apps/FastDeploy
1
0
Fork 0
mirror of https://github.com/PaddlePaddle/FastDeploy.git synced 2025-10-05 16:48:03 +08:00
Code Issues Actions 2 Packages Projects Releases Wiki Activity

MTP rejection_topp add topk input (#3031)

Browse Source
This commit is contained in:
chen
2025-07-28 13:58:45 +08:00
committed by GitHub
parent dd877f38b1
commit 01485cd28b
1 changed files with 2 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
fastdeploy/spec_decode/mtp.py
View File

@@ -275,6 +275,7 @@ class MTPProposer(Proposer):
# self.model_inputs["caches"] = self.cache_kvs
# Inherit generation hyperparameters from the main model for consistency
self.model_inputs["top_p"] = self.main_model_inputs["top_p"]
self.model_inputs["top_k"] = self.main_model_inputs["top_k"]
self.model_inputs["temperature"] = self.main_model_inputs["temperature"]
self.model_inputs["eos_token_id"] = self.main_model_inputs["eos_token_id"]
self.model_inputs["penalty_score"] = self.main_model_inputs["penalty_score"]
@@ -528,6 +529,7 @@ class MTPProposer(Proposer):
self.sampling_metadata = SamplingMetadata(
temperature=self.model_inputs["temperature"],
top_p=self.model_inputs["top_p"],
top_k=self.model_inputs["top_k"],
step_idx=self.model_inputs["step_idx"],
pre_token_ids=self.model_inputs["pre_ids"],
frequency_penalties=self.model_inputs["frequency_score"],