cvis-tmu/compute_canada_sif_files

Description

This is the repo to collect all relevant sif image files, convenient for Compute Canada job running

1. easyr1.sif: The sif image for EasyR1, convert if from Docker file
2. llamafactory.sif: The sif image for LlamaFactory, convert it from Docker image
3. verl-ngc-th2.7.1-cu12.6-vllm0.10.0.sif: The sif image for verl, convert it from Docker image
4. vlmevalkit.sif: The sif image for VLMEvalKit, build it from def file built from its README document installation section.
5. flow-factory.sif: The sif image for Flow-Factory, built from an Apptainer def file following its README Installation section and including additional dependencies (deepspeed, quantization, wandb, swanlab, nvidia extras).
6. worldforge_cuda24.sif: The sif image for WorldForge, built from an Apptainer def file following its README.
7. flow-factory.sif: The sif image for Flow Factory, built from an Apptainer def file following its README.
8. ms-swift402.sif: The sif image for ModelScope SWIFT, converted with Apptainer from the newest Docker Hub tag opencsghq/ms-swift:v3.11.1-cu126 (linux/amd64). Replaces the previous SWIFT 4.0.3 image. Verified contents: ms-swift 3.11.1, PyTorch 2.7.1, CUDA 12.6.3, vLLM 0.10.1.1, ModelScope 1.29.2, Transformers 4.56.2, on Ubuntu 22.04. Intended for Apptainer runs with --nv on hosts with CUDA 12.6 drivers.
9. worldscore_eval.sif: The sif image for WorldScore, built from an Apptainer def file following its README.
10. vsibench_eval.sif: The sif image for Thinking-in-Space, built from an Apptainer def file following its README.
11. SpatialLadder.sif: The sif image for SpatialLadder, built from an Apptainer def file based on nvidia/cuda:12.8.1-devel-ubuntu24.04, including PyTorch 2.8.0 (CUDA 12.8), flash-attention, vllm, deepspeed, transformers, trl, and all project dependencies.
12. vlmevalkit_easi.sif: The sif image for EASI, converted from the Docker image built with its Dockerfile (based on pytorch/pytorch:2.6.0-cuda12.4-cudnn9-devel), including VLMEvalKit, lmms-eval, PyTorch 2.7.1, flash-attn, and all EASI evaluation dependencies.
13. vgllm.sif: The sif image for VG-LLM, built from an Apptainer def file following its README, including Qwen2.5-VL, VGGT, and all dependencies for 3D scene understanding and spatial reasoning tasks.
14. RLCSD.sif: The sif image for RLCSD, built from an Apptainer def file based on nvidia/cuda:13.0.3-cudnn-devel-ubuntu24.04, including PyTorch 2.10.0 (CUDA 13.0), vLLM 0.18.0, flash-attn 2.8.3, deepspeed, transformers, and all project dependencies. Intended for Apptainer runs with --nv on hosts with CUDA 13 drivers.
15. diffusionopsd.sif: The sif image for DiffusionOPSD (upstream: worldbench/DiffusionOPSD), built from containers/diffusionopsd.def based on pytorch/pytorch:2.14.0-cuda13.0-cudnn9-devel, including PyTorch 2.14.0 (CUDA 13.0), Transformers 4.51.3, Diffusers 0.39.0, and the project reward/video stack. Weights are not baked in. Intended for Apptainer runs with --nv on hosts with CUDA 13.0 drivers.
16. think3d-metric-opsd.sif: The sif image for Think3D / SPAgent, built from environment.def on nvidia/cuda:13.4.2-cudnn-devel-ubuntu24.04. Includes Python 3.12, PyTorch 2.14.0+cu130, torchvision 0.29.0+cu130, and the project requirements.txt stack. nvcc reports CUDA 13.4. PyTorch kernels cover compute capabilities 7.5, 8.0, 8.6, 9.0, 10.0, and 12.0. Intended for Apptainer runs with --nv on hosts whose NVIDIA driver supports CUDA 13 (driver branches 580, 590, 595, and 610).