quickstart: install latest vllm (#3)

vllm setup: install torch first (#4)

Updates to build_vllm_from_source.sh (#2)

Add link to trained model on HuggingFace

Add summary explaining why this repo exists and key challenges

Add memory utilities and documentation structure (#1)

Add comprehensive OOM prevention guide and memory profiling results

Add gloo backend flags for single-GPU GRPO

Use vLLM v0.13.0 stable instead of nightly, add build-from-source docs

Pin vLLM to specific cu130 wheel for reproducibility

Initial DGX Spark ML setup guide