Focus project on JAX PPO

web/README.md

@@ -1,11 +1,38 @@
 # COOLRL Lost Cities Web
 
 Browser-only Lost Cities client. The rules engine, observation builder, and PPO
-inference all run on the device. There is no application server.
+inference all run on the device; there is no application server.
+
+The verified final JAX PPO policy is committed at
+`public/models/jax-ppo.onnx` (3.1 MB). Vite copies it to the static build, and
+the app resolves the asset relative to the deployed site so it works on GitHub
+Pages, GitLab Pages, or a normal web root. Its size and SHA-256 are recorded in
+[`public/models/jax-ppo.json`](public/models/jax-ppo.json).
+
+Pushes to `main` build and publish the client through the repository's GitHub
+Pages and GitLab Pages workflows. Each workflow supplies the correct base URL
+for its host.
 
 ## Setup
 
-From the repository root, export an Orbax checkpoint to the browser model:
+Install and run the checked-in final policy:
+
+```bash
+cd web
+npm ci
+npm run dev
+```
+
+Build the same static bundle used by a host:
+
+```bash
+npm run build
+npm run preview
+```
+
+To replace the shipped policy, export a verified Orbax checkpoint from the
+repository root. The exporter writes both the ONNX file and its public
+metadata manifest:
 
 ```bash
 uv run --with onnx scripts/export_jax_ppo_onnx.py \
@@ -13,14 +40,6 @@ uv run --with onnx scripts/export_jax_ppo_onnx.py \
   --output web/public/models/jax-ppo.onnx
 ```
 
-Then install and run the web app:
-
-```bash
-cd web
-npm install
-npm run dev
-```
-
 The policy tries WebGPU first and falls back to ONNX Runtime WebAssembly. If
 the model asset is absent, the UI remains playable using a simple local
 heuristic and reports that fallback in the header.

web/public/models/jax-ppo.json

@@ -0,0 +1,14 @@
+{
+  "format": "coolrl-lost-cities-jax-ppo-onnx-v1",
+  "model_file": "jax-ppo.onnx",
+  "model_size_bytes": 3230780,
+  "model_sha256": "e8241e305c01ea450e92a6178002a22db96710fa94e238bba57953743eb285b2",
+  "source_checkpoint": "final_candidate",
+  "source_config": "main_ppo_config.json",
+  "observation_size": 454,
+  "action_size": 96,
+  "hidden_size": 512,
+  "num_layers": 3,
+  "dtype": "float32",
+  "validation_max_abs_error": 9.918212890625e-05
+}

web/src/model/policy.ts

@@ -7,6 +7,8 @@ import { cardColor, cardRank, isHandshake } from "../game/cards";
 
 export type ExecutionProvider = "webgpu" | "wasm" | "heuristic";
 
+const MODEL_URL = `${import.meta.env.BASE_URL}models/jax-ppo.onnx`;
+
 export interface Policy {
   readonly provider: ExecutionProvider;
   action(state: GameState): Promise<number>;
@@ -66,7 +68,7 @@ class HeuristicPolicy implements Policy {
 }
 
 async function createSession(provider: "webgpu" | "wasm"): Promise<ort.InferenceSession> {
-  return ort.InferenceSession.create("/models/jax-ppo.onnx", {
+  return ort.InferenceSession.create(MODEL_URL, {
     executionProviders: [provider],
     graphOptimizationLevel: "all",
   });