Upload README.md with huggingface_hub

README.md

@@ -21,10 +21,11 @@ All variants take and return fp32 tensors — swap the `.pte` file, keep your ap
 
 | build | file | size (MB) | parity vs fp32 eager (worst corr) | Mac median (ms)* |
 |-----------|------|-----------|------------------------------------|------------------|
-| fp32 | `raft_small_xnnpack_fp32.pte` | 4.4 | 1.000000 | 169.6 |
+| fp32 | `raft_small_xnnpack_fp32.pte` | 4.4 | 1.000000 | 173.0 |
+| Core ML (fp32, iOS) | `raft_small_coreml_all_fp32.pte` | 10.6 | 1.000000 | 12.0 |
 
 \*Mac arm64, single process, median of 10 — a reference point for relative cost
-only, not a device number (torch eager fp32 on the same machine: 135.2 ms).
+only, not a device number (torch eager fp32 on the same machine: 140.6 ms).
 
 ## Verification (executorch 1.4.0, torch 2.13.0)
 
@@ -35,21 +36,9 @@ the correlation over all elements of each output tensor.
 |--------|-------|--------------|------|
 | 0 | [1, 2, 384, 512] | 4.792e-04 | 1.000000 |
 
-XNNPACK delegate coverage (fp32): 62.9% (1208/1921 ops); ops left on the portable kernels: `dim_order_ops._to_dim_order_copy.default` x435, `aten.split_with_sizes_copy.default` x50, `aten.grid_sampler_2d.default` x48, `aten.expand_copy.default` x30, `aten.arange.start_step` x28, `aten.lt.Scalar` x24, `aten.sub.Tensor` x24, `aten.where.self` x24, `aten._native_batch_norm_legit.no_stats` x21, `aten.alias_copy.default` x12, `aten.view_copy.default` x6, `aten.avg_pool2d.default` x3, `aten.cat.default` x2, `aten.unsqueeze_copy.default` x2, `aten.repeat.default` x2, `dim_order_ops._clone_dim_order.default` x1, `aten.sqrt.default` x1
+XNNPACK delegate coverage (fp32): 63.0% (1209/1920 ops); ops left on the portable kernels: `dim_order_ops._to_dim_order_copy.default` x435, `aten.split_with_sizes_copy.default` x50, `aten.grid_sampler_2d.default` x48, `aten.expand_copy.default` x30, `aten.arange.start_step` x28, `aten.lt.Scalar` x24, `aten.sub.Tensor` x24, `aten.where.self` x24, `aten._native_batch_norm_legit.no_stats` x21, `aten.alias_copy.default` x12, `aten.view_copy.default` x6, `aten.avg_pool2d.default` x3, `aten.cat.default` x2, `aten.unsqueeze_copy.default` x2, `aten.repeat.default` x2
 
 ## Conversion
 
 torch.export -> to_edge_transform_and_lower(partitioner) -> .pte
 (conversion scripts: [executorch-models](https://github.com/john-rocky/executorch-models))
-
-**Notes**: No Core ML build: RAFT's correlation volume is a rank-6 tensor and Core ML caps at rank 5. No fp16 build either — this is a conv-only model, where XNNPACK serializes convolution weights as fp32 whatever the graph dtype, so fp16 would be the same size with extra casts.
-
-<!-- funnel:v1 -->
-
----
-
-**More models in this format:** [ExecuTorch Model Zoo](https://huggingface.co/collections/mlboydaisuke/executorch-model-zoo-6a7ff328390b63075ffeae5e) — 31 models, each with the recipe that produced it.
-
-**Want a different model on-device?** [Open a request](https://github.com/john-rocky/on-device-requests) — free, open weights only; the export and its measured numbers get published publicly.
-
-<!-- /funnel:v1 -->