Rewrite README in English; add embeddings note

README.md

@@ -8,25 +8,35 @@ library_name: mlx
 
 # Voxtral 3B — Quantized (MLX)
 
-Ce dépôt regroupe des variantes quantifiées du modèle Voxtral 3B pour MLX (Apple Silicon).
+Public quantized weights of the Voxtral 3B model for Apple MLX. This repo contains MLX-ready variants only.
 
-## Variantes
-- MLX Q4: dossier `mlx-q4/`
-- MLX Q8: dossier `mlx-q8/`
+## Variants
+- MLX Q4: `mlx-q4/`
+- MLX Q8: `mlx-q8/`
 
-## Intégrité (SHA256)
+## Integrity (SHA256)
 - MLX Q4 `model-00001-of-00001.safetensors`:
   - `eec98aef078b3db2c226943d38558d814b10ec387dc5359d333eeed4be5298d2`
 - MLX Q8 `model-00001-of-00001.safetensors`:
   - `37999e4a9dda52a0aedb593636be6c12e69dd8b8457f15ce48134f88b1ccebd3`
 
-## Utilisation rapide (MLX)
+## Quickstart (MLX)
 ```python
-from mlx_lm import load
-# Exemple: charger les poids quantifiés MLX Q4
+from mlx_lm import load, generate
+
+# Load quantized weights (Q4 or Q8 folders are included in the repo)
 model, tokenizer = load("NeoRoth/voxtral-3b-quantized")
+
+prompt = "Hello!"
+print(generate(model, tokenizer, prompt, max_tokens=64))
 ```
 
-## Notes
-- Ces fichiers sont des quantifications dérivées du modèle Voxtral 3B. Respectez la licence du modèle d’origine.
-- Ouvrez une issue si vous repérez un problème (poids manquants, checksum incorrect, etc.).
+## Quantization notes
+- Only inference weights are quantized (Q4/Q8 depending on the folder).
+- Embeddings are NOT quantized to preserve shape compatibility. As a result, any "bits per weight" metric may exceed the nominal target. This is informational, not an error.
+
+## License
+- See `LICENSE.txt`. Also ensure you comply with the original Voxtral model license.
+
+## Issues
+If you notice any mismatch (missing files, wrong checksum), please open an issue.