85b364bf09
Merge pull request #73 from BaiRiDreamer/main
...
Merge VERL RL training + BringUpBench evaluation pipeline
2026-02-12 11:02:03 +08:00
239cba2673
feat(sk2decompile): add BringUpBench evaluation pipeline and results
...
Integrate BringUpBench evaluation into sk2decompile/evaluation/bringupbench/,
corresponding to Section A.6 of the paper (arXiv:2509.22114).
BringUpBench is a benchmark suite of 90 self-contained C programs (505 functions,
O0-O3). SK2Decompile achieves 42.3% compilation rate and 27.0% re-executability
rate, compared to IDA Pro's 23.6% / 21.7%.
Contents:
- scripts/: 5-step reproduction pipeline (compile, decompile, map, infer, eval)
- data/func_maps/: pre-built function-level mappings (source <-> pseudo <-> asm)
- data/infer_results/: SK2Decompile inference outputs for all opt levels
- reports/: per-opt-level evaluation result summaries (Markdown)
- config.env: template environment configuration
- README.md: comprehensive documentation with reproduction guide
Also updated sk2decompile/README.md to reference BringUpBench evaluation.
2026-02-12 00:02:25 +08:00
e33b3e7829
feat(sk2decompile): integrate VERL RL training pipeline into sk2decompile/verl/
...
Add the complete RL (Reinforcement Learning) training content for SK2Decompile,
based on the VERL framework v0.4.1 with GRPO algorithm.
New files added:
- verl/SK2DECOMPILE/README.md: Detailed RL documentation including paper reward
formulations (Eq.3 & Eq.4), step-by-step reproduction guide (VERL installation,
reward function integration, data preparation, training launch), paper
configurations table, hyperparameters, and troubleshooting
- verl/SK2DECOMPILE/reward_functions/: 4 reference reward implementations
- exe_type.py: compilability + placeholder Jaccard (Structure Recovery, Eq.3)
- sim_exe.py: compilability + word-level similarity (Structure Recovery)
- embedding_gte.py: tree-sitter identifier extraction + GTE embedding cosine
similarity (Identifier Naming, Eq.4)
- embedding_qwen3.py: same as GTE variant but with Qwen3-Embedding-0.6B
- verl/SK2DECOMPILE/scripts/: 2 reference training scripts
- run_struct_rl.sh: Structure Recovery RL training
- run_ident_rl.sh: Identifier Naming RL training
Updated files:
- sk2decompile/README.md: Updated RL section to reference VERL v0.4.1, GRPO,
paper Section 3.5, correct script paths, and link to detailed RL README
All reward functions and scripts are marked as reference implementations,
directing readers to the paper (arXiv:2509.22114) for precise formulations.
2026-02-11 21:47:19 +08:00
1c164e21cd
fix model name error
2026-01-25 17:38:20 +08:00
410e3d1f09
detailed usage instructions for sk2decompile models
...
Added detailed instructions for using sk2decompile, including installation steps, data preparation, and inference commands.
2026-01-25 17:32:29 +08:00
2169186d5b
Update citation paper
2025-12-05 16:23:46 +08:00
f5a6f43eae
Refactor section headings in README.md
...
Updated section headings in README.md for consistency and clarity.
2025-12-05 16:20:11 +08:00
061b8e87cd
Update README.md
2025-12-05 16:17:52 +08:00
86857f4f87
Update README with installation and usage instructions
...
Added installation instructions for vllm and clang-format.
2025-12-05 16:15:53 +08:00
71b120a58e
inference
2025-10-16 23:32:08 +08:00
b82f0e511e
Add files via upload
...
inference
2025-10-16 23:29:55 +08:00
cff0c7c3fc
Update README.md
2025-10-15 20:52:04 +08:00
357705de0d
sk2decompile
2025-10-08 18:15:57 +08:00
2098a01c8f
Update release date for SK²Decompile in README
2025-10-04 22:22:51 +08:00
766f3ec79f
SK2Decompile
2025-10-04 22:22:24 +08:00
cf685cdcbb
add docker support
2025-08-22 14:54:09 +08:00
3eee3690d4
Merge pull request #61 from hanXen/main
...
feat: add Dockerfile for ready-to-use environment
2025-08-22 14:47:31 +08:00
4001fbd45c
build: pin dependencies in requirements-demo.txt
2025-08-22 13:08:35 +09:00
473973cf79
build: update pytorch and cuda versions in dockerfile
2025-08-22 13:06:19 +09:00
7ed839a8a2
feat: add Dockerfile for ready-to-use environment
2025-08-20 18:53:31 +09:00
e0f3f65459
Merge pull request #55 from 7Sageer/main
...
Update README for LLaMA-Factory training example
2025-06-30 20:24:20 +08:00
6553713e41
docs: update README for LLaMA-Factory training example
2025-06-30 20:21:50 +08:00
41bd69aae0
Update readability_template.txt
2025-06-20 17:36:41 +08:00
36a2d57a3e
Merge pull request #50 from 7Sageer/main
...
fix: add missing training script
2025-06-12 19:52:58 +08:00
4c38c50505
fix: add missing training script
2025-06-12 19:51:16 +08:00
f46ef2ae47
Update readme.md
2025-06-06 14:59:22 +08:00
494e81735e
Update README.md
2025-05-29 15:10:07 +08:00
717032a0b0
Merge pull request #48 from 7Sageer/main
...
Add LLaMA-Factory training example
2025-05-29 15:01:54 +08:00
e30c73cb92
Add LLaMA-Factory training example with README, requirements, and dataset files
2025-05-29 12:57:28 +08:00
99b209388a
Update readme.md
2025-05-23 16:02:31 +08:00
cec9426c04
Update readme.md
2025-05-22 11:29:34 +08:00
341371b124
Update readme.md
2025-05-22 11:28:02 +08:00
6d6023770b
Update readme.md
2025-05-21 19:40:12 +08:00
26b451f5ee
Update README.md
2025-05-21 19:36:00 +08:00
bdabad7543
Update readme.md
2025-05-21 19:34:40 +08:00
757b7a28fb
legacy
2025-05-21 19:31:00 +08:00
f91058ce48
Update readme.md
2025-05-21 19:26:43 +08:00
16d6f4903d
Add files via upload
2025-05-21 19:25:04 +08:00
54c59d79e4
Add files via upload
2025-05-21 19:24:45 +08:00
e6551ae8c3
evaluation
2025-05-21 19:08:31 +08:00
15c3a0785f
upload mbpp
2025-05-21 17:33:59 +08:00
9eb2fdcda2
Create readme.md
2025-05-21 17:07:54 +08:00
0d663e528e
Add files via upload
2025-05-21 17:02:28 +08:00
44f530d2fb
Add files via upload
2025-05-21 16:54:54 +08:00
02f240f6bd
Add files via upload
2025-05-21 11:05:39 +08:00
61265b5619
Create humaneval-decompile.json
2025-05-21 10:53:05 +08:00
156dea65b5
Create cal_edit_sim.py
2025-05-21 10:52:07 +08:00
1354ed4f6e
Create cal_execute_rate.py
2025-05-21 10:48:21 +08:00
0a5aafca28
Update decompile-bench
2025-05-21 10:44:53 +08:00
72d403f3d7
Update youtube link
2024-10-28 17:39:24 +08:00
albertan017