Delulu: A Verified Multi-Lingual Benchmark for Code Hallucination Detection in Fill-in-the-Middle Tasks

• URL: http://arxiv.org/abs/2605.07024v1
• Date: Thu, 07 May 2026 23:12:00 GMT
• Title: Delulu: A Verified Multi-Lingual Benchmark for Code Hallucination Detection in Fill-in-the-Middle Tasks
• Abstract summary: Large Language Models for code generation frequently produce hallucinations in Fillin-theMiddle (FIM) tasks.<n>We introduce Delulu, a verified multi-lingual benchmark of 1,951 FIM samples across 7 languages and 4 hallucination types.<n>We evaluate 11 openweight FIM models from five families spanning 0.5B-32B parameters.
• Score: 4.089259624354187
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Large Language Models for code generation frequently produce hallucinations in Fill-in-the-Middle (FIM) tasks -- plausible but incorrect completions such as invented API methods, invalid parameters, undefined variables, or non-existent imports. These failures pass superficial review yet introduce runtime errors. We introduce Delulu, a verified multi-lingual benchmark of 1,951 FIM samples across 7 languages and 4 hallucination types. Samples are curated through an adversarial pipeline: a frontier LLM generates plausible hallucinations, four diverse judge models evaluate them, embedding-based clustering mines progressively harder examples, self-contained Docker containers verify that golden completions compile while hallucinated variants produce the expected runtime error, and a final human-expert review removes any remaining biased or trivially decidable samples. We evaluate 11 open-weight FIM models from five families spanning 0.5B-32B parameters: a six-point Qwen2.5-Coder scaling slate, plus a cross-family slate (CodeLlama, DeepSeek-Coder-V2, StarCoder2). The strongest model reaches only 84.5% pass@1, no family exceeds 0.77 Edit Similarity, and every family produces hallucination-aligned completions on a non-trivial share of samples, confirming that the difficulty exposed by Delulu is task-intrinsic rather than family-specific. We release the benchmark, containers, and evaluation framework at https://github.com/microsoft/delulu.

Related papers

• The Librarian Who Refused to Code: Model-Dependent Identity Enactment in LLM Code Generation [0.0]
  We tested four prompt conditions (no persona, two engineer personas, and a research-librarian persona), 12 code-generation tasks, two frontier models, and five runs per cell.<n>Under pre-registered mixed-effects analysis, the condition-by-model interaction was significant for provider-reported output tokens.<n>On Claude Opus, the minimalist engineer persona reduced visible output by 30% without improving correctness, while the thorough engineer persona increased output without a correctness gain.
  arXiv  Detail & Related papers  (2026-07-19T21:47:58Z)
• Beyond Pass Rate: A Multilingual, Execution-Grounded Evaluation of Open Code LLMs [0.0]
  Best model reaches 23.64% mean correctness, compared with a 57.2% human acceptance baseline.<n>Qwen2.5-Coder-14B-Instruct is strongest on hard problems and distinct-problem coverage.<n>Gemma-2-27B-IT achieves the highest all-language lint pass rate.
  arXiv  Detail & Related papers  (2026-06-07T21:10:30Z)
• Synthetic Hallucinations, Real Gains: Hard Negatives from Frontier Models for FIM Hallucination Mitigation [4.089259624354187]
  Small open-source code models that power autocomplete still emit Fill-in-the-Middle (FIM) completions.<n>We propose an execution-free alternative: use frontier code models to synthesise plausible-but-wrong completions as hard negatives.
  arXiv  Detail & Related papers  (2026-06-02T04:15:50Z)
• Measuring Maximum Activations in Open Large Language Models [60.3514350516308]
  We measure global and layerwise maxima on 27 checkpoints from 8 open families spanning dense, MoE, vision-language, intermediate-training, and instruction-tuned variants.<n>We conclude that maximum activation magnitude is a model property tied to family, architecture, and training stage - not a simple byproduct of size.
  arXiv  Detail & Related papers  (2026-05-15T03:31:51Z)
• WildClawBench: A Benchmark for Real-World, Long-Horizon Agent Evaluation [88.10947115397971]
  This work presents WildClawBench, a native-runtime benchmark of 60 human-authored, bilingual, multimodal tasks spanning six thematic categories.<n>Each task averages roughly 8 minutes of wall-clock time and over 20 tool calls.<n> Grading is hybrid, combining deterministic rule-based checks, environment-state auditing of side effects, and an LLM/VLM judge for semantic verification.
  arXiv  Detail & Related papers  (2026-05-11T17:49:43Z)
• Correct Code, Vulnerable Dependencies: A Large Scale Measurement Study of LLM-Specified Library Versions [52.50730821321986]
  We present the first large-scale measurement study of version-level risk in large language models (LLMs)<n>We evaluate 10 LLMs on PinTrace, a curated benchmark of 1,000 Stack Overflow programming tasks.<n>Our findings reveal LLM version selection as a first-class, previously overlooked risk surface in LLM-based development.
  arXiv  Detail & Related papers  (2026-05-07T13:52:59Z)
• Self-reflection in Automated Qualitative Coding: Improving Text Annotation through Secondary LLM Critique [1.5749416770494706]
  Large language models (LLMs) allow for sophisticated qualitative coding of large datasets.<n>We present a simple, generalizable two-stage workflow: an LLM applies a human-designed codebook; a secondary LLM critic performs self-reflection on each positive label.<n>We evaluate this approach on six qualitative codes over 3,000 high-content emails from Apache Software Foundation project evaluation discussions.
  arXiv  Detail & Related papers  (2026-01-14T22:27:13Z)
• Fine-Grained Detection of Context-Grounded Hallucinations Using LLMs [16.173245551933178]
  Context-grounded hallucinations are cases where model outputs contain information not verifiable against the source text.<n>We study the applicability of LLMs for localizing such hallucinations, as a more practical alternative to existing complex evaluation pipelines.
  arXiv  Detail & Related papers  (2025-09-26T17:03:24Z)
• Eliminating Hallucination-Induced Errors in LLM Code Generation with Functional Clustering [0.0]
  We present functional clustering, a black-box wrapper that eliminates nearly all hallucination-induced errors while providing a tunable confidence score.<n>Our verifier preserves baseline pass@1 on solvable tasks yet slashes the error rate of returned answers from 65% to 2%.<n>Because the method requires only sampling and sandbox execution, it applies unchanged to closed-source APIs and future models.
  arXiv  Detail & Related papers  (2025-05-16T18:19:38Z)
• LongHalQA: Long-Context Hallucination Evaluation for MultiModal Large Language Models [96.64960606650115]
  LongHalQA is an LLM-free hallucination benchmark that comprises 6K long and complex hallucination text. LongHalQA is featured by GPT4V-generated hallucinatory data that are well aligned with real-world scenarios.
  arXiv  Detail & Related papers  (2024-10-13T18:59:58Z)
• DebugBench: Evaluating Debugging Capability of Large Language Models [80.73121177868357]
  DebugBench is a benchmark for Large Language Models (LLMs) It covers four major bug categories and 18 minor types in C++, Java, and Python. We evaluate two commercial and four open-source models in a zero-shot scenario.
  arXiv  Detail & Related papers  (2024-01-09T15:46:38Z)
• "Knowing When You Don't Know": A Multilingual Relevance Assessment Dataset for Robust Retrieval-Augmented Generation [90.09260023184932]
  Retrieval-Augmented Generation (RAG) grounds Large Language Model (LLM) output by leveraging external knowledge sources to reduce factual hallucinations. NoMIRACL is a human-annotated dataset for evaluating LLM robustness in RAG across 18 typologically diverse languages. We measure relevance assessment using: (i) hallucination rate, measuring model tendency to hallucinate, when the answer is not present in passages in the non-relevant subset, and (ii) error rate, measuring model inaccuracy to recognize relevant passages in the relevant subset.
  arXiv  Detail & Related papers  (2023-12-18T17:18:04Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.