sglangv0.5.2 & support Qwen3-Next-80B-A3B-Instruct

2025-09-13 17:00:20 +08:00
commit 118f1fc726
2037 changed files with 515371 additions and 0 deletions
--- a/sgl-kernel/python/sgl_kernel/memory.py
+++ b/sgl-kernel/python/sgl_kernel/memory.py
@@ -0,0 +1,18 @@
+import torch
+
+
+def set_kv_buffer_kernel(
+    k_cache: torch.Tensor,
+    v_cache: torch.Tensor,
+    loc: torch.Tensor,
+    k: torch.Tensor,
+    v: torch.Tensor,
+    fallback: bool = False,
+):
+    try:
+        if fallback:
+            raise RuntimeError("Fallback to torch implementation")
+        torch.ops.sgl_kernel.store_kv_cache(k_cache, v_cache, loc, k, v)
+    except RuntimeError:  # ok, fallback to torch implementation
+        k_cache[loc] = k
+        v_cache[loc] = v