Fix multi-image and 2x speed improvements (DS-VL2) (#157)

* fix multi-image and use .tolist() for (2.16× prompt and 1.83x for generation speedup)

* format

mlx_vlm/models/deepseek_vl_v2/deepseek_vl_v2.py

@@ -251,14 +251,14 @@ def process_image_features(
                 if num_width_tiles == 0 or num_height_tiles == 0:
                     break
 
-                num_tiles_in_image = num_width_tiles * num_height_tiles
+                num_tiles_in_image = (num_width_tiles * num_height_tiles).tolist()
 
                 # Get global features [hw, D]
                 global_features = images_embeds[tile_index]
 
                 # Get local features [num_height_tiles * num_width_tiles, hw, D]
                 local_features = images_embeds[
-                    tile_index + 1 : tile_index + 1 + int(num_tiles_in_image)
+                    tile_index + 1 : tile_index + 1 + num_tiles_in_image
                 ]
 
                 tile_index += num_tiles_in_image + 1
@@ -378,14 +378,18 @@ def get_input_embeddings(
 
         batch_num_tiles = [0 for _ in range(bs)]
         total_tiles = []
+
+        # Total number of tiles in each batch
         for idx in range(bs):
             for jdx in range(max_n_images):
                 num_width_tiles, num_height_tiles = images_spatial_crop[idx][jdx]
                 if num_width_tiles == 0 or num_height_tiles == 0:
                     break
-                batch_num_tiles[idx] += 1 + num_width_tiles * num_height_tiles
+                batch_num_tiles[idx] += (
+                    1 + num_width_tiles * num_height_tiles
+                ).tolist()
 
-            total_tiles.append(pixel_values[idx, : int(batch_num_tiles[idx])])
+            total_tiles.append(pixel_values[idx, : batch_num_tiles[idx]])
 
         total_tiles = mx.concatenate(total_tiles, axis=0)
         assert total_tiles.shape[0] == sum(batch_num_tiles)