AMD-AGI/Primus
Jul 2026 – Jul 2026
1 Month active
Languages Used
No languages
Technical Skills
PyTorchPythonYAML configurationbackend developmentdistributed systemsmachine learning
zirui
PROFILE
Zirui
Developed and integrated a diffusion training backend for the Primus framework, enabling scalable single-node and multi-node training with WAN2.1 and WAN2.2 model support. Leveraged PyTorch and Python to implement a modular architecture for model, dataset, and trainer configurations, facilitating extendibility and maintainability. Enhanced the backend with distributed utilities and custom attention backends to optimize diffusion workloads, and added video generation capabilities for WAN2.1 and WAN2.2. Conducted single-node SFT validation and initiated multi-node benchmarking, capturing performance metrics across multiple GPUs. Documented all changes, provided example scripts, and began refactoring configuration management to streamline diffusion-specific parameter handling.
Overall Statistics
Feature vs Bugs
100%Features
Repository Contributions
1Total
Bugs
0
Commits
1
Features
1
Lines of code
2,686
Activity Months1
Your Network
44 peopleWork History
July 2026
1 Commits • 1 Features
Jul 1, 2026July 2026 — Primus: Integrated diffusion training backend with WAN2.1/WAN2.2 support, enabling scalable single-node and multi-node training within the Primus framework. Implemented modular architecture for model, dataset, and trainer configurations; added WAN2.1/2.2 video generation support and distributed utilities to optimize diffusion workloads. Completed single-node SFT validation for WAN2.1-1.3B and WAN2.2-5B; ongoing 2-node validation. Documented changes and added example scripts. Benchmarks captured for WAN2.2 TI2V 5B; results logged and reviewed. Ongoing cleanup includes dataset preprocessing improvements and a config refactor for diffusion-specific parameters.
1 Commits • 1 Features
Jul 1, 2026July 2026 — Primus: Integrated diffusion training backend with WAN2.1/WAN2.2 support, enabling scalable single-node and multi-node training within the Primus framework. Implemented modular architecture for model, dataset, and trainer configurations; added WAN2.1/2.2 video generation support and distributed utilities to optimize diffusion workloads. Completed single-node SFT validation for WAN2.1-1.3B and WAN2.2-5B; ongoing 2-node validation. Documented changes and added example scripts. Benchmarks captured for WAN2.2 TI2V 5B; results logged and reviewed. Ongoing cleanup includes dataset preprocessing improvements and a config refactor for diffusion-specific parameters.
July 2026
Activity
Loading activity data...
Quality Metrics
Correctness80.0%
Maintainability80.0%
Architecture100.0%
Performance80.0%
AI Usage80.0%
Skills & Technologies
Programming Languages
No languages yet
Technical Skills
PyTorchPythonYAML configurationbackend developmentdistributed systemsmachine learning
Repositories Contributed To
Overview of all repositories you've contributed to across your timeline