3 Commits

Include renames

Author	SHA1	Message	Date
iGavroche	281472e0e4	v2.1.1: shared arch detection, sampler auto-detect for z-image/ernie/ideogram4 ... - Extract detect_architecture() to shared rocm_nodes/utils/architecture.py - ROCm KSampler now auto-detects GPU architecture + model type at runtime: - Flow-matching models (z-image, ernie, ideogram4) logged and optimized - Pixel-space models (z-image-turbo) recognized — no VAE decode needed - High-memory models (ideogram4, factor 11.6x) get automatic cleanup - Architecture-aware backend settings (TF32, fp16 accumulation) - ROCmSamplerPerformanceMonitor now model-type + architecture aware - ROCmVAEPerformanceMonitor uses shared architecture detection - Fix ProgressBar mock in test conftest - Add flow-matching and pixel-space model detection tests - Update KSampler/PerfMonitor README docs	2026-06-12 17:49:56 -04:00
iGavroche	c989a48aab	v2.1.0: LTX/z-image VAE support, temporal tiling, fp16 default, APU tuning ... - LTX Video VAE detection (128 channels, 32x spatial compression, 8x temporal) - z-image / z-image-turbo pixel-space VAE passthrough (no decode needed) - fp16 default on all AMD GPUs (Strix Halo, RDNA 2/3, CDNA) - Per-architecture tuning with automatic detection (gfx1151/gfx1100/gfx1030/CDNA) - Temporal tiling for long videos: overlapping chunks with linear blend, zero frame loss, progress bar in UI (disabled by default) - Chunked IO bypass for LTX — fixes OOM with max_split_size_mb caps - Fixed memory estimation: correct for all dims (including 5D temporal) and dtypes - APU-aware memory management (psutil for unified memory on Strix Halo) - Tile size floor for high-compression VAEs (min 64 latent pixels) - 25 unit tests covering all new features	2026-06-12 15:54:56 -04:00
Nino	5d515f035e	docs: add Upgrade to v2 section; fix ROCm branding in title\nchore: add tools/upgrade_to_v2.py migration helper\nrefactor: finalize v2 structure and cleanup wording	2025-10-30 14:47:52 -04:00