Upload USAGE.md with huggingface_hub

USAGE.md

@@ -108,7 +108,7 @@ sh ./opencoti-llamafile-0.10.3-c3-x86_64.llamafile --server --port 8080 \
 |---|---:|---|---|
 | `…-x86_64.llamafile` | ~4.8 GB | Linux/macOS/BSD, x86_64 + aarch64 | CUDA DSO **embedded** (x86_64-linux, sm_75/80/86/89/90/120f); self-extracts on first `-ngl` run |
 | `…-win-x86_64.llamafile.exe` | ~55 MB | **Windows** + Linux/macOS/BSD, x86_64 + aarch64 | none embedded — CPU works everywhere out of the box; GPU via side-load (below) |
-| `…-aarch64.llamafile` *(planned)* | — | same platforms | CUDA sbsa DSO embedded (sm_110f Jetson Thor, sm_121a DGX Spark GB10) |
+| `…-aarch64.llamafile` | ~1.9 GB | same platforms | CUDA sbsa DSO **embedded** (aarch64-linux, sm_110f DGX Spark GB10 / Jetson Thor + sm_121a); GPU out of the box on aarch64-linux CUDA hosts |
 
 The host binary inside every variant is byte-for-byte the same APE
 (same patches, same features); they differ only in the embedded GPU
@@ -134,6 +134,10 @@ to `.exe` (Windows needs the extension) can.
   cp ggml-cuda-x86_64.so ~/.llamafile/v/0.10.3/ggml-cuda.so
   ```
 
+- **Linux aarch64 (DGX Spark / GB10 class):** same mechanism with the
+  sbsa DSO — `dso/<ver>-<tag>/ggml-cuda-sbsa-aarch64.so` → 
+  `~/.llamafile/v/0.10.3/ggml-cuda.so`. Or just download the
+  `…-aarch64.llamafile` artifact, which has it embedded.
 - **No CUDA available:** nothing to do — it falls back to CPU
   inference automatically.