zeromodels/qwen3-30b-a3b-thinking-2507

Run Qwen3 with Keras 3: JAX, PyTorch, or TensorFlow

zeromodels/qwen3-30b-a3b-thinking-2507

Pure-Keras 3 conversion of Qwen/Qwen3-30B-A3B-Thinking-2507 for zeromodels. One implementation runs unmodified on TensorFlow / Torch / JAX. This is a Mixture-of-Experts Qwen3; weights are stored in bfloat16.

For model details, license, and usage terms, see the upstream model card.

Paper: Qwen3 Technical Report (arXiv:2505.09388) · HF Papers

Paper: Qwen2.5-1M Technical Report (arXiv:2501.15383) · HF Papers

Paper: MInference 1.0: Accelerating Pre-filling for Long-Context LLMs via Dynamic Sparse Attention (arXiv:2407.02490) · HF Papers

Paper: RULER: What's the Real Context Size of Your Long-Context Language Models? (arXiv:2404.06654) · HF Papers

Paper: Training-Free Long-Context Scaling of Large Language Models (arXiv:2402.17463) · HF Papers

✨ Quick start

import os
os.environ["KERAS_BACKEND"] = "torch"  # or "jax" / "tensorflow"

from zeromodels.models.qwen3_moe import Qwen3MoeTextGenerate, Qwen3MoeTokenizer

model = Qwen3MoeTextGenerate.from_weights("zeromodels/qwen3-30b-a3b-thinking-2507")
tokenizer = Qwen3MoeTokenizer.from_weights("zeromodels/qwen3-30b-a3b-thinking-2507")

inputs = tokenizer("Give me a short introduction to large language models.")
outputs = model.generate(**inputs, max_new_tokens=128)
print(tokenizer.decode(outputs[0]))

Special Thanks

A huge thank you to the Qwen team at Alibaba for creating and releasing these models.

License: Apache 2.0.