fix: wrong parameter for flash attention in pickle __getstate__

Author: abetlen

llama_cpp/llama.py

@@ -1777,7 +1777,7 @@ def __getstate__(self):
             logits_all=self.context_params.logits_all,
             embedding=self.context_params.embeddings,
             offload_kqv=self.context_params.offload_kqv,
-            flash_offload=self.context_params.flash_offload,
+            flash_attn=self.context_params.flash_attn,
             # Sampling Params
             last_n_tokens_size=self.last_n_tokens_size,
             # LoRA Params