sync-dev-org/tensoracle

Documented CI docs build (.github/workflows/pages.yml: uv sync --locked --only-group docs; uv run --no-sync mkdocs build --strict --config-file ../mkdocs.yml), run from the repo root with the locked docs environment. Also verified the README's CPU-only converter command: python -m tensoracle.convert model.pth model.safetensors (torch 2.14.1+cpu), which produced a safetensors file read back as state_dict.bias=[0.5,-1.0]. The converter prints nothing on success, so it was not used as the probe.

cd <host-path> && git log -1 --format=%H 2>/dev/null; ls -la; cat README.md 2>/dev/null | head -200
cd <host-path> && grep -n '^#' README.md; echo ---; grep -n -i -E 'pip|uv |python -m|mcp|quickstart|nvidia-smi' README.md | head -60; echo ---; ls .github/workflows 2>/dev/null; nvidia-smi 2>&1 | head -5; python3 --version
cd <host-path> && sed -n 276,340p README.md; echo ---; ls python; cat python/pyproject.toml 2>/dev/null | head -80; echo; ls <host-path>; echo $CDP_URL $<env-var>
cd <host-path> && sed -n 1,200p <host-path> | head -150; echo ---; cat .github/workflows/pages.yml | head -40; ls python/src/tensoracle; which uv pip3; ls /dev | grep -i nvidia; lspci 2>/dev/null | grep -i nvidia
cd <host-path> && (sudo apt-get install -y -q python3-venv >/tmp/apt.log 2>&1 || (sudo apt-get update -q >/dev/null 2>&1 && sudo apt-get install -y -q python3-venv >/tmp/apt.log 2>&1)); python3 -m venv <host-path> && export NODE_USE_ENV_PROXY=1 && timeout 900 <host-path> install -q . 2>&1 | tail -20; echo rc=$?; <host-path> -c "import tensoracle,tensorrt;print(tensoracle.__file__, tensorrt.__version__)" 2>&1 | tail -3
cd <host-path> && (timeout 600 <host-path> install -q torch --index-url https://download.pytorch.org/whl/cpu 2>&1 | tail -3); <host-path> -c "import torch;print(torch.__version__)" 2>&1 | tail -1
mkdir -p <host-path> && cd <host-path> && rm -f model.safetensors && <host-path> -c "
import torch
torch.save({'state_dict': {'bias': torch.tensor([0.5, -1.0], dtype=torch.float32)}}, 'model.pth')
print('wrote model.pth')" && <host-path> -m tensoracle.convert <host-path> <host-path>; echo rc=$?; ls -la <host-path>; <host-path> -c "
from safetensors.numpy import load_file; print(load_file('<host-path>'))"
grep -n -i -E 'output|expected_text|stays_running|def ' <host-path> | sed -n 1,80p
sed -n 377,545p <host-path>
cd <host-path> && sed -n '/dependency-groups/,/^\[tool/p' python/pyproject.toml | head -20; <host-path> install -q 'mkdocs>=1.6.1' 'mkdocs-material>=9.7.7' 2>&1 | tail -3; ls -d site 2>/dev/null; timeout 300 <host-path> build --strict --config-file mkdocs.yml 2>&1 | tail -8; echo rc=${PIPESTATUS[0]}
cd <host-path> && for p in mkdocs mkdocs-material mkdocstrings mkdocstrings-python; do grep -A1 "^name = \"$p\"" uv.lock | tail -1; done
cd <host-path> && <host-path> install -q 'mkdocs==1.6.1' 'mkdocs-material==9.7.7' 'mkdocstrings==1.0.6' 'mkdocstrings-python==2.0.9' 2>&1 | tail -3; rm -rf site; timeout 600 <host-path> build --strict --config-file mkdocs.yml > /tmp/mkdocs.log 2>&1; echo rc=$?; tail -5 /tmp/mkdocs.log; ls site | head -5
cd <host-path> && <host-path> install -q uv==0.12.5 2>&1 | tail -2; export UV_PROJECT_ENVIRONMENT=<host-path>; timeout 900 <host-path> sync --locked --only-group docs 2>&1 | tail -4; echo rc=$?; cd <host-path> && rm -rf site && timeout 600 <host-path> build --strict --config-file mkdocs.yml > /tmp/mkdocs.log 2>&1; echo rc=$?; tail -3 /tmp/mkdocs.log
cd <host-path> && rm -rf site && mkdir -p .coldpass && cat > .coldpass/probe.json <<'EOF'
{"kind":"cli","argv":["<host-path>","env":{"NODE_USE_ENV_PROXY":"1"}}
EOF
cat > .coldpass/report.json <<'EOF'
{"outcome":"working",
"operation":"Documented CI docs build (.github/workflows/pages.yml: uv sync --locked --only-group docs; uv run --no-sync mkdocs build --strict --config-file ../mkdocs.yml), run from the repo root with the locked docs environment. Also verified the README's CPU-only converter command: python -m tensoracle.convert model.pth model.safetensors (torch 2.14.1+cpu), which produced a safetensors file read back as state_dict.bias=[0.5,-1.0]. The converter prints nothing on success, so it was not used as the probe.",
"operation_scope":"Covers packaging/install, the docs build, and the CPU-only .pth->safetensors conversion. NOT verified: engine build, reuse, and execution. These need an NVIDIA GPU, driver, and CUDA context; this VM has no GPU (nvidia-smi not found, no /dev/nvidia*). Test suites were not run, per instructions.",
"steps":["apt-get install python3-venv","python3 -m venv <host-path>; pip install . in python/ (tensoracle 0.0-from-vcs, tensorrt-cu13 11.4.0.106 installed; import OK)","pip install torch --index-url https://download.pytorch.org/whl/cpu (for the [convert] extra; torch 2.14.1+cpu)","python -m tensoracle.convert model.pth model.safetensors (rc=0, verified by reading back)","uv 0.12.5 via pip; uv sync --locked --only-group docs with UV_PROJECT_ENVIRONMENT=<host-path>","mkdocs build --strict --config-file mkdocs.yml (rc=0; output in site/, removed after verification so the checker gets a fresh path)"],
"deviations":["The README's tensoracle[convert] extra was installed as pip install torch from the PyTorch CPU index plus the base package, not the default CUDA torch wheel, to save space. The product is the same.","Plain pip install of mkdocs/mkdocstrings failed in strict build (mkdocstrings Highlight.md TypeError). Used the CI path instead: uv 0.12.5 with the committed uv.lock, which resolved correctly.","The GPU-dependent README operations (open_engine, build_serialized_engine, execution) were not run: no NVIDIA GPU in the guest."],
"stuck_step":null,
"main_blocker":"No NVIDIA GPU or driver in the guest. Engine build and execution cannot run here.",
"suggested_fix":"Run on a Linux x86_64 host or VM with an sm86/sm89/sm120 NVIDIA GPU, driver, and TensorRT 11 (tensorrt-cu13), then run the README 'Define and build a standard graph' example with open_engine to cover the GPU path."}
EOF
python3 -m json.tool .coldpass/probe.json >/dev/null && python3 -m json.tool .coldpass/report.json >/dev/null && echo ok; rm -rf site; <env-var>=<host-path> python3 -I <host-path> --validate
cd <host-path> && rm -f .coldpass/model.pth .coldpass/model.safetensors && ls -la .coldpass && ls -d site 2>/dev/null; <env-var>=<host-path> python3 -I <host-path> --validate