Merge pull request huggingface#223 from eliebak/fix-torch-compile

remove torch compile

src/nanotron/models/llama.py

@@ -165,7 +165,7 @@ def __init__(
             bias=False,
             async_communication=tp_linear_async_communication and tp_mode is TensorParallelLinearMode.REDUCE_SCATTER,
         )
-        self.split_silu_mul = torch.compile(GLUActivation(config.hidden_act))
+        self.split_silu_mul = GLUActivation(config.hidden_act)
 
     def forward(self, hidden_states):  # [seq_length, batch_size, hidden_dim]
         merged_states = self.gate_up_proj(hidden_states)