Template updates

FLUX 3 Action DROID

flux-3-action-droid · policy template by Black Forest Labs · Franka Panda · 7 joints · 15 Hz · A100-40GB. Each version below is fixed once published, newest first.

seq init flux-3-action-droid gives the stable version, seq init flux-3-action-droid@beta the beta, and seq init flux-3-action-droid@v<N> any version below that is not withdrawn. A deployment made from this template names the version it runs in seq policy versions and on its page in the console; seq policy upgrade moves it to the newer stable one.

Performance

No runs yet.

Versions

  1. v6stablebeta

    FLUX 3 Action as v5; policy.py says, as the README does, how a worker's CPU and memory are billed.

    Needs sequence-ai 0.56.0 or later.

    • v5's files; only policy.py's comment on CPU and memory changed: the machine its card sets (an A100-40GB's 12 cores and 85 GiB), or what it declares where that is more.
    Use v6
    flux-3-action-droid v6shell
    # start a policy from this version, then deploy it
    # my_policy_user: the name you choose
    seq init flux-3-action-droid@v6 my_policy_user
    seq deploy my_policy_user
    
    # or move a deployment made from this template to it
    seq policy upgrade my_policy_user --to flux-3-action-droid@v6
    
    # on the robot's machine: this version's client.py and its README
    pip install -U sequence-ai
    seq init flux-3-action-droid@v6 my_policy_user

    In the console, Deploy deploys this version: the console deploys the stable one.

  2. v5

    FLUX 3 Action as v4; the README gives its card's price, and its whole machine's CPU and memory apart.

    Needs sequence-ai 0.56.0 or later.

    • v4's files; only the README's price changed: $4.19 an hour for the A100-40GB, and its machine's 12 cores and 85 GiB at the serving rates for CPU and memory.
    Use v5
    flux-3-action-droid v5shell
    # start a policy from this version, then deploy it
    # my_policy_user: the name you choose
    seq init flux-3-action-droid@v5 my_policy_user
    seq deploy my_policy_user
    
    # or move a deployment made from this template to it
    seq policy upgrade my_policy_user --to flux-3-action-droid@v5
    
    # on the robot's machine: this version's client.py and its README
    pip install -U sequence-ai
    seq init flux-3-action-droid@v5 my_policy_user
  3. v4

    FLUX 3 Action as before; its README priced an A100-40GB with its CPU and memory included, as prices were then.

    Needs sequence-ai 0.56.0 or later.

    • A worker was billed its card alone then, its price holding its CPU and memory; v5's README gives the prices that replaced it.
    Use v4
    flux-3-action-droid v4shell
    # start a policy from this version, then deploy it
    # my_policy_user: the name you choose
    seq init flux-3-action-droid@v4 my_policy_user
    seq deploy my_policy_user
    
    # or move a deployment made from this template to it
    seq policy upgrade my_policy_user --to flux-3-action-droid@v4
    
    # on the robot's machine: this version's client.py and its README
    pip install -U sequence-ai
    seq init flux-3-action-droid@v4 my_policy_user
  4. v3

    FLUX 3 Action as before; preprocess names the exterior cameras exterior_1 and exterior_2, as cosmos-droid does.

    Needs sequence-ai 0.56.0 or later.

    • preprocess(wrist=..., exterior_1=..., exterior_2=..., ...) replaces left= and right=; pass each DROID camera as RGB, img[..., :3][..., ::-1]: a 4-channel frame is refused.
    • The loop sleeps to each step's tick rather than a fixed 1/15 s, so the arm keeps 15 steps a second.
    • Run client.py from Python 3.10 or later on the robot's laptop: DROID's own environment is Python 3.7.
    Use v3
    flux-3-action-droid v3shell
    # start a policy from this version, then deploy it
    # my_policy_user: the name you choose
    seq init flux-3-action-droid@v3 my_policy_user
    seq deploy my_policy_user
    
    # or move a deployment made from this template to it
    seq policy upgrade my_policy_user --to flux-3-action-droid@v3
    
    # on the robot's machine: this version's client.py and its README
    pip install -U sequence-ai
    seq init flux-3-action-droid@v3 my_policy_user
  5. v2

    FLUX 3 Action as before, with no Hugging Face token of yours: each deployment reads the weights itself.

    Needs sequence-ai 0.56.0 or later.

    Use v2
    flux-3-action-droid v2shell
    # start a policy from this version, then deploy it
    # my_policy_user: the name you choose
    seq init flux-3-action-droid@v2 my_policy_user
    seq deploy my_policy_user
    
    # or move a deployment made from this template to it
    seq policy upgrade my_policy_user --to flux-3-action-droid@v2
    
    # on the robot's machine: this version's client.py and its README
    pip install -U sequence-ai
    seq init flux-3-action-droid@v2 my_policy_user
  6. v1

    Black Forest Labs' FLUX 3 Action for a Franka Panda on DROID, on an A100, with client.py for the robot.

    Needs sequence-ai 0.56.0 or later, and a Hugging Face token of your own that has accepted the model's license: seq secret set hf --env HF_TOKEN.

    Use v1
    flux-3-action-droid v1shell
    # start a policy from this version, then deploy it
    # my_policy_user: the name you choose
    seq init flux-3-action-droid@v1 my_policy_user
    seq deploy my_policy_user
    
    # or move a deployment made from this template to it
    seq policy upgrade my_policy_user --to flux-3-action-droid@v1
    
    # on the robot's machine: this version's client.py and its README
    pip install -U sequence-ai
    seq init flux-3-action-droid@v1 my_policy_user