Add model card: provenance, files, precision guidance

README.md

@@ -1,10 +1,37 @@
 ---
 license: apache-2.0
+base_model: alibabasglab/FRCRN_SE_16K
+tags:
+  - mlx
+  - audio
+  - audio-to-audio
+  - speech-enhancement
 ---
 
 # FRCRN SE 16K (MLX)
 
-This model was converted to MLX format from
+Speech enhancement at 16 kHz. This model was converted to MLX format from
 [alibabasglab/FRCRN_SE_16K](https://huggingface.co/alibabasglab/FRCRN_SE_16K).
 
 **Code:** [starkdmi/frcrn_mlx](https://github.com/starkdmi/frcrn_mlx)
+
+## Files
+
+| file | size | notes |
+| --- | ---: | --- |
+| `model_fp32.safetensors` | 56 MB | the weights |
+| `FRCRN_SE_16K.yaml` | 614 B | upstream model configuration |
+| `weights.npz` | 110 MB | superseded, see below |
+
+fp32 only; no half-precision or quantized conversion is published.
+
+`weights.npz` predates the safetensors conversion and is kept only so that
+anything already referring to it keeps working. It is not what loads: use
+`model_fp32.safetensors`.
+
+## Use
+
+Consumed by [AudioToolSwift](https://github.com/starkdmi) as
+`ModelRepository.frcrnSE16K`, which pins this repository to an exact revision and
+verifies the SHA-256 of every file it downloads. The loader reads the safetensors
+and nothing else — the YAML is provided for reference.