Data & Analysis ✓ NVIDIA · Official video-question-answeringcosmos3-nanosupervised-fine-tuninglorafsdpcosmos-rlhuggingfaceautoml

Cosmos Reason Video QA Fine-Tuning

A Cosmos-RL playbook for FSDP-based SFT, LoRA, evaluation, and AutoML of Cosmos3-Nano video QA models.

FollowSkills review · FSRS-2.0
Not recommended
53/ 100 5-point scale 2.7 / 5
Trust18 / 25 · 3.6/5

The documentation discloses HF_TOKEN use, gated-model access, Docker environment passing, data paths, model provenance, and preflight requirements, with some user confirmation points. However, Bash is broadly allowed, there is no complete least-privilege boundary, universal launch confirmation, rollback procedure, or detailed token-lifecycle guidance, so points are deducted. No malware, covert exfiltration, credential theft, or destructive default was shown.

Reliability7 / 20 · 1.8/5

Training, evaluation, dataset validation, error patterns, and checkpoint handling are documented in substantial detail. Static calibration caps the score at 10, and the materials contain important inconsistencies, including model_max_length values of 40960 versus 81920, llava versus daft format claims, and references to schemas, skill_info, scripts, and platform skills not supplied here, so points are deducted.

Adaptability10 / 15 · 3.3/5

Trigger phrases, target model, train/evaluate/inference/quantize scenarios, and some non-fit boundaries are reasonably clear, with hardware, Docker, and dataset requirements stated. Chinese-language interaction and localization are not documented, and the core workflow depends on Hugging Face, Docker, S3, and high-end NVIDIA GPUs without evidence of mainland-China reachability, so points are deducted.

Convention8 / 15 · 2.7/5

The material uses progressive disclosure and covers credentials, data contracts, parameters, error patterns, AutoML, and checkpoint handling; it also states Apache-2.0 licensing, version 0.1.0, and NVIDIA authorship. However, standard Instructions and Examples sections are missing, the author format is incomplete, hierarchy and file naming are inconsistent, and no clear changelog, maintenance owner, or update path is provided, so points are deducted.

Effectiveness6 / 15 · 2.0/5

The skill covers Cosmos3-Nano video-QA SFT/LoRA, evaluation, AutoML, dataset preflight, and troubleshooting, so it could produce directly useful configuration and workflow guidance. Under static calibration the score cannot exceed 7; the benchmark contains only one positive task, reports weak Codex correctness, discoverability, and efficiency, and provides no representative verified training outputs, so points are deducted.

Verifiability4 / 10 · 2.0/5

The revision-pinned repository context, one evaluation task, expected behavior, and NVSkills-Eval report provide limited audit evidence. Coverage is only one task with no negative-trigger cases, the report does not independently reproduce key training paths, and no committed test suite or CI coverage is supplied, so points are deducted.

Evidence confidence:Low Reviewed Jul 20, 2026 Reviewed revision 55f18499943e
The upstream repository has new commits since this review. The score still applies to the reviewed revision shown and may not cover the latest changes.
Before you use it
  • Different reference files conflict on model_max_length, dataset formats, and template behavior; unify the authoritative configuration and add consistency checks before publication.
  • Training depends on gated Hugging Face access, HF_TOKEN, Docker/NVIDIA Container Toolkit, S3 or high-end GPUs; reachability and resource compatibility should be validated separately for mainland-China users and target platforms.
  • The documentation recommends downloads and training but does not define a unified confirmation flow for irreversible actions, token least privilege, or rollback after failure.
  • The benchmark covers only one positive task, with weak Codex correctness, discoverability, and efficiency results, so it is not comprehensive execution validation.
See the full review method →

What does this skill do, and when should you use it?

This skill guides training and evaluation of Cosmos3-Nano and compatible Cosmos Reason video question-answering models with Cosmos-RL. It covers dataset preflight, nested spec construction, FSDP parallelism, SFT, LoRA evaluation, checkpoint conversion, and AutoML/HPO. The packaged default base model is nvidia/Cosmos3-Nano from Hugging Face, which may require HF_TOKEN access. Docker, nvidia-container-toolkit, and suitable NVIDIA GPUs are required.

Loads packaged action templates, schemas, and reference guidance; applies user overrides to nested YAML specifications; validates annotation JSON and visible media paths; runs Cosmos-RL training, evaluation, inference, or quantization; routes eligible training runs through AutoML; converts native Cosmos3 checkpoints to a Qwen3-VL Hugging Face safetensors directory with the packaged Python helper when required by the selected image; and supports checkpoint, result, and metric handling.

  1. A team fine-tuning Cosmos3-Nano for video question answering on one or more NVIDIA GPU nodes.
  2. A researcher applying SFT or LoRA to a Cosmos Reason video QA model and evaluating the result.
  3. An engineer launching training or evaluation against llava or daft VLM datasets stored locally, on S3, or on a mounted filesystem.
  4. A practitioner searching training hyperparameters with the skill's AutoML/HPO workflow.
  5. An operator whose Cosmos-RL image cannot load the native Cosmos3 checkpoint format and needs the upstream conversion path.

What are this skill's strengths and limitations?

Pros
  • Covers launch intake, preflight, training, evaluation, inference, quantization, LoRA evaluation, AutoML, and checkpoint conversion.
  • Documents the nonstandard Cosmos-RL FSDP parameters dp_shard_size and dp_replicate_size for GPU and node sizing.
  • Includes concrete requirements for dataset formats, annotation validation, media visibility, model paths, frame handling, and common failure patterns.
  • Can be installed as a single skill through the NVIDIA skills CLI.
Limitations
  • Requires Docker, nvidia-container-toolkit, and appropriate NVIDIA hardware; the guidance recommends 8×80GB A100 or H100 GPUs for the 8B-parameter models.
  • Gated Hugging Face models require an accepted model agreement and a read-capable HF_TOKEN.
  • Some images require a separate Cosmos3-to-Qwen3-VL checkpoint conversion before training or evaluation.
  • The supplied material does not provide standalone benchmark results or independent platform-validation evidence for this skill.

How do you install this skill?

Install the individual skill with the repository's documented skills CLI flow:

npx skills add nvidia/skills --skill tao-finetune-cosmos-reason --yes

The supplied README does not document a separate manual installation procedure for this skill.

How do you use this skill?

After installation, prompt an agent with a supported request such as "fine-tune Cosmos3 Nano on my video QA dataset", "Run Cosmos-RL SFT for nvidia/Cosmos3-Nano", or "Evaluate my Cosmos3-Nano video QA checkpoint". Provide the Docker/GPU environment, dataset locations, and HF_TOKEN when gated Hugging Face access is needed. Training defaults to AutoML when the packaged train schema and template are available; explicitly requesting disabled AutoML or plain training applies an off override for that run.

FAQ

Does the skill require the default NVIDIA model?
No. nvidia/Cosmos3-Nano is the packaged default. The skill says to override it only when the user explicitly requests another base model, with access approval required for that target model.
Why might checkpoint conversion be necessary?
Some Cosmos-RL images cannot load the native Cosmos3 Omni checkpoint format. In that case, the documented upstream Cosmos Framework conversion path produces a Qwen3-VL Hugging Face safetensors directory used consistently as the training and evaluation model path.
What data formats are supported?
The skill specifies the vlm dataset type with llava and daft formats, and accepts training, evaluation, and testing intents. It supports dataset-root mode or direct annotation and media paths.
Can AutoML be disabled?
Yes. AutoML is enabled at the model layer and is the default for eligible train runs, but an explicit automl_policy: off or a request such as "disable AutoML" selects direct model training for that run.

More skills from this repository

All from NVIDIA/skills

Data & Analysis ✓ NVIDIA · Official

Hugging Face GPU Fine-Tuning

Build reproducible Hugging Face fine-tuning pipelines on local NVIDIA GPUs.

Data & Analysis ✓ NVIDIA · Official

Cosmos-Embed Video Retrieval & Fine-Tuning

Use Cosmos-Embed1 to embed, retrieve, deduplicate, and fine-tune video-text data.

Data & Analysis ✓ NVIDIA · Official

TAO MAL Weakly Supervised Segmentation

Train segmentation models from point or box annotations and produce masks with minimal labeling.

Data & Analysis ✓ NVIDIA · Official

TAO MAE Vision Pretraining

Guides NVIDIA TAO through masked-autoencoder pretraining and fine-tuning for visual representations.

Data & Analysis ✓ NVIDIA · Official

TAO Video Action Recognition

Train, evaluate, export, and run inference for RGB, optical-flow, and multi-stream video action models with NVIDIA TAO.

Dev & Engineering ✓ NVIDIA · Official

NeMo MBridge GPU Memory Tuning

Diagnose GPU OOMs in Megatron Bridge training and reduce fragmentation, activation, and PEFT memory usage.

Data & Analysis ✓ NVIDIA · Official

TAO DINO Object Detection

Run NVIDIA TAO workflows for training and deploying DINO 2D object detectors.

Dev & Engineering ✓ NVIDIA · Official

NeMo MBridge MoE VLM Training Guide

Guides FSDP and 3D-parallel choices and tuning for MoE vision-language model training.

Data & Analysis ✓ NVIDIA · Official

TAO Grounding DINO Open-Vocabulary Detection

Train and deploy text-guided object detectors without a fixed class vocabulary.

Data & Analysis ✓ NVIDIA · Official

TAO AutoML Tuning

Run governed hyperparameter optimization for NVIDIA TAO networks across supported SDK platforms.

Data & Analysis ✓ NVIDIA · Official

TAO OneFormer Universal Segmentation

Train and deploy one task-conditioned model for panoptic, instance, and semantic image segmentation.

Data & Analysis ✓ NVIDIA · Official

TAO SegFormer Semantic Segmentation

Guidance for training, evaluating, exporting, and deploying TAO SegFormer models.

Data & Analysis ✓ NVIDIA · Official

TAO Standard Training Workflow

Run a controlled TAO train, evaluate, and export workflow on labeled data.

Dev & Engineering ✓ NVIDIA · Official

TAO Hugging Face Model Integrator

Connect Hugging Face computer-vision models to NVIDIA TAO training, ONNX export, and TensorRT deployment workflows.

Data & Analysis ✓ NVIDIA · Official

TAO AutoML + DEFT Three-Phase Pipeline

Chains hyperparameter tuning, iterative data improvement, and final refinement for AOI and other DEFT applications.

Data & Analysis ✓ NVIDIA · Official

Mask Grounding DINO Training & Deployment

Train and deploy text-guided open-vocabulary instance segmentation models.

Dev & Engineering ✓ NVIDIA · Official

Earth2Studio Forecast Wrapper Builder

Build and validate Earth2Studio time-stepping forecast wrappers.

Data & Analysis ✓ NVIDIA · Official

TAO CenterPose Pose Estimation Skill

Guides agents through training, evaluation, export, and deployment of CenterPose models for 6-DoF object pose estimation.

Data & Analysis ✓ NVIDIA · Official

TAO NVPanoptix3D

Train and run 3D panoptic reconstruction with occupancy completion from posed RGB images.

Data & Analysis ✓ NVIDIA · Official

TAO Person Re-Identification

Train and deploy TAO ReID models for cross-camera person matching.

Related skills