build(deps): update dependencies

[DRL-NextGen]

Removed packages

Package	Version
filetype	1.2.0
minify-html	0.18.1
nvidia-cublas-cu12	12.8.4.1
nvidia-cuda-cupti-cu12	12.8.90
nvidia-cuda-nvrtc-cu12	12.8.93
nvidia-cuda-runtime-cu12	12.8.90
nvidia-cudnn-cu12	9.10.2.21
nvidia-cufft-cu12	11.3.3.83
nvidia-cufile-cu12	1.13.1.3
nvidia-curand-cu12	10.3.9.90
nvidia-cusolver-cu12	11.7.3.90
nvidia-cusparse-cu12	12.5.8.93
nvidia-cusparselt-cu12	0.7.1
nvidia-nvjitlink-cu12	12.8.93
nvidia-nvshmem-cu12	3.4.5
nvidia-nvtx-cu12	12.8.90

Added packages

Package	Version
cuda-tile	1.3.0
cuda-toolkit	13.0.2
fastsafetensors	0.3
ml-dtypes	0.5.4
nvidia-cublas	13.1.0.3
nvidia-cuda-cupti	13.0.85
nvidia-cuda-nvrtc	13.0.88
nvidia-cuda-runtime	13.0.96
nvidia-cudnn-cu13	9.19.0.56
nvidia-cufft	12.0.0.61
nvidia-cufile	1.15.1.6
nvidia-curand	10.4.0.35
nvidia-cusolver	12.0.4.66
nvidia-cusparse	12.6.3.3
nvidia-cusparselt-cu13	0.8.0
nvidia-nccl-cu13	2.28.9
nvidia-nvjitlink	13.0.88
nvidia-nvshmem-cu13	3.4.5
nvidia-nvtx	13.0.85
tilelang	0.1.9
z3-solver	4.15.4.0

Downgraded packages

Package	From	To	Type
apache-tvm-ffi	0.1.10	0.1.9	🔴
detect-secrets	0.13.1+ibm.64.dss (76a765c2)	0.13.1+ibm.64.dss (38ba7e33)	🔴
imagehash	4.3.2	4.3.1	🟢
numpy	2.2.6	2.1.3	🟡
ydata-profiling	4.18.4	4.14.0	🟡

Upgraded packages

Package	From	To	Type
https://github.com/ibm/detect-secrets.git	(38ba7e335083e0a4db8c53e9be414795b27891e9)	—	🔴
backrefs	6.2	7.0	🔴
boto3	1.42.96	1.43.2	🟡
botocore	1.42.96	1.43.2	🟡
build	1.4.4	1.5.0	🟡
cachetools	7.0.6	7.1.1	🟡
cbor2	5.9.0	6.0.1	🔴
cloudpathlib	0.23.0	0.24.0	🔴
cuda-bindings	12.9.4	13.2.0	🔴
cuda-pathfinder	1.5.3	1.5.4	🟢
cuda-python	12.9.4	13.2.0	🔴
fastapi-cloud-cli	0.17.0	0.17.1	🔴
fastcore	1.12.43	1.12.44	🟢
flashinfer-cubin	0.6.6	0.6.8.post1	🔴
flashinfer-python	0.6.6	0.6.8.post1	🔴
google-auth	2.49.2	2.50.0	🟡
greenlet	3.4.0	3.5.0	🟡
huggingface-hub	1.12.0	1.13.0	🟡
llvmlite	0.44.0	0.47.0	🔴
mako	1.3.11	1.3.12	🟢
mistral-common	1.11.0	1.11.1	🟢
numba	0.61.2	0.65.0	🔴
nvidia-nccl-cu12	2.27.5	2.30.4	🟡
openai	2.32.0	2.33.0	🟡
outlines-core	0.2.11	0.2.14	🔴
plum-dispatch	2.8.0	2.9.0	🟡
pymysql	1.1.2	1.1.3	🟢
python-multipart	0.0.26	0.0.27	🔴
pytz	2026.1.post1	2026.2	🟡
quack-kernels	0.3.11	0.4.1	🔴
s3transfer	0.16.1	0.17.0	🔴
scipy	1.15.3, v1.16.3	1.15.3	🟢
tombi	0.9.24	0.10.2	🔴
torch	2.10.0	2.11.0	🟡
torchaudio	2.10.0	2.11.0	🟡
torchvision	0.25.0	0.26.0	🔴
tox	4.53.0	4.53.1	🟢
transformers	5.6.2	5.7.0	🟡
typer	0.25.0	0.25.1	🔴
uv	0.11.7	0.11.8	🔴
virtualenv	21.2.4	21.3.0	🟡
vllm	0.19.1	0.20.1	🔴
wcwidth	0.6.0	0.7.0	🔴
xgrammar	0.1.33	0.2.0	🔴

[AlessandroPomponio]

@christian-pinto vllm has released v0.20.0 and v0.20.1 which seems to have big changes: https://github.com/vllm-project/vllm/releases/tag/v0.20.0.

• CUDA 13.0 default: Default CUDA wheel on PyPI and vllm/vllm-openai:v0.20.0 image switched to CUDA 13.0; architecture lists and build-args cleaned up ([Build] Switch default CUDA to 13.0, update CUDA architecture lists, clean up stale build-args vllm-project/vllm#39878), and CUDA bumped to 13.0.2 to match PyTorch 2.11.0 ([Build] Bump CUDA to 13.0.2 to match PyTorch 2.11.0 vllm-project/vllm#40669). As a general rule of thumb, our CUDA version policy follows PyTorch's. We highly recommend to install vLLM with uv and use --torch-backend=cu129 if you are on CUDA 12.9.
• PyTorch 2.11 upgrade ([release 2.11] Update to torch 2.11 vllm-project/vllm#34644): vLLM ships on torch 2.11 for CUDA, and XPU is now also on torch 2.11 ([XPU] Upgrade torch 2.11 for xpu vllm-project/vllm#37947) — XPU is no longer pinned to 2.10. This is a breaking change for environment dependency.
• Python 3.14: Added to the supported Python version list ([Build] Add Python 3.14 to supported version list. vllm-project/vllm#34770).

Please test it out and let us know if it works

[AlessandroPomponio]

Note that according to vllm-project/vllm#39878

Volta (SM70 / SM75) architectures are removed since they are no longer supported by PyTorch 2.11 with newer CUDA.

This means that newer versions of vLLM will not work with V100s anymore. CC @mgazz as well

[AlessandroPomponio]

I see that they're publishing wheels for the cu129 variant in https://github.com/vllm-project/vllm/releases but they don't appear to be available on PyPI at the time of writing this.

[AlessandroPomponio]

Superseded by #943