Add int8 MLX variant; remove bf16 (OOM-killed on device)

.gitattributes

@@ -37,3 +37,4 @@ xnnpack/whisper_small_xnnpack.pte filter=lfs diff=lfs merge=lfs -text
 xnnpack/whisper_small_xnnpack_fp32.pte filter=lfs diff=lfs merge=lfs -text
 coreml/whisper_small_coreml_fp16.pte filter=lfs diff=lfs merge=lfs -text
 mlx/whisper_small_mlx_bf16.pte filter=lfs diff=lfs merge=lfs -text
+mlx/whisper_small_mlx_int8.pte filter=lfs diff=lfs merge=lfs -text

mlx/config.json

@@ -1,40 +1,25 @@
 {
   "$schema": "https://huggingface.co/software-mansion/react-native-executorch-spec/resolve/main/config.schema.json",
-  "model": "whisper_small",
+  "model": "whisper",
   "family": "whisper",
   "capabilities": [
     "speech-to-text"
   ],
-  "backend": "coreml",
-  "license": "apache-2.0",
+  "backend": "mlx",
+  "license": "mit",
   "tokenizer": "../tokenizer.json",
   "variants": [
     {
-      "file": "whisper_small_mlx_bf16.pte",
-      "precision": "bf16",
-      "quantized": false,
+      "file": "whisper_small_mlx_int8.pte",
+      "precision": "int8",
+      "quantized": true,
       "default": true,
       "methods": {
-        "decode": {
+        "encode": {
           "inputs": [
             {
               "shape": [
-                1,
-                128
-              ],
-              "dtype": "int64"
-            },
-            {
-              "shape": [
-                128
-              ],
-              "dtype": "int64"
-            },
-            {
-              "shape": [
-                1,
-                1500,
-                768
+                480000
               ],
               "dtype": "float32"
             }
@@ -43,18 +28,33 @@
             {
               "shape": [
                 1,
-                128,
-                51865
+                1500,
+                768
               ],
               "dtype": "float32"
             }
           ]
         },
-        "encode": {
+        "decode": {
           "inputs": [
             {
               "shape": [
-                480000
+                1,
+                1
+              ],
+              "dtype": "int64"
+            },
+            {
+              "shape": [
+                1
+              ],
+              "dtype": "int64"
+            },
+            {
+              "shape": [
+                1,
+                1500,
+                768
               ],
               "dtype": "float32"
             }
@@ -63,8 +63,8 @@
             {
               "shape": [
                 1,
-                1500,
-                768
+                1,
+                51865
               ],
               "dtype": "float32"
             }
@@ -73,4 +73,4 @@
       }
     }
   ]
-}
+}

mlx/{whisper_small_mlx_bf16.pte → whisper_small_mlx_int8.pte}

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7c51f14b5dd3f204f1e3a8aa4c4bcfe24a4ce490bc6512aaab85b462db0eaeee
-size 484420096
+oid sha256:0454d8f29289de86f8016b165a89bd3307e5e5437e854216c6ae35f8a94fa273
+size 289455744