当前环境

Collecting environment information...
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 3.27.6
Libc version: glibc-2.35

Python version: 3.10.12 (main, Jun 11 2023, 05:26:28) [GCC 11.4.0] (64-bit runtime)
Python platform: Linux-5.15.0-125.006-shopee-x86_64-with-glibc2.35
Is CUDA available: True
CUDA runtime version: 12.2.140
CUDA_MODULE_LOADING set to: LAZY
GPU models and configuration: 
GPU 0: NVIDIA H100 80GB HBM3
GPU 1: NVIDIA H100 80GB HBM3
GPU 2: NVIDIA H100 80GB HBM3
GPU 3: NVIDIA H100 80GB HBM3
GPU 4: NVIDIA H100 80GB HBM3
GPU 5: NVIDIA H100 80GB HBM3
GPU 6: NVIDIA H100 80GB HBM3
GPU 7: 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.8.9.5
/usr/lib/x86_64-linux-gnu/libcudnn_adv_infer.so.8.9.5
/usr/lib/x86_64-linux-gnu/libcudnn_adv_train.so.8.9.5
/usr/lib/x86_64-linux-gnu/libcudnn_cnn_infer.so.8.9.5
/usr/lib/x86_64-linux-gnu/libcudnn_cnn_train.so.8.9.5
/usr/lib/x86_64-linux-gnu/libcudnn_ops_infer.so.8.9.5
/usr/lib/x86_64-linux-gnu/libcudnn_ops_train.so.8.9.5
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 8468
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:                           4200.00
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 avx512_fp16 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:                           210 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==1.22.2
[pip3] nvidia-nccl-cu12==2.20.5
[pip3] onnx==1.14.0
[pip3] pytorch-quantization==2.1.2
[pip3] torch==2.3.1
[pip3] torch-tensorrt==0.0.0
[pip3] torchdata==0.7.0a0
[pip3] torchtext==0.16.0a0
[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: 0.5.3
vLLM Build Flags:
CUDA Archs: 7.0 7.5 8.0 8.6 8.9 9.0+PTX; ROCm: Disabled; Neuron: Disabled
GPU Topology:
�[4mGPU0	GPU1	GPU2	GPU3	GPU4	GPU5	GPU6	GPU7	NIC0	NIC1	NIC2	NIC3	NIC4	CPU Affinity	NUMA Affinity	GPU NUMA ID�[0m
GPU0	 X 	NV18	NV18	NV18	NV18	NV18	NV18	NV18	PIX	NODE	SYS	SYS	NODE	0-47,96-143	0		N/A
GPU1	NV18	 X 	NV18	NV18	NV18	NV18	NV18	NV18	PXB	NODE	SYS	SYS	NODE	0-47,96-143	0		N/A
GPU2	NV18	NV18	 X 	NV18	NV18	NV18	NV18	NV18	NODE	PXB	SYS	SYS	NODE	0-47,96-143	0		N/A
GPU3	NV18	NV18	NV18	 X 	NV18	NV18	NV18	NV18	NODE	PIX	SYS	SYS	NODE	0-47,96-143	0		N/A
GPU4	NV18	NV18	NV18	NV18	 X 	NV18	NV18	NV18	SYS	SYS	PXB	NODE	SYS	48-95,144-191	1		N/A
GPU5	NV18	NV18	NV18	NV18	NV18	 X 	NV18	NV18	SYS	SYS	PIX	NODE	SYS	48-95,144-191	1		N/A
GPU6	NV18	NV18	NV18	NV18	NV18	NV18	 X 	NV18	SYS	SYS	NODE	PXB	SYS	48-95,144-191	1		N/A
GPU7	NV18	NV18	NV18	NV18	NV18	NV18	NV18	 X 	SYS	SYS	NODE	PIX	SYS	48-95,144-191	1		N/A
NIC0	PIX	PXB	NODE	NODE	SYS	SYS	SYS	SYS	 X 	NODE	SYS	SYS	NODE				
NIC1	NODE	NODE	PXB	PIX	SYS	SYS	SYS	SYS	NODE	 X 	SYS	SYS	NODE				
NIC2	SYS	SYS	SYS	SYS	PXB	PIX	NODE	NODE	SYS	SYS	 X 	NODE	SYS				
NIC3	SYS	SYS	SYS	SYS	NODE	NODE	PXB	PIX	SYS	SYS	NODE	 X 	SYS				
NIC4	NODE	NODE	NODE	NODE	SYS	SYS	SYS	SYS	NODE	NODE	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_0
  NIC1: mlx5_1
  NIC2: mlx5_4
  NIC3: mlx5_5
  NIC4: mlx5_bond_0

🐛 描述问题

我按照文档( https://github.com/vllm-project/vllm/blob/main/docs/source/quantization/fp8.rst 和 https://github.com/vllm-project/vllm/tree/main/examples/fp8 )使用以下命令在1*H100上将vicuna-13b-v1.5量化为fp8精度，并成功获取了fp8模型和kv_cache_scales.json文件：

• FP8模型

import fire
from datasets import load_dataset
from transformers import AutoTokenizer
from auto_fp8 import AutoFP8ForCausalLM, BaseQuantizeConfig

# Credit to: https://github.com/vllm-project/vllm/blob/main/docs/source/quantization/fp8.rst
class AutoFP8:
    def __init__(self,
                 model_path: str,
                 saved_path: str,
                 calib_size: int = 512,
                 activation_scheme: str = "static"):
        self.saved_path = saved_path
        self.calib_size = calib_size

        self.quantize_config = BaseQuantizeConfig(
            quant_method="fp8", activation_scheme=activation_scheme)

        self.tokenizer = AutoTokenizer.from_pretrained(
            model_path, use_fast=True)
        self.tokenizer.pad_token = self.tokenizer.eos_token

        self.model = AutoFP8ForCausalLM.from_pretrained(model_path,
                                                        self.quantize_config)

    def apply_fp8(self):
        # Load and tokenize 512 dataset samples for calibration of activation scales
        ds = load_dataset("mgoin/ultrachat_2k", split="train_sft").select(
            range(self.calib_size))
        examples = [self.tokenizer.apply_chat_template(
            batch["messages"], tokenize=False) for batch in ds]
        examples = self.tokenizer(examples, padding=True, truncation=True,
                                  return_tensors="pt").to("cuda")

        # quantize, and save checkpoint
        self.model.quantize(examples)
        self.model.save_quantized(self.saved_path)

def main(model_path: str,
         saved_path: str,
         calib_size: int = 512,
         ):
    fp8_helper = AutoFP8(model_path, saved_path, calib_size)
    fp8_helper.apply_fp8()

if __name__ == "__main__":
    fire.Fire(main)

• FP8 KV Cache scales

set -euxo pipefail

# https://github.com/vllm-project/vllm/tree/main/examples/fp8

if [ $# = 4 ]; then
  model_path=$1
  output_model_path=$2
  output_kv_cache_path=$3
  device_id=$4

  gpu_num=$(echo "$device_id" |grep -o "[0-9]" |grep -c "")
  export CUDA_VISIBLE_DEVICES=$device_id

  output_model_path+="-tp${gpu_num}"
  output_name="kv_cache_fp8_scales_tp${gpu_num}.json"

  if [ "$(pip list | grep nvidia-ammo | wc -l)" -eq "0" ]; then
    pip install --no-cache-dir --extra-index-url https://pypi.nvidia.com nvidia-ammo==0.7.1
  fi

  # 1. Convert HF model into a quantized HF model.
  if [ ! -d ${output_model_path} ]; then
    python fp8/quantizer/quantize.py \
      --model-dir ${model_path} \
      --dtype float16 \
      --qformat fp8 \
      --kv-cache-dtype fp8 \
      --calib-size 512 \
      --tp-size ${gpu_num} \
      --output-dir ${output_model_path}
  else
    echo "The quantized hf model already exits in ${output_model_path}"
  fi

  # 2. Extract KV Cache Scaling Factors from quantized HF model.
  python3 fp8/extract_scales.py \
    --quantized-model ${output_model_path} \
    --tp-size ${gpu_num} \
    --output-name ${output_name} \
    --output-dir $output_kv_cache_path

else
  echo "Usage: $0 hf_model_path quantized_hf_model_path kv_cache_path device_id(0,1)"
  exit
fi

然而，当我使用 --quantization fp8 、 --kv-cache-dtype fp8 和 --quantization-param-path ${output_kv_cache_scales_file} 启动openai服务器时，出现了以下问题：

但是，如果我们只传递 --quantization fp8 来启动服务器，一切都可以正常工作。
我认为 nvidia-ammo(现在称为 modelopt )工具和最新的 vllm 之间存在一些差距，需要及时更新 FP8 KV Cache 的文档。