fix: Correct hardware configuration to 4x-l4 (4x NVIDIA L4 GPUs)

Previous: hardware: a10g-large (WRONG - single A10G)
Corrected: hardware: 4x-l4 (CORRECT - 4x L4 GPUs)

Hardware specs:
- 4x NVIDIA L4 GPUs
- 24GB VRAM per GPU (96GB total)
- Ampere architecture with bf16 support
- Multi-GPU training enabled

The multi-GPU code will now properly utilize all 4 GPUs:
- Per-device batch: 2
- Gradient accumulation: 1
- Effective batch size: 8 (2 × 1 × 4 GPUs)
- Expected: ~25% utilization per GPU

README.md

@@ -8,7 +8,7 @@ sdk_version: 5.0.0
 app_file: app.py
 pinned: false
 license: mit
-hardware: a10g-large
+hardware: 4x-l4
 python_version: "3.10"
 disable_embedding: false
 ---