use flashinfer.sampling (#18696)

2026-02-26 10:02:38 +08:00
parent 2739d7df62
commit 2ad475b4ed
8 changed files with 14 additions and 431 deletions
--- a/sgl-kernel/python/sgl_kernel/init.py
+++ b/sgl-kernel/python/sgl_kernel/init.py
@@ -101,13 +101,9 @@ from sgl_kernel.quantization import (
    ggml_mul_mat_vec_a8,
 )
 from sgl_kernel.sampling import (
-    min_p_sampling_from_probs,
    top_k_mask_logits,
    top_k_renorm_prob,
-    top_k_top_p_sampling_from_logits,
-    top_k_top_p_sampling_from_probs,
    top_p_renorm_prob,
-    top_p_sampling_from_probs,
 )
 from sgl_kernel.speculative import (
    build_tree_kernel_efficient,