[Proposal] Add HunYuan dense adapter (HunYuanDenseV1ForCausalLM)

[jlarson4]

Proposal

Add a TransformerBridge adapter for HunYuanDenseV1ForCausalLM (Tencent HunYuan dense family).

Motivation

Tencent ships small dense HunYuan models at 0.5B, 1.8B, 4B, and 7B, all runnable on a single consumer GPU. They fit interpretability compute budgets well, which makes the family high-leverage despite modest downloads.

Gap scan (2026-06-18): 13 models, ~114K downloads.

Pitch

The architecture is close to a standard dense decoder with GQA. Check the normalization and the dual fast/slow ("thinking") mode against the HF config while implementing.

• Claude Code users can scaffold with /add-model-support tencent/Hunyuan-1.8B-Instruct.
• Register at the four sites listed in contributing.md.
• Verify smallest-first: tencent/Hunyuan-0.5B, then tencent/Hunyuan-1.8B-Instruct.

Additional context

• Collection: https://huggingface.co/collections/tencent/hunyuan-dense-model
• Checkpoint: https://huggingface.co/tencent/Hunyuan-1.8B-Instruct
• Found via the hf_scraper architecture-gaps pass (2026-06-18).

Checklist

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

[TensorCruncher]

I'd like to work on this issue please

[jlarson4]

Assigned! This one is all yours

[TensorCruncher]

Hey @jlarson4, could you clarify the "dual fast/slow ("thinking") mode"? As far as I can tell, it's a tokeniser thing and not an adapter thing.

[jlarson4]

@TensorCruncher you are correct that it is a tokenizer thing, I just wanted to make sure there was nothing in the architecture config itself that referenced it.

[TensorCruncher]

I tried to verify the adapter after adding registration in the following places as per the adapter creation guide:

1. transformer_lens/model_bridge/supported_architectures/__init__.py
2. transformer_lens/factories/architecture_adapter_factory.py
3. transformer_lens/tools/model_registry/__init__.py

I ran the following in my terminal:

adil@Syeds-MacBook-Pro TransformerLens % uv run python -m transformer_lens.tools.model_registry.verify_models \
  --model tencent/Hunyuan-7B-Instruct \
  --max-memory 16 \  
  --device cpu \
  --dtype float32 \
  --no-ht-reference
Model 'tencent/Hunyuan-7B-Instruct' not found in supported_models.json, skipping
No valid models found in registry

Could point out where I went wrong? @jlarson4