-deaufosse- - Malignant -v0.2-

| Parameter | Likely value | |--------------------|----------------------------------------| | Base model | Llama 2 7B, Mistral 7B, or Gemma 2B | | Fine‑tuning method | QLoRA, LoRA, or full fine‑tune | | Context length | 4k–8k tokens (extendable via RoPE) | | Quantization | Likely none (v0.2 is full fp16) | | Training data | Adversarial prompts, dark fiction, exploits | | License | CC‑BY‑NC‑SA or Apache 2.0 |