rnow-cli

name: rnow-cli description: Use the ReinforceNow CLI for RLHF training. Use when running rnow commands, initializing projects, submitting training runs, testing rollouts, running evals, or downloading models. Triggers on "rnow", "rnow init", "rnow run", "rnow test", "rnow eval", "rnow download", "rnow deploy", "rnow login", "training run".

ReinforceNow CLI Reference

The rnow CLI manages RLHF training projects on the ReinforceNow platform.

Installation

pip install rnow

Command Overview

rnow login

Authenticate using OAuth device flow.

rnow login [OPTIONS]

Example:

rnow login
# Opens browser for authentication
# Stores credentials in ~/.reinforcenow/credentials.json

rnow logout

Remove stored credentials.

rnow logout

rnow status

Check authentication status and running jobs.

rnow status

Output:

Logged in as: user@example.com
Organization: My Team (org_abc123)
Active runs: 2
  - run_xyz789 (running) - Math Training
  - run_def456 (queued) - Code Agent

rnow orgs

List or select organizations.

# List all organizations
rnow orgs

# Select an organization
rnow orgs ORG_ID

Example:

rnow orgs
# Output:
# * org_abc123 - My Team (owner)
#   org_def456 - Other Team (member)

rnow orgs org_def456
# Switched to: Other Team

rnow init

Initialize a new project from a template.

rnow init [OPTIONS]

Available Templates

Examples:

# Create SFT project
rnow init --template sft --name "my-sft-project"

# Create RL project with tools
rnow init --template rl-tools

# Create from tutorial
rnow init --template tutorial-reward

Generated Files

rnow run

Submit project for training. Supports inline overrides for any config.yml setting.

rnow run [OPTIONS] [OVERRIDES...]

Options

Inline overrides

Any config.yml setting can be overridden with key=value arguments. These are applied to the uploaded config (not just in-memory), so the trainer sees the overridden values.

# Model override (same as -m flag)
rnow run model.path=Qwen/Qwen3-4B-Instruct-2507
rnow run -m Qwen/Qwen3-4B-Instruct-2507  # equivalent shorthand

# Training overrides
rnow run data.batch_size=8 data.group_size=16 trainer.num_epochs=5
rnow run algorithm.adv_estimator=grpo trainer.learning_rate=0.0002

# Rollout overrides (RL only)
rnow run rollout.max_turns=3 rollout.max_context_window=16384
rnow run rollout.reasoning_mode=disabled  # Turn off reasoning for hybrid models

# LoRA rank override
rnow run model.qlora_rank=16

Common overrides

Multi-model training

If model.path is a list in config.yml, a separate run is submitted for each model:

model:
  path:
    - Qwen/Qwen3-8B
    - Qwen/Qwen3-4B-Instruct-2507
    - meta-llama/Llama-3.1-8B-Instruct

Required files

• config.yml - Configuration
• train.jsonl - Training data
• rewards.py - Reward functions (RL only)

Optional files

• tools.py - Tool definitions
• requirements.txt - Python dependencies

Examples

# Basic run
rnow run

# Override model and epochs
rnow run -m Qwen/Qwen3-4B-Instruct-2507 -e 3

# Train a finetuned model (continue training from checkpoint)
rnow run -m 37e6f995-0efd-4fc6-beaa-badb7af94054

# Override multiple settings inline
rnow run data.batch_size=8 trainer.learning_rate=0.0002 rollout.max_turns=5

rnow stop

Cancel an active training run.

rnow stop RUN_ID [OPTIONS]

Example:

rnow stop run_abc123xyz
# Are you sure you want to stop this training run? [y/N]: y
# Save model checkpoint? [y/N]: n
# Run stopped.

rnow stop run_abc123xyz --save-model -y
# Stopping and saving checkpoint...

rnow test

Test RL rollouts before submitting. Runs rollouts via the cloud (Modal), not locally.

rnow test [OPTIONS]

Supported models for --model

• OpenAI models: gpt-5-nano (default, fastest), gpt-5-mini, gpt-5.2, gpt-5.2-pro
• GPU models: Qwen/Qwen3-8B, Qwen/Qwen3-4B-Instruct-2507, etc.
• Finetuned models: UUID (e.g., 37e6f995-0efd-4fc6-beaa-badb7af94054)

Examples

# Basic test (1 rollout, random entry, gpt-5-nano)
rnow test

# Multiple rollouts
rnow test -n 5

# Test specific entries
rnow test -e 0,3,7

# Test all entries
rnow test --all

# Override model
rnow test --model gpt-5.2 -n 3

# Test finetuned model
rnow test --model 37e6f995-0efd-4fc6-beaa-badb7af94054 -n 3

# Test GPU model
rnow test --model Qwen/Qwen3-8B -n 3

rnow eval

Run model evaluation and calculate pass@k metrics.

rnow eval [OPTIONS]

Two modes:

1. From project directory (uploads files)

rnow eval --model Qwen/Qwen3-8B --pass1 --pass8

Uses config.yml, train.jsonl, rewards.py, tools.py from the project directory.

2. From existing eval (reuses files)

rnow eval --eval-id cmle8ma5h000004l44thj3t3a --model gpt-5-nano --pass1

Reuses all files from the source eval. Only need to specify --model.

Options

Reasoning Modes

When --reasoning-mode is not set (null), each model uses its default:

Use disabled as the universal "turn off reasoning" value (works for all model types).

OpenAI Model Compatibility

Examples

# Eval finetuned model on existing eval
rnow eval --eval-id cmle8ma5h000004l44thj3t3a \
  --model 37e6f995-0efd-4fc6-beaa-badb7af94054 --pass1

# Eval with reasoning mode
rnow eval --eval-id cmle8ma5h000004l44thj3t3a \
  --model gpt-5.2 --reasoning-mode high --pass1 --pass8

# Eval from project directory with overrides
rnow eval --model Qwen/Qwen3-8B --pass1 --pass8 \
  --max-turns 5 --temperature 0.8

# Limit to first 50 samples
rnow eval --eval-id cmxyz123 --model gpt-5-nano \
  --pass1 --max-samples 50

rnow download

Download a trained model checkpoint.

rnow download MODEL_ID [OPTIONS]

Example:

rnow download abc123 -o ./my-model
# Downloading checkpoint...
# Progress: 100%
# Saved to: ./my-model/

rnow download abc123 --keep-archive
# Keeps both extracted files and .tar archive