Add int8 MLX variant alongside bf16

.gitattributes

@@ -37,3 +37,4 @@ xnnpack/whisper_tiny_xnnpack.pte filter=lfs diff=lfs merge=lfs -text
 xnnpack/whisper_tiny_xnnpack_fp32.pte filter=lfs diff=lfs merge=lfs -text
 coreml/whisper_tiny_coreml_fp16.pte filter=lfs diff=lfs merge=lfs -text
 mlx/whisper_tiny_mlx_bf16.pte filter=lfs diff=lfs merge=lfs -text
+mlx/whisper_tiny_mlx_int8.pte filter=lfs diff=lfs merge=lfs -text

mlx/config.json

@@ -1,12 +1,12 @@
 {
   "$schema": "https://huggingface.co/software-mansion/react-native-executorch-spec/resolve/main/config.schema.json",
-  "model": "whisper_tiny",
+  "model": "whisper",
   "family": "whisper",
   "capabilities": [
     "speech-to-text"
   ],
   "backend": "mlx",
-  "license": "apache-2.0",
+  "license": "mit",
   "tokenizer": "../tokenizer.json",
   "variants": [
     {
@@ -15,18 +15,38 @@
       "quantized": false,
       "default": true,
       "methods": {
+        "encode": {
+          "inputs": [
+            {
+              "shape": [
+                480000
+              ],
+              "dtype": "float32"
+            }
+          ],
+          "outputs": [
+            {
+              "shape": [
+                1,
+                1500,
+                384
+              ],
+              "dtype": "float32"
+            }
+          ]
+        },
         "decode": {
           "inputs": [
             {
               "shape": [
                 1,
-                128
+                1
               ],
               "dtype": "int64"
             },
             {
               "shape": [
-                128
+                1
               ],
               "dtype": "int64"
             },
@@ -43,13 +63,21 @@
             {
               "shape": [
                 1,
-                128,
+                1,
                 51865
               ],
               "dtype": "float32"
             }
           ]
-        },
+        }
+      }
+    },
+    {
+      "file": "whisper_tiny_mlx_int8.pte",
+      "precision": "int8",
+      "quantized": true,
+      "default": false,
+      "methods": {
         "encode": {
           "inputs": [
             {
@@ -69,8 +97,43 @@
               "dtype": "float32"
             }
           ]
+        },
+        "decode": {
+          "inputs": [
+            {
+              "shape": [
+                1,
+                1
+              ],
+              "dtype": "int64"
+            },
+            {
+              "shape": [
+                1
+              ],
+              "dtype": "int64"
+            },
+            {
+              "shape": [
+                1,
+                1500,
+                384
+              ],
+              "dtype": "float32"
+            }
+          ],
+          "outputs": [
+            {
+              "shape": [
+                1,
+                1,
+                51865
+              ],
+              "dtype": "float32"
+            }
+          ]
         }
       }
     }
   ]
-}
+}

mlx/whisper_tiny_mlx_int8.pte

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ce7945ecce2eebde1645f3e1d98a6fdddeccc5e02afbbf961c0806bd62f59569
+size 59889792