. Such anomalies make it impossible to verify the '35/48' claim or the headline 8.19% EM / 7.51% ES averages. Please supply a clean, machine-readable table and state the aggregation formula used for the headline numbers.
  4. [§3.1 (RepoEval-Updated)] The paper evaluates on a modified benchmark, RepoEval-Updated, but does not release it. The described modification (removing repositories created before 2022-03-31 and adding new Python/Java projects up to 2023-01-01) is not accompanied by a repository list, split definitions, or a version hash. Since this modified dataset is one of only two benchmarks supporting the central comparison, and no results on the original RepoEval are reported, readers cannot reproduce the evaluation or audit the claim that leakage was mitigated. The dataset should be released, or the evaluation should also be run on the original public RepoEval.
minor comments (6)
  1. [Throughout] Typos and inconsistent naming: 'Vallina RAG' in Table 2, 'Hybird Graph Retriver' in Figure 2, and inconsistent use of 'reranker' vs 're-ranker'.
  2. [Figures 3 and 4] Both figures appear as garbled unicode/font artifacts in the manuscript text, with no readable axes or curves. They should be regenerated. The text claims 'six Qwen variants' in Figure 3 but the figure is unreadable.
  3. [§3.1] The adaptive fusion weight alpha is said to be tuned on a held-out validation split, but no split sizes or validation details are given. The relationship between this tuned alpha and the 'learnable' W_alpha/b_alpha is also unclear.
  4. [§3.2] The word 'significantly' is used repeatedly, but no confidence intervals, standard deviations, or significance tests are reported. Please either add statistical analysis or temper the wording.
  5. [Abstract/§3.2] The headline gains of +8.19% EM and +7.51% ES are reported only as averages. Please include per-dataset or per-language breakdowns so the reader can verify the claim.
  6. [Front matter] The CCS Concepts block still contains the placeholder 'Do Not Use This Code' and must be filled in before publication.

Circularity Check

0 steps flagged

No circularity: GRACE's reported gains are empirical comparisons against external baselines; the paper's omissions are reproducibility gaps, not definitional forcing.

full rationale

I find no circular step in the paper. GRACE's central claim is an empirical performance comparison against external baselines (RepoFuse, RLCoder, GraphCoder, Vanilla RAG, No RAG) on public benchmarks (CrossCodeEval and RepoEval-Updated), and the LLM completions are generated by fixed backbone models from constructed prompts; they are not computed from any fitted parameter of the proposed system. The only learnable quantities mentioned, such as the adaptive fusion weight alpha and W_alpha/b_alpha, are either tuned on a held-out validation split or described without a training objective, so they cannot force test-set results by construction. The lack of a specified GNN training procedure or graph-level readout (Section 2.2.2) is a serious reproducibility and mechanism gap, and the author-modified RepoEval-Updated benchmark is not released, but these are external-validity and completeness concerns rather than circularity: no equation or definition in the paper makes the predicted output equal to an input of the method. The paper's own Section 4.1 acknowledges these threats candidly. There are no load-bearing self-citations, no imported uniqueness theorems, and no fitted quantity renamed as a prediction. Thus the correct circularity finding is no significant circularity (score 0).

Axiom & Free-Parameter Ledger

5 free parameters · 5 axioms · 0 invented entities

The central claim depends on domain assumptions about the usefulness of graph structure and on hyperparameters tuned on a validation split. No new physical or conceptual entities are introduced beyond the fused graph data structure, which is not a postulated entity with independent falsifiable handles.

free parameters (5)
  • alpha (adaptive fusion weight) = not reported; tuned on validation split
    Balances semantic vs structural similarity in reranking; defined with learnable W_alpha and b_alpha but no training procedure is given.
  • theta (cross-edge attention threshold) = 0.4
    Controls whether cross-graph edges are added during fusion; fixed by hand.
  • k (retrieval depth) = 3
    Selected after sensitivity analysis in RQ4; the default value.
  • lambda (MMR diversity weight) = not reported
    Controls the diversity-relevance tradeoff in reranking; the value is not specified.
  • HNSW ef_search = 256
    Semantic ANN search parameter used for retrieval.
axioms (5)
  • domain assumption Code graph structure (calls, inheritance, dataflow) carries information useful for code completion beyond textual similarity.
    Foundational premise of GRACE; used to justify graph retrieval and fusion (Sections 1, 2.2, 2.3).
  • domain assumption codet5p-110m-embedding produces semantically meaningful code vectors without task-specific fine-tuning.
    Semantic encoding path relies entirely on this pre-trained model (Eq. 1, Section 2.2.1).
  • domain assumption Laplacian positional encoding plus node text embeddings yields a useful subgraph similarity measure.
    Structural retrieval path; no training or validation of this measure is described (Sections 2.2.1 and 2.2.2).
  • domain assumption Cross-attention between query and retrieved nodes identifies code entities that should be structurally linked.
    Underpins the fusion mechanism (Eq. 6, Section 2.3.2); the threshold theta is fixed at 0.4.
  • domain assumption The AST of the incomplete snippet is a sufficient query graph for retrieving useful context.
    Query graph is built solely from the AST (Algorithm 1, Phase 1).

pith-pipeline@v1.3.0-alltime-deepseek · 19896 in / 15883 out tokens · 154221 ms · 2026-08-05T04:41:41.256762+00:00 · methodology

0 comments
Cite this review

Pith. "Pith review of GRACE: Graph-Guided Repository-Aware Code Completion through Hierarchical Code Fusion." pith.science (2026). https://pith.science/paper/7IF3KHKQ

@misc{pith2026250905980,
  author       = {Pith},
  title        = {Pith review of: GRACE: Graph-Guided Repository-Aware Code Completion through Hierarchical Code Fusion},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/7IF3KHKQ}},
  note         = {Machine review of arXiv:2509.05980}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

LLMs excel in localized code completion but struggle with repository-level tasks due to limited context windows and complex semantic and structural dependencies across codebases. While Retrieval-Augmented Generation (RAG) mitigates context scarcity by retrieving relevant code snippets, current approaches face significant limitations. They overly rely on textual similarity for retrieval, neglecting structural relationships such as call chains and inheritance hierarchies, and lose critical structural information by naively concatenating retrieved snippets into text sequences for LLM input. To address these shortcomings, GRACE constructs a multi-level, multi-semantic code graph that unifies file structures, abstract syntax trees, function call graphs, class hierarchies, and data flow graphs to capture both static and dynamic code semantics. For retrieval, GRACE employs a Hybrid Graph Retriever that integrates graph neural network-based structural similarity with textual retrieval, refined by a graph attention network-based re-ranker to prioritize topologically relevant subgraphs. To enhance context, GRACE introduces a structural fusion mechanism that merges retrieved subgraphs with the local code context and preserves essential dependencies like function calls and inheritance. Extensive experiments on public repository-level benchmarks demonstrate that GRACE significantly outperforms state-of-the-art methods across all metrics. Using DeepSeek-V3 as the backbone LLM, GRACE surpasses the strongest graph-based RAG baselines by 8.19% EM and 7.51% ES points on every dataset. The code is available at https://anonymous.4open.science/r/grace_icse-C3D5.

Figures

Figures reproduced from arXiv: 2509.05980 by Baoyi Wang, Chen Zhi, Jianwei Yin, Junxiao Han, Shuiguang Deng, Xingliang Wang, Xinkui Zhao.

Figure 1
Figure 1. Figure 1: Comparison of standard RAG-based code completion with GRACE. Conventional RAG retrieves code solely by semantic [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Framework of GRACE, which consists of three main components: (a) multi-level code graph construction, (b) hybrid [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Impact of Qwen backbone size on the performance [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Effect of retrieval depth𝑘 on the macro performance of GRACE. mainly from Python and Java projects. Results may not generalise to other languages (e.g., C/C++, Go, Rust) or industrial monore￾pos with proprietary frameworks. Extending GRACE to additional ecosystems forms part of future work. (2) Repository scale. Although the chosen corpora contain hundreds of files, very large codebases (millions of LOC) c… view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 2 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Better Call Grep: Evaluating and Improving Grep-Like Lexical Retrieval for Repository-Level Code Completion

    cs.SE 2026-01 conditional novelty 6.0

    LLM-generated ripgrep queries plus BM25 re-ranking and line-interval de-duplication outperform graph- and RL-based retrievers for repository-level code completion on CrossCodeEval and RepoEval-Updated.

  2. Formal Architecture Descriptors as Navigation Primitives for AI Coding Agents

    cs.SE 2026-04 unverdicted novelty 5.0

    Formal architecture descriptors reduce AI coding agent navigation steps by 33-44% and behavioral variance by 52% in controlled and observational studies.

Reference graph

Works this paper leans on

60 extracted references · 20 canonical work pages · cited by 2 Pith papers · 2 internal anchors

  1. [1]

    Josh Achiam, Steven Adler, Sandhini Agarwal, Lama Ahmad, Ilge Akkaya, Floren- cia Leoni Aleman, Diogo Almeida, Janko Altenschmidt, Sam Altman, Shyamal Anadkat, et al. 2023. Gpt-4 technical report. arXiv preprint arXiv:2303.08774 (2023)

    Pith/arXiv arXiv 2023

  2. [2]

    Lakshya A Agrawal, Aditya Kanade, Navin Goyal, Shuvendu K Lahiri, and Sri- ram K Rajamani. 2023. Guiding language models of code with global context using monitors. arXiv preprint arXiv:2306.10763 (2023)

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  3. [3]

    Loubna Ben Allal, Raymond Li, Denis Kocetkov, Chenghao Mou, Christopher Akiki, Carlos Munoz Ferrandis, Niklas Muennighoff, Mayank Mishra, Alex Gu, Manan Dey, et al. 2023. Santacoder: don’t reach for the stars! arXiv preprint arXiv:2301.03988 (2023)

    Pith/arXiv arXiv 2023

  4. [4]

    Anthropic. 2024. Claude 3.5 Sonnet. https://www.anthropic.com/news/claude-3- 5-sonnet. 2024

    work page 2024

  5. [5]

    Jacob Austin, Augustus Odena, Maxwell Nye, Maarten Bosma, Henryk Michalewski, David Dohan, Ellen Jiang, Carrie Cai, Michael Terry, Quoc Le, et al. 2021. Program synthesis with large language models. arXiv preprint arXiv:2108.07732 (2021)

    Pith/arXiv arXiv 2021

  6. [6]

    Ramakrishna Bairi, Atharv Sonwane, Aditya Kanade, Vageesh D C, Arun Iyer, Suresh Parthasarathy, Sriram Rajamani, Balasubramanyan Ashok, and Shashank Shet. 2024. Codeplan: Repository-level coding using llms and planning. Proceed- ings of the ACM on Software Engineering 1, FSE (2024), 675–698

    work page 2024

  7. [7]

    Mohammad Bavarian, Heewoo Jun, Nikolas Tezak, John Schulman, Christine McLeavey, Jerry Tworek, and Mark Chen. 2022. Efficient training of language models to fill in the middle. arXiv preprint arXiv:2207.14255 (2022)

    Pith/arXiv arXiv 2022

  8. [8]

    Mark Chen, Jerry Tworek, Heewoo Jun, Qiming Yuan, Henrique Ponde De Oliveira Pinto, Jared Kaplan, Harri Edwards, Yuri Burda, Nicholas Joseph, Greg Brockman, et al. 2021. Evaluating large language models trained on code. arXiv preprint arXiv:2107.03374 (2021)

    Pith/arXiv arXiv 2021

  9. [9]

    Wei Cheng, Yuhan Wu, and Wei Hu. 2024. Dataflow-guided retrieval augmen- tation for repository-level code completion. arXiv preprint arXiv:2405.19782 (2024)

    Pith/arXiv arXiv 2024

  10. [10]

    Aakanksha Chowdhery, Sharan Narang, Jacob Devlin, Maarten Bosma, Gaurav Mishra, Adam Roberts, Paul Barham, Hyung Won Chung, Charles Sutton, Se- bastian Gehrmann, et al. 2023. Palm: Scaling language modeling with pathways. Journal of Machine Learning Research 24, 240 (2023), 1–113

    work page 2023

  11. [11]

    Fenia Christopoulou, Gerasimos Lampouras, Milan Gritta, Guchun Zhang, Yin- peng Guo, Zhongqi Li, Qi Zhang, Meng Xiao, Bo Shen, Lin Li, et al. 2022. Pangu- coder: Program synthesis with function-level language modeling. arXiv preprint arXiv:2207.11280 (2022)

    Pith/arXiv arXiv 2022

  12. [12]

    Colin B Clement, Shuai Lu, Xiaoyu Liu, Michele Tufano, Dawn Drain, Nan Duan, Neel Sundaresan, and Alexey Svyatkovskiy. 2021. Long-range modeling of source code files with eWASH: Extended window access by syntax hierarchy. arXiv preprint arXiv:2109.08780 (2021)

    Pith/arXiv arXiv 2021

  13. [13]

    Yangruibo Ding, Zijian Wang, Wasi Ahmad, Hantian Ding, Ming Tan, Nihal Jain, Murali Krishna Ramanathan, Ramesh Nallapati, Parminder Bhatia, Dan Roth, et al. 2023. Crosscodeeval: A diverse and multilingual benchmark for cross-file code completion. Advances in Neural Information Processing Systems 36 (2023), 46701–46723

    2023

  14. [14]

    Yangruibo Ding, Zijian Wang, Wasi Uddin Ahmad, Murali Krishna Ramanathan, Ramesh Nallapati, Parminder Bhatia, Dan Roth, and Bing Xiang. 2022. Cocomic: Code completion by jointly modeling in-file and cross-file context. arXiv preprint arXiv:2212.10007 (2022)

    Pith/arXiv arXiv 2022

  15. [15]

    Aaron Grattafiori, Abhimanyu Dubey, Abhinav Jauhri, Abhinav Pandey, Abhishek Kadian, Ahmad Al-Dahle, Aiesha Letman, Akhil Mathur, Alan Schelten, Alex Vaughan, et al. 2024. The llama 3 herd of models. arXiv preprint arXiv:2407.21783 (2024)

    Pith/arXiv arXiv 2024

  16. [16]

    Daya Guo, Qihao Zhu, Dejian Yang, Zhenda Xie, Kai Dong, Wentao Zhang, Guanting Chen, Xiao Bi, Yu Wu, YK Li, et al. 2024. DeepSeek-Coder: When the Large Language Model Meets Programming–The Rise of Code Intelligence.arXiv preprint arXiv:2401.14196 (2024)

    Pith/arXiv arXiv 2024

  17. [17]

    Koray Kavukcuoglu. 2025. Gemini 2.5: Our most intelligent AI model. https://blog.google/technology/google-deepmind/gemini-model-thinking- updates-march-2025/. 2025

    work page 2025

  18. [18]

    Urvashi Khandelwal, Omer Levy, Dan Jurafsky, Luke Zettlemoyer, and Mike Lewis. 2019. Generalization through memorization: Nearest neighbor language models. arXiv preprint arXiv:1911.00172 (2019)

    Pith/arXiv arXiv 2019

  19. [19]

    Seohyun Kim, Jinman Zhao, Yuchi Tian, and Satish Chandra. 2021. Code pre- diction by feeding trees to transformers. In 2021 IEEE/ACM 43rd International Conference on Software Engineering (ICSE) . IEEE, 150–162

    work page 2021

  20. [20]

    Patrick Lewis, Ethan Perez, Aleksandra Piktus, Fabio Petroni, Vladimir Karpukhin, Naman Goyal, Heinrich Küttler, Mike Lewis, Wen-tau Yih, Tim Rocktäschel, et al. 2020. Retrieval-augmented generation for knowledge-intensive nlp tasks. Advances in neural information processing systems 33 (2020), 9459–9474

    2020

  21. [21]

    Jian Li, Yue Wang, Michael R Lyu, and Irwin King. 2017. Code completion with neural attention and pointer networks. arXiv preprint arXiv:1711.09573 (2017)

    Pith/arXiv arXiv 2017

  22. [22]

    Jia Li, Yunfei Zhao, Yongmin Li, Ge Li, and Zhi Jin. 2023. Acecoder: Utilizing existing code to enhance code generation. arXiv preprint arXiv:2303.17780 (2023)

    Pith/arXiv arXiv 2023

  23. [23]

    Raymond Li, Loubna Ben Allal, Yangtian Zi, Niklas Muennighoff, Denis Kocetkov, Chenghao Mou, Marc Marone, Christopher Akiki, Jia Li, Jenny Chim, et al. 2023. Starcoder: may the source be with you! arXiv preprint arXiv:2305.06161 (2023)

    Pith/arXiv arXiv 2023

  24. [24]

    Ming Liang, Xiaoheng Xie, Gehao Zhang, Xunjin Zheng, Peng Di, Hongwei Chen, Chengpeng Wang, Gang Fan, et al. 2024. Repofuse: Repository-level code completion with fused dual context. arXiv preprint arXiv:2402.14323 (2024)

    Pith/arXiv arXiv 2024

  25. [25]

    Dianshu Liao, Shidong Pan, Qing Huang, Xiaoxue Ren, Zhenchang Xing, Huan Jin, and Qinying Li. 2023. Context-aware code generation framework for code repositories: Local, global, and third-party library awareness. CoRR (2023)

    work page 2023

  26. [26]

    Aixin Liu, Bei Feng, Bing Xue, Bingxuan Wang, Bochao Wu, Chengda Lu, Cheng- gang Zhao, Chengqi Deng, Chenyu Zhang, Chong Ruan, et al. 2024. Deepseek-v3 technical report. arXiv preprint arXiv:2412.19437 (2024)

    Pith/arXiv arXiv 2024

  27. [27]

    Fang Liu, Ge Li, Yunfei Zhao, and Zhi Jin. 2020. Multi-task learning based pre- trained language model for code completion. In Proceedings of the 35th IEEE/ACM international conference on automated software engineering . 473–485

    work page 2020

  28. [28]

    Nelson F Liu, Kevin Lin, John Hewitt, Ashwin Paranjape, Michele Bevilacqua, Fabio Petroni, and Percy Liang. 2023. Lost in the middle: How language models use long contexts. arXiv preprint arXiv:2307.03172 (2023)

    Pith/arXiv arXiv 2023

  29. [29]

    Tianyang Liu, Canwen Xu, and Julian McAuley. 2023. Repobench: Benchmarking repository-level code auto-completion systems. arXiv preprint arXiv:2306.03091 (2023)

    Pith/arXiv arXiv 2023

  30. [30]

    Wei Liu, Ailun Yu, Daoguang Zan, Bo Shen, Wei Zhang, Haiyan Zhao, Zhi Jin, and Qianxiang Wang. 2024. Graphcoder: Enhancing repository-level code completion via code context graph-based retrieval and language model. arXiv preprint arXiv:2406.07003 (2024)

    Pith/arXiv arXiv 2024

  31. [31]

    Anton Lozhkov, Raymond Li, Loubna Ben Allal, Federico Cassano, Joel Lamy- Poirier, Nouamane Tazi, Ao Tang, Dmytro Pykhtar, Jiawei Liu, Yuxiang Wei, et al. 2024. Starcoder 2 and the stack v2: The next generation. arXiv preprint arXiv:2402.19173 (2024)

    Pith/arXiv arXiv 2024

  32. [32]

    Ziyang Luo, Can Xu, Pu Zhao, Qingfeng Sun, Xiubo Geng, Wenxiang Hu, Chongyang Tao, Jing Ma, Qingwei Lin, and Daxin Jiang. 2023. Wizardcoder: Empowering code large language models with evol-instruct. arXiv preprint arXiv:2306.08568 (2023)

    Pith/arXiv arXiv 2023

  33. [33]

    Erik Nijkamp, Hiroaki Hayashi, Caiming Xiong, Silvio Savarese, and Yingbo Zhou. 2023. Codegen2: Lessons for training llms on programming and natural languages. arXiv preprint arXiv:2305.02309 (2023)

    Pith/arXiv arXiv 2023

  34. [34]

    OpenAI. 2024. Hello GPT-4o. https://openai.com/index/hello-gpt-4o/. 2024

    work page 2024

  35. [35]

    Md Rizwan Parvez, Wasi Uddin Ahmad, Saikat Chakraborty, Baishakhi Ray, and Kai-Wei Chang. 2021. Retrieval augmented code generation and summarization. arXiv preprint arXiv:2108.11601 (2021)

    Pith/arXiv arXiv 2021

  36. [36]

    Boci Peng, Yun Zhu, Yongchao Liu, Xiaohe Bo, Haizhou Shi, Chuntao Hong, Yan Zhang, and Siliang Tang. 2024. Graph retrieval-augmented generation: A survey. arXiv preprint arXiv:2408.08921 (2024)

    Pith/arXiv arXiv 2024

  37. [37]

    Huy Nhat Phan, Hoang Nhat Phan, Tien N Nguyen, and Nghi DQ Bui. 2024. Repohyper: Better context retrieval is all you need for repository-level code completion. CoRR (2024)

    work page 2024

  38. [38]

    Qwen, :, An Yang, Baosong Yang, Beichen Zhang, Binyuan Hui, Bo Zheng, Bowen Yu, Chengyuan Li, Dayiheng Liu, Fei Huang, Haoran Wei, Huan Lin, Jian Yang, Jianhong Tu, Jianwei Zhang, Jianxin Yang, Jiaxi Yang, Jingren Zhou, Junyang Lin, Kai Dang, Keming Lu, Keqin Bao, Kexin Yang, Le Yu, Mei Li, Mingfeng Xue, Pei Zhang, Qin Zhu, Rui Men, Runji Lin, Tianhao Li,...

    Pith/arXiv arXiv 2025

  39. [39]

    Stephen Robertson, Hugo Zaragoza, et al . 2009. The probabilistic relevance framework: BM25 and beyond. Foundations and Trends® in Information Retrieval 3, 4 (2009), 333–389

    2009

  40. [40]

    Baptiste Roziere, Jonas Gehring, Fabian Gloeckle, Sten Sootla, Itai Gat, Xiao- qing Ellen Tan, Yossi Adi, Jingyu Liu, Romain Sauvestre, Tal Remez, et al. 2023. Code llama: Open foundation models for code. arXiv preprint arXiv:2308.12950 (2023)

    Pith/arXiv arXiv 2023

  41. [41]

    Freda Shi, Xinyun Chen, Kanishka Misra, Nathan Scales, David Dohan, Ed H Chi, Nathanael Schärli, and Denny Zhou. 2023. Large language models can be easily distracted by irrelevant context. In International Conference on Machine Learning . PMLR, 31210–31227

    work page 2023

  42. [42]

    Disha Shrivastava, Denis Kocetkov, Harm de Vries, Dzmitry Bahdanau, and Torsten Scholak. 2023. Repofusion: Training code models to understand your repository. arXiv preprint arXiv:2306.10998 (2023)

    Pith/arXiv arXiv 2023

  43. [43]

    Disha Shrivastava, Hugo Larochelle, and Daniel Tarlow. 2023. Repository-level prompt generation for large language models of code. In International Conference on Machine Learning. PMLR, 31693–31715

    work page 2023

  44. [44]

    Hanzhuo Tan, Qi Luo, Ling Jiang, Zizheng Zhan, Jing Li, Haotian Zhang, and Yuqun Zhang. 2024. Prompt-based code completion via multi-retrieval augmented generation. ACM Transactions on Software Engineering and Methodology (2024)

    work page 2024

  45. [45]

    Ze Tang, Jidong Ge, Shangqing Liu, Tingwei Zhu, Tongtong Xu, Liguo Huang, and Bin Luo. 2023. Domain adaptive code completion via language models and GRACE: Graph-Guided Repository-Aware Code Completion through Hierarchical Code Fusion Conference ICSE ’26, April 12-18, 2026, Rio de Janeiro, Brazil decoupled domain databases. In 2023 38th IEEE/ACM Internati...

    work page 2023

  46. [46]

    Hongyuan Tao, Ying Zhang, Zhenhao Tang, Hongen Peng, Xukun Zhu, Bingchang Liu, Yingguang Yang, Ziyin Zhang, Zhaogui Xu, Haipeng Zhang, et al

    work page

  47. [47]

    Yue Wang, Hung Le, Akhilesh Deepak Gotmare, Nghi DQ Bui, Junnan Li, and Steven CH Hoi. 2023. Codet5+: Open code large language models for code understanding and generation. arXiv preprint arXiv:2305.07922 (2023)

    Pith/arXiv arXiv 2023

  48. [48]

    Yanlin Wang, Yanli Wang, Daya Guo, Jiachi Chen, Ruikai Zhang, Yuchi Ma, and Zibin Zheng. 2024. Rlcoder: Reinforcement learning for repository-level code completion. arXiv preprint arXiv:2407.19487 (2024)

    Pith/arXiv arXiv 2024

  49. [49]

    Sushma Reddy Yadavalli, Lokesh Chandra Das, and Myounggyu Won. 2023. Rlpg: Reinforcement learning approach for dynamic intra-platoon gap adaptation for highway on-ramp merging. In 2023 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) . IEEE, 5514–5521

    work page 2023

  50. [50]

    Daoguang Zan, Bei Chen, Zeqi Lin, Bei Guan, Yongji Wang, and Jian-Guang Lou

    work page

  51. [51]

    Daoguang Zan, Bei Chen, Dejian Yang, Zeqi Lin, Minsu Kim, Bei Guan, Yongji Wang, Weizhu Chen, and Jian-Guang Lou. 2022. CERT: continual pre-training on sketches for library-oriented code generation. arXiv preprint arXiv:2206.06888 (2022)

    Pith/arXiv arXiv 2022

  52. [52]

    Daoguang Zan, Bei Chen, Fengji Zhang, Dianjie Lu, Bingchao Wu, Bei Guan, Yongji Wang, and Jian-Guang Lou. 2022. Large language models meet NL2Code: A survey. arXiv preprint arXiv:2212.09420 (2022)

    Pith/arXiv arXiv 2022

  53. [53]

    Fengji Zhang, Bei Chen, Yue Zhang, Jacky Keung, Jin Liu, Daoguang Zan, Yi Mao, Jian-Guang Lou, and Weizhu Chen. 2023. Repocoder: Repository-level code com- pletion through iterative retrieval and generation.arXiv preprint arXiv:2303.12570 (2023)

    Pith/arXiv arXiv 2023

  54. [54]

    Xiangyu Zhang, Yu Zhou, Guang Yang, and Taolue Chen. 2023. Syntax-aware retrieval augmented code generation. In Findings of the Association for Computa- tional Linguistics: EMNLP 2023 . 1291–1302

    work page 2023

  55. [55]

    Ziyin Zhang, Chaoyu Chen, Bingchang Liu, Cong Liao, Zi Gong, Hang Yu, Jianguo Li, and Rui Wang. 2023. Unifying the perspectives of nlp and software engineering: A survey on language models for code. arXiv preprint arXiv:2311.07989 (2023)

    Pith/arXiv arXiv 2023

  56. [56]

    Qinkai Zheng, Xiao Xia, Xu Zou, Yuxiao Dong, Shan Wang, Yufei Xue, Lei Shen, Zihan Wang, Andi Wang, Yang Li, et al. 2023. Codegeex: A pre-trained model for code generation with multilingual benchmarking on humaneval-x. InProceedings of the 29th ACM SIGKDD Conference on Knowledge Discovery and Data Mining . 5673–5684

    work page 2023

  57. [57]

    Zibin Zheng, Kaiwen Ning, Yanlin Wang, Jingwen Zhang, Dewu Zheng, Mingxi Ye, and Jiachi Chen. 2023. A survey of large language models for code: Evolution, benchmarking, and future trends. arXiv preprint arXiv:2311.10372 (2023)

    Pith/arXiv arXiv 2023

  58. [58]

    Qihao Zhu, Daya Guo, Zhihong Shao, Dejian Yang, Peiyi Wang, Runxin Xu, Y Wu, Yukun Li, Huazuo Gao, Shirong Ma, et al . 2024. Deepseek-coder-v2: Breaking the barrier of closed-source models in code intelligence. arXiv preprint arXiv:2406.11931 (2024)

    Pith/arXiv arXiv 2024

  59. [2022]

    When language model meets private library.arXiv preprint arXiv:2210.17236 (2022)

    work page internal anchor Pith review Pith/arXiv arXiv 2022

  60. [2025]

    arXiv preprint arXiv:2505.16901 (2025)

    Code Graph Model (CGM): A Graph-Integrated Large Language Model for Repository-Level Software Engineering Tasks. arXiv preprint arXiv:2505.16901 (2025)

    Pith/arXiv arXiv 2025