search

intel / intel-extension-for-pytorch

A Python package for extending the official PyTorch that can easily obtain performance on Intel platform
Apache License 2.0
1.59k stars 244 forks source link

FP16 output on PVC has randomness #526

Open rnwang04 opened 8 months ago

rnwang04 commented 8 months ago

Describe the bug

I found that on PVC GPU, output of fp16 has randomness. The same llm model, if I run two times, the output will be different, below is the reproduce script:

import torch
import intel_extension_for_pytorch
import os
import pytest
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer

device = 'xpu'
prompt = "Once upon a time, there existed a little girl who liked to have adventures. She wanted to go to places and meet new people, and have fun"

@pytest.mark.parametrize('Model, Tokenizer, model_path',[
    (AutoModelForCausalLM, AutoTokenizer, "meta/Llama-2-7b-chat-hf"),
    ])
def test_model(Model, Tokenizer, model_path):
    with torch.inference_mode():
        tokenizer = Tokenizer.from_pretrained(model_path, trust_remote_code=True)
        input_ids = tokenizer.encode(prompt+prompt+prompt, return_tensors="pt").to(device)
        print(input_ids.shape)

        model = Model.from_pretrained(model_path,
                                      trust_remote_code=True)
        model = model.half().to(device)
        logits_base_model = (model(input_ids)).logits
        torch.xpu.synchronize()
        model.to('cpu')  # deallocate gpu memory

        model = Model.from_pretrained(model_path,
                                      trust_remote_code=True)
        model = model.half().to(device)
        logits_optimized_model = (model(input_ids)).logits
        torch.xpu.synchronize()
        model.to('cpu')

        tol = 1e-03
        num_false = torch.isclose(logits_optimized_model, logits_base_model, rtol=tol, atol=tol)\
            .flatten().tolist().count(False)
        percent_false = num_false / logits_optimized_model.numel()
        print(percent_false)
        assert percent_false < 1e-02

After running pytest logits.py -v -s, the test will fail.

FAILED logits.py::test_model[AutoModelForCausalLM-AutoTokenizer-/home/llm-models/Llama-2-7b-chat-hf] - assert 0.47967353723404255 < 0.01
====================================== 1 failed, 3 warnings in 126.80s (0:02:06) =======================================

Versions

Collecting environment information...
PyTorch version: 2.1.0a0+cxx11.abi
PyTorch CXX11 ABI: Yes
IPEX version: 2.1.10+xpu
IPEX commit: a12f9f650
Build type: Release

OS: Ubuntu 22.04.3 LTS (x86_64)
GCC version: (Ubuntu 11.4.0-1ubuntu1~22.04) 11.4.0
Clang version: N/A
IGC version: 2024.0.2 (2024.0.2.20231213)
CMake version: version 3.24.0
Libc version: glibc-2.35

Python version: 3.9.18 (main, Sep 11 2023, 13:41:44)  [GCC 11.2.0] (64-bit runtime)
Python platform: Linux-5.19.0-41-generic-x86_64-with-glibc2.35
Is XPU available: True
DPCPP runtime version: 2024.0
MKL version: 2024.0
GPU models and configuration: 
[0] _DeviceProperties(name='Intel(R) Data Center GPU Max 1100', platform_name='Intel(R) Level-Zero', dev_type='gpu, support_fp64=1, total_memory=49152MB, max_compute_units=448, gpu_eu_count=448)
Intel OpenCL ICD version: 23.30.26918.50-736~22.04
Level Zero version: 1.3.26918.50-736~22.04

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):                          224
On-line CPU(s) list:             0-223
Vendor ID:                       GenuineIntel
Model name:                      Intel(R) Xeon(R) Platinum 8480+
CPU family:                      6
Model:                           143
Thread(s) per core:              2
Core(s) per socket:              56
Socket(s):                       2
Stepping:                        6
Frequency boost:                 enabled
CPU max MHz:                     2001.0000
CPU min MHz:                     800.0000
BogoMIPS:                        4000.00
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 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 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 ibt amx_bf16 avx512_fp16 amx_tile amx_int8 flush_l1d arch_capabilities
Virtualization:                  VT-x
L1d cache:                       5.3 MiB (112 instances)
L1i cache:                       3.5 MiB (112 instances)
L2 cache:                        224 MiB (112 instances)
L3 cache:                        210 MiB (2 instances)
NUMA node(s):                    2
NUMA node0 CPU(s):               0-55,112-167
NUMA node1 CPU(s):               56-111,168-223
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 store bypass: Mitigation; Speculative Store Bypass disabled via prctl
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] intel-extension-for-pytorch==2.1.10+xpu
[pip3] numpy==1.26.0
[pip3] torch==2.1.0a0+cxx11.abi
[pip3] torchvision==0.16.0a0+cxx11.abi
[conda] intel-extension-for-pytorch 2.1.10+xpu               pypi_0    pypi
[conda] numpy                     1.26.0                   pypi_0    pypi
[conda] torch                     2.1.0a0+cxx11.abi          pypi_0    pypi
[conda] torchvision               0.16.0a0+cxx11.abi          pypi_0    pypi
jgong5 commented 8 months ago

@arthuryuan1987 Seems related to non-determinism of oneDNN kernels?