Wang Yang, Yipu Dou
We lifted 13 functions out of this paper's own repositories and ran 13 of them in a sandbox. "Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results.
| Repository | Role | Ran |
|---|---|---|
| douyipu/mosaic | canonical | 13 of 13 |
| Function | Status | Where it lives |
|---|---|---|
| agg_attack_complexity | Ran | douyipu/mosaic/score/xguard_agg.py code served (permissive licence) · get_code("192df5ea0a2af774") |
| agg_concealment | Ran | douyipu/mosaic/score/xguard_agg.py code served (permissive licence) · get_code("a97bd64c3b347ea9") |
| agg_pressure | Ran | douyipu/mosaic/score/xguard_agg.py code served (permissive licence) · get_code("6e7ff173309098ff") |
| allocate_budget | Ran | douyipu/mosaic/score/need.py code served (permissive licence) · get_code("3cd829d3fc86577c") |
| compute_bucket_need | Ran | douyipu/mosaic/score/need.py code served (permissive licence) · get_code("74e6be00213605fb") |
| compute_if_score | Ran | douyipu/mosaic/score/ifeval_score.py code served (permissive licence) · get_code("7466abd666c8d9d7") |
| compute_inst_complexity | Ran | douyipu/mosaic/score/ifeval_score.py code served (permissive licence) · get_code("ba87b941afaede2b") |
| compute_sample_need | Ran | douyipu/mosaic/score/need.py code served (permissive licence) · get_code("1d398c911017fc67") |
| orbench_boundary_level | Ran | douyipu/mosaic/score/orbench_agg.py code served (permissive licence) · get_code("dc879b52216afc2f") |
| orbench_valid | Ran | douyipu/mosaic/score/orbench_agg.py code served (permissive licence) · get_code("1fb94f52d201c5f2") |
| orbench_weight | Ran | douyipu/mosaic/score/orbench_agg.py code served (permissive licence) · get_code("9b7d21698963c704") |
| process_batch | Ran | douyipu/mosaic/score/experiment_utils.py code served (permissive licence) · get_code("58f8a5668faf2811") |
| process_batch | Ran | douyipu/mosaic/score/util.py code served (permissive licence) · get_code("cf9275b0b5918737") |
Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.
We study how to allocate a fixed supervised fine-tuning budget when three objectives must be balanced at once: multi-turn safety alignment, low over-refusal on benign boundary queries, and instruction following under verifiable constraints. We propose MOSAIC (Multi-Objective Slice-Aware Iterative Curation for Alignment), a multi-objective framework for closed-loop data mixture search built on a unified L1-L3 evaluation interface. MOSAIC turns slice-level failure profiles into executable data actions, including dataset-level mixture ratios, bucket-level weights, and focus criteria. Under a fixed 1M-token budget and five rounds of independent fine-tuning from the same base model, MOSAIC improves internal XGuard from 2.76 to 4.67 while keeping OrBench at 4.41 and IFEval at 3.65. The final Pareto solution also generalizes better than a random static LoRA baseline on independent attack, over-refusal, and capability tests, suggesting that structured failure diagnosis can serve as a practical control signal for budgeted data construction. Code is available at https://github.com/douyipu/mosaic.
The same record, over MCP at https://syntology.ai/mcp:
get_harvested_code_for_paper("2603.18637")
get_code_for_paper("2603.18637")
have("2603.18637")
Connect an agent — have() is free.