Some configs missing dataset.vocab_size in MockIterableDataset

[torsli]

Describe the bug

Some benchmarking configs, like https://github.com/NVIDIA-NeMo/Automodel/blob/main/examples/benchmark/configs/qwen3_moe_30b_te_deepep.yaml, use the MockIterableDataset but do not specify a vocab_size. This leads to TypeError: MockIterableDataset.__init__() missing 1 required positional argument: 'vocab_size'
.
Steps/Code to reproduce bug

Run a recipe with qwen3_moe_30b_te_deepep.yaml

Additional context

There are multiple configs missing dataset.vocab_size, haven't found them all.

[qiaochuz-nv]

QA RCCA Analysis

Fix PR: #1404
Root Cause: Several benchmark YAML configs missing dataset.vocab_size field for MockIterableDataset, causing crash on dataset init.

Trigger Config: Benchmark + MockIterableDataset + affected configs

Coverage Assessment:

Dimension	Bug Config	Nearest Test	Match
Benchmark	Required	10 benchmark tests	Yes
MockIterableDataset	Required	Benchmark tests use this	Yes
vocab_size field	Was missing	Fixed in YAML configs	Yes

Status: COVERED — Multiple benchmark tests use MockIterableDataset. If vocab_size is removed, tests would crash.

Conclusion: No new test needed.