adapt to dsv32 on dcu

852a49c5 · maxiao · 8f7453e3 · 852a49c5 · 852a49c5 · 852a49c5
Commit 852a49c5 authored Sep 30, 2025 by maxiao
20 changed files
--- a/python/sglang/srt/layers/attention/trtllm_mla_backend.py
+++ b/python/sglang/srt/layers/attention/trtllm_mla_backend.py
--- a/python/sglang/srt/layers/attention/vision.py
+++ b/python/sglang/srt/layers/attention/vision.py
--- a/python/sglang/srt/layers/communicator.py
+++ b/python/sglang/srt/layers/communicator.py
--- a/python/sglang/srt/layers/elementwise.py
+++ b/python/sglang/srt/layers/elementwise.py
@@ -187,9 +187,7 @@ fused_dual_residual_rmsnorm_kernel_autotune = rmsnorm_autotune(
 def fused_dual_residual_rmsnorm(x, residual, weight1, weight2, eps, autotune=False):
    assert len(x.shape) == 2
-    assert (
+    assert x.shape == residual.shape and x.dtype == residual.dtype
-        x.shape == residual.shape and x.dtype == residual.dtype
-    ), f"{x.shape=} {residual.shape=} {x.dtype=} {residual.dtype=}"
    output, mid = torch.empty_like(x), torch.empty_like(x)
    bs, hidden_dim = x.shape
    if autotune:

--- a/python/sglang/srt/layers/layernorm.py
+++ b/python/sglang/srt/layers/layernorm.py
--- a/python/sglang/srt/layers/linear.py
+++ b/python/sglang/srt/layers/linear.py
--- a/python/sglang/srt/layers/logits_processor.py
+++ b/python/sglang/srt/layers/logits_processor.py
--- a/python/sglang/srt/layers/moe/ep_moe/layer.py
+++ b/python/sglang/srt/layers/moe/ep_moe/layer.py
--- a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=256,N=256,device_name=NVIDIA_H800,dtype=fp8_w8a8,block_shape=[128, 128].json
+++ b/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=256,N=256,device_name=NVIDIA_H800,dtype=fp8_w8a8,block_shape=[128, 128].json
--- a/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=512,N=128,device_name=NVIDIA_H800,dtype=fp8_w8a8,block_shape=[128, 128].json
+++ b/python/sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=512,N=128,device_name=NVIDIA_H800,dtype=fp8_w8a8,block_shape=[128, 128].json
--- a/python/sglang/srt/layers/moe/fused_moe_triton/fused_moe_triton_config.py
+++ b/python/sglang/srt/layers/moe/fused_moe_triton/fused_moe_triton_config.py
--- a/python/sglang/srt/layers/moe/fused_moe_triton/layer.py
+++ b/python/sglang/srt/layers/moe/fused_moe_triton/layer.py
--- a/python/sglang/srt/layers/parameter.py
+++ b/python/sglang/srt/layers/parameter.py
--- a/python/sglang/srt/layers/quantization/compressed_tensors/compressed_tensors.py
+++ b/python/sglang/srt/layers/quantization/compressed_tensors/compressed_tensors.py
--- a/python/sglang/srt/layers/quantization/compressed_tensors/schemes/__init__.py
+++ b/python/sglang/srt/layers/quantization/compressed_tensors/schemes/__init__.py
--- a/python/sglang/srt/layers/quantization/compressed_tensors/schemes/compressed_tensors_w8a8_int8.py
+++ b/python/sglang/srt/layers/quantization/compressed_tensors/schemes/compressed_tensors_w8a8_int8.py
--- a/python/sglang/srt/layers/quantization/deep_gemm_wrapper/configurer.py
+++ b/python/sglang/srt/layers/quantization/deep_gemm_wrapper/configurer.py
--- a/python/sglang/srt/layers/quantization/mxfp4.py
+++ b/python/sglang/srt/layers/quantization/mxfp4.py
--- a/python/sglang/srt/layers/quantization/quark/quark_moe.py
+++ b/python/sglang/srt/layers/quantization/quark/quark_moe.py
--- a/python/sglang/srt/layers/quantization/w4afp8.py
+++ b/python/sglang/srt/layers/quantization/w4afp8.py