Datasetpublic

ByteDance-Seed/EdgeBench

Overview EdgeBench is a benchmark of 134 real-world tasks for evaluating how autonomous AI agents learn from real-world environments. Instead of measuring one-shot performance, EdgeBench places agents in executable task environments with realistic, multi-level feedback and lets them iterate for 12+ hours per task — tracking the full trajectory of improvement, not just the final score. We publicly release 51 tasks… See the full description on the dataset page: https://huggingface.co/datasets/ByteDance-Seed/EdgeBench.

sourceHugging Facecc-by-4.0updated 2mo agoView on Hugging Face

84likes7.5kdownloads

rootjagua_nesting_optimization.json

jagua_nesting_optimization.json35 linesDownload Raw Back to root

1{2  "task_id": "jagua_nesting_optimization",3  "name": "Jagua Nesting Optimization",4  "category": "Combinatorial Optimization",5  "base_image": "rust",6  "platform": "linux/amd64",7  "internet": false,8  "cwd": "/home/workspace/jagua-rs",9  "submit_paths": [10    "lbf/src/",11    "jagua-rs/src/",12    "lbf/Cargo.toml",13    "jagua-rs/Cargo.toml",14    "Cargo.toml"15  ],16  "submit_exclude": [17    "target/",18    ".git/",19    "assets/"20  ],21  "work": {22    "image_tag": "992b06cb3f8e",23    "specs_dir": "/home/workspace/jagua-rs",24    "agent_query": "## Role\n\nYou are an expert Rust optimization engineer working on a 2D irregular nesting solver built on top of jagua-rs.\n\n## Task\n\nRead `SEBENCH_TASK.md`, inspect the real `jagua-rs` / `lbf` codebase, and improve the `lbf` optimizer. Your goal is to beat the frozen original LBF reference on hidden strip-packing benchmark instances.\n\n## Rules\n\n- Preserve the existing `cargo run --release --bin lbf -- -i ... -p spp -c ... -s ...` CLI behavior.\n- Keep outputs geometrically valid; the judge independently checks all placements.\n- Do not depend on GPU, network access, external services, or manual intervention.\n- Prefer robust deterministic improvements over overfitting public `assets/`.\n- Submit after meaningful changes so you can use iterative feedback.\n"25  },26  "judge": {27    "image_tag": "bba393b2dd53",28    "eval_cmd": "cd /home/workspace/jagua-rs && JAGUA_EVAL_CASES=16 JAGUA_EVAL_SAMPLES=900 JAGUA_CASE_TIMEOUT=20 JAGUA_BUILD_TIMEOUT=300 CANDIDATE_ROOT=/home/workspace/jagua-rs REFERENCE_ROOT=/home/workspace/jagua-rs-ref bash /home/workspace/jagua_task_scorer/score.sh",29    "eval_timeout": 1200,30    "parser": "score_sum",31    "score_direction": "maximize",32    "selection": "score_first"33  }34}35