add test_eagle_get_hidden_states.py

Author: co63oc

tests/operators/test_eagle_get_hidden_states.py

@@ -0,0 +1,68 @@
+# Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+#
+# 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 unittest
+
+import numpy as np
+import paddle
+
+from fastdeploy.model_executor.ops.gpu import eagle_get_hidden_states
+
+
+class TestEagleGetHiddenStates(unittest.TestCase):
+    def test_eagle_get_hidden_states(self):
+        np.random.seed(2023)
+        paddle.seed(2023)
+        bs = np.random.randint(1, 8 + 1, dtype=np.int32)
+        input_token_num = np.random.randint(2 * 1024, 4 * 1024 + 1, dtype=np.int32)
+        dim_embed = np.array(1024, dtype=np.int32)
+        actual_draft_token_num = np.random.randint(2, 6, dtype=np.int32)
+
+        seq_lens_this_time = np.random.randint(0, 2, bs, dtype=np.int32)
+        seq_lens_encoder = np.random.randint(0, input_token_num // bs + 1, bs, dtype=np.int32)
+        accept_nums = np.random.randint(0, actual_draft_token_num + 1, bs, dtype=np.int32)
+        base_model_seq_lens_this_time = np.random.randint(0, input_token_num // bs + 1, bs, dtype=np.int32)
+        base_model_seq_lens_encoder = np.random.randint(0, 2, bs, dtype=np.int32)
+
+        seq_lens_decoder = np.random.randint(0, input_token_num // bs + 1, bs, dtype=np.int32)
+        stop_flags = np.random.randint(0, 2, bs, dtype=np.int32)
+
+        seq_lens_this_time_tensor = paddle.to_tensor(seq_lens_this_time, dtype=paddle.int32)
+        seq_lens_encoder_tensor = paddle.to_tensor(seq_lens_encoder, dtype=paddle.int32)
+        accept_nums_tensor = paddle.to_tensor(accept_nums, dtype=paddle.int32)
+        base_model_seq_lens_this_time_tensor = paddle.to_tensor(base_model_seq_lens_this_time, dtype=paddle.int32)
+        base_model_seq_lens_encoder_tensor = paddle.to_tensor(base_model_seq_lens_encoder, dtype=paddle.int32)
+
+        seq_lens_decoder_tensor = paddle.to_tensor(seq_lens_decoder, dtype=paddle.int32)
+        stop_flags_tensor = paddle.to_tensor(stop_flags, dtype=paddle.int32)
+
+        input = np.random.randint(0, 10, (input_token_num, dim_embed), dtype=np.int32)
+        input_tensor = paddle.to_tensor(input, dtype=paddle.float16)
+        gpu_out = eagle_get_hidden_states(
+            input_tensor,
+            seq_lens_this_time_tensor,
+            seq_lens_encoder_tensor,
+            seq_lens_decoder_tensor,
+            stop_flags_tensor,
+            accept_nums_tensor,
+            base_model_seq_lens_this_time_tensor,
+            base_model_seq_lens_encoder_tensor,
+            actual_draft_token_num,
+        )
+        out_ref = np.array([6, 4, 3, 3], dtype=np.float16)
+        np.testing.assert_allclose(gpu_out.numpy()[0][0:4], out_ref)
+
+
+if __name__ == "__main__":
+    unittest.main()

tests/operators/test_eagle_get_self_hidden_states.py

@@ -0,0 +1,52 @@
+# Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
+#
+# 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 unittest
+
+import numpy as np
+import paddle
+
+from fastdeploy.model_executor.ops.gpu import eagle_get_self_hidden_states
+
+
+class TestEagleGetSelfHiddenStates(unittest.TestCase):
+    def test_eagle_get_self_hidden_states(self):
+        paddle.seed(2023)
+        np.random.seed(2023)
+        bs = np.random.randint(1, 8 + 1, dtype=np.int32)
+        input_token_num = np.random.randint(2 * 1024, 4 * 1024 + 1, dtype=np.int32)
+        dim_embed = np.array(1024, dtype=np.int32)
+
+        last_seq_lens_this_time = np.random.randint(0, input_token_num // bs, bs, dtype=np.int32)
+        seq_lens_this_time = np.random.randint(0, input_token_num // bs, bs, dtype=np.int32)
+        step_idx = np.arange(0, bs, dtype=np.int32)
+
+        last_seq_lens_this_time_tensor = paddle.to_tensor(last_seq_lens_this_time, dtype=paddle.int32)
+        seq_lens_this_time_tensor = paddle.to_tensor(seq_lens_this_time, dtype=paddle.int32)
+        step_idx_tensor = paddle.to_tensor(step_idx, dtype=paddle.int64)
+
+        input = np.random.randint(0, 10, (input_token_num, dim_embed), dtype=np.int32)
+        input_tensor = paddle.to_tensor(input, dtype=paddle.float16)
+        gpu_out = eagle_get_self_hidden_states(
+            input_tensor,
+            last_seq_lens_this_time_tensor,
+            seq_lens_this_time_tensor,
+            step_idx_tensor,
+        )
+        out_ref = np.array([5, 4, 2, 8], dtype=np.float16)
+        np.testing.assert_allclose(gpu_out.numpy()[0][0:4], out_ref)
+
+
+if __name__ == "__main__":
+    unittest.main()