Add OLMo Hybrid model (#43358)
* Add OLMo 3.5 Hybrid model
* Address review feedback
* make RoPE optional
Signed-off-by: Yanhong Li <“yanhongl@allenai.org”>
* Remove GPT2Tokenizer and use default TokenizersBackend
* Address code review feedback
* Rename OLMo 3.5 Hybrid → OLMo 3.2 Hybrid
* Address review feedback
* address review feedback
* small fixes
* fixup in modular
* address feedback + rename to olmo hybrid
* some small fixups with modular and simplify
* fix date
* oops
* remove from list on docstring excceptions
* move repo to internal testing and fix mdoel type
* style
* update official model path
---------
Signed-off-by: Yanhong Li <“yanhongl@allenai.org”>
Co-authored-by: Yanhong Li <“yanhongl@allenai.org”>
Co-authored-by: Anton Vlasjuk <73884904+vasqu@users.noreply.github.com>
Co-authored-by: vasqu <antonprogamer@gmail.com>
