change precision to fp32

mlc-chat-config.json

@@ -27,7 +27,8 @@
     "context_window_size": 768,
     "prefill_chunk_size": 0,
     "max_batch_size": 80,
-    "tensor_parallel_shards": 1
+    "tensor_parallel_shards": 1,
+    "dtype": "float32"
   },
   "vocab_size": 50265,
   "context_window_size": 768,

roberta-cls-model-q0f32-android.tar

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:aed76df6f74c17a1d9bc067b97bceb4ad064b0d8b05db9de9c061d2ee85e73d0
-size 112431
+oid sha256:127c68c8bcf0923fa75081acea5197cb48137a5a74f1f35c3c80c8533754ab4a
+size 111882