Dataset Viewer

The dataset viewer is not available because its heuristics could not detect any supported data files. You can try uploading some data files, or configuring the data files location manually.

Nimble-v4-max on the Decision Index

Full public-suite Decision Index 0.3 runs of Nimble-v4-max:

  • runs/nimble-v4.1-max/: v4.1 (model tag v4.1), 72.74.
  • runs/nimble-v4-max/: v4.0 (model tag v4.0), 68.46.

Each run folder holds the results and scores in the layout the Decision Index kit uploads. code/nimble/ is the engine that produced both runs. With the kit (apolinario/decision-index) installed:

hf download bespokelabs/Nimble-v4-max --revision v4.1 --local-dir Nimble-v4-max
PYTHONPATH=code python -m decision_index run --engine nimble.evaluation.decision_index_engine:NimbleEngine \
    --option adapter=Nimble-v4-max --option max_tails=48 --rows <suite rows> --out <run dir>

It needs a CUDA GPU with BF16, torch 2.8, transformers 5.17.0, accelerate, flash-linear-attention 0.5.2 and causal-conv1d 1.7.0. The model needs about 55 GB in BF16 and runs on one RTX PRO 6000. shared_prefix is on by default: a request's shared part runs once, then each question's short tail. Requests longer than the model's 32,768-token context are refused as unsupported, and nothing is truncated.

The engine checks the model's schema_config.json before it runs: if the prompt files or the option-code tokens differ from the ones the model was trained with, it refuses to load.

Released under CC BY-NC 4.0, like the model.

Citation

@misc{bespokelabs2026nimblev4max,
  title        = {Nimble-v4-max},
  author       = {Shreyas Pimpalgaonkar and Anirudha Ramesh and Sid Devic and Mrinal Kumar and Pratik Dalal and Mahesh Sathiamoorthy},
  year         = {2026},
  publisher    = {Bespoke Labs},
  howpublished = {\url{https://huggingface.co/bespokelabs/Nimble-v4-max}}
}
Downloads last month
419