Supported Models#
Generative Models#
Model |
Support |
W8A8 |
LoRA |
Tensor Parallel |
Expert Parallel |
Data Parallel |
Piecewise Kunlun Graph |
|---|---|---|---|---|---|---|---|
Qwen3 |
✅ |
✅ |
✅ |
✅ |
✅ |
✅ |
|
Qwen3-Moe |
✅ |
✅ |
✅ |
✅ |
✅ |
✅ |
✅ |
Qwen3-Next |
✅ |
✅ |
✅ |
✅ |
✅ |
✅ |
✅ |
Qwen3.8 |
✅ |
✅ |
✅ |
||||
Deepseek v3.2 |
✅ |
✅ |
✅ |
✅ |
✅ |
Qwen3.8 (27B-W8A8): verified on a single XPU (tensor_parallel_size=1) with a
compressed-tensors W8A8 checkpoint; LoRA and multi-XPU tensor parallel
are pending verification. See the Qwen3.8 tutorial.
Multimodal Language Models#
Model |
Support |
W8A8 |
LoRA |
Tensor Parallel |
Expert Parallel |
Data Parallel |
Piecewise Kunlun Graph |
|---|---|---|---|---|---|---|---|
Qwen3-VL |
✅ |
✅ |
✅ |
✅ |
✅ |