Diffs vs. Whole Files: An Empirical Comparison of Iterative Edit-Based and Direct Generation for Flutter/Dart Code Models

• URL: http://arxiv.org/abs/2609.05779v1
• Date: Sat, 05 Sep 2026 00:13:49 GMT
• Title: Diffs vs. Whole Files: An Empirical Comparison of Iterative Edit-Based and Direct Generation for Flutter/Dart Code Models
• Abstract summary: We train two code models in two output regimes: direct generation and diff-based generation.<n>Direct generation substantially outperforms diff-based generation on every metric we measure.<n>We identify a single, architecture-independent mechanism behind the conditions where diff-based generation does win.
• Score: 0.0
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Large language models used for code editing can be trained and deployed in at least two output regimes: direct generation, where the model emits the entire modified file in one shot, and iterative diff-based generation ("steps"), where the model emits a sequence of localized search/replace edits applied one at a time until it signals completion or a step budget is exhausted. The diff-based regime is attractive because it mirrors how developers edit code and should require far fewer generated tokens per turn. We train two code models - a 100M-parameter model trained from scratch (Rainbow-Pony-100M) and a fine-tuned Qwen2.5-Coder-0.5B - in both regimes on a shared Flutter/Dart dataset, and evaluate all four resulting models on a held-out set of approx 1,790 tasks per model. Direct generation substantially outperforms diff-based generation on every metric we measure - compilation/static-analysis pass rate, bits-per-byte, character-level similarity to the reference, and blinded LLM-judge ratings of goal fulfillment, correctness, and code quality - and the gap persists after controlling for task difficulty via a matched-ID comparison and when restricting to code that compiles on both sides. We then identify a single, architecture-independent mechanism behind the conditions where diff-based generation does win: it is competitive on short, spatially localized edits, and its category-level wins concentrate in exactly the two task categories - refactoring and error-handling/edge-case fixes - with the lowest mean edit-step count in our dataset. We term this task locality and discuss its implications for when an edit-based training regime is and is not the right choice for a code-editing model.

Related papers

• Encoder-Decoder Manifold Alignment for Idempotent Generation [51.959208893243776]
  We argue that a key reason for this failure is a geometric mismatch between the manifold learned by the encoder and decoder.<n>We propose a new training framework that explicitly closes this gap by forcing the encoder and decoder to learn consistent representations of the same underlying data manifold.
  arXiv  Detail & Related papers  (2026-06-21T02:14:37Z)
• To Diff or Not to Diff? Structure-Aware and Adaptive Output Formats for Efficient LLM-based Code Editing [54.399335188667045]
  BlockDiff and FuncDiff are structure-aware diff formats that represent changes as block-level rewrites of syntactically coherent units.<n>AdaEdit is a general adaptive edit strategy that trains LLMs to dynamically choose the most token-efficient format between a given diff format and full code.
  arXiv  Detail & Related papers  (2026-04-30T01:14:13Z)
• Every Step Counts: Decoding Trajectories as Authorship Fingerprints of dLLMs [63.82840470917859]
  We show that the decoding mechanism of dLLMs can be used as a powerful tool for model attribution.<n>We propose a novel information extraction scheme called the Directed Decoding Map (DDM), which captures structural relationships between decoding steps and better reveals model-specific behaviors.
  arXiv  Detail & Related papers  (2025-10-02T06:25:10Z)
• Datarus-R1: An Adaptive Multi-Step Reasoning LLM for Automated Data Analysis [0.0]
  We present Datarus-R1-14B, a language model fine-tuned from Qwen 2.5-14B-Instruct to act as a virtual data analyst and graduate-level problem solver.<n>Datarus is trained not on isolated question-answer pairs but on full analytical trajectories including reasoning steps, code execution, error traces, self-corrections, and final conclusions.
  arXiv  Detail & Related papers  (2025-08-18T21:58:18Z)
• LLM-Based Detection of Tangled Code Changes for Higher-Quality Method-Level Bug Datasets [8.166584296080805]
  We investigate the utility of Large Language Models for detecting tangled code changes by leveraging both commit messages and method-level code diffs.<n>Our results demonstrate that combining commit messages with code diffs significantly enhances model performance.<n>Applying our approach to 49 open-source projects improves the distributional separability of code metrics between buggy and non-buggy methods.
  arXiv  Detail & Related papers  (2025-05-13T06:26:13Z)
• Robust Learning of Diverse Code Edits [10.565439872488328]
  Software engineering activities frequently involve edits to existing code.<n>Code language models (LMs) lack the ability to handle diverse types of code-edit requirements.<n>We propose a novel synthetic data generation pipeline and an adaptation algorithm.
  arXiv  Detail & Related papers  (2025-03-05T16:39:04Z)
• COrAL: Order-Agnostic Language Modeling for Efficient Iterative Refinement [80.18490952057125]
  Iterative refinement has emerged as an effective paradigm for enhancing the capabilities of large language models (LLMs) on complex tasks. We propose Context-Wise Order-Agnostic Language Modeling (COrAL) to overcome these challenges. Our approach models multiple token dependencies within manageable context windows, enabling the model to perform iterative refinement internally.
  arXiv  Detail & Related papers  (2024-10-12T23:56:19Z)
• ELDER: Enhancing Lifelong Model Editing with Mixture-of-LoRA [55.697627106315004]
  Large language models (LLMs) require model editing to efficiently update specific knowledge within them and avoid factual errors.<n>Previous approaches manage sequential edits by freezing original parameters and discretely allocating new parameters for each knowledge update.<n>We propose ELDER, a novel approach to create a continuous association between data and adapters.
  arXiv  Detail & Related papers  (2024-08-19T02:27:00Z)
• Single-Stage Visual Relationship Learning using Conditional Queries [60.90880759475021]
  TraCQ is a new formulation for scene graph generation that avoids the multi-task learning problem and the entity pair distribution. We employ a DETR-based encoder-decoder conditional queries to significantly reduce the entity label space as well. Experimental results show that TraCQ not only outperforms existing single-stage scene graph generation methods, it also beats many state-of-the-art two-stage methods on the Visual Genome dataset.
  arXiv  Detail & Related papers  (2023-06-09T06:02:01Z)

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.