fix: pass rope_theta argument when initializing LlamaLikeBlock for models like qwen2, mistral, etc.

[Shuai-Xie]

This PR aims to fix this issue #567.

    def fuse_transformer(self):
            ...
            blocks.append(
                LlamaLikeBlock(
                    hidden_size=self.model.config.hidden_size,
                    n_heads=self.model.config.num_attention_heads,
                    n_kv_heads=self.model.config.num_key_value_heads,
                    qkv_layer=qkv,
                    o_proj=module.self_attn.o_proj,
                    mlp=module.mlp,
                    norm_1=norm_1,
                    norm_2=norm_2,
                    dev=device,
                    max_seq_len=self.model.config.max_seq_len,
                    rope_theta=self.model.config.rope_theta,                # only add this line.
                )
            )

[casper-hansen]

Thanks for the fix!