- activation-recompute-fusion
- awq-dequant-gemm
- blockwise-fp8-gemm
- deepseek-mla-lowrank-proj
- deepseek-mla-proj-backward
- dora-magnitude-decompose
- fp4-stochastic-round
- fp8-attention-backward
- fp8-blockwise-gemm
- fp8-flash-attention
- fp8-kv-decode-attention
- fp8-per-token-quant-transpose
- fp8-quant-dequant-fused
- fsdp-mixed-precision-cast
- fused-rmsnorm-fp8-quant
- gptq-actorder-gemm
- int4-kv-cache-quant
- int4-weight-only-decode-gemv
- int8-w8a8-gemm
- kv-cache-4bit-groupwise
- kv-cache-fp8-quant-paged
- layernorm-residual-fp8-backward
- logits-gather-lastpos
- lora-backward-fused
- lora-fused-gemm
- mixed-precision-gemm-split
- mtp-head-forward
- mxfp4-dequant-gemm
- mxfp4-training-backward
- mxfp8-blockwise-gemm
- mxfp8-dual-axis-quantize
- nvfp4-dequant-gemm
- nvfp4-gemm-backward
- nvfp4-quantize
- pipeline-microbatch-stash
- qlora-nf4-dequant-gemm
- quantized-optimizer-state
- qwen3-next-gdn-inproj-backward
- qwen3-next-output-gate-backward
- rmsnorm-quant-transpose-fused
- smoothquant-fused-scale
- w4a16-dequant-gemm
- zamba-shared-attn-lora-backward
- zamba-shared-attn-lora-qkv
- 9.92 kB