# engine: sglang==0.5.4
# engine: flashinfer-python==0.4.1
# sglang 0.5.4 on CUDA 12 (torch 2.8). Used by the Qwen3-VL SGLang reward
# server, which needs --torch 2.8.0 --sglang 0.5.4 --transformers 4.57.1.
#
# sglang and flashinfer are both installed --no-deps. flashinfer 0.4.1 pins
# apache-tvm-ffi==0.1.0b15, a pre-release that has since been withdrawn from
# PyPI, so the published dependency set can no longer be resolved at all. Their
# runtime dependencies are listed here instead, with apache-tvm-ffi held to a
# released 0.1.x -- the range flashinfer itself moved to from 0.5.0 onwards.
#
# The torch family is left out: pyproject.toml pins it.
IPython
aiohttp
anthropic>=0.20.0
apache-tvm-ffi==0.1.9
blobfile==3.0.0
build
click
compressed-tensors
cuda-python
datasets
decord2
einops
fastapi
grpcio==1.75.1
grpcio-health-checking==1.75.1
grpcio-reflection==1.75.1
grpcio-tools==1.75.1
hf_transfer
huggingface_hub
interegular
llguidance<0.8.0,>=0.7.11
modelscope
msgspec
ninja
numpy
nvidia-cudnn-frontend>=1.13.0
nvidia-cutlass-dsl==4.2.1
nvidia-ml-py
openai==1.99.1
openai-harmony==0.0.4
orjson
outlines==0.1.11
packaging
partial_json_parser
pillow
prometheus-client>=0.20.0
psutil
py-spy
pybase64
pydantic
python-multipart
pyzmq>=25.1.2
requests
scipy
sentencepiece
setproctitle
sgl-kernel==0.3.16.post3
sglang-router
soundfile==0.13.1
tabulate
tiktoken
timm==1.0.16
torch_memory_saver==0.0.9
torchao==0.9.0
torchvision
tqdm
transformers==4.57.1
uvicorn
uvloop
xgrammar==0.1.25
