search

IsaacRe / vllm-kvcompress

KV cache compression for high-throughput LLM inference
Apache License 2.0
93 stars 5 forks source link

[Benchmark]: The throughput benchmark result does not match the one of paper #4

Open DriverSong opened 1 month ago

DriverSong commented 1 month ago

Your current environment

Collecting environment information...
PyTorch version: 2.4.0+cu121
Is debug build: False
CUDA used to build PyTorch: 12.1
ROCM used to build PyTorch: N/A

OS: Ubuntu 22.04.4 LTS (x86_64)
GCC version: (Ubuntu 11.4.0-1ubuntu1~22.04) 11.4.0
Clang version: Could not collect
CMake version: version 3.29.2
Libc version: glibc-2.35

Python version: 3.10.12 (main, Nov 20 2023, 15:14:05) [GCC 11.4.0] (64-bit runtime)
Python platform: Linux-5.15.0-91-generic-x86_64-with-glibc2.35
Is CUDA available: True
CUDA runtime version: 12.4.131
CUDA_MODULE_LOADING set to: LAZY
GPU models and configuration: GPU 0: NVIDIA H100 80GB HBM3
Nvidia driver version: 550.90.07
cuDNN version: Probably one of the following:
/usr/lib/x86_64-linux-gnu/libcudnn.so.9.1.0
/usr/lib/x86_64-linux-gnu/libcudnn_adv.so.9.1.0
/usr/lib/x86_64-linux-gnu/libcudnn_cnn.so.9.1.0
/usr/lib/x86_64-linux-gnu/libcudnn_engines_precompiled.so.9.1.0
/usr/lib/x86_64-linux-gnu/libcudnn_engines_runtime_compiled.so.9.1.0
/usr/lib/x86_64-linux-gnu/libcudnn_graph.so.9.1.0
/usr/lib/x86_64-linux-gnu/libcudnn_heuristic.so.9.1.0
/usr/lib/x86_64-linux-gnu/libcudnn_ops.so.9.1.0
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:                      52 bits physical, 57 bits virtual
Byte Order:                         Little Endian
CPU(s):                             192
On-line CPU(s) list:                0-191
Vendor ID:                          GenuineIntel
Model name:                         Intel(R) Xeon(R) Platinum 8468V
CPU family:                         6
Model:                              143
Thread(s) per core:                 2
Core(s) per socket:                 48
Socket(s):                          2
Stepping:                           8
CPU max MHz:                        3800.0000
CPU min MHz:                        800.0000
BogoMIPS:                           4800.00
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 tsc_known_freq pni pclmulqdq dtes64 monitor 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 cat_l2 cdp_l3 invpcid_single intel_ppin cdp_l2 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 split_lock_detect avx_vnni avx512_bf16 wbnoinvd dtherm ida arat pln pts hwp hwp_act_window hwp_epp hwp_pkg_req avx512vbmi umip pku ospke waitpkg avx512_vbmi2 gfni vaes vpclmulqdq avx512_vnni avx512_bitalg tme avx512_vpopcntdq la57 rdpid bus_lock_detect cldemote movdiri movdir64b enqcmd fsrm md_clear serialize tsxldtrk pconfig arch_lbr amx_bf16 avx512_fp16 amx_tile amx_int8 flush_l1d arch_capabilities
Virtualization:                     VT-x
L1d cache:                          4.5 MiB (96 instances)
L1i cache:                          3 MiB (96 instances)
L2 cache:                           192 MiB (96 instances)
L3 cache:                           195 MiB (2 instances)
NUMA node(s):                       2
NUMA node0 CPU(s):                  0-47,96-143
NUMA node1 CPU(s):                  48-95,144-191
Vulnerability Gather data sampling: Not affected
Vulnerability Itlb multihit:        Not affected
Vulnerability L1tf:                 Not affected
Vulnerability Mds:                  Not affected
Vulnerability Meltdown:             Not affected
Vulnerability Mmio stale data:      Not affected
Vulnerability Retbleed:             Not affected
Vulnerability Spec rstack overflow: 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] numpy==1.26.4
[pip3] nvidia-cublas-cu12==12.1.3.1
[pip3] nvidia-cuda-cupti-cu12==12.1.105
[pip3] nvidia-cuda-nvrtc-cu12==12.1.105
[pip3] nvidia-cuda-runtime-cu12==12.1.105
[pip3] nvidia-cudnn-cu12==9.1.0.70
[pip3] nvidia-cudnn-frontend==1.3.0
[pip3] nvidia-cufft-cu12==11.0.2.54
[pip3] nvidia-curand-cu12==10.3.2.106
[pip3] nvidia-cusolver-cu12==11.4.5.107
[pip3] nvidia-cusparse-cu12==12.1.0.106
[pip3] nvidia-dali-cuda120==1.37.1
[pip3] nvidia-ml-py==12.560.30
[pip3] nvidia-nccl-cu12==2.20.5
[pip3] nvidia-nvimgcodec-cu12==0.2.0.7
[pip3] nvidia-nvjitlink-cu12==12.6.77
[pip3] nvidia-nvtx-cu12==12.1.105
[pip3] nvidia-pyindex==1.0.9
[pip3] onnx==1.16.0
[pip3] optree==0.11.0
[pip3] pytorch-quantization==2.1.2
[pip3] pytorch-triton==3.0.0+989adb9a2
[pip3] pyzmq==26.0.3
[pip3] torch==2.4.0
[pip3] torch-tensorrt==2.4.0a0
[pip3] torchvision==0.19.0
[pip3] transformers==4.45.2
[pip3] triton==3.0.0
[conda] Could not collect
ROCM Version: Could not collect
Neuron SDK Version: N/A
vLLM Version: 0.6.0
vLLM Build Flags:
CUDA Archs: 9.0; ROCm: Disabled; Neuron: Disabled
GPU Topology:
GPU0    NIC0    NIC1    NIC2    NIC3    NIC4    NIC5    NIC6    NIC7    NIC8    NIC9    CPU Affinity    NUMA Affinity   GPU NUMA ID
GPU0     X      SYS     SYS     SYS     SYS     SYS     SYS     NODE    NODE    PXB     PIX     48-95,144-191   1               N/A
NIC0    SYS      X      PXB     NODE    NODE    NODE    NODE    SYS     SYS     SYS     SYS
NIC1    SYS     PXB      X      NODE    NODE    NODE    NODE    SYS     SYS     SYS     SYS
NIC2    SYS     NODE    NODE     X      PXB     NODE    NODE    SYS     SYS     SYS     SYS
NIC3    SYS     NODE    NODE    PXB      X      NODE    NODE    SYS     SYS     SYS     SYS
NIC4    SYS     NODE    NODE    NODE    NODE     X      PIX     SYS     SYS     SYS     SYS
NIC5    SYS     NODE    NODE    NODE    NODE    PIX      X      SYS     SYS     SYS     SYS
NIC6    NODE    SYS     SYS     SYS     SYS     SYS     SYS      X      PXB     NODE    NODE
NIC7    NODE    SYS     SYS     SYS     SYS     SYS     SYS     PXB      X      NODE    NODE
NIC8    PXB     SYS     SYS     SYS     SYS     SYS     SYS     NODE    NODE     X      PXB
NIC9    PIX     SYS     SYS     SYS     SYS     SYS     SYS     NODE    NODE    PXB      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_0
  NIC1: mlx5_1
  NIC2: mlx5_2
  NIC3: mlx5_3
  NIC4: mlx5_4
  NIC5: mlx5_5
  NIC6: mlx5_6
  NIC7: mlx5_7
  NIC8: mlx5_8
  NIC9: mlx5_9

How would you like to use vllm

I‘ve run the benchmark script benchmark_llama3_8b.sh on 1xH100 with model Meta-Llama-3.1-8B-Instruct. With the params unchanged but the --num-prompts. I set --num-prompts 1000 to run the script and the result is as follow: image The throughput does not show a significant or regular pattern with changes in the --compression-rate. Is there a limit to the improvement in throughput from KV compression?

Before submitting a new issue...

IsaacRe commented 4 weeks ago

Hi there 👋

We don't actually report throughput benchmarks for the 8B model on an H100--all 8B throughput benchmarks were run on an NVIDIA L4. We did run throughput benchmarks for the 70B model on an H100, so you should be able to reproduce those on your hardware.

But to answer your question, yes there absolutely is. The improvement in thoughput is very small when the amount of allocable cache space is large (eg. when running small models on GPUs with lots of vRAM) since large decoding batches can then fit in cache even without any compression. Because of this, our evaluations focus on model/hardware configurations where most of the available memory is allocated to the model parameters.

An additional factor is the input token length. The shorter the inputs are, the easier it is to fit large batches into limited cache space without compression. As input length increases, eventually the limited cache space will impose a limit on the maximum batch size that can be used during decoding, hurting throughput. At this point, compression can be used to free up cache space and improve throughput.

Thanks for taking the time to reproduce the results. If you have any with input_len > 500 I'd be interested to see them!