Recipe · Declare Lab (SUTD)

NORA

Commands and version pins exactly as executed for our measurements. Replace <hardware> with the card id (rtx-4090 / rtx-5090 / a800-80g). Common steps (mirrors, disk, nohup) are on the recipes index.

1 · Environment

transformers 4.57.1 · torch system 2.5.1 (5090 requires 2.8) · accelerate · qwen-vl-utils · einops

python -m venv --system-site-packages /root/envs/qwen
/root/envs/qwen/bin/pip install transformers==4.57.1 accelerate qwen-vl-utils einops

2 · Weights

declare-lab/nora · 7.5 GB · Qwen2.5-VL 3B backbone

export HF_ENDPOINT=https://hf-mirror.com HF_HOME=/root/autodl-tmp/hf HF_HUB_DISABLE_XET=1
python -c "from huggingface_hub import snapshot_download as s; s('declare-lab/nora')"

{T.run}

{r.run}

{r['load_' + L] &&

{T.load} · {r['load_' + L]}

}

{T.pit}

{(r.pitfalls || []).map((p: any) => ())}

{T.sym}	{T.fix}
{p[L]}	{p['fix_' + L]}

{T.meas}

{T.src}

{rows.length ?

{rows.map((x) => ())}

{T.hwc}	{T.prec}	{T.p50}	{T.p95}	{T.vram}	{T.n}	{T.date}
{x.h.name}	{x.prec}	{fmtMs(x.p50)}	{fmtMs(x.p95)}	{x.vram != null ? x.vram + ' GB' : '—'}	{x.n}	{x.date}

{T.nomeas}

}