datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
vla-framework-repro-blackwell-20260707
vla_framework — Blackwell 复现结果 (2026-07-07)
vla_framework(develop) 在 RTX PRO 4000 Blackwell ×2 (sm_120, driver 595) 上从零复现的评测结果与视频。
conda env vla_framework，torch 2.11.0+cu128。详见仓库 docs/REPRODUCTION_REPORT_2026-07-07.md。
1. SmolVLA 纯 VLA 基线 — place_bread_skillet
模型 arrow-hf/smolvla-robotwin-place-bread-skillet-50ep-multi，passthrough(无 MPC)，robotwin_gt，chunk_exec=50，max_steps=600，subprocess-per-episode。
**env_success 4/10 (40%)**，对照文档 vtc benchmark multi 45% ——… See the full description on the dataset page: https://huggingface.co/datasets/arrow-hf/vla-framework-repro-blackwell-20260707.Frameworker_User_Studyvla-framework-poselib-8task-eval-v3