Commits · c0a970b093efab5a78d43a3022e6810ee7a42a51 · nexlab / coderai · GitLab

Jul 01, 2026
- frontproxy: intelligent per-shared-GPU model-swap queue (batch, then swap) · c0a970b0
  Stefy Lanza (nextime / spora ) authored Jul 01, 2026
  c0a970b0
- manager: clean cross-engine VRAM swap (evict a busy sibling at its unit boundary) · c9791579
  Stefy Lanza (nextime / spora ) authored Jul 01, 2026
  c9791579
Jun 30, 2026
- video: fix dual-expert OOM regression + harden pipeline VRAM teardown · e1ab02b1
  Stefy Lanza (nextime / spora ) authored Jun 30, 2026
  e1ab02b1
- township: make the tool web UI mobile-friendly · e63b008a
  Stefy Lanza (nextime / spora ) authored Jun 30, 2026
  e63b008a
- packaging: build SageAttention into the OCI image (non-fatal, arch-gated) · c4b4b884
  Stefy Lanza (nextime / spora ) authored Jun 30, 2026
  c4b4b884
Jun 29, 2026
- video: gate 'auto' attention on diffusers' real Sage usability · fcd74214
  Stefy Lanza (nextime / spora ) authored Jun 29, 2026
  fcd74214
- video: resident all-GPU experts + faster attention backend (flash/sage) · fe2cdee1
  Stefy Lanza (nextime / spora ) authored Jun 29, 2026
  fe2cdee1
- township: gate enhance + odds + ZIP + upload behind full-match checkboxes · f4ce8539
  Stefy Lanza (nextime / spora ) authored Jun 29, 2026
  f4ce8539
- video: malloc_trim host RAM after each clip (stop glibc arena from hoarding decode buffers) · 6bbc2d31
  Stefy Lanza (nextime / spora ) authored Jun 29, 2026
  6bbc2d31
- manager: universal footprint rule — already-quantized (cached) models measured... · 88c1e787
  Stefy Lanza (nextime / spora ) authored Jun 29, 2026
  88c1e787
- manager: measured_vram_gb = FULL model footprint (GPU + offloaded), not the resident slice · 273eb23c
  Stefy Lanza (nextime / spora ) authored Jun 29, 2026
  273eb23c
- video: OOM fallback LADDER (no 5xx until exhausted) + monolithic cache finalize · b8bcdfea
  Stefy Lanza (nextime / spora ) authored Jun 29, 2026
  b8bcdfea
- manager: sweep orphan VRAM + untracked-VRAM diagnostic before every video load · 2ab70301
  Stefy Lanza (nextime / spora ) authored Jun 29, 2026
  2ab70301
- video: revert to model offload for injected components (the leak, not model offload, was the OOM) · 1dbf15ed
  Stefy Lanza (nextime / spora ) authored Jun 29, 2026
  1dbf15ed
- video: fix generation-OOM VRAM leak (traceback pin) + use sequential offload... · bc31e2d5
  Stefy Lanza (nextime / spora ) authored Jun 29, 2026
  bc31e2d5
Jun 28, 2026
- video: hook-based offload for injected cache components (fix device mismatch); gen retry safety net · 9ed35326
  Stefy Lanza (nextime / spora ) authored Jun 28, 2026
  9ed35326
- video: incremental per-component pipeline cache (load/build/cache one component at a time) · 93e2a4b9
  Stefy Lanza (nextime / spora ) authored Jun 28, 2026
  93e2a4b9
- video: stop the offload retry death-spiral; log swallowed load/gen errors; robust pipeline cache · e48b90e0
  Stefy Lanza (nextime / spora ) authored Jun 28, 2026
  e48b90e0
- manager: don't strand VRAM or churn the live model on eviction · bb95ab1d
  Stefy Lanza (nextime / spora ) authored Jun 28, 2026
  bb95ab1d
Jun 27, 2026
- router: pin video generation to a capable engine (never the gguf-only sibling) · cc6db025
  Stefy Lanza (nextime / spora ) authored Jun 27, 2026
  cc6db025
- gpu: scope per-engine VRAM to the engine's own card (fix Radeon showing NVIDIA) · ff39ee42
  Stefy Lanza (nextime / spora ) authored Jun 27, 2026
  ff39ee42
- gpu: context-free VRAM query so idle/GGUF engine pins no CUDA context · 9c150b3e
  Stefy Lanza (nextime / spora ) authored Jun 27, 2026
  9c150b3e
Jun 26, 2026
- fix: drop stale per-request LoRA adapters before re-applying · 59416db8
  Stefy Lanza (nextime / spora ) authored Jun 26, 2026
  59416db8
- fix: apply peft<->gptqmodel AWQ shim on the inference LoRA path too · 3fb6c9b9
  Stefy Lanza (nextime / spora ) authored Jun 26, 2026
  3fb6c9b9
- proxy: fix double-prefix when outer proxy mounts a tool at the same name · 2b6942d7
  Stefy Lanza (nextime / spora ) authored Jun 26, 2026
  2b6942d7
- proxy+training: chain-aware nginx for double-proxy; report training model as loaded · f177d01d
  Stefy Lanza (nextime / spora ) authored Jun 26, 2026
  f177d01d
- tasks page: engine indicator, live image-gen progress, per-task cooling · 931b5b85
  Stefy Lanza (nextime / spora ) authored Jun 26, 2026
  931b5b85
- loras: peft ↔️gptqmodel AWQ compat shim (fixes add_adapter ImportError) · df8939fe
  Stefy Lanza (nextime / spora ) authored Jun 26, 2026
  df8939fe
- loras: match VAE input dtype in Z-Image trainer (bf16 pipeline VAE) · ff6002d6
  Stefy Lanza (nextime / spora ) authored Jun 26, 2026
  ff6002d6
Jun 25, 2026
- loras: load Z-Image via ZImagePipeline (fix 4-bit text-encoder quant-state crash) · 6d83cb5e
  Stefy Lanza (nextime / spora ) authored Jun 25, 2026
  6d83cb5e
- loras: train Z-Image LoRA via 4-bit QLoRA (fast, uses the cached turbo build) · 26bdb59e
  Stefy Lanza (nextime / spora ) authored Jun 25, 2026
  26bdb59e
- loras: route Z-Image to _train_dit by name; train quantized base via full model · 7e4e7eba
  Stefy Lanza (nextime / spora ) authored Jun 25, 2026
  7e4e7eba
- quant: fix settings showing "GPTQModel not installed" when it is installed · e72e33eb
  Stefy Lanza (nextime / spora ) authored Jun 25, 2026
  e72e33eb
- fix: model load/unload 401 (front auth probe); township refs honor run-page res/steps · e3d0d35c
  Stefy Lanza (nextime / spora ) authored Jun 25, 2026
  e3d0d35c
- loras: native Z-Image (DiT) LoRA training (_train_dit) · 1f586669
  Stefy Lanza (nextime / spora ) authored Jun 25, 2026
  1f586669
- engines: cross-engine VRAM eviction for co-located GGUF/torch engines · 88063940
  Stefy Lanza (nextime / spora ) authored Jun 25, 2026
  88063940
- front: route GGUF pinned to torch engine -> gguf sibling; sync cached-models refresh · ec1f7218
  Stefy Lanza (nextime / spora ) authored Jun 25, 2026
  ec1f7218
- release: 0.1.12 (township match-delete 404 fix) · e715cc49
  Stefy Lanza (nextime / spora ) authored Jun 25, 2026
  e715cc49
- township: fix 404 after match delete behind reverse-proxy mount · 79bd4666
  Stefy Lanza (nextime / spora ) authored Jun 25, 2026
  79bd4666
Jun 24, 2026
- engines: process-isolate GGUF from torch on NVIDIA (fix CUDA-context poisoning) · 2fae0beb
  Stefy Lanza (nextime / spora ) authored Jun 25, 2026
  2fae0beb