chenchenghao/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 11 Packages Projects Releases Wiki Activity

[Auto Sync] Update weight_utils.py (20260212) (#18692)

Browse Source 
Co-authored-by: github-actions[bot] <github-actions[bot]@users.noreply.github.com>
Co-authored-by: Dan Zheng <dzheng@x.ai>
This commit is contained in:
Lianmin Zheng
2026-02-12 16:26:05 -08:00
committed by GitHub
parent b48fe1d95e
commit 9815ee934c
3 changed files with 126 additions and 16 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
python/sglang/srt/layers/logits_processor.py
View File

@@ -55,7 +55,7 @@ from sglang.srt.model_executor.forward_batch_info import (
ForwardMode,
)
from sglang.srt.server_args import get_global_server_args
from sglang.srt.utils import is_npu, use_intel_amx_backend
from sglang.srt.utils.common import is_npu, use_intel_amx_backend
logger = logging.getLogger(__name__)
@@ -89,8 +89,8 @@ class LogitsProcessorOutput:
# The logprobs of input tokens. shape: [#token]
input_token_logprobs: Optional[torch.Tensor] = None
# The logprobs and ids of the top-k tokens in input positions. shape: [#seq, #token, k]
input_top_logprobs_val: List = None
input_top_logprobs_idx: List = None
input_top_logprobs_val: Optional[List] = None
input_top_logprobs_idx: Optional[List] = None
# The logprobs and ids of the requested token ids in input positions. shape: [#seq, n] (n is the number of requested token ids)
# Can contain either lists or GPU tensors (for delayed GPU-to-CPU transfer optimization)
input_token_ids_logprobs_val: Optional[List[Union[List[float], torch.Tensor]]] = (