[Performance]: Flashinfer backend's improvement is marginal compared to FlashAttention backend for long context Qwen2-72b-instruct-128k #6617
Description
Proposal to improve performance
My model is public version of Qwen2-72b-Instruct-128k, which has GQA and enables YARN to support longer contexts.
"rope_scaling": {
"factor": 4.0,
"original_max_position_embeddings": 32768,
"type": "yarn"
}
I noticed that flashinfer's blog said that Notably, FlashInfer achieves up to 2-3x speedup for Grouped-Query Attention on A100 & H100, compared to vLLM implementation in https://flashinfer.ai/2024/02/02/introduce-flashinfer.html,
so i decided to use -e VLLM_ATTENTION_BACKEND="FLASHINFER" to accelerate the long context inference of vLLM.
Report of performance regression
I used a modified version of benchmark_serving.py script to test my own ruler_128k dataset and send 8 requests of 128K tokens concurrently.
The result turns out as bellow.
FlashAttention
FlashInfer
From the benchmark result above, the improvement from FlashInfer is quite marginal.
Misc discussion on performance
I cannot find performance benchmarks of flashinfer compared to flashattention with cudagraph.
Maybe flashinfer has not been completely supported in vllm v0.5.2, or is there some any other reasons make flashinfer's performance not as good as it described in the blog above?
Your current environment (if you think it is necessary)
The output of python collect_env.py
PyTorch version: 2.3.1+cu121
Is debug build: False
CUDA used to build PyTorch: 12.1
ROCM used to build PyTorch: N/A
OS: Ubuntu 22.04.3 LTS (x86_64)
GCC version: (Ubuntu 11.4.0-1ubuntu1~22.04) 11.4.0
Clang version: Could not collect
CMake version: version 3.30.0
Libc version: glibc-2.35
Python version: 3.10.12 (main, Mar 22 2024, 16:50:05) [GCC 11.4.0] (64-bit runtime)
Python platform: Linux-5.4.0-171-generic-x86_64-with-glibc2.35
Is CUDA available: True
CUDA runtime version: Could not collect
CUDA_MODULE_LOADING set to: LAZY
GPU models and configuration:
GPU 0: NVIDIA A100-SXM4-80GB
GPU 1: NVIDIA A100-SXM4-80GB
GPU 2: NVIDIA A100-SXM4-80GB
GPU 3: NVIDIA A100-SXM4-80GB
GPU 4: NVIDIA A100-SXM4-80GB
GPU 5: NVIDIA A100-SXM4-80GB
GPU 6: NVIDIA A100-SXM4-80GB
GPU 7: NVIDIA A100-SXM4-80GB
Nvidia driver version: 545.23.08
cuDNN version: Could not collect
HIP runtime version: N/A
MIOpen runtime version: N/A
Is XNNPACK available: True
CPU:
Architecture: x86_64
CPU op-mode(s): 32-bit, 64-bit
Address sizes: 46 bits physical, 57 bits virtual
Byte Order: Little Endian
CPU(s): 128
On-line CPU(s) list: 0-127
Vendor ID: GenuineIntel
Model name: Intel(R) Xeon(R) Platinum 8358 CPU @ 2.60GHz
CPU family: 6
Model: 106
Thread(s) per core: 2
Core(s) per socket: 32
Socket(s): 2
Stepping: 6
Frequency boost: enabled
CPU max MHz: 3400.0000
CPU min MHz: 800.0000
BogoMIPS: 5200.00
Flags: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe syscall nx pdpe1gb rdtscp lm constant_tsc art arch_perfmon pebs bts rep_good nopl xtopology nonstop_tsc cpuid aperfmperf pni pclmulqdq dtes64 ds_cpl vmx smx est tm2 ssse3 sdbg fma cx16 xtpr pdcm pcid dca sse4_1 sse4_2 x2apic movbe popcnt tsc_deadline_timer aes xsave avx f16c rdrand lahf_lm abm 3dnowprefetch cpuid_fault epb cat_l3 invpcid_single ssbd mba ibrs ibpb stibp ibrs_enhanced tpr_shadow vnmi flexpriority ept vpid ept_ad fsgsbase tsc_adjust bmi1 avx2 smep bmi2 erms invpcid cqm rdt_a avx512f avx512dq rdseed adx smap avx512ifma clflushopt clwb intel_pt avx512cd sha_ni avx512bw avx512vl xsaveopt xsavec xgetbv1 xsaves cqm_llc cqm_occup_llc cqm_mbm_total cqm_mbm_local wbnoinvd dtherm ida arat pln pts avx512vbmi umip pku ospke avx512_vbmi2 gfni vaes vpclmulqdq avx512_vnni avx512_bitalg tme avx512_vpopcntdq rdpid md_clear pconfig flush_l1d arch_capabilities
Virtualization: VT-x
L1d cache: 3 MiB (64 instances)
L1i cache: 2 MiB (64 instances)
L2 cache: 80 MiB (64 instances)
L3 cache: 96 MiB (2 instances)
NUMA node(s): 2
NUMA node0 CPU(s): 0-31,64-95
NUMA node1 CPU(s): 32-63,96-127
Vulnerability Gather data sampling: Mitigation; Microcode
Vulnerability Itlb multihit: Not affected
Vulnerability L1tf: Not affected
Vulnerability Mds: Not affected
Vulnerability Meltdown: Not affected
Vulnerability Mmio stale data: Mitigation; Clear CPU buffers; SMT vulnerable
Vulnerability Retbleed: Not affected
Vulnerability Spec store bypass: Mitigation; Speculative Store Bypass disabled via prctl and seccomp
Vulnerability Spectre v1: Mitigation; usercopy/swapgs barriers and __user pointer sanitization
Vulnerability Spectre v2: Mitigation; Enhanced IBRS, IBPB conditional, RSB filling, PBRSB-eIBRS SW sequence
Vulnerability Srbds: Not affected
Vulnerability Tsx async abort: Not affected
Versions of relevant libraries:
[pip3] flashinfer==0.0.9+cu121torch2.3
[pip3] numpy==1.26.4
[pip3] nvidia-nccl-cu12==2.20.5
[pip3] torch==2.3.1
[pip3] torchvision==0.18.1
[pip3] transformers==4.42.4
[pip3] triton==2.3.1
[conda] Could not collect
ROCM Version: Could not collect
Neuron SDK Version: N/A
vLLM Version: 0.5.2
vLLM Build Flags:
CUDA Archs: Not Set; ROCm: Disabled; Neuron: Disabled
GPU Topology:
GPU0 GPU1 GPU2 GPU3 GPU4 GPU5 GPU6 GPU7 NIC0 NIC1 NIC2 NIC3 NIC4 NIC5 NIC6 NIC7 CPU Affinity NUMA Affinity GPU NUMA ID
GPU0 X NV12 NV12 NV12 NV12 NV12 NV12 NV12 SYS SYS PXB PXB SYS SYS SYS PXB 0-31,64-95 0 N/A
GPU1 NV12 X NV12 NV12 NV12 NV12 NV12 NV12 SYS SYS PXB PXB SYS SYS SYS PXB 0-31,64-95 0 N/A
GPU2 NV12 NV12 X NV12 NV12 NV12 NV12 NV12 SYS SYS SYS SYS SYS SYS PXB SYS 0-31,64-95 0 N/A
GPU3 NV12 NV12 NV12 X NV12 NV12 NV12 NV12 SYS SYS SYS SYS SYS SYS PXB SYS 0-31,64-95 0 N/A
GPU4 NV12 NV12 NV12 NV12 X NV12 NV12 NV12 SYS SYS SYS SYS SYS PXB SYS SYS 32-63,96-127 1 N/A
GPU5 NV12 NV12 NV12 NV12 NV12 X NV12 NV12 SYS SYS SYS SYS SYS PXB SYS SYS 32-63,96-127 1 N/A
GPU6 NV12 NV12 NV12 NV12 NV12 NV12 X NV12 PXB PXB SYS SYS PXB SYS SYS SYS 32-63,96-127 1 N/A
GPU7 NV12 NV12 NV12 NV12 NV12 NV12 NV12 X PXB PXB SYS SYS PXB SYS SYS SYS 32-63,96-127 1 N/A
NIC0 SYS SYS SYS SYS SYS SYS PXB PXB X PIX SYS SYS PXB SYS SYS SYS
NIC1 SYS SYS SYS SYS SYS SYS PXB PXB PIX X SYS SYS PXB SYS SYS SYS
NIC2 PXB PXB SYS SYS SYS SYS SYS SYS SYS SYS X PIX SYS SYS SYS PIX
NIC3 PXB PXB SYS SYS SYS SYS SYS SYS SYS SYS PIX X SYS SYS SYS PIX
NIC4 SYS SYS SYS SYS SYS SYS PXB PXB PXB PXB SYS SYS X SYS SYS SYS
NIC5 SYS SYS SYS SYS PXB PXB SYS SYS SYS SYS SYS SYS SYS X SYS SYS
NIC6 SYS SYS PXB PXB SYS SYS SYS SYS SYS SYS SYS SYS SYS SYS X SYS
NIC7 PXB PXB SYS SYS SYS SYS SYS SYS SYS SYS PIX PIX SYS SYS SYS X
Legend:
X = Self
SYS = Connection traversing PCIe as well as the SMP interconnect between NUMA nodes (e.g., QPI/UPI)
NODE = Connection traversing PCIe as well as the interconnect between PCIe Host Bridges within a NUMA node
PHB = Connection traversing PCIe as well as a PCIe Host Bridge (typically the CPU)
PXB = Connection traversing multiple PCIe bridges (without traversing the PCIe Host Bridge)
PIX = Connection traversing at most a single PCIe bridge
NV# = Connection traversing a bonded set of # NVLinks
NIC Legend:
NIC0: mlx5_cx5_0
NIC1: mlx5_cx5_1
NIC2: mlx5_cx5_2
NIC3: mlx5_cx5_3
NIC4: mlx5_cx6_0
NIC5: mlx5_cx6_1
NIC6: mlx5_cx6_2
NIC7: mlx5_cx6_3