envs

Environments

This directory contains reproducible runtime environment definitions for executing vLLM inference workflows.

Available Environments

uv Environments

Name	Python	vLLM	CUDA	Notes
u260423_vllm_compiled	3.12.11	0.19.2rc1.dev171+ge9f331d72 (built from source)	12.9.1	100% compiled from source. No precompiled wheel or bundled kernels — every .cu / .cpp is built locally against the FASRC CUDA 12.9 toolchain.
u260324_vllm_source	3.12.11	0.18.1rc1.dev101+ga32783bb3.precompiled	12.8	Editable install of a vLLM git tree using VLLM_USE_PRECOMPILED=1 — Python-only edits, kernels come from upstream tarball.
u260304_vllm	3.12.11	0.11.2	12.9.1	Precompiled vLLM 0.11.2 wheel. Fastest path to a working inference env.

Conda Environments

Name	Python	vLLM	CUDA	Notes
c250609_vllm085	3.12	0.8.5.post1	12.4	PyTorch 2.6.0

Docker Environments

No docker environments available yet.

Singularity Environments

No singularity environments available yet.

---

Purpose

Environments provide the foundation for all workflows, reports, and workshops in this repository. Each environment is a complete, reproducible setup that includes:

• Exact package versions with lock files
• Installation instructions
• Verification procedures
• Hardware and software specifications
• Troubleshooting guidance

Directory Structure

envs/
  README.md (this file)
  conda/          # Conda environments
  docker/         # Docker containers
  uv/             # uv-based Python environments
  singularity/    # Singularity containers

Naming Convention

Environment folders follow this naming pattern:

<type-prefix><date>_<short-description>

Environment Type Prefixes:

Prefix	Type	Example
c	Conda	c260304_vllm
u	uv	u260304_vllm
d	Docker	d260304_cuda124
s	Singularity	s260304_vllm

Date Format: YYMMDD (year, month, day)

Using Environments

Each environment directory contains:

1. README.md - Complete setup guide with:

   • Specifications
   • Installation instructions
   • Verification steps
   • Usage examples
   • Troubleshooting

2. Environment Definition Files:

   • Conda: environment.yml
   • uv: pyproject.toml, uv.lock
   • Docker: Dockerfile
   • Singularity: *.def

3. Additional Resources: (when applicable)

   • Test scripts
   • Setup scripts
   • Configuration files
   • Documentation

Environment Versioning

For Minor Updates (Bug Fixes/Patches)

Keep the same date and append version suffix:

• Example: u260304_vllm → u260304_vllm_v1

For Major Changes

Create a new environment with a new date:

• New dependencies added/removed
• Version changes (Python, CUDA, PyTorch, vLLM)
• Major configuration changes
• Example: u260304_vllm → u260315_vllm

Document the relationship in the new environment's README under a "History" section.

Contributing

To contribute a new environment, see the detailed guidelines in CONTRIBUTING.md.

Quick checklist:

1. Create appropriately named directory
2. Include all environment definition files
3. Write comprehensive README
4. Test environment creation from scratch
5. Submit pull request

External Resources

Important: Environment files (conda .tar.gz, Docker images, Singularity .sif files) should NOT be committed to this repository. Instead:

• Document exact build/creation steps
• Provide external links (Google Drive, AWS S3, Docker Hub)
• Include checksums for verification
• Docker images should reference Docker Hub repositories

Questions?

For questions about environments or contributing:

• Check existing environments for examples
• Review CONTRIBUTING.md
• Open an issue for discussion
• Contact repository maintainers