tzchen07/rai-n3-nano-4b-safety-lora-r64-ipfix

Add USAGE.md with verified working vLLM inference recipe + critical gotchas

Add missing chat_template.jinja (vLLM/transformers need this for chat formatting)

Add missing adapter_config.json (peft requires this to load)

Add prompt template: v3_1_CategoryOnly.jinja

Add prompt template: v3_1_VerdictReason.jinja

Add script: analyze_fps.py

Add script: patch_knorm.py

Add script: eval_patched.sh

Add script: eval_g4_terse.py

Add script: eval_n3_raw_v14_VR.py

Add script: binary_metrics_v2.py

Add script: train_n3_advanced.py

Add script: train_n3_DPO.py

Add script: train_n3_NVrec.py

Add script: score_aligned.py

Add script: merge_n3.py

Update model card with comprehensive training details, eval setup, and prompt template

RAI safety classifier: N3 Nano 4B LoRA r=64 + IP-fix (Block%=0.170%, P=100%)

initial commit