ReinforceNowReinforceNow

Quickstart

We will finetune a Qwen/Qwen3-8B using supervised finetuning (SFT) on example conversations. SFT trains the model to mimic high-quality responses from your dataset.

After completing the installation, paste this prompt into Claude Code:

Claude Code Prompt
Finetune Qwen3-8B with RL on this math reasoning dataset:
https://huggingface.co/datasets/ReinforceNow/rl-single-math-reasoning

Option B: Manual Setup

Step 1: Create a new folder and fetch the template:

rnow init --template sft

Step 2: Start the training run:

rnow run

Your terminal should look similar to this:

Terminal Run

Step 3: View metrics and traces:

Navigate to your experiment link to view your model's performance. You can share your traces by clicking the share button, like this one: https://www.reinforcenow.ai/shared/runs/cmixqdocj000004l8cbkg2bkp

Quickstart Graphs
Quickstart Traces

You can share your experiment traces by clicking the share button.

Next steps

Create Your First Reward

Write your first reward function to train a model

Train Your First Agent

Train an agent that can use tools with RL

CLI Reference

Complete reference for all CLI commands

Overview

Install the ReinforceNow CLI

Create Your First Reward

Write your first reward function to train a model on OpenMathReasoning

On this page

Option A: Claude Code (Recommended)Option B: Manual SetupStep 1: Create a new folder and fetch the template:Step 2: Start the training run:Step 3: View metrics and traces:Next steps