Upload fp32.sft

More promising layerskip-aware weights (it actually seems to perform better than the base model, atlhough with just one voice I'm comparing so far)

models/ckpt/ar+nar-layerskip-llama-8/fp32.sft

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3f71dd7fb7cdd9ea646dc6e82ab63fcf281e13be8021247c87524c7faca540bc
-size 456274658
+oid sha256:2feb27e01d483cdd3f1e55c6f5fafeb257b2af791f01791623289ea1174d4cbb
+size 456274522