Datasets:

laion
/

mv-posttrain-eval-responses

Modalities:

Formats:

Size:

Libraries:

Dataset card Data Studio Files Files and versions

mv-posttrain-eval-responses

9.9 GB

Ctrl+K

Ctrl+K

1 contributor

History: 39 commits

marianna13's picture

GSM8K 3-seed sampled responses + per-sample is_correct (8/11 tasks now per-sample graded)

b2d8371 verified about 12 hours ago

.gitattributes

2.5 kB
initial commit 5 months ago
eval_responses.parquet
9.89 GB
xet

GSM8K 3-seed sampled responses + per-sample is_correct (8/11 tasks now per-sample graded) about 12 hours ago
eval_scores.parquet

61.3 kB
xet

GSM8K: 3-seed sampled (temp0.7, do_sample) mean@3 + per-seed + std; replaces greedy single-seed about 12 hours ago
eval_scores_dual.csv

20.7 kB
Add dual-regime (greedy/sampling) evals @ max_tokens=32000 for 80 checkpoints (6 tasks); add decoding+max_tokens cols; legacy rows tagged decoding=legacy; +eval_scores_dual.csv 2 months ago
eval_scores_wide.csv

6.58 kB
Scores: +MV-100B ablation, corrected code caps. 2 months ago
mv_aime24_amc23_best_responses.parquet

12.9 MB
xet

Add temp-0.7 x3 AIME24+AIME25 responses (6 MV stages) alongside existing; new `decoding` column distinguishes (sample_temp0.7 vs original). 2 months ago