GMP CLI

Install

Install the genmanip-client package in your client environment:

pip install -e standalone_tools/packages/genmanip_client/
gmp --help

Core commands

Command	Purpose
gmp submit	Submit or reconnect benchmark tasks on the eval server.
gmp status	Inspect progress and metrics for the current run.
gmp eval	Run client workers and interact with server episodes.
gmp plot	Post-process episode outputs into visualization artifacts.
gmp clean	Remove generated caches, logs, eval outputs, and temporary leftovers.
gmp visualize	Browse eval results and replay episodes in the Rerun viewer.

Submit, status, and eval

gmp submit

Benchmark family + split:

gmp submit ebench/mobile_manip/test --run_id mobile_test
gmp submit ebench/table_top_manip/val_unseen --run_id tabletop_val_unseen
gmp submit ebench/generalist/val_train --run_id generalist_val_train

Benchmark alias:

gmp submit ebench --run_id full_benchmark

Supported task-setting paths:

Task settings:

• mobile_manip
• table_top_manip
• generalist

Splits:

• val_train
• val_unseen
• test

gmp status

gmp status --host 127.0.0.1 --port 8087
gmp submit ebench --run_id history_id
gmp status

gmp eval

gmp eval -a r5a -g lift2 --worker_ids 0 --frame_save_interval 10
gmp eval --worker_ids 0,1 --chunk_size 8 --host 127.0.0.1 --port 8087

For custom model integration, see Integrate Your Own Model.

Clean, plot, and visualize

gmp plot

gmp plot client_results/<benchmark>/<run_id>/<task>/<seed>

gmp clean

Use gmp clean to remove generated artifacts from local runs.

Preview what would be removed:

gmp clean --dry-run

Remove generated mesh cache, eval results, logs, and leftover lock/tmp files:

gmp clean

Also remove downloaded benchmark package cache:

gmp clean --all

gmp visualize

gmp visualize starts a local HTTPS viewer for browsing runs, task success rates, and per-episode replays.

Install the visualize extra:

pip install -e "standalone_tools/packages/genmanip_client/[visualize]"

Basic usage:

gmp visualize
gmp visualize --port 55088

Cache management:

gmp visualize --flush-cache --dry-run
gmp visualize --flush-cache

Notes:

• gmp visualize expects evaluation outputs under saved/eval_results/.
• The viewer uses HTTPS and may show a one-time browser certificate warning.
• The current rerun-sdk path used by visualize requires Python 3.11+.

Common options

• --run_id: identifies and resumes a run.
• --host, --port: eval server target (default is local 127.0.0.1:8087).
• --worker_ids: worker allocation in gmp eval.
• --frame_save_interval: client-side frame saving frequency.
• --chunk_size: action chunk length when your model predicts chunked actions.