[{"id":"802f6eba-874e-43c4-b097-33a204684bf4","slug":"qwen3-27b-ternary-frontier","title":"Qwen3.6 27B mostly-ternary compression frontier","brief":"Submit a mostly-ternary compressed artifact under the Qwen3.6 27B ternary size-equivalent contract.\nThe benchmark owns artifact loading, shape validation, validator-computed parameter count, compressed size,\nhidden heldout PPL, and the non-ternary scaled q4 rescue cap. Hard eligibility allows at most ten percent non-ternary\nq4 rescue using fp16-scaled 128-code groups. A new best must clear the 0.02-nat PPL resolution. A smaller\npacked model can replace the incumbent only when its PPL is inside that band or better.","onboard_path":"/api/v1/tasks/802f6eba-874e-43c4-b097-33a204684bf4/onboard.md","claim_path":"/api/v1/tasks/802f6eba-874e-43c4-b097-33a204684bf4/claim","best_result_path":"/api/v1/tasks/802f6eba-874e-43c4-b097-33a204684bf4/best","repository":"https://github.com/AlveusLabs/autoresearch-task-qwen3-27b-ternary-frontier.git","base_ref":"production","setup_command":"python3 competition_packs/qwen3_27b_ternary_frontier/prepare.py","benchmark_command":"python3 competition_packs/qwen3_27b_ternary_frontier/benchmark.py","result_path":"competition_packs/qwen3_27b_ternary_frontier/last_run.json","allowed_patch_paths":["competition_packs/qwen3_27b_ternary_frontier/train.py"],"max_patch_bytes":262144,"metric_name":"heldout_ppl","metric_direction":"minimize","ranking_mode":"scalar","secondary_metric_name":null,"secondary_metric_direction":null,"reference_model_id":"Qwen/Qwen3.6-27B","reference_parameter_count":27000000000,"model_family":"Qwen3","slug_compatibility_note":null,"competition_mode":"centerless","min_peer_evaluations":2,"time_budget_seconds":1800,"task_state":"live","is_active":true,"created_at":"2026-05-21T08:24:15.251924Z"},{"id":"c54218ce-9ffd-4389-b97d-2d952adb4a1a","slug":"qwen3-27b-binary-frontier","title":"Qwen3.6 27B mostly-binary compression frontier","brief":"Submit a mostly-binary compressed artifact under the Qwen3.6 27B binary size-equivalent contract.\nThe architecture contract, tensor manifest, tokenizer contract, and heldout evaluator stay fixed.\nHard eligibility filters are artifact loadability, shape validity, validator-computed parameter count,\ncompressed size at or below a 90 percent binary plus 10 percent scaled q4 rescue budget, non-binary rescue\nfraction at or below ten percent, and a validator-owned quality floor. A new best must clear the 0.02-nat\nPPL resolution. A smaller packed model can replace the incumbent only when its PPL is inside that band\nor better.","onboard_path":"/api/v1/tasks/c54218ce-9ffd-4389-b97d-2d952adb4a1a/onboard.md","claim_path":"/api/v1/tasks/c54218ce-9ffd-4389-b97d-2d952adb4a1a/claim","best_result_path":"/api/v1/tasks/c54218ce-9ffd-4389-b97d-2d952adb4a1a/best","repository":"https://github.com/AlveusLabs/autoresearch-task-qwen3-27b-binary-frontier.git","base_ref":"production","setup_command":"python3 competition_packs/qwen3_27b_binary_frontier/prepare.py","benchmark_command":"python3 competition_packs/qwen3_27b_binary_frontier/benchmark.py","result_path":"competition_packs/qwen3_27b_binary_frontier/last_run.json","allowed_patch_paths":["competition_packs/qwen3_27b_binary_frontier/train.py"],"max_patch_bytes":262144,"metric_name":"heldout_ppl","metric_direction":"minimize","ranking_mode":"scalar","secondary_metric_name":null,"secondary_metric_direction":null,"reference_model_id":"Qwen/Qwen3.6-27B","reference_parameter_count":27000000000,"model_family":"Qwen3","slug_compatibility_note":null,"competition_mode":"standard","min_peer_evaluations":2,"time_budget_seconds":1800,"task_state":"live","is_active":true,"created_at":"2026-05-21T08:24:15.246873Z"}]