Upload README.md with huggingface_hub

README.md

@@ -28,21 +28,6 @@ Used by **[wipe.photos](https://wipe.photos)** — a fully private, in-browser o
 
 **`lama_512_fp16.onnx` is recommended when you can fix the input to 512×512** (as wipe.photos does): the whole graph computes in fp16, and the FFTs are replaced by fixed-size matmuls so every op runs on the GPU with no CPU fallbacks — 0.43s vs 2.18s per 512×512 inference on WebGPU (M-series Mac), at the same quality as the weight-only variants (zero pixels differing by more than 50/255 from fp32). For arbitrary input sizes, use `lama_fp16.onnx` — half the size of fp32 with a max pixel difference of just 3–11/255.
 
-## Benchmark
-
-Tested on two images (Node.js, CPU):
-
-| | fp32 | fp16 | int8 |
-|---|---|---|---|
-| File size | 209 MB | **110 MB** | 61.5 MB |
-| Load time | 0.69–0.81s | **0.40–0.41s** | 0.29–0.30s |
-| Inference time | 13–19s | 13–19s | 13–19s |
-| Max pixel diff (example.jpg) | — | **11** | 24 |
-| Max pixel diff (download.png) | — | **3** | 48 |
-| Pixels > 50 diff | — | **0** | 0 |
-
-Inference time is identical across all three — the compressed weights are dequantized to fp32 before each Conv so the CPU does the same work. The savings are purely storage and download size.
-
 ## Input
 
 | Name  | Shape           | Description |