Fix infinite page growth: remove fill_height (feedback loop with HF iframe auto-resizer)

Replace animated scene with draw-once static background (no animation loop, no flicker possible)

Fix severe flicker: remove backdrop-filter blur over animated canvas, static grain, reduced-motion + slow-GPU freeze

Perf: pre-render nebula/glow sprites, cap scene at 30fps (fixes browser lag)

Redesign UI: cosmic Wheeler-DeWitt scene (parallax stars, orrery, ripples), compact glass layout

Show search progress via gr.Progress (no chat-message streaming)

Revert searching indicator: regressed generation

Stream a searching indicator while candidates are drawn

Remove type= from Chatbot (removed in gradio 6)

Coerce gradio-6 content parts in render() so structure never reaches the prompt

Set Chatbot type='messages' to match on_submit dict format

Handle list-valued chat content in effort scoring

Use effort-scaled decoding (best-of-N + rerank)

Update app.py

Upload app.py with huggingface_hub

Upload app.py with huggingface_hub

Upload app.py with huggingface_hub

Serve WheelerV2 (new arch code + morphable/ELO/fractal-rope); nrng decoding (rep 1.15 + top-k 40); tested example prompts; accurate description

Add no-repeat-ngram(3): A/B on real weights at deployed t0.7/top_p0.9, rep-4 0.181 -> 0.000, distinct-2 0.745 -> 0.913

Revert "Decoding upgrade from sampling sweep: no-repeat-ngram(3) + reply-only repetition penalty + min_p(0.08) tail trim"

Decoding upgrade from sampling sweep: no-repeat-ngram(3) + reply-only repetition penalty + min_p(0.08) tail trim

Upload folder using huggingface_hub

Fix startup: add transformers to requirements; KV-cached decode with per-request ring-memory reset (matches generate.py harness)

Upload folder using huggingface_hub

Upload folder using huggingface_hub