v0.2.2

• Update torch to align with upstream project, torch==2.2.2 torchvision==0.17.2 torchaudio==2.2.2
• Implement nvidia variant for CPUs without AVX2 support, default-nvidia-noavx2
• Implement nvidia variant with TensorRT-LLM backend, default-nvidia-tensorrtllm

v0.2.1

Minor updates to documentation and compose example:

• Created config folder: cache
• Created config folder: instruction-templates
• Provided example for ROCM and Arc compose deployment
• Modified docker-entrypoint.sh to populate instruction-templates

v0.2

Refactor project to reflect upstream development.

• Introduce variants: base, default
• Introduce platforms: nvidia, cpu, rocm, arc
• Change base image from nvidia/cuda to ubuntu:22.04 to slim docker images
• Deprecate variants: cuda, llama-cpu, triton, monkey-patch
  • Use default-cpu instead of llama-cpu for new deployments
  • Use default-nvidia instead of cuda, triton
  • Please raise an issue if new version has missing functionality

v0.1

Last version before v0.2 refactor.