Qwen3-0.6B-3bit-g256

Qwen3-0.6B quantized to 3-bit (group_size 256, symmetric) with an 8-bit embedding; ≈4.40 bits/weight.

Weights are provided dequantized in fp16 for direct loading and evaluation — the quantization is already baked into the weights, so evaluate the model as-is (no further quantization).