Fix utf8_check_validity (pytorch#6543)

index is already increased in the loop, so we should not over advance

extension/android/jni/jni_layer_llama.cpp

@@ -42,21 +42,21 @@ bool utf8_check_validity(const char* str, size_t length) {
       uint8_t next_byte = static_cast<uint8_t>(str[i + 1]);
       if ((byte & 0xE0) == 0xC0 &&
           (next_byte & 0xC0) == 0x80) { // 2-byte sequence
-        i += 2;
+        i += 1;
       } else if (
           (byte & 0xF0) == 0xE0 && (next_byte & 0xC0) == 0x80 &&
           (i + 2 < length) &&
           (static_cast<uint8_t>(str[i + 2]) & 0xC0) ==
               0x80) { // 3-byte sequence
-        i += 3;
+        i += 2;
       } else if (
           (byte & 0xF8) == 0xF0 && (next_byte & 0xC0) == 0x80 &&
           (i + 2 < length) &&
           (static_cast<uint8_t>(str[i + 2]) & 0xC0) == 0x80 &&
           (i + 3 < length) &&
           (static_cast<uint8_t>(str[i + 3]) & 0xC0) ==
               0x80) { // 4-byte sequence
-        i += 4;
+        i += 3;
       } else {
         return false; // Invalid sequence
       }