apps/FastDeploy
1
0
Fork 0
mirror of https://github.com/PaddlePaddle/FastDeploy.git synced 2026-04-23 00:17:25 +08:00
Code Issues Actions 19 Packages Projects Releases Wiki Activity

[XPU]Set top_p=0.0 by default on XPU to optimize performance (#5686)

Browse Source 
Co-authored-by: ddchenhao66 <dhaochen163.com>
This commit is contained in:
ddchenhao66
2025-12-23 11:01:01 +08:00
committed by GitHub
parent 3aee5c4bf5
commit 4a74f5ab9b
1 changed files with 3 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files
fastdeploy/worker/xpu_model_runner.py
+3 -1
View File
@@ -704,7 +704,9 @@ class XPUModelRunner(ModelRunnerBase):
dtype="int64",
)
self.share_inputs["eos_token_id"] = paddle.full([self.model_config.eos_tokens_lens, 1], 0, dtype="int64")
self.share_inputs["top_p"] = paddle.full([max_num_seqs, 1], self.model_config.top_p, dtype="float32")
# self.share_inputs["top_p"] = paddle.full([max_num_seqs, 1], self.model_config.top_p, dtype="float32")
# self.share_inputs["top_p"] default to 0.0 on XPU for consideration of the performance
self.share_inputs["top_p"] = paddle.full([max_num_seqs, 1], 0.0, dtype="float32")
self.share_inputs["top_k"] = paddle.full([max_num_seqs, 1], 0, dtype="int64")
self.share_inputs["top_k_list"] = [0] * max_num_seqs
self.share_inputs["min_p"] = paddle.full([max_num_seqs, 1], 0.0, dtype="float32")