[MetaxGPU] Support FastDeploy on metax gpu (#3241)

* [MetaxGPU] Support FastDeploy on metax gpu

* Update metax_worker.py

1. change worker log;
2. remove custom allreduce, adapt it later;
3. remove cuda graph;

* Update __init__.py

1. remove metax's key work comment

* Update __init__.py

1. remove metax's key word comment;
2. add fused_moe_kernel_paddle import

---------

Co-authored-by: yongqiangma <xing.wo@163.com>

fastdeploy/model_executor/pre_and_post_process.py

@@ -45,6 +45,14 @@ elif current_platform.is_dcu():
         step_paddle,
         update_inputs,
     )
+elif current_platform.is_maca():
+    from fastdeploy.model_executor.ops.gpu import (
+        get_padding_offset,
+        save_output,
+        set_stop_value_multi_ends,
+        step_paddle,
+        update_inputs,
+    )
 else:
     from fastdeploy.model_executor.ops.gpu import (
         get_padding_offset,
@@ -225,6 +233,19 @@ def post_process_normal(
             model_output.stop_seqs_len,
             False,
         )  # multi ends
+    elif current_platform.is_maca():
+        set_stop_value_multi_ends(
+            sampler_output.sampled_token_ids,
+            model_output.stop_flags,
+            model_output.seq_lens_this_time,
+            model_output.eos_token_id,
+            model_output.next_tokens,
+            model_output.pre_ids,
+            model_output.step_idx,
+            model_output.stop_token_ids,
+            model_output.stop_seqs_len,
+            False,
+        )  # multi ends
     else:
         set_stop_value_multi_ends(
             sampler_output.sampled_token_ids,
@@ -573,6 +594,18 @@ def rebuild_padding(
             output_padding_offset,
             max_input_length,
         )
+    elif current_platform.is_maca():
+        from fastdeploy.model_executor.ops.gpu import rebuild_padding
+
+        hidden_states = rebuild_padding(
+            tmp_out,
+            cum_offsets,
+            seq_len_this_time,
+            seq_lens_decoder,
+            seq_lens_encoder,
+            output_padding_offset,
+            max_input_length,
+        )
     else:
         raise RuntimeError("Not supported platform")
     return hidden_states