Add FP16 conversion option and update UI description

app.py

@@ -113,7 +113,10 @@ def process_model(model_id, q_method, oauth_token: gr.OAuthToken | None):
         with tempfile.TemporaryDirectory(dir="converted") as tmpdir:
             # The target dir must not exist
             mlx_path = os.path.join(tmpdir, "mlx")
-            convert(model_id, mlx_path=mlx_path, quantize=True, q_bits=QUANT_PARAMS[q_method])
+            if q_method == "FP16":
+                convert(model_id, mlx_path=mlx_path, quantize=False, dtype="float16")
+            else:
+                convert(model_id, mlx_path=mlx_path, quantize=True, q_bits=QUANT_PARAMS[q_method])
             print("Conversion done")
             upload_to_hub(path=mlx_path, upload_repo=upload_repo, hf_path=model_id, oauth_token=oauth_token)
             print("Upload done")
@@ -142,9 +145,9 @@ with gr.Blocks(css=css) as demo:
     )
 
     q_method = gr.Dropdown(
-        ["Q2", "Q3", "Q4", "Q6", "Q8"],
-        label="Quantization Method",
-        info="MLX quantization type",
+        ["FP16", "Q2", "Q3", "Q4", "Q6", "Q8"],
+        label="Conversion Method",
+        info="MLX conversion type (FP16 for float16, Q2–Q8 for quantized models)",
         value="Q4",
         filterable=False,
         visible=True
@@ -160,8 +163,8 @@ with gr.Blocks(css=css) as demo:
             gr.Markdown(label="output"),
             gr.Image(show_label=False),
         ],
-        title="Create your own MLX Quants, blazingly fast ⚡!",
-        description="The space takes an HF repo as an input, quantizes it and creates a Public/ Private repo containing the selected quant under your HF user namespace.",
+        title="Create your own MLX Models, blazingly fast ⚡!",
+        description="The space takes an HF repo as an input, converts it to MLX format (FP 16 or quantized), and creates a Public/Private repo under your HF user namespace.",
         api_name=False
     )