bumping sgl-deep-gemm to 0.2.0 (#39371)

python/pyproject.toml

@@ -76,7 +76,7 @@ dependencies = [
   "sentencepiece",
   "setproctitle",
   "sgl-deep-ep==0.1.2",
-  "sgl-deep-gemm==0.1.7",
+  "sgl-deep-gemm==0.2.0",
   "sglang-kernel==0.4.7",
   "smg-grpc-servicer>=0.9.0",
   "soundfile==0.13.1",

python/sglang/srt/batch_invariant_ops/batch_invariant_ops.py

@@ -255,6 +255,12 @@ def _matmul_persistent_deepgemm(
     dtype = a.dtype
     out = torch.empty((M, N), device=a.device, dtype=dtype)
 
+    # DeepGEMM 0.2 defaults BF16 GEMMs to cuBLASLt, whose reduction can
+    # depend on the batch size. Older wheels always use the invariant kernel.
+    get_deterministic = getattr(deep_gemm, "get_deterministic_algorithms", None)
+    deterministic = get_deterministic() if get_deterministic else True
+    if not deterministic:
+        deep_gemm.use_deterministic_algorithms(True)
     try:
         deep_gemm.bf16_gemm_nn(a, b, out)
     except RuntimeError as e:
@@ -264,6 +270,9 @@ def _matmul_persistent_deepgemm(
             f"Consider increasing MIN_DEEPGEMM_DIM in matmul_persistent() or disabling DeepGEMM "
             f"for small matrices. Original error: {e}"
         ) from e
+    finally:
+        if not deterministic:
+            deep_gemm.use_deterministic_algorithms(False)
 
     # TODO can this be put in DeepGEMM's `c`?
     if bias is not None: