datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
DLLM-Planing-Task
DLLM-Planning-Task
Benchmark datasets for evaluating planning capabilities of Diffusion Language Models (DLLMs).
Dataset Description
This dataset contains multiple planning and combinatorial reasoning tasks designed to evaluate discrete diffusion language models. Each task has train/test splits in either CSV or JSONL format.
Tasks
Task
Format
Description
Sudoku
CSV
9x9 Sudoku puzzles. Columns: quizzes, solutions.
Path Finding (path-2-6)… See the full description on the dataset page: https://huggingface.co/datasets/zeyuzy/DLLM-Planing-Task.dllm-prm-llada-eval-gsm8k
LLaDA-8B-Base PRM-Guided Evaluation (GSM8K)
PRM-Guided generation outputs on GSAI-ML/LLaDA-8B-Base, full GSM8K test (1,319 problems), K=8, 16 configurations: {bidir, causal} × branch_every {16, 32, 48, 64} × seeds {42, 43}.
Summary (sample std)
Method
n
mean ± std
LLaDA bidir PRM-Guided
8
0.3164 ± 0.0075
LLaDA causal PRM-Guided
8
0.2225 ± 0.0090
LLaDA Vanilla K=1
1
0.2077
Bidir-over-causal gap: +9.4 pp, 95% CI [+8.5, +10.3] pp.
Schema
{… See the full description on the dataset page: https://huggingface.co/datasets/AnonyRepo/dllm-prm-llada-eval-gsm8k.dllm-qwen38-ar-baseline
AR baseline for the Qwen3.8-27B → block-diffusion conversion (GSM8K, pinned 500-problem subset)
日本語要約: Qwen/Qwen3.8-27B を Fast-dLLM v2 で
block-diffusion dLLM 化する計画の AR 参照スコアです。seed 固定の GSM8K 500 問・4-shot・
greedy で acc 0.968（484/500、skipped 0）。H100 1 枚で 42 分 ≈ $2.8。停止条件
（stop literal）として「block-diffusion 訓練 0.3B tokens の後、この subset で acc ≥ 0.918」
を要求し、届かなければ変換を続けません。訓練 run 自体はこの判断待ちで held です。
Why this exists — the stop literal
We are converting Qwen/Qwen3.8-27B into… See the full description on the dataset page: https://huggingface.co/datasets/com-junkawasaki/dllm-qwen38-ar-baseline.