apps/FastDeploy
1
0
Fork 0
mirror of https://github.com/PaddlePaddle/FastDeploy.git synced 2025-10-05 08:37:06 +08:00
Code Issues Actions 2 Packages Projects Releases Wiki Activity

【Inference Optimize】DeepSeek-V3-model MLA Optimize (#3886)

Browse Source 
* support MLA chunk_size auto search & cuda_graph
This commit is contained in:
AIbin
2025-09-11 10:46:09 +08:00
committed by GitHub
parent 637d96c6ae
commit a7392a0ff9
23 changed files with 375 additions and 310 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
tests/layers/test_append_attention.py
View File

@@ -380,10 +380,12 @@ class TestAppendGroupQueryAttnWithRope(unittest.TestCase):
self.max_enc_len_this_time = paddle.to_tensor([self.max_enc_len_this_time], "int32", place=paddle.CPUPlace())
self.max_dec_len_this_time = paddle.to_tensor([self.max_dec_len_this_time], "int32", place=paddle.CPUPlace())
self.seq_lens_this_time = self.seq_lens_encoder
self.decoder_batch_ids = paddle.full([self.batch_size], 0, dtype="int32")
self.decoder_tile_ids_per_batch = paddle.full([self.batch_size], 0, dtype="int32")
decode_max_tile_size = 1024 * self.batch_size * np.ceil((2 * 10) / 12)
self.decoder_batch_ids = paddle.full([int(decode_max_tile_size)], 0, dtype="int32")
self.decoder_tile_ids_per_batch = paddle.full([int(decode_max_tile_size)], 0, dtype="int32")
self.decoder_num_blocks_cpu = paddle.full([1], 0, dtype="int32").pin_memory()
self.decoder_num_blocks_device = paddle.full([1], 0, dtype="int32")
self.decoder_chunk_size_device = paddle.full([1], 64, dtype="int32")
self.max_len_tensor_cpu = paddle.full([8], 0, dtype="int32").cpu()
self.encoder_batch_ids = paddle.full([self.batch_size], 0, dtype="int32")
@@ -484,6 +486,8 @@ class TestAppendGroupQueryAttnWithRope(unittest.TestCase):
self.decoder_batch_ids,
self.decoder_tile_ids_per_batch,
self.decoder_num_blocks_cpu,
self.decoder_num_blocks_device,
self.decoder_chunk_size_device,
self.max_len_tensor_cpu,
self.encoder_batch_ids,
self.encoder_tile_ids_per_batch,