Quickstart Guide

This page will walk you through running inference with the pre-trained Cosmos-Reason1-7B model on a sample video.

Basic Example

Run the inference_sample.py script to instruct the Cosmos-Reason1-7B model to describe what happens in the sample video.

./scripts/inference_sample.py

Complete Examples

The inference.py example script allows you to perform the following inference tasks:

• Caption the sample video:

  ./scripts/inference.py --prompt prompts/caption.yaml --videos assets/sample.mp4 -v
  

• Ask a question about the sample video with reasoning:

  ./scripts/inference.py --prompt prompts/question.yaml --question 'What are the potential safety hazards?' --reasoning --videos assets/sample.mp4 -v
  

• Temporally caption the sample video and save the input frames to outputs/temporal_caption_text for debugging:

  ./scripts/inference.py --prompt prompts/temporal_caption_text.yaml --videos assets/sample.mp4 --timestamp -v -o outputs/temporal_caption_text
  

Configuring Inference

You can configure inference by editing the following files:

• Prompts: prompts/README.md

• Sampling Parameters: configs/sampling_params.yaml

• Vision Processor Configuration: configs/vision_config.yaml

Next Steps

The Using Cosmos-Reason1 for Rejection Sampling page outlines how to use Cosmos-Reason1 to perform rejection sampling on videos produced by Cosmos-Predict.

Refer to the Reason1 Post-Training Guide to learn more about post-training the Reason1 model.