- gqa_decode_attn: rewrite to per-KV-head, K in smem - gqa_prefill_attn: new kernel for Q_len > 1 with GQA |
||
|---|---|---|
| .. | ||
| gqa_decode_attn.cu | ||
| gqa_prefill_attn.cu | ||
- gqa_decode_attn: rewrite to per-KV-head, K in smem - gqa_prefill_attn: new kernel for Q_len > 1 with GQA |
||
|---|---|---|
| .. | ||
| gqa_decode_attn.cu | ||
| gqa_prefill_attn.cu | ||