# Serving never uses this environment. Torch/CUDA come from the pinned base image.
llmcompressor==0.14.0
compressed-tensors==0.19.0
transformers==5.17.0
datasets==5.0.1
numpy==2.5.3