Research map / Large language models / Instruction and fine-tuning
On-the-Fly Adapting Code Summarization on Trainable Cost-Effective Language Models
Yufan Cai, Yun Lin, Chenyan Liu, Jinglian Wu, Yifan Zhang, Yiming Liu, Yeyun Gong, Jin Song Dong
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add AdaCom on-the-fly adaptation | codeT5-small · CodeKG | BLEU4: 34.89 -> 49.05 | Improved |
| add AdaCom on-the-fly adaptation | codeT5-large · CodeKG | BLEU4: 45.99 -> 49.87 | Improved |
| add AdaCom on-the-fly adaptation | seven code language models · CodeKG | BLEU4 average boost: 14.9% | Improved |
| use smaller model instead of larger model with AdaCom | AdaCom · CodeKG | relative BLEU4 boost: 40.6 -> 8.4 | Mixed |
| add AdaCom across datasets | codeT5-small · nine datasets | BLEU4 average boost: 8.3% | Improved |
| add AdaCom | codeT5-small · CSN-ruby | METEOR: 25.11 -> 25.05 | Hurt |
| add AdaCom | codeT5-small · CSN-python | BLEU4: 19.71 -> 19.92 | No clear effect |
| replace Retro with AdaCom | CodeT5-base · CSN overall | BLEU4: 19.19 -> 20.34 | Improved |
| replace CLS-cosine retrieval with AdaCom | CodeT5-base · CSN overall | BLEU4: 19.52 -> 20.34 | Improved |
| use retrieved helpful comment alone instead of adapting model (ablation) | AdaCom · CSN overall | BLEU4: 20.34 -> 12.73 | Hurt |
| disable adaptation (model only) (ablation) | AdaCom-base · CSN overall | BLEU4: 20.34 -> 18.73 | Hurt |
| replace CodeT5+ larger model with AdaCom-base (ablation) | CodeT5 · CSN overall | BLEU4: 20.15 -> 20.34 | Improved |
Closest papers by meaning
- AlchemistCoder: Harmonizing and Eliciting Code Capability by Hindsight Tuning on Multi-source Data (NeurIPS 2024 · similarity 0.86)
- NextCoder: Robust Adaptation of Code LMs to Diverse Code Edits (ICML 2025 · similarity 0.85)
- Hot or Cold? Adaptive Temperature Sampling for Code Generation with Large Language Models (AAAI 2024 · similarity 0.85)
- Unsupervised Evaluation of Code LLMs with Round-Trip Correctness (ICML 2024 · similarity 0.85)
- InverseCoder: Self-improving Instruction-Tuned Code LLMs with Inverse-Instruct (AAAI 2025 · similarity 0.85)
- ChatGPT-Generated Code Assignment Detection Using Perplexity of Large Language Models (Student Abstract) (AAAI 2024 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
