Restore working ZeroGPU probe pending CPU hosting decision

README.md

@@ -1,14 +1,33 @@
 ---
-title: AI Learns To Drive — TiGa interface
+title: TAO ZeroGPU Qualification
+emoji: 🏎️
+colorFrom: green
+colorTo: blue
 sdk: gradio
 sdk_version: 6.29.1
-python_version: "3.10"
+python_version: '3.10'
 app_file: app.py
+pinned: false
 ---
 
-Private CPU integration candidate. Custom FastAPI interface on port 7860.
-Simulator: https://github.com/Zorro909/driving_sim_rs
-Hosting adaptation and UI: TiGa https://github.com/seeker-cyber-maker/driving_sim_rs
-GPU is deliberately unavailable in this candidate. No bundled tracks or networks.
-One server process; no persistent storage. See DATA_POLICY.md for retention limits.
-Not yet verified in the hosted Linux runtime; do not publish as a stable release.
+Private native CUDA compatibility probe and bounded upload smoke test, not a qualified trainer.
+Uses included ZeroGPU allowance through a 45-second leased call. No paid Jobs
+or dedicated hardware requested. Check the authenticated caller's included quota
+before each call; the app itself does not know account billing balances.
+
+Original Docker source and dashboard remain in this repository but are inactive.
+Rollback: restore README.md from revision ba07e35a39211af7a0deda0ebd9adb2c554b4f38
+and request CPU Basic. BINARY_PROVENANCE.json pins the two exact HF-built binaries
+to that revision and upstream d8ab74c25df70b6a86ed101080e33d5fb785ea6f.
+
+No CUDA initialization outside the lease. First check device discovery and a
+three-value sine kernel in process, then the normal Rust gpu-info subprocess.
+No fallback, trajectory parity, speedup or H200 claim. A second lease
+must independently pass before testing training. Retained PACKAGE_MANIFEST.json
+describes the prior Docker snapshot only; ZEROGPU_MANIFEST.json covers this overlay.
+
+The upload test validates the exact supplied A01 track and a 6-4-2 Formula network
+before leasing a GPU for at most 90 seconds. It repeats qualification, then runs
+4 vehicles for 1 generation of 60 physics ticks, seed 42, proton math profile.
+The subprocess has a 25-second timeout. It exports a candidate and receipt, not
+an improvement claim. Real-game import compatibility and physics parity remain unverified.

app.py

@@ -1,33 +1,36 @@
-"""HF CPU release entrypoint. One process owns all ephemeral results."""
-import hashlib
-import json
-from pathlib import Path
-from server import create_app
-from worker import run_native
+import spaces
+import gradio as gr
+from cloud_probe import run_probe
 
-ROOT=Path(__file__).resolve().parent
+def cpu_smoke():
+    return run_probe(False)
 
-def verify_package(root=ROOT):
-    manifest=json.loads((root/'RELEASE_MANIFEST.json').read_text())
-    for name,expected in manifest['files'].items():
-        path=(root/name).resolve()
-        if not path.is_relative_to(root.resolve()) or not path.is_file():
-            raise RuntimeError('Invalid release member')
-        if hashlib.sha256(path.read_bytes()).hexdigest()!=expected:
-            raise RuntimeError('Release integrity check failed: '+name)
-    return manifest
+@spaces.GPU(duration=90)
+def gpu_smoke():
+    return run_probe(True)
 
-def cpu_worker(track,network):
-    return run_native(track,network,binary=ROOT/'bin/altd-sim',
-        importer=ROOT/'bin/extract_saved_track',
-        scene_template=ROOT/'assets/formula_template.json',with_report=True)
+def selected_smoke(backend):
+    if backend == 'cpu':
+        return cpu_smoke()
+    if backend == 'gpu':
+        return gpu_smoke()
+    raise ValueError('Choose CPU or GPU')
 
-def main():
-    verify_package()
-    for name in ('altd-sim','extract_saved_track'):
-        (ROOT/'bin'/name).chmod(0o755)
-    import uvicorn
-    uvicorn.run(create_app(worker=cpu_worker),host='0.0.0.0',port=7860,
-        workers=1,access_log=False)
+with gr.Blocks() as demo:
+    gr.Markdown('''# AI Learns To Drive — private compute test
+Simulator by [Zorro909](https://github.com/Zorro909/driving_sim_rs).
+UI and HF implementation by TiGa (seeker-cyber-maker).
 
-if __name__=='__main__': main()
+Fixed synthetic network, 4 cars, 1 generation, 60 ticks. CPU runs without a GPU lease.
+GPU uses up to 90 seconds of included allowance; check quota first. No automatic runs.
+No user uploads; temporary model files are deleted before the receipt is returned.
+This tests backend execution, not useful learning, performance, or full UI integration.
+''')
+    backend=gr.Radio(choices=[('CPU — no GPU allowance needed','cpu'),('GPU — uses included GPU allowance','gpu')],value='cpu',label='Compute')
+    report=gr.JSON(label='Execution receipt')
+    gr.Button('Run selected backend',variant='primary').click(selected_smoke,inputs=backend,outputs=report,api_name='selected_smoke',concurrency_limit=1,concurrency_id='native')
+    gr.Button('Run CPU-only test').click(cpu_smoke,outputs=report,api_name='cpu_smoke',concurrency_limit=1,concurrency_id='native')
+    gr.Button('Run GPU test').click(gpu_smoke,outputs=report,api_name='gpu_smoke',concurrency_limit=1,concurrency_id='native')
+
+if __name__=='__main__':
+    demo.queue(max_size=1).launch()

requirements.txt

@@ -1,2 +1 @@
-fastapi==0.141.1
-uvicorn==0.52.1
+spaces==0.51.3