No specific extraction method could be identified for this model. The producer either did not document it or used a proprietary pipeline.
What is a refusal direction? →The same weights this author released in different packaging. Pick the format that matches your runtime.
Calibrated W4A4 NVFP4: AWQ+GPTQ on routed experts, FP8 attention/shared/lm_head,
BF16 keep-list (incl. PLE table), static FP8 KV scheme (dropped at serve time:
vLLM v0.29 QSA does not support KV quantization).
Status: serving under debug (NaN logits on vLLM v0.29 TP8; benchmark vs BF16
and reference NVFP4 in progress).
The author's README evolved over time. Click a version to see its content at that point.
Every model page has an "Open in app" button that hands off directly to a local runtime of your choice - Infrahuman, LM Studio, or Ollama. No API keys, no subscription, no prompt leakage.