Reconverted in SFT mode & Add 4B LLM

Basically this merge model can be converted using either the turbo config or the sft config.
I found the sft version to have better results especially at higher step counts, this lines up with my experiences in the gradio.
It will still work with few steps in this mode.

.gitattributes

@@ -40,3 +40,5 @@ ace-vae-BF16.gguf filter=lfs diff=lfs merge=lfs -text
 Ace-Qwen3-Embedding-0.6B-Q8_0.gguf filter=lfs diff=lfs merge=lfs -text
 Ace-Qwen3-Embedding-0.6B-BF16.gguf filter=lfs diff=lfs merge=lfs -text
 acestep-5Hz-lm-0.6B-BF16.gguf filter=lfs diff=lfs merge=lfs -text
+acestep-5Hz-lm-4B-BF16.gguf filter=lfs diff=lfs merge=lfs -text
+acestep-5Hz-lm-4B-Q8_0.gguf filter=lfs diff=lfs merge=lfs -text

acestep-5Hz-lm-4B-BF16.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9fca9915919a1108da33d8d98fbe6c1a07bfee101d27098416650dcbde037b72
+size 8384454592

acestep-5Hz-lm-4B-Q8_0.gguf

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:972f91147a167f0c041f1b158d67985a82c0f6a852e68cdf70e46030cf08b1bc
+size 4457323648

acestep-v15-sftturbo50-BF16.gguf

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:96c19802c1928e263e59a4fc01fe9bb4f3af24efaf4582c66b643a3fd38e3b36
-size 4791642080
+oid sha256:af24191ae7748e16e5d1860dbe37e581baea5a5d39a7bd479ca09c84fa023fc7
+size 4791642016