Commit Graph
22 Commits
Author SHA1 Message Date
lihaoyun6 3b0699ead9 Merge remote-tracking branch 'upstream/nightly' 2025-08-12 22:28:36 +08:00
Adrien Toupet d075dde799 feat: Extract model caching to main node + add hierarchical debug system
- Move cache_model and enable_debug options from BlockSwap to main SeedVR2 node

- cache_model: Keep models in RAM between runs (skip reload for faster iterations)

- enable_debug: Show detailed memory/timing tracking with hierarchical display

- Replace hardcoded debug statements with unified Debug class

- Add visual icons for log categorization

- Implement parent-child timing relationships for operation breakdown

- Add VRAM/RAM usage tracking with before/after comparisons

Model caching now available for all workflows, not just BlockSwap users.

Debug system provides granular performance insights when enabled.
2025-08-08 00:49:46 +02:00
Adrien Toupet 4118595968 Revert to 5c65c44 and fix integration issues
- Resolved merge conflicts

- Added missing constants.py and model_registry.py files

- Added graceful error handling for failed model downloads
2025-08-08 00:49:46 +02:00
lihaoyun6 3b530dc983 Added MPS backend support (for running on macOS) 2025-08-06 18:52:33 +08:00
Adrien Toupet 7b63eeea4c Revert "Models update - FP8, GGUF, Artifacts" 2025-07-24 12:53:46 -04:00
Adrien Toupet bc8f0aa6d0 Optimized RoPE stability fix and Patch RoPE for blockswap 2025-07-24 09:32:18 -04:00
Adrien Toupet b551e32ae1 Fixed scoping in _stabilize_rope_computations 2025-07-24 01:05:49 -04:00
Adrien Toupet 99938bf175 Apply RoPE stability fix universally to prevent artifacts in all precision modes 2025-07-23 22:01:55 -04:00
Adrien Toupet 09a9bf081d Fix artifacts when using mixed precision model without blockswap 2025-07-23 21:09:27 -04:00
Adrien Toupet bc292f26bd Fix artifacts when using mixed precision model without blockswap 2025-07-23 20:56:02 -04:00
Adrien Toupet 63defbd150 Tentative fix for artifacts without blockswap 2025-07-23 17:36:08 -04:00
Adrien Toupet d38a3956b9 Tentative fix for artifacts without blockswap 2025-07-23 17:21:45 -04:00
Adrien Toupet d87fca3ee2 Tentative fix for artifacts without blockswap 2025-07-23 17:10:59 -04:00
Adrien Toupet 5b38f1d9af Tentative fix for artifacts without blockswap 2025-07-23 17:01:36 -04:00
Adrien Toupet 2f02db68b4 Tentative fix for artifacts without blockswap 2025-07-23 16:52:51 -04:00
Adrien Toupet 468d8a29ba Tentative fix for artifacts without blockswap 2025-07-23 16:41:21 -04:00
Adrien Toupet 6feb33ac23 Tentative fix for artifacts without blockswap 2025-07-23 15:46:49 -04:00
Adrien Toupet 5fa2383b6a Added logging for artifact debugging 2025-07-23 12:10:42 -04:00
Adrien Toupet e30fce3a35 Optimize FP8 handling to preserve native FP8 weights for memory efficiency while converting only tensors during computation 2025-07-22 23:09:34 -04:00
Adrien Toupet a0bf32cb97 Reverted bf16 conversion logic - don't force it 2025-07-09 09:29:56 -04:00
Adrien Toupet 50dfb771e6 Add BlockSwap support 2025-07-04 21:28:47 -04:00
NumZ 2786c05faa Speed Up 30 to 50%, fix Memory Leak, refacto) 2025-06-30 12:46:41 +02:00