Research map / Large language models / Language, safety and interpretability
LLMs Unleashed: Generating Protocol Code from RFC Specifications
Junfeng Long, Jinshu Su, Biao Han
Read the paperCheck your idea against 137,217 papers
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 |
|---|---|---|---|
| provide full RFC in one input instead of chapter-wise or fixed-length chunking (ablation) | RFC component extraction · ICMP, TCP, QUIC functional component extraction | F1 | Improved |
| chapter-wise multi-turn chunking of RFC (ablation) | RFC component extraction · TCP, QUIC extraction | precision | Hurt |
| add summarized RFC context to prompt (ablation) | Qwen2.5 TCP generation · TCP code generation | protocol behavior: 0.6095 -> 0.8571 | Improved |
| add raw RFC excerpts to prompt (ablation) | Qwen-coder TCP generation · TCP code generation | protocol behavior: 0.9524 -> 0.5619 | Hurt |
| add summarized RFC context to prompt (ablation) | Qwen-coder TCP generation · TCP code generation | Composite Success Rate: 0.93 -> 0.70 | Hurt |
| add context to prompt for GPT-4o (ablation) | GPT-4o TCP generation · TCP code generation | protocol behavior: 0.6571 -> 0.7143 | Improved |
| inject diagram-derived handshake sequence into prompt (ablation) | TCP generation with summaries · TCP teardown flag coverage | flag coverage: 0.66 -> 0.86 (Qwen2.5) | Improved |
| use Qwen-Coder instead of general Qwen2.5 | ICMP generation · ICMP code generation | Composite Success Rate (%): 95.0 -> 60.0 | Hurt |
| use APG LLM pipeline instead of rule-based Sage | ICMP generation · ICMP generation time | minutes: 19 min -> 6 min | Improved |
| use APG instead of Sage for IGMP | IGMP generation · IGMP generation time | minutes: 2 min -> 3 min (Qwen2.5), 2 min -> 1 min (Qwen-coder) | Mixed |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- Sage — SAGE: A Dataflow-Native Framework for Modular, Controllable, and Transparent LLM-Augmented Reasoning (ICML 2026) · better on 3 of 3 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- DOMAINEVAL: An Auto-Constructed Benchmark for Multi-Domain Code Generation (AAAI 2025 · similarity 0.82)
- How efficient is LLM-generated code? A rigorous & high-standard benchmark (ICLR 2025 · similarity 0.81)
- TSNBench: Benchmarking LLM Proficiency in Time-Sensitive Networking (NeurIPS 2026 · similarity 0.81)
- IaC-Eval: A Code Generation Benchmark for Cloud Infrastructure-as-Code Programs (NeurIPS 2024 · similarity 0.81)
- Can LLM Replace Stack Overflow? A Study on Robustness and Reliability of Large Language Model Code Generation (AAAI 2024 · similarity 0.81)
- CodeChain: Towards Modular Code Generation Through Chain of Self-revisions with Representative Sub-modules (ICLR 2024 · similarity 0.81)
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.
