[BUG] Deepspeed Inference Error For ~3800 Context Length and Higher

Describe the bug
When the context length exceeds 3778, the code breaks. Using 64 generated tokens. Deepspeed version 0.8.0.

Error:

[launch.py:318:sigkill_handler] Killing subprocess

According to an early investigation, attn_softmax_v2 seems to be the kernel causing errors

To Reproduce
Steps to reproduce the behavior:

GPT2 model with random weights and long context

Expected behavior
The code should run without process being killed

ds_report output

DeepSpeed C++/CUDA extension op report

NOTE: Ops not installed will be just-in-time (JIT) compiled at
runtime if needed. Op compatibility means that your system
meet the required dependencies to JIT install the op.

JIT compiled ops requires ninja
ninja .................. [OKAY]

op name ................ installed .. compatible

[WARNING] async_io requires the dev libaio .so object and headers but these were not found.
[WARNING] async_io: please install the libaio-devel package with yum
[WARNING] If libaio is already installed (perhaps from source), try setting the CFLAGS and LDFLAGS environment variables to where it can be found.
async_io ............... [NO] ....... [NO]
cpu_adagrad ............ [NO] ....... [OKAY]
cpu_adam ............... [NO] ....... [OKAY]
fused_adam ............. [NO] ....... [OKAY]
fused_lamb ............. [NO] ....... [OKAY]
quantizer .............. [NO] ....... [OKAY]
random_ltd ............. [NO] ....... [OKAY]
sparse_attn ............ [NO] ....... [OKAY]
spatial_inference ...... [NO] ....... [OKAY]
transformer ............ [NO] ....... [OKAY]
stochastic_transformer . [NO] ....... [OKAY]
transformer_inference .. [NO] ....... [OKAY]
utils .................. [NO] ....... [OKAY]

DeepSpeed general environment info:
torch install path ............... ['/home/ec2-user/anaconda3/envs/vectorinf3/lib/python3.8/site-packages/torch']
torch version .................... 1.13.1
deepspeed install path ........... ['/home/ec2-user/research/DeepSpeedInt8Upgrade/deepspeed']
deepspeed info ................... 0.8.0+unknown, unknown, unknown
torch cuda version ............... 11.7
torch hip version ................ None
nvcc version ..................... 11.7
deepspeed wheel compiled w. ...... torch 1.13, cuda 11.7

Screenshots
If applicable, add screenshots to help explain your problem.

System info (please complete the following information):

• OS: [e.g. Ubuntu 18.04] Amazon Linux 2
• GPU count and types [e.g. two machines with x8 A100s each] 1 GPU
• (if applicable) what DeepSpeed-MII version are you using
• (if applicable) Hugging Face Transformers/Accelerate/etc. versions
• Python version
• Any other relevant info about your setup

Docker context
Are you using a specific docker image that you can share?

Additional context
Add any other context about the problem here.

Reproduce the failure with a GPT2 model using random weights and a context length above 3778, with 64 generated tokens. Start by investigating the attn_softmax_v2 kernel mentioned in the report and the process-killed error. Done means the same long-context inference completes without the process being killed.