docs: plan Flash-Next tiered inference for RTX 4070 Ti SUPER and 32GB RAM

scripts/capacity.py

@@ -0,0 +1,28 @@
+#!/usr/bin/env python3
+"""Planning arithmetic only: no model loading, hardware emulation or speed claims."""
+import argparse
+import json
+import math
+from pathlib import Path
+
+
+def main():
+    parser = argparse.ArgumentParser(description=__doc__)
+    parser.add_argument('--checkpoint-gib', type=float, default=123.568)
+    args = parser.parse_args()
+    if not math.isfinite(args.checkpoint_gib) or args.checkpoint_gib <= 0:
+        parser.error('--checkpoint-gib must be finite and positive')
+    cfg = json.loads((Path(__file__).resolve().parents[1] / 'configs/target.json').read_text())
+    combined = cfg['inference_vram_budget_gib'] + cfg['process_tree_and_page_cache_budget_gib']
+    print(json.dumps({
+        'status': 'planning_arithmetic_only',
+        'checkpoint_gib': args.checkpoint_gib,
+        'combined_runtime_budget_gib': combined,
+        'optimistic_nonresident_checkpoint_gib': round(max(0, args.checkpoint_gib - combined), 3),
+        'warning': 'Budgets include runtime overhead. File size is not runtime size. No loadability or speed prediction.',
+        'hardware_test_executed': False
+    }, ensure_ascii=False, indent=2))
+
+
+if __name__ == '__main__':
+    main()