Upload README.md with huggingface_hub

README.md

@@ -116,11 +116,15 @@ The ONNX wrapper inverts `mask_in` internally before passing it to the encoder,
 
 ## Outputs (ONNX)
 
-| File | Output shape | Aggregation |
-|------|-------------|-------------|
-| `astromer2_mean.onnx` | `[batch, 256]` | Masked mean pooling: `sum(z * mask_in) / sum(mask_in)` |
-| `astromer2_max.onnx`  | `[batch, 256]` | Masked max pooling over valid timesteps |
-| `astromer2_full.onnx` | `[batch, 200, 256]` | Full per-timestep sequence; consumer aggregates |
+Single file `astromer2.onnx` with three named outputs:
+
+| Output name | Shape | Aggregation |
+|-------------|-------|-------------|
+| `mean` | `[batch, 256]` | Masked mean pooling: `sum(z * mask_in) / sum(mask_in)` |
+| `max`  | `[batch, 256]` | Masked max pooling over valid timesteps |
+| `sequence` | `[batch, 200, 256]` | Per-timestep features |
+
+Request only the output(s) you need via `session.run(["mean"], feed)` — onnxruntime will prune unused computation.
 
 ONNX opset: 13.