A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 4, 17:30 UTC
[NVIDIA] Add sm_110 (Jetson Thor) to CUDA 13.0 build targets (#39233)
Johnny committed
Apr 23, 2026 at 21:42 UTC
7f95a66cbffcd111c6d37abdcb7ca297cec47b78
4 files changed
+5
-5
docker/Dockerfile
+2
-2
@@ -188,7 +188,7 @@ RUN --mount=type=cache,target=/root/.cache/uv \
188
# Explicitly set the list to avoid issues with torch 2.2
189
# See https://github.com/pytorch/pytorch/pull/123243
190
# From versions.json: .torch.cuda_arch_list
191
-ARG torch_cuda_arch_list='7.5 8.0 8.6 8.9 9.0 10.0 12.0+PTX'
191
+ARG torch_cuda_arch_list='7.5 8.0 8.6 8.9 9.0 10.0 11.0 12.0+PTX'
192
ENV TORCH_CUDA_ARCH_LIST=${torch_cuda_arch_list}
193
#################### BUILD BASE IMAGE ####################
194
@@ -765,7 +765,7 @@ ARG PIP_EXTRA_INDEX_URL UV_EXTRA_INDEX_URL
765
ENV UV_HTTP_TIMEOUT=500
766
767
# install kv_connectors if requested
768
-ARG torch_cuda_arch_list='7.5 8.0 8.6 8.9 9.0 10.0 12.0+PTX'
768
+ARG torch_cuda_arch_list='7.5 8.0 8.6 8.9 9.0 10.0 11.0 12.0+PTX'
769
ENV TORCH_CUDA_ARCH_LIST=${torch_cuda_arch_list}
770
RUN --mount=type=cache,target=/root/.cache/uv \
771
--mount=type=bind,source=requirements/kv_connectors.txt,target=/tmp/kv_connectors.txt,ro \
docker/docker-bake.hcl
+1
-1
@@ -20,7 +20,7 @@ variable "NVCC_THREADS" {
20
}
21
22
variable "TORCH_CUDA_ARCH_LIST" {
23
- default = "8.0 8.9 9.0 10.0"
23
+ default = "8.0 8.9 9.0 10.0 11.0 12.0"
24
}
25
26
variable "COMMIT" {
docker/versions.json
+1
-1
@@ -32,7 +32,7 @@
32
"default": "false"
33
},
34
"TORCH_CUDA_ARCH_LIST": {
35
- "default": "7.5 8.0 8.6 8.9 9.0 10.0 12.0+PTX"
35
+ "default": "7.5 8.0 8.6 8.9 9.0 10.0 11.0 12.0+PTX"
36
},
37
"MAX_JOBS": {
38
"default": "2"
tools/flashinfer-build.sh
+1
-1
@@ -35,7 +35,7 @@ elif [[ "${CUDA_VERSION}" == 12.[8-9]* ]]; then
35
FI_TORCH_CUDA_ARCH_LIST="7.5 8.0 8.9 9.0a 10.0a 10.3a 12.0"
36
else
37
# CUDA 13.0+
38
- FI_TORCH_CUDA_ARCH_LIST="7.5 8.0 8.9 9.0a 10.0f 12.0"
38
+ FI_TORCH_CUDA_ARCH_LIST="7.5 8.0 8.9 9.0a 10.0f 11.0 12.0f"
39
fi
40
41
echo "🏗️ Building FlashInfer AOT for arches: ${FI_TORCH_CUDA_ARCH_LIST}"