apps/FastDeploy
1
0
Fork 0
mirror of https://github.com/PaddlePaddle/FastDeploy.git synced 2025-10-30 03:22:05 +08:00
Code Issues Actions 6 Packages Projects Releases Wiki Activity

[FIX 2.0.2] Topk topp sampling fix (#2805)

Browse Source 
* fix topk-topp

* fix
This commit is contained in:
Sunny-bot1
2025-07-10 21:15:03 +08:00
committed by GitHub
parent e681e1e719
commit 4025ea7e5b
6 changed files with 27 additions and 70 deletions
Download Patch File Download Diff File Expand all files Collapse all files

24
fastdeploy/model_executor/layers/sample/ops/top_k_top_p_sampling.py
View File

@@ -64,9 +64,20 @@ def top_k_top_p_sampling(
seed=seed,
k=k,
mode=mode)
# rejection
elif top_p_class == "rejection":
ids = rejection_top_p_sampling(x, top_p, top_k, seed, order)
_ = None
# base non-truncated
elif top_p_class == "base_non_truncated":
_, ids = paddle.tensor.top_p_sampling(x,
top_p,
threshold=threshold,
topp_seed=topp_seed,
seed=seed,
k=k,
mode="non-truncated")
# base truncated
else:
_, ids = paddle.tensor.top_p_sampling(x,
top_p,
@@ -74,7 +85,7 @@ def top_k_top_p_sampling(
topp_seed=topp_seed,
seed=seed,
k=k,
mode=mode)
mode="truncated")
return _, ids
@@ -102,26 +113,25 @@ def air_top_p_sampling(
def rejection_top_p_sampling(
x: paddle.Tensor,
top_p: paddle.Tensor,
top_k: Optional[paddle.Tensor] = None,
top_k: paddle.Tensor,
seed: int = -1,
order: Literal['top_k_first', 'joint'] = "top_k_first",
) -> paddle.Tensor:
"""
rejection_top_p_sampling
"""
assert top_p is not None, "Top_p should not be none when FD_SAMPLING_CLASS is rejection"
try:
from fastdeploy.model_executor.ops.gpu import (
rejection_top_p_sampling, top_k_renorm_probs)
if top_k is None:
if paddle.count_nonzero(top_k) == 0:
ids = rejection_top_p_sampling(
x,
top_p,
None,
seed,
)
elif top_k is not None and top_p is not None:
else:
if order == "top_k_first":
renorm_probs = top_k_renorm_probs(x, top_k)
ids = rejection_top_p_sampling(
@@ -137,10 +147,6 @@ def rejection_top_p_sampling(
top_k,
seed,
)
else:
raise ValueError(
"Top_p cannot be none."
)
except ImportError:
raise RuntimeError("Cannot import rejection_top_p_sampling op.")
return ids