no response when send too big request

[Sherlock-Holo]

Describe the bug

RUST_LOG=trace mistralrs-server --port 20005 --isq Q4K --truncate-sequence plain -m /data/ai/huggingface/DeepSeek-R1-Distill-Qwen-14B

use this to start a mistrals-server, then send a big chat completion request

http POST http://gpu:20005/v1/chat/completions < /tmp/big-data.json

big-data.json

Expect

mistrals-server truncate the request, then reply the response

Happened

print this log, eat 100% CPU and 10.716G VRAM, but doesn't reply any response

mistral.log

Latest commit or version

b73e2e9

build with

export CUDA_NVCC_FLAGS=-fPIE
cargo b -r --features='cuda cudnn'

GPU is Tesla T4