AI reviews for all 22,977 AAAI-26 papers were preferred by authors and PC members over human reviews on accuracy and suggestions and outperformed baselines at spotting weaknesses.
super hub Canonical reference
DeepSeek-Coder: When the Large Language Model Meets Programming -- The Rise of Code Intelligence
Canonical reference. 74% of citing Pith papers cite this work as background.
191
Pith papers citing it
Background
74% of classified citations
abstract
The rapid development of large language models has revolutionized code intelligence in software development. However, the predominance of closed-source models has restricted extensive research and development. To address this, we introduce the DeepSeek-Coder series, a range of open-source code models with sizes from 1.3B to 33B, trained from scratch on 2 trillion tokens. These models are pre-trained on a high-quality project-level code corpus and employ a fill-in-the-blank task with a 16K window to enhance code generation and infilling. Our extensive evaluations demonstrate that DeepSeek-Coder not only achieves state-of-the-art performance among open-source code models across multiple benchmarks but also surpasses existing closed-source models like Codex and GPT-3.5. Furthermore, DeepSeek-Coder models are under a permissive license that allows for both research and unrestricted commercial use.
hub tools
citation-role summary
background 30
method 6
dataset 2
baseline 1
citation-polarity summary
claims ledger
- abstract The rapid development of large language models has revolutionized code intelligence in software development. However, the predominance of closed-source models has restricted extensive research and development. To address this, we introduce the DeepSeek-Coder series, a range of open-source code models with sizes from 1.3B to 33B, trained from scratch on 2 trillion tokens. These models are pre-trained on a high-quality project-level code corpus and employ a fill-in-the-blank task with a 16K window to enhance code generation and infilling. Our extensive evaluations demonstrate that DeepSeek-Coder
authors
co-cited works
representative citing papers
The first SoK on LLM-based AutoPT frameworks provides a six-dimension taxonomy of agent designs and a unified empirical benchmark evaluating 15 frameworks via over 10 billion tokens and 1,500 manually reviewed logs.
BOUND refines LLMs' package-validity boundary via targeted editing to cut package hallucination rates by 79.9% on edit prompts and 65.4% on unseen prompts in recommendation tasks while generalizing to code generation.
Preregistered placebo-controlled decomposition shows external executable counterexamples drive self-repair gains in small code models more than re-exposure or self-critique.
NL2Scratch supplies an executable benchmark of 311,648 NL-Scratch pairs and the SAC metric, showing LLMs with high lexical F1 often fail semantic alignment on actions, conditions, and numbers.
Mat-Pref benchmark shows GRPO after SFT lets Qwen3-8B reach 65-72% on compositional materials reasoning tasks, exceeding zero-shot 235B models on held-out structure families and cross-property transfer.
Incomplete constrainers in constrained decoding push LLMs into low-probability program regions, making unconstrained decoding outperform constrained decoding on functional correctness across seven models and three benchmarks.
CodeBlock partitions code responses into syntactically coherent blocks, scores them with generalized cross-entropy and data-flow signals, and applies sparse supervision to achieve higher pass@1 than full SFT using 1.9% of tokens on six benchmarks.
OpenRTLSet supplies 131k+ Verilog samples with AI-generated descriptions to enable fine-tuning of LLMs for hardware module design.
PrivCode++ introduces the first DP code generation method protecting both prompts and code via latent-conditioned two-stage training, claiming higher utility and stronger privacy than prior baselines.
Multilingual execution-grounded benchmark finds top open code LLM at 23.64% correctness versus 57.2% human baseline, with compile errors dominating 63% of failures.
SkelDPO improves code generation efficiency by 2-7% over prior DPO methods via joint preference losses on full code and efficiency-critical skeletons.
Using frontier models to synthesize plausible-but-wrong FIM completions as hard negatives for SFT improves Delulu exact match by +18.8 and edit similarity by +0.22 on Qwen2.5-Coder-7B while also lifting HumanEval-Infilling and SAFIM.
An LLM agent with Rocq backend automatically builds a verified RISC-V RV32I interpreter (1859 lines Rocq, 2848 lines extracted C++) that passes 265 tests and 12-hour fuzzing, while a Dafny backend fails.
CopT reverses CoT by eliciting a draft answer first then using continuous-embedding contrastive verification and on-policy thinking to reflect and correct, yielding up to 23% higher accuracy and 57% fewer tokens without training.
Constrained Diffusion for Code (CDC) integrates constraint satisfaction into the reverse denoising process of discrete diffusion models via constraint-aware operators that use optimization and program analysis to steer generation toward feasible programs.
Phoenix-bench shows agentic AI systems lose 37-58% resolved rate when moving from SWE-bench Verified to hardware tasks because bugs spread across parallel modules via signal flow, with testbench feedback lifting performance by 42-45% while file-level oracles add only 1.4%.
StepCodeReasoner aligns code reasoning with verifiable stepwise execution traces via print anchors and bi-level GRPO reinforcement learning, reaching SOTA results on CRUXEval (91.1%) and LiveCodeBench (86.5%) for a 7B model.
MeshFIM enables local low-poly mesh editing by autoregressively filling target regions conditioned on context, using boundary markers, positional embeddings, and a gated geometry encoder to enforce attachment, topology, and region limits.
POPO uses bounded importance sampling on positive rollouts and a siamese policy network to achieve implicit negative gradients and stable optimization, matching or exceeding GRPO on math benchmarks such as 36.67% on AIME 2025.
Fine-tuned 7B LLMs generating unified diffs for neural architecture refinement achieve 66-75% valid rates and 64-66% mean first-epoch accuracy, outperforming full-generation baselines by large margins while cutting output length by 75-85%.
PuzzleMark provides a robust and imperceptible watermarking method for code datasets using adaptive variable name concatenation and statistical verification, achieving perfect detection rates with minimal performance impact.
RepoDoc uses a repository knowledge graph with module clustering and semantic impact propagation to generate more complete documentation 3x faster with 85% fewer tokens and handle incremental updates 73% faster than prior LLM-based tools.
EXPO-SQL improves Text-to-SQL by using clause-level rewards derived from execution error messages and incremental clause execution instead of uniform query-level rewards.
citing papers explorer
-
Hackers or Hallucinators? A Comprehensive Analysis of LLM-Based Automated Penetration Testing
The first SoK on LLM-based AutoPT frameworks provides a six-dimension taxonomy of agent designs and a unified empirical benchmark evaluating 15 frameworks via over 10 billion tokens and 1,500 manually reviewed logs.
-
Mitigating Package Hallucinations in Large Language Models via Model Editing
BOUND refines LLMs' package-validity boundary via targeted editing to cut package hallucination rates by 79.9% on edit prompts and 65.4% on unseen prompts in recommendation tasks while generalizing to code generation.
-
Falsification, Not Exposure: An Internally Preregistered Placebo-Controlled Decomposition of Self-Repair Feedback in Frozen Small Code Models
Preregistered placebo-controlled decomposition shows external executable counterexamples drive self-repair gains in small code models more than re-exposure or self-critique.
-
NL2Scratch: An Executable Benchmark and Evaluation for Block-Based Programming
NL2Scratch supplies an executable benchmark of 311,648 NL-Scratch pairs and the SAC metric, showing LLMs with high lexical F1 often fail semantic alignment on actions, conditions, and numbers.
-
Mat-Pref: Verifiable-Reward Training Improves Compositional Reasoning in Inorganic Materials
Mat-Pref benchmark shows GRPO after SFT lets Qwen3-8B reach 65-72% on compositional materials reasoning tasks, exceeding zero-shot 235B models on held-out structure families and cross-property transfer.
-
The Alignment Problem in Constrained Code Generation
Incomplete constrainers in constrained decoding push LLMs into low-probability program regions, making unconstrained decoding outperform constrained decoding on functional correctness across seven models and three benchmarks.
-
CODEBLOCK: Learning to Supervise Code at the Right Granularity
CodeBlock partitions code responses into syntactically coherent blocks, scores them with generalized cross-entropy and data-flow signals, and applies sparse supervision to achieve higher pass@1 than full SFT using 1.9% of tokens on six benchmarks.
-
OpenRTLSet: A Fully Open-Source Dataset for Large Language Model-based Verilog Module Design
OpenRTLSet supplies 131k+ Verilog samples with AI-generated descriptions to enable fine-tuning of LLMs for hardware module design.
-
PrivCode++: Latent-Conditioned Differentially Private Code Generation for Comprehensive Guarantees
PrivCode++ introduces the first DP code generation method protecting both prompts and code via latent-conditioned two-stage training, claiming higher utility and stronger privacy than prior baselines.
-
Beyond Pass Rate: A Multilingual, Execution-Grounded Evaluation of Open Code LLMs
Multilingual execution-grounded benchmark finds top open code LLM at 23.64% correctness versus 57.2% human baseline, with compile errors dominating 63% of failures.
-
SkelDPO: A Skeleton-Guided Direct Preference Optimization Framework for Efficient Code Generation
SkelDPO improves code generation efficiency by 2-7% over prior DPO methods via joint preference losses on full code and efficiency-critical skeletons.
-
Synthetic Hallucinations, Real Gains: Hard Negatives from Frontier Models for FIM Hallucination Mitigation
Using frontier models to synthesize plausible-but-wrong FIM completions as hard negatives for SFT improves Delulu exact match by +18.8 and edit similarity by +0.22 on Qwen2.5-Coder-7B while also lifting HumanEval-Infilling and SAFIM.
-
CopT: Contrastive On-Policy Thinking with Continuous Spaces for General and Agentic Reasoning
CopT reverses CoT by eliciting a draft answer first then using continuous-embedding contrastive verification and on-policy thinking to reflect and correct, yielding up to 23% higher accuracy and 57% fewer tokens without training.
-
Constrained Code Generation with Discrete Diffusion
Constrained Diffusion for Code (CDC) integrates constraint satisfaction into the reverse denoising process of discrete diffusion models via constraint-aware operators that use optimization and program analysis to steer generation toward feasible programs.
-
Is Agentic AI Ready for Real-World Hardware Engineering? A Deep Dive with Phoenix-bench
Phoenix-bench shows agentic AI systems lose 37-58% resolved rate when moving from SWE-bench Verified to hardware tasks because bugs spread across parallel modules via signal flow, with testbench feedback lifting performance by 42-45% while file-level oracles add only 1.4%.
-
StepCodeReasoner: Aligning Code Reasoning with Stepwise Execution Traces via Reinforcement Learning
StepCodeReasoner aligns code reasoning with verifiable stepwise execution traces via print anchors and bi-level GRPO reinforcement learning, reaching SOTA results on CRUXEval (91.1%) and LiveCodeBench (86.5%) for a 7B model.
-
MeshFIM: Local Low-Poly Mesh Editing via Fill-in-the-Middle Autoregressive Generation
MeshFIM enables local low-poly mesh editing by autoregressively filling target regions conditioned on context, using boundary markers, positional embeddings, and a gated geometry encoder to enforce attachment, topology, and region limits.
-
Beyond Negative Rollouts: Positive-Only Policy Optimization with Implicit Negative Gradients
POPO uses bounded importance sampling on positive rollouts and a siamese policy network to achieve implicit negative gradients and stable optimization, matching or exceeding GRPO on math benchmarks such as 36.67% on AIME 2025.
-
Delta-Based Neural Architecture Search: LLM Fine-Tuning via Code Diffs
Fine-tuned 7B LLMs generating unified diffs for neural architecture refinement achieve 66-75% valid rates and 64-66% mean first-epoch accuracy, outperforming full-generation baselines by large margins while cutting output length by 75-85%.
-
PuzzleMark: Implicit Jigsaw Learning for Robust Code Dataset Watermarking in Neural Code Completion Models
PuzzleMark provides a robust and imperceptible watermarking method for code datasets using adaptive variable name concatenation and statistical verification, achieving perfect detection rates with minimal performance impact.
-
RepoDoc: A Knowledge Graph-Based Framework to Automatic Documentation Generation and Incremental Updates
RepoDoc uses a repository knowledge graph with module clustering and semantic impact propagation to generate more complete documentation 3x faster with 85% fewer tokens and handle incremental updates 73% faster than prior LLM-based tools.
-
EXPO-SQL: Execution-based Clause-level Policy Optimization for Text-to-SQL
EXPO-SQL improves Text-to-SQL by using clause-level rewards derived from execution error messages and incremental clause execution instead of uniform query-level rewards.
-
When Prompt Under-Specification Improves Code Correctness: An Exploratory Study of Prompt Wording and Structure Effects on LLM-Based Code Generation
Structurally rich task descriptions make LLMs robust to prompt under-specification, and under-specification can enhance code correctness by disrupting misleading lexical or structural cues.
-
Aligned Multi-View Scripts for Universal Chart-to-Code Generation
Introduces an aligned multi-language dataset and a language-conditioned low-rank adapter for generating executable plotting code in Python, R, and LaTeX from chart images.
-
Cascaded Code Editing: Large-Small Model Collaboration for Effective and Efficient Code Editing
A cascaded large-small model system generates edit sketches with the large model and applies them with the small model to make code editing both accurate and token-efficient.
-
IG-Search: Step-Level Information Gain Rewards for Search-Augmented Reasoning
IG-Search computes step-level information gain rewards from policy probabilities to improve credit assignment in RL training for search-augmented QA, yielding 1.6-point gains over trajectory-level baselines on multi-hop tasks.
-
Evaluating LLMs Code Reasoning Under Real-World Context
R2Eval is a new benchmark with 135 real-world code reasoning problems from Python projects that preserves complex data structures for more realistic LLM evaluation.
-
Structural Anchors and Reasoning Fragility:Understanding CoT Robustness in LLM4Code
CoT prompting in LLM4Code shows mixed robustness that depends on model family, task structure, and perturbations destabilizing structural anchors, leading to trajectory deformations like lengthening, branching, and simplification.
-
Think Anywhere in Code Generation
Think-Anywhere lets LLMs invoke on-demand reasoning at any token during code generation via cold-start imitation followed by outcome-based RL, reaching state-of-the-art results on LeetCode, LiveCodeBench, HumanEval, and MBPP.
-
Steerable Instruction Following Coding Data Synthesis with Actor-Parametric Schema Co-Evolution
IFCodeEvolve synthesizes coding data via actor-schema co-evolution with MCTS, boosting a 32B model's performance to match proprietary SOTA on instruction following.
-
RACC: Representation-Aware Coverage Criteria for LLM Safety Testing
RACC defines six representation-aware coverage criteria that score jailbreak test suites by measuring activation of safety concepts extracted from LLM hidden states on a calibration set.
-
CodeOCR: On the Effectiveness of Vision Language Models in Code Understanding
Multimodal LLMs process code as images to achieve up to 8x token compression, with visual cues like syntax highlighting aiding tasks and clone detection remaining resilient or even improving under compression.
-
Can LLMs Compress (and Decompress)? Evaluating Code Understanding and Execution via Invertibility
LLMs lack internal coherence for reliable bidirectional code reasoning, as they fail round-trip consistency on compression tasks even after training.
-
In Line with Context: Repository-Level Code Generation via Context Inlining
InlineCoder reframes repository-level code generation as function-level coding by using a draft anchor to inline the target function into its call graph for upstream usage and downstream dependency context.
-
MemSearcher: Training LLMs to Reason, Search and Manage Memory via End-to-End Reinforcement Learning
MemSearcher trains LLMs to manage compact memory in multi-turn searches via multi-context GRPO for end-to-end RL, outperforming ReAct-style baselines with stable token counts.
-
EyeMulator: Improving Code Language Models by Mimicking Human Visual Attention
EyeMulator augments CodeLLM fine-tuning loss with token weights derived from human eye-tracking scan paths, producing large gains on code translation and summarization across StarCoder, Llama-3.2 and DeepSeek-Coder.
-
VocabTailor: Dynamic Vocabulary Selection for Downstream Tasks in Small Language Models
VocabTailor introduces a decoupled dynamic vocabulary selection framework that reduces vocabulary-related memory in SLMs by up to 99% with minimal task performance loss.
-
Data Mixing Agent: Learning to Re-weight Domains for Continual Pre-training
An RL agent learns domain re-weighting policies from evaluation feedback to improve balanced performance in continual pre-training of LLMs across source and target domains.
-
SWE-agent: Agent-Computer Interfaces Enable Automated Software Engineering
SWE-agent introduces a custom agent-computer interface that lets LM agents solve software engineering tasks, reaching 12.5% pass@1 on SWE-bench and 87.7% on HumanEvalFix, exceeding prior non-interactive approaches.
-
Large Language Models for Multi-Lingual Equivalent Mutant Detection: An Extended Empirical Study
LLM-based methods achieve higher F1-scores than traditional approaches for equivalent mutant detection in Java and C, with fine-tuned code embeddings performing best and showing cross-lingual generalization.
-
Do Machines Struggle Where Humans Do? LLM and Human Comprehension of Obfuscated Code
Reasoning-tuned LLMs align with human comprehension failure patterns under code obfuscation using the Block Model, unlike instruction-tuned variants.
-
AI-Generated PowerShell Malware: An Experimental Framework and Dataset
An experimental framework and annotated dataset show LLM-generated PowerShell malware triggers OS events with median 84.5% Jaccard overlap to real malware and 48.4% complete matches.
-
Towards Knowledge Alignment in Code LLMs: Contrastive Unlearning for Evolving APIs
CURE applies contrastive unlearning to reduce deprecated API usage in code LLMs and improve correct replacements on a benchmark dataset while preserving general performance.
-
SrDetection: A Self-Referential Framework for Data Leakage Detection in Code Large Language Models
SrDetection detects data leakage in Code LLMs via contrast between original benchmark samples and their semantic variants, reporting F1 gains of 21.52 (gray-box) and 14.46 (black-box) over baselines in a controlled testbed.
-
Citation Discipline in Spec-Driven Development: A Cross-Model Empirical Study of Output Determinism and Automated Hallucination Detection in LLM-Generated Code
Mandatory per-line citations in SDD frameworks reduce LLM output determinism but enable reliable automated hallucination detection (TDR 86-88%, FPR 0%), a trade-off replicated across Claude and GLM models.
-
Breaking the Rounding Trap: Securing LLMs against Quantization-Conditioned Backdoors
QuantGuard is a pre-quantization method using differentiable rounding controls, error-guided reversal constraints, output consistency, and weight regularization on a small calibration set to suppress quantization-conditioned backdoors while preserving performance.
-
OASIF: An Efficient Obfuscation-Aware Self-Improving Framework for LLM-Based Assembly Code Instruction Following and Comprehension
OASIF improves open-source LLMs on obfuscated assembly comprehension by 5-17 percentage points on commercial VM obfuscators via a three-phase self-evolving training pipeline.
-
KernelSight-LM: A Kernel-Level LLM Inference Simulator
KernelSight-LM simulates LLM inference at kernel granularity with cross-generation (12.1% per-kernel error) and target-measured (3.8% error) tiers, yielding end-to-end median errors of 15.4%/12.8%/3.0% and 14.3%/6.2%/2.7% for TTFT/TPOT/throughput across six model families.
-
When AI Reviews Its Own Code: Recursive Self-Training Collapse in Code LLMs
Experiments across code LLMs show no-review collapses fastest, human-gated filters slow collapse, and AI self-gates lose effect over time, degenerating to ungated self-training under self-confirming acceptance as proven via gated distributional reweighting and spectral analysis.
-
To Run or Not to Run: Analyzing the Cost-Effectiveness of Code Execution in LLM-Based Program Repair
Empirical analysis of LLM repair agents shows execution provides concentrated benefits, with restrictions causing only a 1.25 pp non-significant drop in resolve rate while cutting token and time costs.