asingh15/tcs-qwen36-27b-direction-rollouts-partial-1017-mc8
TCS Qwen3.6-27B Direction Beam Partial 1017 MC8 Verified compact export for tcs_qwen36_27b_direction_beam_full1801_mc8_20260815. Source dataset: TCS train-00000-of-00001.parquet Problems: 1,017 Displayed trajectories: 4,068 Chunk probes: 32,544 Terminal answers and rubric grades: 260,352 Policy: Qwen/Qwen3.6-27B Judge: openai/gpt-oss-20b (low reasoning) At each displayed chunk, the policy proposes four directions plus a null continuation. A width-four stochastic beam retains… See the full description on the dataset page: https://huggingface.co/datasets/asingh15/tcs-qwen36-27b-direction-rollouts-partial-1017-mc8.
TCS Qwen3.6-27B Direction Beam Partial 1017 MC8
Verified compact export for tcs_qwen36_27b_direction_beam_full1801_mc8_20260815.
- Source dataset:
TCS train-00000-of-00001.parquet - Problems: 1,017
- Displayed trajectories: 4,068
- Chunk probes: 32,544
- Terminal answers and rubric grades: 260,352
- Policy:
Qwen/Qwen3.6-27B - Judge:
openai/gpt-oss-20b(low reasoning)
At each displayed chunk, the policy proposes four directions plus a null continuation. A width-four stochastic beam retains four leaves, with 2 terminal sample(s) generated from each leaf. Only terminal samples are graded; the chunk value is their mean normalized rubric score. Browser bundles under web/ are split by problem and probe, with each probe embedding its 8 terminal answers, grades, and ancestry. The explorer therefore fetches one file for all details at a selected chunk.