Add EfRLFN MLX weights

NOTICE

@@ -0,0 +1,3 @@
+efrlfn-mlx — Apple MLX port of EfRLFN. MIT License.
+Derived from EvgeneyBogatyrev/EfRLFN (MIT), "EfRLFN: Efficient Real-time Super-Resolution",
+ICLR 2026. Pretrained x2/x4 weights are the official releases, converted to MLX safetensors.

README.md

@@ -0,0 +1,17 @@
+---
+license: mit
+library_name: mlx
+pipeline_tag: image-to-image
+tags: [mlx, super-resolution, efrlfn]
+base_model: EvgeneyBogatyrev/EfRLFN
+---
+# EfRLFN x4 (MLX)
+Apple MLX port of [EfRLFN](https://github.com/EvgeneyBogatyrev/EfRLFN) (ICLR 2026), x4 super-resolution.
+PT-vs-MLX parity ~1e-6. Realtime on Apple Silicon (270x480 -> 1080x1920 ~0.06s for x4).
+```python
+from efrlfn_mlx import EfRLFNConfig
+from efrlfn_mlx.pipeline import load_model, upscale_to_file
+m = load_model("model.safetensors", EfRLFNConfig.x4())
+upscale_to_file(m, "lr.png", "sr.png")
+```
+MIT, derived from EvgeneyBogatyrev/EfRLFN.

config.json

@@ -0,0 +1,7 @@
+{
+  "in_channels": 3,
+  "out_channels": 3,
+  "feature_channels": 52,
+  "num_blocks": 6,
+  "upscale": 4
+}

model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:742940547cd93bfc8b66627056a08081acbff37c0f5d773ab273f8d2f4ee31ad
+size 2020675