Benign huggingface Download

mkg20001/ai-wasteland-local main

Shell changes working directoryExecutes a child process

SHA-25640b6856615c49e8d8600fcdc1bdefa0bfcddc35e24b21a469530e7fb8000997a

URLhttps://huggingface.co/mkg20001/ai-wasteland-local

Evidence

English function-word token "the" lines 1–6

1#!/usr/bin/env bash
2#
3# The offline model, end to end and repeatable.
4#
5#   ./model.sh corpus [n]     compose and score training data
6#   ./model.sh extr …

English function-word token "for" lines 11–15

11:16… rules [file]   publish what the schemas permit, for the decoder
12#   ./model.sh serve [port]   an OpenAI-shaped endpoint in front of it
13#   ./model.sh calibrate [file] correlate rollout events with final completion
14#   ./model.sh test           the trainer's own tests
15#   ./model.sh all [n]   …

Shell changes working directory lines 26–32

26:10… ry of weights says how it was made.
27#
28set -uo pipefail
29
30root="$(cd -- "$(dirname -- "${BASH_SOURCE[0]}")" && pwd)"
31
32# ---------------------------------------------------------------------------

Shell invokes mkdir command lines 155–161

21… — "what was this trained with?" — has no answer.
156record() {
  mkdir -p "$AIW_MODEL_DIR"
  {
      echo "# Written by model.sh. The parameters this model was trained with."
      echo "# Re-run with: env \$(grep -v '^#' run.env | xargs) ./model.sh train"
      for var in A …

Executes a child process lines 246–261

45… $AIW_MODEL_DIR" "${1:-8080}"
      ;;
248
  calibrate)
      exec "$root/calibrate-rollouts.sh" "${1:-$root/build/rollouts.jsonl}"
      ;;
252
  test)
      ( cd "$root/trainer" && in_shell cargo test )
      ;;
256
  all)
      "$0" corpus "${1:-2000}" || exit 1
      "$0" train || exit 1
      samples="$root/trainer/data/samples.jsonl"
      "$0" sample "${AI …