[Bug] Different batch_size return different evaluating result

[SingL3]

Environment

Collecting system information...
---------------------------------
System Environment Report
Created: 2023-08-21 17:44:51 CST
---------------------------------

PyTorch information
-------------------
PyTorch version: 2.0.1+cu117
Is debug build: False
CUDA used to build PyTorch: 11.7
ROCM used to build PyTorch: N/A

OS: Ubuntu 20.04.5 LTS (x86_64)
GCC version: (Ubuntu 9.4.0-1ubuntu1~20.04.1) 9.4.0
Clang version: Could not collect
CMake version: version 3.26.3
Libc version: glibc-2.31

Python version: 3.10.11 (main, May 16 2023, 00:28:57) [GCC 11.2.0] (64-bit runtime)
Python platform: Linux-4.18.0-425.3.1.el8.x86_64-x86_64-with-glibc2.31
Is CUDA available: True
CUDA runtime version: 11.8.89
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

Nvidia driver version: 515.65.01
cuDNN version: Probably one of the following:
/usr/lib/x86_64-linux-gnu/libcudnn.so.8.7.0
/usr/lib/x86_64-linux-gnu/libcudnn_adv_infer.so.8.7.0
/usr/lib/x86_64-linux-gnu/libcudnn_adv_train.so.8.7.0
/usr/lib/x86_64-linux-gnu/libcudnn_cnn_infer.so.8.7.0
/usr/lib/x86_64-linux-gnu/libcudnn_cnn_train.so.8.7.0
/usr/lib/x86_64-linux-gnu/libcudnn_ops_infer.so.8.7.0
/usr/lib/x86_64-linux-gnu/libcudnn_ops_train.so.8.7.0
HIP runtime version: N/A
MIOpen runtime version: N/A
Is XNNPACK available: True

CPU:
Architecture:                    x86_64
L1i cache:                       2 MiB
L2 cache:                        80 MiB
L3 cache:                        96 MiB
NUMA node0 CPU(s):               0-31,64-95
NUMA node1 CPU(s):               32-63,96-127
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
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
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 intel_ppin 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 wbnoinvd dtherm ida arat pln pts avx512vbmi umip pku ospke avx512_vbmi2 gfni vaes vpclmulqdq avx512_vnni avx512_bitalg tme avx512_vpopcntdq la57 rdpid fsrm md_clear pconfig flush_l1d arch_capabilities

Versions of relevant libraries:
[pip3] numpy==1.24.4
[pip3] pytorch-ranger==0.1.1
[pip3] torch==2.0.1
[pip3] torch-optimizer==0.3.0
[pip3] torchdata==0.6.1
[pip3] torchmetrics==0.11.4
[pip3] torchtext==0.15.2
[pip3] torchvision==0.15.2
[conda] numpy                     1.24.4                   pypi_0    pypi
[conda] pytorch-ranger            0.1.1                    pypi_0    pypi
[conda] torch                     2.0.1                    pypi_0    pypi
[conda] torch-optimizer           0.3.0                    pypi_0    pypi
[conda] torchdata                 0.6.1                    pypi_0    pypi
[conda] torchmetrics              0.11.4                   pypi_0    pypi
[conda] torchtext                 0.15.2                   pypi_0    pypi
[conda] torchvision               0.15.2                   pypi_0    pypi


Composer information
--------------------
Composer version: 0.15.1
Composer commit hash: None
Host processor model name: Intel(R) Xeon(R) Platinum 8358P CPU @ 2.60GHz
Host processor core count: 64
Number of nodes: 1
Accelerator model name: NVIDIA A100-SXM4-80GB
Accelerators per node: 1
CUDA Device Count: 4

To reproduce

Steps to reproduce the behavior:

1. Eval a model
2. Modified a the batch_size of evaluating like this:

device_eval_batch_size: 4 # to 32

icl_tasks:
-
  ...
  batch_size: 4 # to 32

3. Get slightly different result: 0.751351 (bs=4) and 0.744144 (bs=32)

Expected behavior

Get same result under different batch_size

Additional context

[mvpatel2000]

Thanks for reporting this! Could you please try a few follow-on things?

1. Can you please try upgrading to latest Composer version (0.16.1)?
2. Can you please see if this same issue happens if you only run on 1 GPU?

In torch, distributed samplers duplicate data to ensure it is even across all ranks. We've added code to correct for this, but it is possible it is buggy. I'd like to ensure you're running on the latest version with these fixes and confirm it's the same issue and not something else