Skip to content

feat: add MolmoAct2-BimanualYAM inference quickstart#2

Merged
AndresNinou merged 1 commit into
mainfrom
mayor/add-molmoact-quickstart
May 23, 2026
Merged

feat: add MolmoAct2-BimanualYAM inference quickstart#2
AndresNinou merged 1 commit into
mainfrom
mayor/add-molmoact-quickstart

Conversation

@AndresNinou
Copy link
Copy Markdown
Collaborator

@AndresNinou AndresNinou commented May 23, 2026

Live-tested today: authorize end-to-end works in 422ms, worker responds in 109ms.

$ export REFLEX_API_KEY=rfx_...
$ python3 quickstart_molmoact.py
✓ session authorized in 422 ms
  session_id:   kn789etxbh93svjrvqr7bxgg6s879yfs
  worker_url:   https://reflex-inc--reflex-inference-molmoact-webrtc-baseline...
  token:        rfx_session_kn789etxbh93svjrvqr7bxgg…
✓ worker responded in 109 ms
  health: {'ok': True, 'engine': 'webrtc-rust', ...}
✓ All checks passed

Proves PROD.1 shipped — new users can sign up, top up, mint key, run inference.


View with Codesmith Autofix with Codesmith
Need help on this PR? Tag @codesmith with what you need. Autofix is disabled.

Summary by CodeRabbit

  • New Features

    • Added a new quickstart script for verifying MolmoAct2-BimanualYAM model access with API key resolution and worker health verification.
  • Documentation

    • Added guide for MolmoAct2-BimanualYAM inference with setup instructions, authentication flow, and closed-loop bimanual demo integration via Reflex API.

Review Change Stack

PROD.1 is live — anyone with an rfx_* API key + balance can now authorize
sessions for the molmoact2-bimanualyam baseModel and get routed to our
Modal BASELINE worker (us-west, B200).

quickstart_molmoact.py:
  - Uses requests/numpy/pillow (no SDK required for the auth-test flow)
  - Calls publicApi:authorizeSession with the user's API key
  - Verifies worker health with the returned session token
  - Live-tested: 422ms authorize + 109ms worker health
  - Points to the full reflex-inc/reflex cli for closed-loop arm demos

README:
  - New section 'MolmoAct2-BimanualYAM Inference (NEW)' inserted before
    Prerequisites
  - Documents pricing ($10/hr per GPU-second, ~$0.001 per inference),
    quality (visually lossless, PSNR 38.8 dB), latency (220ms p50 RTT)
@AndresNinou AndresNinou merged commit 50b53ca into main May 23, 2026
0 of 2 checks passed
@coderabbitai
Copy link
Copy Markdown

coderabbitai Bot commented May 23, 2026

Caution

Review failed

The pull request is closed.

ℹ️ Recent review info
⚙️ Run configuration

Configuration used: Organization UI

Review profile: ASSERTIVE

Plan: Pro Plus

Run ID: ee3be011-fdb3-4da7-8899-24ff09cf5c5e

📥 Commits

Reviewing files that changed from the base of the PR and between 512d8d5 and 9469257.

📒 Files selected for processing (2)
  • README.md
  • quickstart_molmoact.py

📝 Walkthrough

Walkthrough

This PR adds a new quickstart_molmoact.py script that serves as an end-to-end verification tool for MolmoAct2-BimanualYAM access. It resolves an API key, authorizes a session via Convex, confirms worker liveness through a health check, and documents the workflow in the README with setup instructions and next steps for the closed-loop demo.

Changes

MolmoAct2-BimanualYAM Quickstart Script

Layer / File(s) Summary
Module setup and configuration
quickstart_molmoact.py
Module docstring, imports, and constants (Convex URL, model/robot identifiers, image size, state dimension) establish script identity and backend connectivity.
Credential resolution and Convex communication
quickstart_molmoact.py
Helper functions _resolve_api_key() and _convex_mutation() load API keys from environment or local files and POST to Convex with JSON response parsing and error handling.
Session authorization
quickstart_molmoact.py
authorize_session(api_key) calls the Convex publicApi:authorizeSession mutation with model and robot type, handles specific failure reasons with user-facing messages, and prints session metadata before returning the result.
Inference verification and health check
quickstart_molmoact.py
_synthetic_observation() generates deterministic gradient images and JPEG-encodes them for three camera views. run_inference(session) performs an authenticated /health request to verify worker responsiveness and prints guidance for the full WebRTC-based inference path.
Main orchestration and user documentation
quickstart_molmoact.py, README.md
main() orchestrates API key resolution, session authorization, and worker health check, printing final status and next steps. README documents the new script with installation commands, what it verifies, pricing/quality/latency notes, and optional steps for the closed-loop bimanual YAM demo.

Estimated code review effort

🎯 2 (Simple) | ⏱️ ~12 minutes

✨ Finishing Touches
📝 Generate docstrings
  • Create stacked PR
  • Commit on current branch
🧪 Generate unit tests (beta)
  • Create PR with unit tests
  • Commit unit tests in branch mayor/add-molmoact-quickstart

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

Comment @coderabbitai help to get the list of available commands and usage tips.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant