pp-hirano/ofc-solver

Fix Round 1 burn validation: use hand size instead of state round

Show top 3 placements in feedback instead of just rank

Switch to ZeroGPU for fast inference

Use same solver settings for player feedback as AI (20/8/500)

Improve feedback quality: rollout-based lookahead for all rounds

Fix wild card foul bug + exact scoring for Round 5

Add placement feedback in Play vs AI mode

Switch to 2-phase solver: p1=20/k=8 -> p3=500

Upgrade Stage 7 prune model and reduce k=6 -> k=4

Optimize hot loop: combine enumerate + encode into single pass

Revert "Switch from PyTorch to ONNX Runtime for inference"

Switch from PyTorch to ONNX Runtime for inference

Update Stage 9 model v5 -> v8

Reduce rollouts for HF Space (30/20/500)

Update solver to Stage 9 model (v5) with eval_stage=9

Fix web model decryption fallback

Deploy stronger initial solver to web

Deploy stage9 beam search

Deploy encrypted stage-specific models

Fix dark mode: use color-scheme:light on HTML output containers

Show Processing... on Solve button while computing

Fix dark mode: add explicit background and text colors to all HTML

Add Solver tab for 5-card initial placement analysis

Optimize rollout allocation: Phase3=150, Phase1=6, Phase2=12

Show Processing... on New Game button while AI computes

Encrypt model: load from .pt.enc using MODEL_KEY secret

Fix score readability (white background + explicit text color), show burn cards

Replace text input with Radio button GUI for card placement

Fix: show AI board during game (only burns are hidden in OFC)

Add game mode: play OFC poker vs AI with scoring

Fix: use Gradio 5.23.3, remove huggingface_hub pin

Fix: pin huggingface_hub<0.25 for Gradio compatibility

Fix: use Gradio 5.6.0 for huggingface_hub compatibility

Fix: try Gradio 4.36.0

Fix: pin Python 3.10 to avoid audioop removal in 3.13

Fix: set sdk_version to 4.44.1

Fix: remove sdk_version to use latest Gradio

Initial deployment: OFC 5-card placement solver with Gradio UI

initial commit