File size: 259 Bytes
ac53c7a | 1 2 3 4 5 6 | ---
library_name: transformers
---
Tiny random Mistral4, pre-quantized with a per-layer plan: fp8 (block 16x16) on attention and routed experts, nvfp4 on shared experts. `dense_reference.pt` holds the bf16 weights before quantization; `make.py` rebuilds it.
|