0.0.7

What's Changed

• Sync with Triton 24.04
• Bump TRT-LLM version to 0.9.0
• Add support for llama-2-7b-chat, llama-3-8b, and llama-3-8b-instruct for both vLLM and TRT-LLM
• Improve error checking and error messages of building TRT-LLM engines
• Log the underlying convert_checkpoint.py and trtllm-build commands for reproducibility/visibility
• Don't call convert_checkpoint.py if converted weights are already found
• Call convert_checkpoint.py via subprocess to improve total memory usage
• Attempt to cleanup failed trtllm models in model repository if import or engine building fails, rather than leaving the model repository in an unfinished state.
• Update tests to wait for both HTTP and GRPC server endpoints to be ready before testing
  • Fixes intermittent ConnectionRefusedError in CI tests

Full Changelog: 0.0.6...0.0.7