Hao AI Lab
Official@hao-ai-lab · United States of America
Offers specialized engineering frameworks for FastVideo model parity, weight conversion, and diffusion transformer validation.
Agent Skills by Hao AI Lab
Showing 24 vetted skills indexed across 1 GitHub repositories.
hao-ai-labdreamverse-deploy
Redeploys the Dreamverse backend and frontend on a chosen local GPU with readiness checks.
hao-ai-labci-runner
Manages FastVideo's Slurm-based GPU CI lanes, Buildkite pipeline graph, and GB200 validation workflows.
hao-ai-labreseed-performance-baseline
Re-seed rolling performance baselines in the FastVideo HF tracking dataset from reviewed benchmark JSONs.
hao-ai-labadd-model-06-port-generic
Prototype and parity-debug FastVideo scheduler, upsampler, or vocoder components.
hao-ai-labseed-ssim-references
Generate, verify, and upload SSIM reference artefacts to Hugging Face datasets.
hao-ai-labsearch-related-work
Search indexed markdown files for related papers, repositories, and blog posts.
hao-ai-labindex-related-work
Extract key insights from papers, repositories, or articles into a structured markdown index.
hao-ai-labadd-model-08-trace
Detect layer-level numerical divergence between FastVideo and reference models.
hao-ai-lablog-experiment
Update a Markdown journal with experiment parameters, outcomes, and lessons.
hao-ai-labadd-model-05-port-encoder
Prototype and validate custom encoders for the FastVideo framework.
hao-ai-labadd-model-09-pipeline
Automate FastVideo inference pipeline setup and verification for deployment.
hao-ai-labadd-model-02-parity
Generate parity test scaffolds comparing official models with FastVideo implementations.
hao-ai-labadd-model-04-port-vae
Prototype and parity-debug FastVideo VAE modules for video, image, and audio processing.
hao-ai-labadd-model-01-prep
Prepare FastVideo model assets by cloning repositories and configuring environments.
hao-ai-labadd-model-07-conversion
Convert official FastVideo weights into standardized load-ready components with validation.
hao-ai-labreseed-ssim-references
Back up, regenerate, review, and upload SSIM reference videos for specific models.
hao-ai-labadd-model-03-port-dit
Prototype and parity-debug diffusion transformer components within FastVideo.
hao-ai-labmonitor-experiment
Monitor W&B training runs to detect anomalies and generate alerts.
hao-ai-lablaunch-experiment
Generates torchrun launch commands for FastImage training pipelines and datasets.
hao-ai-labsummarize-run
Retrieve W&B or local metrics and generate markdown comparison reports.
hao-ai-labadd-model
Stage reference code, convert weights, and validate compatibility for FastVideo models.
hao-ai-labadd-model-10-pr-review
Review pull requests for model development, configuration, conversion, and pipeline integration.
hao-ai-labdecompose-pipeline-pr
Split large pull requests into smaller reviewable PRs with dependency analysis.
hao-ai-labevaluate-video-quality
Assess generated video quality using SSIM, loss trajectories, and caption consistency.
Frequently Asked Questions About Hao AI Lab
FAQPage SchemaWhat specific tasks can engineers perform using these capabilities?▼
Engineers can perform weight conversion, numerical divergence detection, and parity testing for diffusion-based video models. The system supports prototyping VAE modules, diffusion transformer components, and encoders, while facilitating quality assessment through SSIM metrics and loss trajectory analysis.
Which personas benefit most from these model development skills?▼
These skills are designed for research engineers and machine learning practitioners focused on generative video synthesis. The functionality specifically targets developers responsible for porting official model weights, maintaining parity with reference implementations, and managing complex training pipelines.
What are the prerequisites for running these model integration tasks?▼
Users require an environment configured for FastVideo development, including access to W&B for experiment monitoring and local compute resources for torchrun execution. Dependencies include standard model weight repositories and the specific FastVideo framework structure required for component prototyping and validation.