zhuzilin/ring-flash-attention Issues
ring flash attention with BPT
Closed 3精度问题
Closed 3Perf on 8*H800
Closed 1ring attention实现原理
Closed 12flash attention版本
Closed 11Numerical errors in backward
Updated 4verify causal masking
Updated 6Is it support casual attention?
Closed 1是否需要更新全局最大值？
Closed 2Benchmark Question
Closed 2run the code has error
Updated 4多机训练速度问题
Updated 3mask to zigzag attention
Closed 1多卡qkv维度问题
Closed 2stripe_flash_attn_varlen_func
Closed 1large memory usage
Updated 5test on 8*A800
Closed 3关于tp和分块操作最终聚合的问题
Closed 1Question about updating lse
Closed 24卡 A100 测试 ring attention 性能不太行呢
Closed 1ring attention with varlen
Closed 2Great work
Updated