Verified:GitHub

tao-finetune-cosmos-reason

Automate fine-tuning of video QA models with SFT and LoRA.

83|20|Updated Apr 16, 2026

Search Tags:#LoRA #parallelism #SFT #AutoML #model fine-tuning #video QA #Cosmos3-Nano

One-click install

npx skills add https://github.com/NVIDIA-TAO/tao-skill-bank --skill tao-finetune-cosmos-reason

Or copy as Structured Prompt for Agent▼

Please help me install this Agent Skill.
Skill: tao-finetune-cosmos-reason
Source: https://github.com/NVIDIA-TAO/tao-skill-bank/tree/main/skills/models/tao-finetune-cosmos-reason
Command: npx skills add https://github.com/NVIDIA-TAO/tao-skill-bank --skill tao-finetune-cosmos-reason

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires huggingface_hub, torch, transformers, and includes scripts (resource) and references (resource) and assets (resource) components.

What problem does it solve?

This Skill streamlines the process of fine-tuning video QA models like Cosmos3-Nano with Supervised Fine-Tuning (SFT) and LoRA (Low-Rank Adaptation), reducing the complexity of the workflow for model training and evaluation.

Core Features & Use Cases

Fine-Tuning: Automate the fine-tuning process of video QA models with SFT and LoRA.
Parallelism: Utilize FSDP-based parallelism for GPU and node configurations.
Evaluation: Evaluate the performance of fine-tuned models using provided metrics.
AutoML: Leverage AutoML for hyperparameter optimization and model tuning.
Use Case: A researcher can use this Skill to fine-tune a video QA model on a new dataset, saving time on manual hyperparameter tuning and allowing for faster iterations.

Quick Start

To start the fine-tuning process, use the tao-finetune-cosmos-reason skill with the required parameters, such as the dataset paths and the base model path.

Frequently Asked Questions about tao-finetune-cosmos-reason

High-intent search queries and answers about installing and using this skill.

FAQPage Schema

How do I fine-tune a video QA model with LoRA and SFT?▼

This Skill automates video QA model fine-tuning using SFT and LoRA, requiring dataset paths and a HuggingFace base model path to initiate the training workflow.

Can I use distributed training across multiple GPUs for video QA fine-tuning?▼

Yes, video QA fine-tuning supports distributed training across multiple GPUs and nodes using FSDP-based parallelism for scalable model training.

Do I need Docker and nvidia-container-toolkit to run LoRA fine-tuning?▼

Yes, executing LoRA fine-tuning requires Docker and nvidia-container-toolkit to properly configure the GPU environment and run the training containers.

Does the fine-tuning workflow include hyperparameter optimization?▼

Yes, the fine-tuning workflow includes AutoML features for hyperparameter optimization and model tuning, reducing manual effort and enabling faster iterations.

How are fine-tuned video QA models evaluated?▼

Fine-tuned video QA models are evaluated using provided metrics within the workflow, leveraging HuggingFace models for base training and performance evaluation.