edenzzzz / ring-flash-attention Goto Github PK

This repo implements the RingAttention with FlashAttention. Currently, this repo implements:

• ring_flash_attn_func: ring attention version of flash_attn_func
• ring_flash_attn_varlen_func: ring attention version of flash_attn_varlen_func
• zigzag_ring_flash_attn_func: an optimized version of ring_flash_attn_func, see issue#2
• zigzag_ring_flash_attn_varlen_func: an optimized version of ring_flash_attn_varlen_func
• stripe_flash_attn_func: stripe attention version of ring_flash_attn_func, the block size is set to 1 to use flash_attn api.

Note that

• all function has the *_func, *_kvpacked_func, *_qkvpacked_func variant implemented.
• the varlen versions only support passing one cu_seqlens.

The main idea is to use the softmax_lse output from the flash attention kernels.

The current performance on 8xH800 is (benchmark/benchmark_qkvpacked_func.py):

Note that

• when running the benchmark with with 8 gpu, the flash attn code is running with 1/8 computation of ring attention.
• nvlink between GPUs are required for high performance.
• the varlen versions are slow at the moment, please use the non-varlen version if possible.

There are some arithmetic errors with the current implementation. The reason for them is probably that flash attention will return bf16 value for each block, so we cannot accumluate the values with the original fp32 ones.

And also because we need to save extra fp32 buffer during computation, the memory usage would be higher than theoretic limit.

• Implement ring_flash_attn_varlen_qkvpacked_func
• Implement zigzag_ring_flash_attn_qkvpacked_func issue#2
• Implement stripe_flash_attn_qkvpacked_func
• Implement zigzag_ring_flash_attn_varlen_qkvpacked_func
• Implement *_kvpacked_func and *_func variant for all APIs
• Optimize *_varlen_func
• Try to upstream to flash attention.

torchrun --nproc_per_node 8 test/test_ring_flash_attn_func.py
torchrun --nproc_per_node 8 test/test_ring_flash_attn_varlen_func.py
torchrun --nproc_per_node 8 test/test_zigzag_ring_flash_attn_func.py
torchrun --nproc_per_node 8 test/test_zigzag_ring_flash_attn_varlen_func.py
torchrun --nproc_per_node 8 test/test_stripe_flash_attn_func.py

torchrun --nproc_per_node 8 benchmark/benchmark_qkvpacked_func.py
torchrun --nproc_per_node 8 benchmark/benchmark_varlen_qkvpacked_func.py