[MUSA]: Add flashinfer sampling backend (#24978)

Co-authored-by: github-actions[bot] <github-actions[bot]@users.noreply.github.com>
Co-authored-by: zhiguo.qin <zhiguo.qin@mthreads.com>
Co-authored-by: R0CKSTAR <yeahdongcn@gmail.com>

python/pyproject_other.toml

@@ -115,7 +115,7 @@ srt_musa = [
   "sglang[runtime_common]",
   "torch",
   "torch_musa",
-  "torchada>=0.1.55",
+  "torchada>=0.1.56",
   "mthreads-ml-py",
   "mate>=0.2.0",
   "deep-gemm>=0.1.3",

python/sglang/srt/layers/sampler.py

@@ -20,6 +20,7 @@ from sglang.srt.utils.common import (
     crash_on_warnings,
     get_bool_env_var,
     is_cuda,
+    is_musa,
     is_npu,
 )
 
@@ -33,6 +34,15 @@ if is_cuda():
         top_p_renorm_prob,
     )
 
+if is_musa():
+    from sgl_kernel import (
+        min_p_sampling_from_probs,
+        top_k_renorm_prob,
+        top_k_top_p_sampling_from_probs,
+        top_p_renorm_prob,
+    )
+
+
 if is_npu():
     import torch_npu