Step-3.5-Flash_gpuep_1.0_gpu_128

Source model: stepfun-ai/Step-3.5-Flash

Introduction

This model was created using Olive Mobius Builder.

Quantization Strategy

• Activations: FP16
• MoE experts: Olive GPTQ search / Group 64 / INT2 Weights (asymmetric)
• Attention, dense MLP, and shared experts: INT8 / Group 128 (symmetric)
• LM head: INT4 MatMulNBits / Group 32
• No vision encoder (text-only decoder)

---

License

This model is a quantized derivative of stepfun-ai/Step-3.5-Flash and is distributed under the same license as the source model, the Apache License, Version 2.0. A copy of the license is included in the LICENSE file of this repository.

Modifications copyright (c) 2026 Advanced Micro Devices, Inc. All rights reserved. AMD has modified the original model files (quantization and conversion to ONNX) for optimized deployment on AMD hardware.

Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License.