# Model Blob SHA-256 Hashes — Day 3 Output-DLP-Gate Lab
# This lab uses the same qwen2.5:1.5b model as the other Day-1/2/3 labs (only for
# --live mode). The DEFAULT run scores the bundled corpus and needs NO model.
#
# These are content-addressed Ollama model blob hashes (ADR-6).
# Verify after pulling:
#   docker compose exec ollama ollama show qwen2.5:1.5b

# ── qwen2.5:1.5b ─────────────────────────────────────────────────────────────
# Architecture: qwen2 | Parameters: 1.5B | Quantization: Q4_K_M | License: Apache 2.0
# Use: --live generation of model outputs that the DLP gate then scans.
#
# Main GGUF blob (986 MB, Q4_K_M quantized):
QWEN2_5_1_5B_BLOB_SHA256=183715c435899236895da3869489cc30ac241476b4971a20285b1a462818a5b4
#
# Config/manifest SHA-256:
QWEN2_5_1_5B_CONFIG_SHA256=377ac4d7aeefd5b870c9fccff9a6d4df36901d99fe3277c2f755bc401601ba1c

# ── spaCy en_core_web_lg NER model (Presidio backend) ────────────────────────
# Installed via pip at build time (python -m spacy download en_core_web_lg, ~587 MB).
# Not HuggingFace — no cache / TRANSFORMERS_OFFLINE flag applies to spaCy itself.
# Verify: docker run --rm day3-output-dlp-gate-runner:local \
#           python -c "import en_core_web_lg; print(en_core_web_lg.__version__)"
#
# ── llm-guard (NON-model scanners only) ──────────────────────────────────────
# llm-guard==0.3.16. This lab uses ONLY Secrets / Regex / BanSubstrings — none of
# which load a HuggingFace model. NO model weights are baked for llm-guard here.
