Fetching metadata from the HF Docker repository...
-
footer: raise contrast (opacity .33->.75, lighter parchment) per organizer feedback
-
fix: re-run scribe on cards a failed chunk left blank; never paint an empty (subjectless) prompt
-
fix: re-run scribe on cards a failed chunk left blank; never paint an empty (subjectless) prompt
-
submission: hackathon track + achievement tags, demo/blog/social/traces links
-
picker: 'Open demo pack' button (bundled deck, zero-GPU)
-
add bundled demo deck (breakfast foods · RWS)
-
reading: clear stale reading on entry; faster first-card paint (back last + brisk poll)
-
reading: clear stale reading on entry; faster first-card paint (back last + brisk poll)
-
reading: clear stale reading on entry; faster first-card paint (back last + brisk poll)
-
fix: reading TypeError — drop stale timeout kwarg (removed with EndpointLLM)
-
fix: custom-style box shows on first pick (.input not .change)
-
placeholders: category prompt + instructional reading prompt
-
landing: drop greeting, two-line invocation
-
footer: 'powered by … from …' wording
-
footer: systematic per-line credits, bottom-left, faded
-
strip: local-only backends; remove diagnostic endpoints
-
strip: local-only backends; remove diagnostic endpoints
-
strip: local-only backends; remove diagnostic endpoints
-
strip: local-only backends; remove diagnostic endpoints
-
strip: local-only backends; remove diagnostic endpoints
-
strip: local-only backends; remove diagnostic endpoints
-
Upload README.md with huggingface_hub
-
Upload app.py with huggingface_hub
-
Upload arcana/frame.py with huggingface_hub
-
Upload arcana/frame.py with huggingface_hub
-
Upload app.py with huggingface_hub
-
Upload app.py with huggingface_hub
-
Upload arcana/frame.py with huggingface_hub
-
Sync from dev: Qwen3-14B (single model) + FLUX.2-klein, generative scribe, 4-way parallel scribe/paint, lazy deck view, prompt logging, GUI + card-frame fixes
-
examples conjure live (no pre-baked decks) + two-stage 'Generating/Painting' loader
-
examples conjure live (no pre-baked decks) + two-stage 'Generating/Painting' loader
-
remove stale pre-baked demo deck thermodynamics-rider-waite-smith-1780952458 (examples now regenerate live)
-
remove stale pre-baked demo deck lord_of_the_rings-bundled (examples now regenerate live)
-
remove stale pre-baked demo deck breakfast_foods-playful-1780954994 (examples now regenerate live)
-
promote hybrid: 8B map + 30B read + split build (VRAM) + FLUX preload + CUDA cache
-
promote hybrid: 8B map + 30B read + split build (VRAM) + FLUX preload + CUDA cache
-
promote hybrid: 8B map + 30B read + split build (VRAM) + FLUX preload + CUDA cache
-
promote hybrid: 8B map + 30B read + split build (VRAM) + FLUX preload + CUDA cache
-
llama.cpp: flash_attn (Llama-arch) + n_batch=2048 for faster prompt eval
-
startup CPU preload of FLUX pipe (faster conjure: .to cuda only in GPU call)
-
startup CPU preload of FLUX pipe (faster conjure: .to cuda only in GPU call)
-
hybrid LLM split: 8B mapping + (env) 30B reading; 9B image; disk is ample
-
hybrid LLM split: 8B mapping + (env) 30B reading; 9B image; disk is ample
-
FLUX.2: switch to distilled klein-4B (built for 4 steps) + guidance 1.0
-
FLUX.2: switch to distilled klein-4B (built for 4 steps) + guidance 1.0
-
speed: drop JSON grammar for the 8B mapping (rely on extract_json + retries)
-
promote ZeroGPU local-first build: Nemotron via llama.cpp + FLUX.2 via diffusers
-
baseline: default LLM = Nemotron-3-Nano-30B-A3B via NVIDIA API (≤32B, Nemotron-prize)
-
Coverflow deck view (swipe/flip/fade carousel + overhead toggle + in-page lightbox), zip download, arcana subtitle
-
Generated decks now open in the Deck-View/Overhead browser (Read/Save/Back); removed flaky gen_fork
