mirror of
https://github.com/NVIDIA/TensorRT-LLM.git
synced 2026-02-07 11:41:47 +08:00
70 lines
1.3 KiB
Plaintext
70 lines
1.3 KiB
Plaintext
--extra-index-url https://download.pytorch.org/whl/cu128
|
|
-c constraints.txt
|
|
accelerate>=1.7.0
|
|
build
|
|
colored
|
|
cuda-python>=12
|
|
diffusers>=0.27.0
|
|
lark
|
|
mpi4py
|
|
numpy<2
|
|
onnx>=1.18.0
|
|
onnx_graphsurgeon>=0.5.2
|
|
openai
|
|
polygraphy
|
|
psutil
|
|
nvidia-ml-py>=12
|
|
# Just a wrapper since nvidia-modelopt requires pynvml
|
|
pynvml==12.0.0
|
|
pulp
|
|
pandas
|
|
h5py==3.12.1
|
|
StrEnum
|
|
sentencepiece>=0.1.99
|
|
tensorrt
|
|
# https://docs.nvidia.com/deeplearning/frameworks/pytorch-release-notes/rel-25-06.html#rel-25-06 uses 2.8.0a0.
|
|
torch>=2.7.1,<=2.8.0a0
|
|
torchvision
|
|
nvidia-modelopt[torch]~=0.33.0
|
|
nvidia-nccl-cu13
|
|
nvidia-cuda-nvrtc-cu13
|
|
transformers==4.55.0
|
|
prometheus_client
|
|
prometheus_fastapi_instrumentator
|
|
pydantic>=2.9.1
|
|
pydantic-settings[yaml]
|
|
omegaconf
|
|
pillow==10.3.0
|
|
wheel<=0.45.1
|
|
optimum
|
|
# evaluate needs datasets>=2.0.0 which triggers datasets>3.1.0 which is not stable: https://github.com/huggingface/datasets/issues/7467
|
|
datasets==3.1.0
|
|
evaluate
|
|
mpmath>=1.3.0
|
|
click
|
|
click_option_group
|
|
aenum
|
|
pyzmq
|
|
fastapi==0.115.4
|
|
uvicorn
|
|
setuptools<80
|
|
ordered-set
|
|
peft
|
|
einops
|
|
flashinfer-python @ git+https://github.com/VALLIS-NERIA/flashinfer.git@29e57482ad6410444aa86cf92c669dbbb506e978
|
|
opencv-python-headless
|
|
xgrammar==0.1.21
|
|
llguidance==0.7.29
|
|
jsonschema
|
|
backoff
|
|
nvtx
|
|
matplotlib # FIXME: this is added to make nvtx happy
|
|
meson
|
|
ninja
|
|
etcd3
|
|
blake3
|
|
soundfile
|
|
triton==3.3.1; platform_machine == "x86_64"
|
|
tiktoken
|
|
blobfile
|