Zhongzhu/tunekv-terminalbench-27b

TB2 transfer eval (Harbor 0.1.28 terminus-2): RH vs DH, 2 rounds

tb6: full run trajectories (journal + agent logs) and training bundles (episodes/rows/logs/checkpoints) + SHA256SUMS

tb6: GRPO tuned+base verdicts (grpo/eval), comparisons, summary, README, final scripts (part 6)

tb6: GRPO tuned+base verdicts (grpo/eval), comparisons, summary, README, final scripts (part 5)

tb6: GRPO tuned+base verdicts (grpo/eval), comparisons, summary, README, final scripts (part 4)

tb6: GRPO tuned+base verdicts (grpo/eval), comparisons, summary, README, final scripts (part 3)

tb6: GRPO tuned+base verdicts (grpo/eval), comparisons, summary, README, final scripts (part 2)

tb6: GRPO tuned+base verdicts (grpo/eval), comparisons, summary, README, final scripts

tb6 COMPARISON_27B_CE.json

tb6 COMPARISON_27B_GRPO.json

tb6 27B lean verdicts q27-base80-b4

tb6 27B lean verdicts q27-grpo-tuned80-r2

tb6 27B lean verdicts q27-grpo-tuned80

tb6 27B GRPO: artifact (TP1 hybrid, n_inject 704) + rows report/manifest + config + train steps + canary

tb6 27B CE paired comparison

tb6 27B lean verdicts q27-tp-r4

tb6 27B lean verdicts q27-tp-r3

tb6 27B lean verdicts q27-tp-r2

tb6 27B lean verdicts q27-tp-r1

tb6 27B lean verdicts q27-ce-tuned80-r2

tb6 27B lean verdicts q27-ce-tuned80

tb6 27B lean verdicts q27-base80-b3

tb6 27B lean verdicts q27-base80-b2

tb6 27B lean verdicts q27-grpo-t3

tb6 27B lean verdicts q27-grpo-t2

tb6 27B lean verdicts q27-grpo-t1

tb6 27B lean verdicts q27-base80

27B shared prefix (704 ids)

tb6 pipeline scripts

tb6 27B CE+0.1KL: artifact (TP1 hybrid, n_inject 704) + 148 rows + config + train steps + canary

initial commit