ninja
psutil
ray >= 2.9
sentencepiece
numpy
torch == 2.2.0
transformers @ git+https://github.com/huggingface/transformers.git@0e4a1c3401421d3342938a183c7b021cdac6fd79 # for cohere, remove when a new release drops
uvicorn
openai # for fastapi's openai proxy emulation
xformers >= 0.0.24
einops  # Required for phi-1_5
fschat >= 0.2.23
pydantic >= 2.0
fastapi
colorlog
einops # for phi
prometheus_client # for prometheus metrics
triton >= 2.1.0
lark == 1.1.8 # for grammars
pynvml == 11.5.0
scipy # for quip
rich
cupy-cuda12x == 12.3.0 # install cupy-cuda11x for CUDA 11.8
outlines >= 0.0.27
bitsandbytes >= 0.41.0
loguru
hf_transfer # for faster downloads