Javare/Local_AI_WebGPU

1from fastapi import FastAPI, HTTPException2from fastapi.staticfiles import StaticFiles3from fastapi.responses import FileResponse4from pydantic import BaseModel5import os6from huggingface_hub import HfApi, hf_hub_download7import json8 9app = FastAPI()10 11DATASET_REPO_ID = "Javare/Local_AI_Leaderboard" 12FILENAME = "scores.json"13HF_TOKEN = os.environ.get("HF_TOKEN")14 15# Nouveau modèle étendu selon tes exigences16class ScoreEntry(BaseModel):17    config: str18    browser: str19    power: str20    min_tps: float21    max_tps: float22    avg_tps: float23    total_tokens: int24    duration: float25 26def get_scores():27    try:28        path = hf_hub_download(repo_id=DATASET_REPO_ID, filename=FILENAME, repo_type="dataset", token=HF_TOKEN)29        with open(path, "r", encoding="utf-8") as f:30            return json.load(f)31    except Exception:32        return []33 34@app.get("/api/scores")35def read_scores():36    scores = get_scores()37    # On trie le TOP 10 par la vitesse moyenne (avg_tps)38    scores.sort(key=lambda x: x.get("avg_tps", 0), reverse=True)39    return scores[:10]40 41@app.post("/api/score")42def add_score(entry: ScoreEntry):43    if not HF_TOKEN:44        raise HTTPException(status_code=500, detail="HF_TOKEN manquant")45    46    scores = get_scores()47    scores.append(entry.dict())48    49    local_path = "scores.json"50    with open(local_path, "w", encoding="utf-8") as f:51        json.dump(scores, f, ensure_ascii=False, indent=2)52        53    api = HfApi()54    try:55        api.upload_file(56            path_or_fileobj=local_path,57            path_in_repo=FILENAME,58            repo_id=DATASET_REPO_ID,59            repo_type="dataset",60            token=HF_TOKEN61        )62    except Exception as e:63        raise HTTPException(status_code=500, detail=str(e))64        65    return {"status": "success"}66 67@app.get("/")68def read_index():69    return FileResponse("index.html")70 71app.mount("/assets", StaticFiles(directory="assets"), name="assets")