packed
swa-100mb-after-wc-uniform-newlex-after-ckpt500-packed-bfdiso_seed455nor-100mb-after-wc-uniform-newlex-before-ckpt500-packed-bfdiso_seed3407eng-latn-100mb-after-ppt-Dp-100mb-packed-ckpt500_seed455_seed455eng-latn-100mb-after-ppt-Dp-100mb-packed-ckpt500_seed3407eng-latn-10mb-after-ppt-Dp-100mb-packed-ckpt500_seed3407nor-100mb-after-wc-uniform-newlex-before-ckpt500-packed-bfdiso_seed455tam-100mb-after-wc-uniform-newlex-after-ckpt500-packed-bfdiso_seed3407nld-latn-100mb-after-ppt-Dp-100mb-packed-bfdiso-ckpt500_seed455
Datasets
All datasets matching “packed”DeepSeek-MixedModeReasoning-Logits-Packed-16384sequence_length: 16384
dataset:
train_dataset:
repo_id: arcee-ai/DeepSeek-MixedModeReasoning-Logits-Packed-16384
split: train
prepacked: true
teacher:
kind: dataset
legacy_logit_compression:
exact_k: 32
invert_polynomial: true
k: 32
polynomial_degree: 0
term_dtype: float32
vocab_size: 129280
with_sqrt_term: false
packed_long_20BDeepSeek-MixedModeReasoning-Logits-Packed-16384sequence_length: 16384
dataset:
train_dataset:
repo_id: arcee-ai/DeepSeek-MixedModeReasoning-Logits-Packed-16384
split: train
prepacked: true
teacher:
kind: dataset
legacy_logit_compression:
exact_k: 32
invert_polynomial: true
k: 32
polynomial_degree: 0
term_dtype: float32
vocab_size: 129280
with_sqrt_term: false
transfer-bench-intents-v2-packed
transfer-bench-intents-v2
CUDA→Ascend NPU 迁移评测任务的 intent 清单数据集（第二代）:3619 个真实开源 repo,
每个含 5–25 条经校验的 tier1/tier2 测试意图（intents.json)+ analyst 分析底稿（notes.md),
以及迁移任务元数据与打包的原始 repo 快照。
选集：rosetta-selector（transfer-bench-Rosetta/rosetta-selector/）fastpath
确定性筛选管线产出——口径"PyTorch 项目且未适配 NPU"，零 agent 成本；
候选集 rosetta-selector/runs/20260831-154337/candidates
生成:transfer-bench-Rosetta/intents-generation/gen_intents.py，模型 kimi/kimi-k2.7-code,
source run… See the full description on the dataset page: https://huggingface.co/datasets/foreverCuSO4/transfer-bench-intents-v2-packed.robodojo-assets-packed
RoboDojo Assets — verified transport shards
This public dataset is a byte-preserving transport repack of Assets/ from
RoboDojo-Benchmark/RoboDojo, pinned at source commit
aada55a7a58b520a2004b9db02e884144b6e83e7. It reduces 15,335 source files to 11 independently downloadable and
restorable tar.zst shards. No payload file was changed.
这是 RoboDojo Assets/ 的公开传输分片：固定源 commit、减少小文件数量、每片可独立下载/重试，
完整复原后会逐文件做 SHA-256 校验。
Contents
Source files: 15,335
Source logical… See the full description on the dataset page: https://huggingface.co/datasets/didfd/robodojo-assets-packed.Qwen3-235B-Logits-Packed-8192sequence_length: 8192
dataset:
train_dataset:
repo_id: arcee-ai/Qwen3-235B-Logits-Packed-8192
split: train
prepacked: true
teacher:
kind: dataset
logprob_compressor:
d: 151936
delta_encoding: true
error_diffusion: false
exact_dtype: float32exact_k: 32
k: 128
polynomial_terms: [0, 1, 2]
residual_bins: []
term_dtype: float32