Skip to content

Performance of the C++ interface of flash attention and flash attention v2 in large language model (LLM) inference scenarios.

License

Notifications You must be signed in to change notification settings

Bruce-Lee-LY/flash_attention_inference

Repository files navigation

Flash Attention Inference

Performance of the C++ interface of Flash Attention and Flash Attention v2 in large language model (LLM) inference scenarios. The calculation expression is as follows, where the precision of tensor Q, K, V and O is FP16. Remove redundant code from Flash Attention that has nothing to do with inference, such as backward, dropout, bf16 and torch dependencies, so you can easily integrate this Flash Attention into LLM inference programs. In addition, Flash Attention and Flash Attention v2 have been modified to support group query attention (GQA) / multi query attention (MQA), hybrid by prefill and decoding and attention with linear biases (ALiBi) inference scenarios.

O = Softmax(Q * K^T) * V

fmha

Support

  • GQA/MQA Inference: Group query attention / multi query attention inference
  • Hybrid Inference: Hybrid inference by prefill and decoding
  • ALiBi Inference: Attention with linear biases inference

Compile

Environment

  • OS: Linux
  • Cmake Version: >= 3.16
  • GCC Version: >= 5.0
  • CUDA Version: >= 11.4
  • Others: gflags, ccache
sudo apt-get install libgflags-dev ccache

Clone

git clone https://github.com/Bruce-Lee-LY/flash_attention_inference.git

Build

NVIDIA A100

cd flash_attention_inference
./build.sh -a 80 -t Release -b OFF
./build.sh -a 80 -t Debug -b OFF

RTX3080Ti / RTX3090 / RTX A6000

cd flash_attention_inference
./build.sh -a 86 -t Release -b OFF
./build.sh -a 86 -t Debug -b OFF

Run Sample

./run_sample.sh

Performance

Process the data in the log and plot it as a line chart.

cd tools/performance
./performance.sh

RTX3090

  • CUDA Version: 11.8
  • Head Num: 32
  • Head Dim: 128

Prefill

Seq Len

The performance of both is similar for short sequences and Flash Attention v2 performs well in long sequences. It can increase by about 60%.

  • Batch Size: 1
  • Seq Q: Seq Len
  • Seq K: Seq Len

prefill_seq_throughput

Batch Size

When the batch size is small, the Flash Attention v2 performance is better. When the batch size is large, the performance of the two kernels is comparable.

  • Batch Size: Batch Size
  • Seq Q: 128
  • Seq K: 128

prefill_batch_throughput

Decoding

Seq Len

The performance of both is similar for short sequences and Flash Attention performs well in long sequences. It can increase by about 100%.

  • Batch Size: 1
  • Seq Q: 1
  • Seq K: Seq Len

decoding_seq_throughput

Batch Size

The Flash Attention performance is better than Flash Attention v2 regardless of batch size.

  • Batch Size: Batch Size
  • Seq Q: 1
  • Seq K: 128

decoding_batch_throughput

Hybrid

Regardless of the ratio of Prefill to Decoding, Flash Attention and Flash Attention v2 are similar in performance.

  • Batch Size: 100
  • Seq Q: 128
  • Seq K: 128

hybrid_throughput

Reference

  • Flash Attention: v1.0.9
  • Flash Attention v2: v2.1.0
  • cutlass: v3.1.0