datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
scout_idx1024to2048_trace_align
scout_idx1024to2048_trace_align
DeepSeek-V4-Flash reference-grounded trace-alignment annotations over the
1024 scout trajectories in HerrHruby/scout_idx1024to2048_raw.
Judge: DeepSeek-V4-Flash @ high reasoning effort, --judge-input both
(reference solution + rubric), --prefix-layer 8 (judge sees the full
trajectory). Generated by
synthetic.midtrain_v3.branch_walk.trace_align (commit 2796b2b on
synth-midtrain-v3).
File
trace_align/layer_008.jsonl 1 024 records… See the full description on the dataset page: https://huggingface.co/datasets/HerrHruby/scout_idx1024to2048_trace_align.loo_idx1_5pc_5_aug-step1-masterloo_idx1_5pc_1_aug-step1-mastermodels-loo_idx1_5pc_1-step1