Skip to content

Configure HistoPLUS access

HistoPLUS is hosted as a gated model on Hugging Face. TumorQuantAI can prepare and inspect public slides without it, but inference requires approved access.

1. Request access

Open the HistoPLUS model page, sign in, and request access. Creating a token does not approve the model request; both steps are required.

2. Create a read token

After the model request is approved, create a Hugging Face access token with Read permission.

3. Save the token privately

# Create the private TumorQuantAI configuration directory.
install -d -m 700 "$HOME/.config/tumorquantai"

# Paste the Hugging Face read token without displaying it.
read -rsp "Hugging Face read token: " HF_READ_TOKEN
printf '\n'
printf '%s' "$HF_READ_TOKEN" > "$HOME/.config/tumorquantai/hf_token"
unset HF_READ_TOKEN
chmod 600 "$HOME/.config/tumorquantai/hf_token"

TumorQuantAI automatically reads $HOME/.config/tumorquantai/hf_token. Do not put the token value after --token-file, in an environment file, in a Git command, or in an issue.

4. Check readiness

# Check the installed runtime, public metadata, and local credential file.
tumorquantai doctor --online

Doctor confirms that the token file exists and is private; the first inference confirms that the approved account can download the pinned HistoPLUS artifact.

Authorized local weight

An organization-provided authorized weight can be selected instead of a token:

# Run with an authorized local HistoPLUS weight.
tumorquantai run /path/to/slides \
  --output /path/to/results \
  --local-weight /approved/model-store/histoplus_cellvit_segmentor_20x.pt \
  --preset smoke \
  --source-mpp 0.261780 \
  --cpu

The weight is hashed for provenance and read from its original location. Do not copy it into the repository or result directory.

Replace a compromised token

Revoke it in Hugging Face, repeat the private-file commands with a new read token, and rerun tumorquantai doctor --online.

Next: run the public one-slide QuickStart.