vllm.model_executor.layers.quantization.compressed_tensors.compressed_tensors_moe ¶

Modules:

• compressed_tensors_moe_w4a4_nvfp4 –
• compressed_tensors_moe_w4a8_int8 –
• compressed_tensors_moe_w8a8_fp8 –
• compressed_tensors_moe_w8a8_int8 –
• compressed_tensors_moe_w8a8_mxfp8 –
• compressed_tensors_moe_wna16_marlin –
• compressed_tensors_moe_wna16_rdna3 –

  CompressedTensors MoE W4A16 using the fused RDNA3 (gfx1100) HIP kernel.

• rocm_moe_rdna –

  ROCm MoE kernel dispatcher.