When prompted with the basic arithmetic expression 1+2+3, the LLM responds with the incorrect result 3 instead of the correct sum 6. This suggests an issue in the model’s mathematical reasoning or token interpretation logic.

image: registry.redhat.io/rhelai1/bootc-aws-nvidia-rhel9:1.4.4-1745341457
AWS Instance: g6.48xlarge
ilab, version 0.23.5
granite-3.1-8b-lab-v1

Steps to reproduce:
1. Create a aws instance with ami id ami-05f6fda386d6e6c73.
2.  use podman login registry.stage.redhat.io
3. Execute "sudo bootc switch registry.redhat.io/rhelai1/bootc-aws-nvidia-rhel9:1.4.4-1745341457" or just sudo booct upgarde --apply
4. Execute "ilab config init"
5. Download models
ilab model download --repository docker://registry.redhat.io/rhelai1/skills-adapter-v3 --release 1.4 && ilab model download --repository docker://registry.redhat.io/rhelai1/knowledge-adapter-v3 --release 1.4 && ilab model download --repository docker://registry.redhat.io/rhelai1/granite-3.1-8b-lab-v1 --release 1.4 && ilab model download --repository docker://registry.redhat.io/rhelai1/granite-3.1-8b-starter-v1 --release 1.4 && ilab model download --repository docker://registry.redhat.io/rhelai1/mixtral-8x7b-instruct-v0-1 --release 1.4 && ilab model download --repository docker://registry.redhat.io/rhelai1/prometheus-8x7b-v2-0 --release 1.4

6. Execute 
ilab model serve # Ctrl + C after gunicorn startsilab model chat
7. Use the prompt "1+2+3".
The response received is 3 as below