2 Commits
Author SHA1 Message Date
Gavroche 59c06ab32c Fix workflow UUID validation and update performance results
- Fixed invalid UUID in example_workflow.json (was string, now proper UUID)
- Updated example_workflow_wan_video.json with new UUID
- Added WAN 2.2 video performance results: 163s → 139s (15% improvement)
- Updated README with video workflow performance metrics
- Both workflows now pass ComfyUI Zod schema validation

Performance Summary:
- Flux 1024x1024: 500s → 110s (78% improvement)
- WAN 2.2 Video (2s, 320x320): 163s → 139s (15% improvement)
2025-10-03 10:57:31 -04:00
Nino 1e3d3a1cc3 Initial release: RocM-Nino v1.0.0
- Complete ROCM optimization suite for AMD GPUs (gfx1151)
- 6 optimized nodes: VAE decode, KSampler, and performance monitors
- ROCm 6.4+ specific optimizations and memory management
- Easy installation with install.py script
- Comprehensive documentation and example workflows
- Expected 20-40% performance improvement for gfx1151 architecture

Features:
- ROCMOptimizedVAEDecode: Main optimized VAE decode
- ROCMOptimizedVAEDecodeTiled: Advanced tiled VAE decode
- ROCMOptimizedKSampler: Optimized sampling with ROCm tuning
- ROCMOptimizedKSamplerAdvanced: Advanced sampler with extended control
- ROCMVAEPerformanceMonitor: VAE performance analysis
- ROCMSamplerPerformanceMonitor: Sampler performance analysis
2025-10-03 09:37:23 -04:00