Test3lora / app.py

Commit History

  1. Remove concurrency_limit — no request queue, memory cleanup retained Running

    Verified signature
    995d272
  2. Aggressive memory optimizations: worker-side state release + CUDA cache return after every request, main-process frames/audio cleanup after mux, encode_remote temp file removal, startup pack cleanup, and concurrency_limit=1 on generate

    Verified signature
    1acc2f3

  1. LoRA lifecycle hardening + conditioner retry + output rotation

    Verified signature
    731d4fc

  1. Fix kohya LoRA converter for hybrid ComfyUI-prefix + kohya-suffix format (missionary LoRA IndexError crash)

    Verified signature
    40ddddb
  2. Add OOM diagnostics: memory phase logs (_log_mem), in-generation sampler, bucket debug dumps to /data/debug (startup + oom with torch memory summary)

    Verified signature
    c1cbf61

  1. Add extensive debug logging (_log) + bucket-backed LoRA cache (_fetch_lora_file at /data/loras); generate wrapper logs requests/failures

    Verified signature
    c977a06

  1. Remove _TURBO_MERGE_BONUS from get_duration: the turbo merge is instant, the 120s bonus inflated the duration reservation

    Verified signature
    abc01d9
  2. Remove _LORA_MERGE_BONUS from get_duration: NSFW/custom LoRA merges are instant, the 60s/LoRA bonus inflated the duration reservation

    Verified signature
    f45e1e9
  3. Fix 5 oracle findings: parse tree/blob repo-page and spaces/-prefixed URLs, budget NSFW/custom LoRA merge time in get_duration, show LoRA label for custom-only runs, correct kohya alpha/rank scaling

    Verified signature
    3cd4c9d

  1. Rework custom LoRA to a single URL box + Download & Verify button: paste an HF repo link or direct .safetensors link, download/verify/load on button press, apply on Generate

    Verified signature
    cded1e5
  2. Add Custom LoRAs section: load user-supplied MiniMax-H3 LoRAs from any public HF repo (ComfyUI/kohya auto-detected), per-row scale + optional trigger, AIO auto-loads, applied/removed around the denoise

    Verified signature
    f7f09ad
  3. Remove 'Suggest LoRAs from prompt' button and its keyword map; keep conflict checker

    Verified signature
    bbbbc0b
  4. Fix startup crash: gr.Button does not accept info kwarg in gradio 6.20

    Verified signature
    eb78354
  5. Add 'Suggest LoRAs from prompt' button (keyword-matched top 5), conflict warnings in the LoRA description box, remove gr.Examples

    Verified signature
    220454b
  6. Rename Space to 'minimax h3 turbo + lora's' (README title, browser tab, in-app heading)

    Verified signature
    315c97a

  1. Fix ComfyUI LoRA converter crash on files with kohya-style .alpha scalar keys (blowjob v3.0); fix gr.Examples outputs count (6 return values vs 5 outputs)

    Verified signature
    81634a7
  2. Add per-LoRA descriptions: 2-sentence intended-use + caveat for all 97 LoRAs, shown in a live description box under the selection dropdown

    Verified signature
    e2f15ca
  3. Replace NSFW LoRA checkboxes with searchable multiselect dropdown (97 LoRAs, scales without squish) + LoRA label under video

    Verified signature
    77d08b7

  1. Add Dildoing LoRA (nsfw_dildo) for natural dildo insertion, fix stale penis checkbox label

    Verified signature
    d09d43c

  1. Swap penis LoRA to Penis Lora - H3 v1.1 (kohya format, larger dataset) with kohya->ComfyUI converter

    Verified signature
    b913e22
  2. Lower HMPenis V2 LoRA scale to 0.7 to restore prompt adherence

    Verified signature
    2ed3fcb
  3. Auto-load HMNSFW AIO V2.5 motion adapter when any NSFW anatomy LoRA is selected

    Verified signature
    b1f8352
  4. Add HMPenis V2 NSFW LoRA (nsfw_penis) with UI checkbox

    Verified signature
    c980aa9

  1. Remove NSFW LoRA duration bonus (merge is small, covered by existing pad)

    Verified signature
    b05620a
  2. Auto-append inniepussy trigger token when NSFW LoRA checkbox is checked

    Verified signature
    f292fc9
  3. Switch NSFW LoRA to HMInnie V1 (innie pussy)

    Verified signature
    4557d53
  4. Add NSFW Vagina (HMPussy V1) LoRA support

    Verified signature
    20689d3

  1. feat: show toast notification when download button is clicked

    Verified signature
    9a6b509
  2. fix: use gr.DownloadButton for proper browser download, no JS hacks

    Verified signature
    8f41237
  3. fix: download button reads video path directly from video component, opens /file= URL in new tab

    Verified signature
    4021070
  4. feat: download button uses JS anchor click for real browser download, path stored in hidden textbox

    Verified signature
    444a27c
  5. fix: gr.File 6.20 only accepts str, not tuple

    Verified signature
    7c10805
  6. fix: remove visible=False from Image and File components so they can receive updates

    Verified signature
    f6a2960
  7. fix: download button returns (path, filename) tuple to trigger browser download, last frame reveals image

    Verified signature
    bb6b028
  8. feat: add Export Last Frame and Download Video buttons below generated video

    Verified signature
    ebd17a9

  1. docs: fix stale comments (20GB→1.2GB, startup merge→on-demand), update codemap.md with turbo

    Verified signature
    721686d
  2. docs: update UI text and README for turbo LoRA, fix placement default, fix step counts

    Verified signature
    a654b44
  3. fix: remove stale _TURBO_Deltas global declaration

    Verified signature
    58a0771
  4. fix: store LoRA A/B factors (~1.2GB) instead of precomputed deltas (~20GB) — fits in GPU free memory, compute B@A on-the-fly

    Verified signature
    e6b2d43
  5. fix: add 120s turbo merge bonus to GPU duration + debug prints in _generate

    Verified signature
    53c3920
  6. fix: deltas stay on CPU at startup (avoid 132GB packing overfill), lazy-move to CUDA on first merge + debug prints

    Verified signature
    7f0f4e1
  7. fix: force deltas to CUDA (packed placeholders are CPU) + add debug prints

    Verified signature
    297fb55
  8. perf: pre-move LoRA deltas to CUDA at parse time — merge/unmerge now zero-transfer in-place += / -=

    Verified signature
    0abb027
  9. fix: move LoRA merge/unmerge inside @spaces.GPU where real weights are live on CUDA

    Verified signature
    e48f34e
  10. fix: keep LoRA deltas on CPU to avoid ZeroGPU packing 132GB — move to CUDA inside @spaces.GPU

    Verified signature
    f95b21c
  11. fix: on-demand LoRA merge with pre-moved CUDA deltas (zero-allocation in-place)

    Verified signature
    864d8e9
  12. fix: revert LoRA to startup merge to avoid OOM during generation

    Verified signature
    c43e4d2
  13. fix: use InstantX diffusers LoRA conversion of larryvrh, strip transformer. prefix in key nav

    Verified signature
    74fdf62
  14. upgrade: swap lightx2v LoRA for larryvrh v4-step600-EMA at 7 steps (better quality)

    Verified signature
    c944330
  15. fix: override Gradio CSS variables to eliminate gray backgrounds on textbox/checkbox/accordion

    Verified signature
    03f1d3d