diff --git a/gguf-py/gguf/tensor_mapping.py b/gguf-py/gguf/tensor_mapping.py
index ff22a60..783b148 100644
--- a/gguf-py/gguf/tensor_mapping.py
+++ b/gguf-py/gguf/tensor_mapping.py
@@ -1355,10 +1355,12 @@ class TensorNameMap:
 
         MODEL_TENSOR.ATTN_SUB_NORM: (
             "model.layers.{bid}.self_attn.inner_attn_ln",  # bitnet
+            "model.layers.{bid}.self_attn.attn_sub_norm",  # bitnet (transformers >= 4.57)
         ),
 
         MODEL_TENSOR.FFN_SUB_NORM: (
             "model.layers.{bid}.mlp.ffn_layernorm",  # bitnet
+            "model.layers.{bid}.mlp.ffn_sub_norm",  # bitnet (transformers >= 4.57)
         ),
 
         MODEL_TENSOR.DEC_ATTN_NORM: (
diff --git a/src/models/bitnet.cpp b/src/models/bitnet.cpp
index a08881b..e9f0124 100644
--- a/src/models/bitnet.cpp
+++ b/src/models/bitnet.cpp
@@ -129,7 +129,7 @@ llama_model_bitnet::graph::graph(const llama_model & model, const llm_graph_para
                 model.layers[il].ffn_gate, NULL, model.layers[il].ffn_gate_s,
                 NULL,                      NULL, NULL,
                 NULL,
-                LLM_FFN_SILU, LLM_FFN_PAR, il);
+                LLM_FFN_RELU_SQR, LLM_FFN_PAR, il);
         cb(cur, "ffn_sub_out", il);
 
         cur = build_norm(cur,