Files
watermarks-remover/compose.yaml
T
Guillaume Meyer (The Opinionated Man)andGitHub d5f4f03f85 feat: add reproducible SynthID-text removal benchmark (#145)
* feat: add reproducible SynthID-text removal benchmark

bench_synthid_text.py orchestrates the existing Layer B machinery into a
controlled, shareable experiment: generate watermarked + unwatermarked
samples with the MarkLLM SynthID scheme, run removal variants (strength x
candidates) plus controls (no-removal, Layer-A-only, optional re-stamp),
and report clear rate, score suppression, quality, and cost (tokens,
wall time, USD) with a clears-per-MTok efficiency ratio.

Emits report.md / results.json / results.csv with the exact reproduction
command and pinned commits; optional Gemini official-detector tier when
WATERMARKS_GEMINI_API_KEY is set. Mock-based tests, no torch in CI.

* docs: add README section on running the SynthID-text benchmark

Explains what LLM performs the Layer B rewrite (an external model configured
via WATERMARKS_REWRITE_* env vars or --rewrite-* flags; MarkLLM's opt-1.3b is
only the watermark generator/detector) and how to run a benchmark with Ollama
or an OpenAI-compatible endpoint, plus the non-origin-model re-stamp caveat.

* fix: honor WATERMARKS_REWRITE_ALLOW_REMOTE in the SynthID-text benchmark

The --rewrite-allow-remote flag now defaults from the env var (matching
rewrite_text.py and the other WATERMARKS_REWRITE_* settings), so a
non-loopback rewrite endpoint works after sourcing .env without an extra
flag.

* fix: MarkLLM sparse checkout and deps for the SynthID harness

- setup_markllm.sh sparse-checkout omitted '/visualize/', which
  watermark/base.py imports at module load — every scheme (incl. SynthID)
  failed with 'No module named visualize' during generation/detection.
- requirements-markllm.txt omitted scikit-learn, imported by the SynthID
  detector (watermark/synthid/detector_bayesian_torch.py).

Both broke the MarkLLM harness at runtime; the benchmark's sanity gate then
excluded every sample, producing empty per-variant results.

* fix: drop 4 GiB RLIMIT_AS on benchmark subprocesses

_run_cmd applied the common child RLIMIT_AS (default 4 GiB) via
subprocess_preexec_fn to every MarkLLM/rewrite child. torch needs a much
larger address space: CUDA init failed with 'out of memory' at
cudaGetDeviceCount and the 5.2 GB fp32 opt-1.3b could not load, so every
sample was excluded at generation. text_detectors.py already applies no
address-space cap to MarkLLM by default; the benchmark now matches.

* perf: keep MarkLLM resident via a serve worker (624 cold starts -> 1)

The benchmark spawned a fresh torch + opt-1.3b process per operation
(~60-90s each); a full run needs ~624 of them. detect_text_watermark.py
gains a 'serve' mode (JSON-lines over stdin/stdout, ready handshake) that
loads the model once; bench_synthid_text.py uses it via MarkLLMWorker with
automatic fallback to one-shot subprocesses (--no-worker to force).
Turns ~8h runs into ~40-60min.

* perf: skip per-candidate Gemini detections in rewrite subprocess

* feat: run the SynthID-text benchmark from the wr-markllm compose service

- Dockerfile.markllm: add '/visualize/' to the sparse checkout (same fix as
  setup_markllm.sh) and COPY the benchmark + rewrite scripts (stdlib-only).
- compose.yaml: wr-markllm gets the WATERMARKS_REWRITE_* and
  WATERMARKS_GEMINI_* env wiring, a bench-out volume for --out-dir, and a
  read-only mount of the bundled corpus (build context is service/, so the
  corpus cannot be COPY'd).
- docs: docker compose run example.

Note: the image ships CPU torch by design, so the container path is for
portability/CI; GPU runs use the host setup_markllm.sh venv.

* feat: per-sample progress logging in the benchmark

The persistent worker returns samples in-memory, so nothing is written
until the end of a run — runs looked stuck. eprint a [gen i/N] line per
generated sample and a [removal] summary per sample.

* chore: migrate Gemini config to gemini-3.6-flash; document SynthID-text retirement

Google retired SynthID text watermarking on the Generative Language API
(Aug 2026): text output is no longer watermarked and DETECT_TEXT_WATERMARK
is rejected on current 3.x models (confirmed by Google AI staff). Migrate
the default detection model to gemini-3.6-flash, document the retirement
in vendor-notes.md and the benchmark report caveat, and keep the detector
seam fail-soft until a vendor endpoint (e.g. Vertex AI) returns.

* feat: remove gemini-synthid-text detector (Google retired text watermarking)

Google removed SynthID text watermarking from the Generative Language API
(Aug 2026): text output is no longer watermarked and DETECT_TEXT_WATERMARK
is rejected on current 3.x models, so the vendor detector had nothing to
detect. Remove GeminiSynthIDTextDetector and its wiring:

- text_detectors.py: drop the Gemini class, HTTP helpers, and constants;
  keep MarkLLM + Claude seams (registry now markllm + claude-text).
- server.py / rewrite_text.py: per-candidate detection now triggers on
  --markllm-scheme only.
- bench_synthid_text.py: remove the Gemini tier (before/after, report
  table, --no-gemini flag); report caveat notes the retirement.
- configs/docs: drop WATERMARKS_GEMINI_* from .env.example / compose /
  README / SKILL.md / vendor-notes.md; keep the retirement note.
- tests: gemini tests removed or converted to MarkLLM (mocked subprocess).
- Dockerfile.markllm: parameterize BASE_IMAGE + TORCH_INDEX_URL so a GPU/
  arm64 image can be built (used for the --gpus all benchmark run).

* fix: harden notes aggregation against non-string notes

A run completed all samples but crashed at the final aggregate step with
'cannot use list as a set element' when a row's notes contained a
non-string value. Filter notes to strings (aggregate + CSV) and add a
regression test.

* perf: let the rewrite subprocess reuse the resident MarkLLM worker

The rewrite subprocess (rewrite_text.py) ran its own before/after MarkLLM
detects, each a ~20s torch+model cold start (~12 per sample = ~5min of the
~6min/sample runtime). Now:

- detect_text_watermark.py serve gains --port N: a loopback TCP JSON-lines
  listener (default -1 = off) sharing the resident model, with a lock so
  stdin and socket requests never run the model concurrently.
- text_detectors.MarkLLMTextDetector checks WATERMARKS_MARKLLM_PORT and
  does a fast loopback detect when a worker is up, falling back to the
  one-shot subprocess otherwise.
- The benchmark worker publishes its port via that env var, so the rewrite
  subprocess inherits it and its detects hit the resident model.

Turns ~6 min/sample into ~1-2 min; a full run drops from ~2h to ~40-50min.
Tests: loopback-client + fallback + env-publish coverage.

* chore: add benchmark-smoke.sh / benchmark-full.sh wrappers

Simple host wrappers: source .env, default MARKLLM_DIR to ~/MarkLLM, use a
repo-local HF cache by default, and run bench_synthid_text.py with a quick
(2 docs, 1 seed, paraphrase:1) or full (8 docs x 3 seeds, three variants,
re-stamp control) configuration. OUT_DIR overrides the output location.
2026-08-18 18:13:56 -07:00

158 lines
5.4 KiB
YAML

# Whole-infra bring-up for watermarks-remover.
#
# docker compose up --build -d # core HTTP service only
# docker compose --profile harness up --build -d # + markllm / markdiffusion harnesses
# docker compose --profile heavy up --build -d # + ctrlregen / synthid (local builds)
#
# The skill and any web app talk to the wr-core service at http://127.0.0.1:8765
# (loopback-only host mapping). The heavy/harness images are one-shot CLIs:
# `up` starts them with `--help` to confirm the image, and real jobs run via
# `docker compose run`, e.g.:
# docker compose run --rm wr-ctrlregen /data/shot.png -o /data/out.png
#
# ctrlregen and synthid bake in upstream code that is not publicly
# redistributable (all-rights-reserved / non-commercial Research License), so
# they build from source locally and are never pushed to GHCR.
name: watermarks-remover
services:
wr-core:
build:
context: service
dockerfile: Dockerfile
image: ghcr.io/guillaumemeyer/watermarks-remover:latest
ports:
- "127.0.0.1:8765:8765"
environment:
# Empty by default (no auth). Set to require `Authorization: Bearer <key>`.
WATERMARKS_SERVER_API_KEY: ${WATERMARKS_SERVER_API_KEY:-}
# Optional SynthID image scorer sidecar (heavy profile).
WATERMARKS_SYNTHID_SCORER_URL: ${WATERMARKS_SYNTHID_SCORER_URL:-}
WATERMARKS_SYNTHID_SCORER_API_KEY: ${WATERMARKS_SYNTHID_SCORER_API_KEY:-}
read_only: true
tmpfs:
- /tmp
init: true
user: "10001:10001"
restart: unless-stopped
wr-markllm:
profiles: [harness]
build:
context: service
dockerfile: Dockerfile.markllm
image: ghcr.io/guillaumemeyer/watermarks-remover:markllm-latest
# One-shot CLI: `up` just confirms the image; run real jobs with
# `docker compose run --rm wr-markllm detect /data/wm.txt --scheme kgw`.
command: ["--help"]
read_only: true
tmpfs:
- /tmp
init: true
user: "10001:10001"
environment:
HF_TOKEN: ${HF_TOKEN:-}
HF_HOME: /home/markllm/.cache/huggingface
# Layer B rewrite backend (openai-compatible / ollama).
WATERMARKS_REWRITE_BACKEND: ${WATERMARKS_REWRITE_BACKEND:-}
WATERMARKS_REWRITE_MODEL: ${WATERMARKS_REWRITE_MODEL:-}
WATERMARKS_REWRITE_BASE_URL: ${WATERMARKS_REWRITE_BASE_URL:-}
WATERMARKS_REWRITE_API_KEY: ${WATERMARKS_REWRITE_API_KEY:-}
WATERMARKS_REWRITE_ALLOW_REMOTE: ${WATERMARKS_REWRITE_ALLOW_REMOTE:-}
WATERMARKS_REWRITE_REASONING_EFFORT: ${WATERMARKS_REWRITE_REASONING_EFFORT:-}
volumes:
- markllm-cache:/home/markllm/.cache/huggingface
# Writable output for the benchmark + the bundled seed corpus (the
# build context is service/, so the corpus is mounted at runtime):
# docker compose run --rm wr-markllm \
# /app/bench_synthid_text.py --markllm-dir /opt/markllm \
# --corpus /bench-corpus --out-dir /data ...
- bench-out:/data
- ./benchmarks/corpus:/bench-corpus:ro
wr-markdiffusion:
profiles: [harness]
build:
context: service
dockerfile: Dockerfile.markdiffusion
image: ghcr.io/guillaumemeyer/watermarks-remover:markdiffusion-latest
command: ["--help"]
read_only: true
tmpfs:
- /tmp
init: true
user: "10001:10001"
environment:
HF_TOKEN: ${HF_TOKEN:-}
HF_HOME: /home/markdiffusion/.cache/huggingface
volumes:
- markdiffusion-cache:/home/markdiffusion/.cache/huggingface
wr-ctrlregen:
profiles: [heavy]
build:
context: service
dockerfile: Dockerfile.ctrlregen
# Local-only tag: upstream `noai-watermark` ships no LICENSE (all-rights-
# reserved), so this image is never published.
image: watermarks-remover-ctrlregen:local
command: ["--help"]
read_only: true
tmpfs:
- /tmp
init: true
user: "10001:10001"
environment:
HF_TOKEN: ${HF_TOKEN:-}
NOAI_WATERMARK_DIR: /opt/noai-watermark
volumes:
- ctrlregen-cache:/home/remover/.cache/huggingface
wr-synthid:
profiles: [heavy]
build:
context: service
dockerfile: Dockerfile.synthid
# Local-only tag: upstream `reverse-SynthID` is under a non-commercial
# Research License, so this image is never published.
image: watermarks-remover-synthid-scorer:local
command: ["--help"]
read_only: true
tmpfs:
- /tmp
init: true
user: "10001:10001"
volumes:
- synthid-cache:/home/scorer/.cache/huggingface
# HTTP SynthID scorer sidecar: lets wr-core score images before/after
# cleaning without bundling the non-commercial reverse-SynthID code in the
# published core image. Reach it via WATERMARKS_SYNTHID_SCORER_URL on
# wr-core (see .env.example) and require the same bearer key on both sides.
wr-synthid-score:
profiles: [heavy]
build:
context: service
dockerfile: Dockerfile.synthid
image: watermarks-remover-synthid-scorer:local
entrypoint: ["python3"]
command: ["/app/synthid_score_server.py", "--host", "0.0.0.0", "--port", "8766"]
read_only: true
tmpfs:
- /tmp
init: true
user: "10001:10001"
environment:
WATERMARKS_SYNTHID_SCORER_API_KEY: ${WATERMARKS_SYNTHID_SCORER_API_KEY:-}
REVERSE_SYNTHID_DIR: /opt/reverse-synthid
volumes:
- synthid-cache:/home/scorer/.cache/huggingface
volumes:
markllm-cache:
bench-out:
markdiffusion-cache:
ctrlregen-cache:
synthid-cache: