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

Enable JIT clamp_position and resolve_future_token_ids on ROCm (#21116)

Browse Source
This commit is contained in:
Lianmin Zheng
2026-03-22 22:33:54 -07:00
committed by GitHub
parent 32a85ef128
commit 7050011dee
4 changed files with 5 additions and 16 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
python/sglang/jit_kernel/csrc/elementwise/clamp_position.cuh
View File

@@ -29,7 +29,7 @@ struct ClampPosition {
SymbolicSize N = {"num_elements"};
SymbolicDevice device_;
device_.set_options<kDLCUDA>();
device_.set_options<kDLCUDA, kDLROCM>();
TensorMatcher({N}) //
.with_dtype<T>()

2
python/sglang/jit_kernel/csrc/elementwise/resolve_future_token_ids.cuh
View File

@@ -34,7 +34,7 @@ struct ResolveFutureTokenIds {
SymbolicSize N = {"num_tokens"};
SymbolicSize M = {"map_size"};
SymbolicDevice device_;
device_.set_options<kDLCUDA>();
device_.set_options<kDLCUDA, kDLROCM>();
TensorMatcher({N}).with_dtype<T>().with_device(device_).verify(input_ids);