From cade85e2d8def6a2d5bf6efff9abad1bf4cb092b Mon Sep 17 00:00:00 2001
From: mrmhodak <mrmhodak@gmail.com>
Date: Sun, 12 Nov 2023 02:20:40 -0500
Subject: [PATCH 1/2] MigraphX version of reference code

---
 language/bert/migraphx_SUT.py | 96 +++++++++++++++++++++++++++++++++++
 1 file changed, 96 insertions(+)
 create mode 100644 language/bert/migraphx_SUT.py

diff --git a/language/bert/migraphx_SUT.py b/language/bert/migraphx_SUT.py
new file mode 100644
index 0000000000..588a738c35
--- /dev/null
+++ b/language/bert/migraphx_SUT.py
@@ -0,0 +1,96 @@
+# coding=utf-8
+# Copyright 2018 The Google AI Language Team Authors.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import array
+import json
+import os
+import sys
+
+sys.path.insert(0, os.getcwd())
+
+import mlperf_loadgen as lg
+import numpy as np
+import migraphx
+from squad_QSL import get_squad_QSL
+
+
+class BERT_migraphx_SUT:
+    def __init__(self, args):
+
+        print("Loading ONNX model...")
+        self.quantized = args.quantized
+        self.batch_size = args.batch_size
+        self.model_path = args.model
+        self.model = migraphx.parse_onnx(
+            self.model_path, default_dim_value=self.batch_size
+        )
+        print("Quantize to fp16...")
+        migraphx.quantize_fp16(self.model)
+        print("Compile for gpu...")
+        self.model.compile(migraphx.get_target("gpu"))
+
+        print("Constructing SUT...")
+        self.sut = lg.ConstructSUT(self.issue_queries, self.flush_queries)
+        print("Finished constructing SUT.")
+
+        print("Warmup...")
+        self.fd = {
+            "input_ids": np.zeros((self.batch_size, 384), dtype=np.int64),
+            "input_mask": np.zeros((self.batch_size, 384), dtype=np.int64),
+            "segment_ids": np.zeros((self.batch_size, 384), dtype=np.int64),
+        }
+        self.model.run(self.fd)
+
+        self.qsl = get_squad_QSL(args)
+
+    def issue_queries(self, query_samples):
+        idx = [q.index for q in query_samples]
+        query_id = [q.id for q in query_samples]
+        bs = self.batch_size
+        for i in range(0, len(idx), bs):
+            eval_features = self.qsl.get_features(idx[i : i + bs])
+            actual_batchsize = len(eval_features)
+            self.fd["input_ids"][:actual_batchsize] = np.array(
+                [eval_feature.input_ids for eval_feature in eval_features]
+            ).astype(np.int64)
+            self.fd["input_mask"][:actual_batchsize] = np.array(
+                [eval_feature.input_mask for eval_feature in eval_features]
+            ).astype(np.int64)
+            self.fd["segment_ids"][:actual_batchsize] = np.array(
+                [eval_feature.segment_ids for eval_feature in eval_features]
+            ).astype(np.int64)
+            results = self.model.run(self.fd)
+            scores = [np.array(result) for result in results]
+            outputs = np.stack(scores, axis=-1)
+
+            response_array_refs = []
+            response = []
+            for i, qid in enumerate(query_id[i : i + bs]):
+                response_array = array.array(
+                    "B", np.array(outputs[i], np.float32).tobytes()
+                )
+                response_array_refs.append(response_array)
+                bi = response_array.buffer_info()
+                response.append(lg.QuerySampleResponse(qid, bi[0], bi[1]))
+            lg.QuerySamplesComplete(response)
+
+    def flush_queries(self):
+        pass
+
+    def __del__(self):
+        print("Finished destroying SUT.")
+
+def get_migraphx_sut(args):
+    return BERT_migraphx_SUT(args)

From c2d5c55d00213a8bfc77b2189bcf1ba01f1ea6d5 Mon Sep 17 00:00:00 2001
From: Miro <mirhodak@amd.com>
Date: Tue, 13 Feb 2024 00:43:57 -0500
Subject: [PATCH 2/2] Update migraphx_SUT.py

---
 language/bert/migraphx_SUT.py | 13 +------------
 1 file changed, 1 insertion(+), 12 deletions(-)

diff --git a/language/bert/migraphx_SUT.py b/language/bert/migraphx_SUT.py
index 588a738c35..c0be2a52c8 100644
--- a/language/bert/migraphx_SUT.py
+++ b/language/bert/migraphx_SUT.py
@@ -1,17 +1,6 @@
 # coding=utf-8
-# Copyright 2018 The Google AI Language Team Authors.
+# Copyright 2023 AMD
 #
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-#     http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
 
 import array
 import json