CharlieLLL/DeepResearch9K-qwen3-1p7b-eval-traces-0917

Qwen3-1.7B worker evaluation traces — DeepResearch9K 12 raw/SFT runs under six orchestrators. Published with user authorization to use public HF storage on 2026-09-17. Original private repositories and historical data remain unchanged; only this new 1.7B campaign is included here. Full raw results, coordinator/worker traces, measured role tokens/cache, episode timings, input snapshot, manual-protocol grading, and separate theoretical within-conversation Max prefix estimates are… See the full description on the dataset page: https://huggingface.co/datasets/CharlieLLL/DeepResearch9K-qwen3-1p7b-eval-traces-0917.