tiny-qwen3-llvq

A one-layer Qwen3 of 148 KB in the LLVQ format, for tests. Its weights mean nothing.

It has lattice layers, 4-bit layers, a rotation and a 4-bit embedding, so loading it goes through every part of the format. It is the fixture llvq-tetra/tests/fixtures/mini of github.com/pjmalandrino/llvq, written by the Rust test the_mini_fixture_describes_a_whole_qwen3_layer.

pip install llvq-tetra

from transformers import AutoModelForCausalLM

model = AutoModelForCausalLM.from_pretrained("Pier-Jean/tiny-qwen3-llvq")