qbench_export.json

{
  "quantization": "nf4",
  "source_base": "McGill-NLP/LLM2Vec-Meta-Llama-3-8B-Instruct-mntp",
  "source_supervised": "McGill-NLP/LLM2Vec-Meta-Llama-3-8B-Instruct-mntp-supervised",
  "torch": "2.6.0+cu124",
  "cuda": "12.4",
  "method": "Snapshot-downloaded base WITHOUT adapter, bnb-quantized via LlamaBiModel.from_pretrained, save_pretrained. MNTP adapter files copied into base/ so reload mirrors Hub flow. Supervised adapter in sibling dir for LLM2VEC_LOCAL_PEFT."
}