File size: 1,358 Bytes
e4805d0 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 | set -x
cd /job/work
python3 -m venv --system-site-packages /job/work/.venv
. /job/work/.venv/bin/activate
pip install -q selectolax warcio onnx onnxruntime onnxscript tldextract aiohttp 2>&1 | tail -3
python3 -c "import selectolax, onnx, onnxruntime, aiohttp, torch; print('ok', torch.cuda.is_available())"
# compare baked overlays in image vs production overlay files
P=/home/user/.local/share/rtx-pro-apps/qwen3.8-flash-next/runtime-speed-20260924
for pair in "qwen_sparse_attn_backend.py:layers/attention/qwen_sparse_attn_backend.py" "qsa_sparse_attn.py:layers/attention/qsa/sparse_attn.py" "qwen4_exp_mtp_hotmap.py:models/qwen4_exp_mtp.py" "eagle_worker_v2.py:speculative/eagle_worker_v2.py" "metadata.py:layers/attention/qsa/metadata.py" "graph_metadata.py:layers/attention/qsa/graph_metadata.py" "qsa_indexer.py:layers/attention/qsa/qsa_indexer.py" "qsa_kv_pool.py:mem_cache/qsa_kv_pool.py" "eagle_utils.py:speculative/eagle_utils.py" "qwenfast.py:qwenfast.py" "qwen4_exp_startup.py:models/qwen4_exp.py" "loader_startup.py:model_loader/loader.py"; do
a=${pair%%:*}; b=${pair#*:}
echo "$a $(sha256sum $P/$a | cut -c1-12) $(sha256sum /sgl-workspace/sglang/python/sglang/srt/$b | cut -c1-12)"
done
ls /opt/ | head; ls /opt/qwen* 2>/dev/null
curl -sI https://data.commoncrawl.org/crawl-data/CC-MAIN-2026-39/warc.paths.gz | head -3
df -h /job | tail -1
|