nikita-savelyev-cerebras/tiny-random-qwen3.8-flash-next-fp8

Tiny random Qwen3.8 Flash-Next FP8 text model

This synthetic Qwen4Exp text model uses the architecture and fine-grained FP8 expert layout of Qwen/Qwen3.8-Flash-Next-FP8, reduced to four decoder blocks and four routed experts per block. Its weights are random and intended for loader tests, not inference quality. A compact tokenizer is adapted from malaiwah/qwen4-exp-tiny-random-bf16.