Convert padding itertools.islice to list

[malsbat]

torch.tensor fails with a ValueError when processing the islice:

(EngineCore_DP3 pid=1421) ValueError: expected sequence of length 128 at dim 1 (got 0)

Convert the islice to a list to be compatible with the PySequence_Fast functions used by pytorch (see recursive_store in pytorch/torch/csrc/utils/tensor_new.cpp).

[malsbat]

I found this while using the --data-parallel-size option to vllm:

python -m vllm.entrypoints.openai.api_server --port 8000 --model google/gemma-3-27b-it --served-model-name model-gemma-3-27b-it --data-parallel-size 4 --dtype bfloat16 --block-size 128 --max-model-len 16384 --max-num-seqs 128 --gpu-memory-util 0.9 --num-lookahead-slots 1 --disable-log-requests --uvicorn-log-level warning

After server is ready, issuing requests with vllm bench serve resulted in the ValueError exception mentioned above.

I am using vault.habana.ai/gaudi-docker/1.22.1/ubuntu24.04/habanalabs/pytorch-installer-2.7.1 with:

| HL-SMI Version:                              hl-1.22.1-fw-61.4.2.1          |
| Driver Version:                                     1.22.1-97ec1a4          |
| Nic Driver Version:                                 1.22.1-97ec1a4          |

[github-actions]

🚧 CI Blocked

The main CI workflow was not started for the following reason:

The DCO sign-off check has failed. Please ensure all commits are signed-off (e.g., 'git commit -s').

[github-actions]

🚧 CI Blocked

The main CI workflow was not started for the following reason:

The DCO sign-off check did not pass (status: action_required). Please ensure all commits are signed-off.