openvla_oft fine-tuned on genesis-hr-bench (training-in-progress, step 50,000)

⚠️ This is a live training snapshot, not a final artifact. The repo is periodically overwritten with whatever checkpoint is currently latest on the training cluster (slurm job 1372375). The step number, loss, and performance reflect the snapshot taken at the last upload — not the converged model.

Provenance

Base model openvla/openvla-7b
Dataset zhouqh/hrbench (curated *_target subset)
Family openvla_oft
Slurm job 1372375
Current step 50,000 (snapshot from openvla-7b+genesis_hr_bench+b8+lr-0.0005+lora-r32+dropout-0.0--image_aug--openvla_oft_target_20260525--50000_chkpt)
Wandb openvla_oft_genesis_hr_bench
Last uploaded at 2026-05-26 09:33:06 UTC

Usage

This is intended as a fast-feedback artifact for collaborators wanting to download and test partial-training snapshots without waiting for the full run. For the final, frozen checkpoint (once training completes), look for a non--latest repo under the same author.

Tracking progress

The repo's last-modified timestamp tells you when the snapshot was taken. The README always reflects the current step. To get the next snapshot, just hf download again — the URL stays the same.

Downloads last month
256
Safetensors
Model size
8B params
Tensor type
BF16
·
Video Preview
loading

Model tree for zimplex/openvla-oft-genesis-hr-bench-target-latest

Finetuned
(20)
this model

Dataset used to train zimplex/openvla-oft-genesis-hr-bench-target-latest