Improve README with usage and benchmarks (still files-only)

README.md

@@ -1,7 +1,8 @@
+# Parakeet‑TDT 0.6B v3 (MLX) — Model Files
 
-# Parakeet‑TDT v3 MLX Weights (files‑only)
-
-Files‑only hosting for MLX weights and config. No Inference Widget, no runnable code in this repo.
+This repository hosts the MLX model files (config + weights + tokenizer) for Parakeet‑TDT v3.
+It is intentionally files‑only (no widget, no runnable code). Use these files with your own
+codebase or with the `parakeetv3_mlx` package in your project.
 
 ## Contents
 - `parakeet-tdt-v3-mlx/config.json`
@@ -12,7 +13,7 @@ Files‑only hosting for MLX weights and config. No Inference Widget, no runnabl
 
 ## How to use
 
-### Option A: Download with huggingface_hub
+### Option A: Download programmatically
 ```python
 from huggingface_hub import snapshot_download
 from parakeetv3_mlx.utils import from_pretrained
@@ -26,14 +27,28 @@ print("".join(t.text for t in res.tokens))
 ```
 
 ### Option B: Download manually
-- Click “Files and versions” and download the 5 files under `parakeet-tdt-v3-mlx/` to a local folder
-- Then:
+1) In “Files and versions”, download the files under `parakeet-tdt-v3-mlx/` into a local folder.
+2) Load with:
 ```python
 from parakeetv3_mlx.utils import from_pretrained
 model = from_pretrained("/path/to/parakeet-tdt-v3-mlx")
 ```
 
 ### Notes
-- Requires the `parakeetv3_mlx` Python package (bundled in your project) and MLX.
-- Audio should be mono 16 kHz WAV; `librosa` will resample if needed.
-- For long audio, enable local attention and chunking in your code.
+- Requires the `parakeetv3_mlx` Python package (your app or local project) and Apple’s MLX.
+- Audio: mono 16 kHz WAV recommended (librosa can resample automatically).
+- Long audio: enable local attention + chunking in your code for best memory/perf trade‑off.
+
+## Benchmarks (Apple Silicon)
+- Settings: chunk=120s, overlap=15s, local attention (256,256), bf16
+- Device: M4 Pro
+
+| Audio (1h) | Wall time | RTF |
+|------------|-----------|-----|
+| English    | 43.6 s    | 82.6× |
+| German     | 59.6 s    | 60.4× |
+
+On M4 Max, throughput is typically ~2× higher under the same settings.
+
+## About the author
+Profile: https://huggingface.co/Jimmi42