Modelpublic

xpuenabler/gpt-oss-15.5b-23E-SFT-v6-gpu

sourceHugging Faceapache-2.0updated 5mo agoView on Hugging Face

0likes3downloads

2 commits on main

1c8e9a95mo ago

v6-gpu: v5-gpu recipe + Expert ReduceSum decomposition baked into the export. INT4 g=64 attn/dense/lm_head/embed; MoE experts and router FP16; SDPA + Softmax + Expert-ReduceSum decomposed; dtype=fp16 + defensive fp32->fp16 rewrite. Stateful, GenAI LLMPipeline-compatible.

xpuenabler

1aea3125mo ago

initial commit

xpuenabler