bug fix: convert-hf-to-gguf plamo2 add chat-template

[mmngays]

plamo2: missing SpecialVocab.add_to_gguf call caused chat_template to be dropped; add it.

[CISC]

Well, it wasn't entirely unintentional, the PLaMo2 models are not instruction tuned, and the included chat template does not add much (if any) value.

Basically you should treat it as a base model and use normal completion, not chat completion.

[grapevine-AI]

PLaMo2 does not have an instruction model, but instead has a translation model.

The chat template will be of great use in this translation derivation model.

[CISC]

The chat template will be of great use in this translation derivation model.

How so? All it does is add <|plamo:op|>, you have to do everything else manually, why not this as well?

To be clear, the reason I'm saying this is that adding a chat template will make people think it's actually usable in chat completion, it's not.

[CISC]

To further this sentiment, the examples in their README.md doesn't even use it.

[mmngays]

If there are (or will be) users who want to run plamo2 with instruction-style prompting or further fine‑tuning, I feel this capability really ought to be exposed as a feature.
That said, if we’re okay with the fact that plamo2’s current integration in llama.cpp cannot use a chat template due to its architecture, should we go ahead and close this PR?

[CISC]

For now I think so, if any instruction-tuned variants show up we can address the issue then.

[mmngays]

OK! Then I’ll go ahead and close this PR.
There are some flexible parts that are hard to express in templates like EN-JP or JP-EN.
I’ll leave a note in a gist or something.