[Proposal] Additional Architecture Adapter tests

[jlarson4]

Proposal

Add config tests for remaining architecture adapters

Motivation

Only about a third of our Architecture Adapters have focused tests in tests/unit/model_bridge/supported_architectures/. Use the existing available test suites as examples for how the new tests should be written.

Pitch

Create a small test suite for each of the remaining architecture adapters that do not have one at present. Feel free to tackle as few or as many at once as you'd like, just leave a comment stating which architecture(s) you're working on to reduce duplication of effort.

Architectures that need tests

• llama
• mistral (@chandrudp29 completed)
• mixtral (@RecreationalMath completed)
• gemma1 (@jlarson4 completed)
• gemma2 (@jlarson4 completed)
• gpt2 (@sunny1401 completed)
• qwen (@has9800 assigned)
• qwen2 (@Rishik00 completed)
• qwen3 (@sunny1401 completed)
• t5 (@RecreationalMath assigned)
• gpt_oss (@RecreationalMath completed)
• bloom (@indrayani21 assigned)
• opt (@willytop8 completed)
• phi
• phi3
• neo (@chandrudp29 completed)
• neox (@chandrudp29 completed)
• gptj (@along-l completed)
• bert
• apertus
• hubert
• olmo (@Rishik00 assigned)
• olmo2 (@RecreationalMath completed)
• olmo3 (@RecreationalMath assigned)
• olmoe (@RecreationalMath completed)
• openelm (@chandrudp29 completed)
• granite
• granite_moe
• granite_moe_hybrid
• qwen3_next
• stablelm (@RecreationalMath completed)
• llava_next (@sunny1401 completed)
• llava_onevision (@chandrudp29 completed)

Checklist

• I have checked that there is no similar issue in the repo (required)

[sunny1401]

I would like to take a stab at this if no one else is working. Can start with gpt2?

[jlarson4]

@sunny1401 Feel free to take on GPT2!

This issue is designed with multiple people being able to work on it at once, I just didn't want to muddy up the issue pool with 20+ identical issues.

[Rishik00]

can I take Qwen2?

[jlarson4]

certainly @Rishik00 !

[willytop8]

I'd like to take opt.