This is the mixed-precision INT4/W4A16 AutoRound checkpoint used in the DynaExQ paper experiments. It is derived from Qwen/Qwen3-30B-A3B-Instruct-2507.

The checkpoint uses symmetric expert-weight quantization with mixed higher-precision fallbacks recorded in quantization_config.json.

python

from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = "Kris2017/Qwen3-30B-A3B-Instruct-2507-W4A16-AutoRound"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    device_map="auto",
    torch_dtype="auto",
)

Code, experiment scripts, manifests, and paper sources are available in DynaQuant. The original model license and usage restrictions continue to apply.