use pydantic.v1 in langchain.py

[andri-jpg]

fix issue #29

[LLukas22]

Thanks 👍. If the CI passes I'll merge it and create a bugfix release.

[andri-jpg]

However, when I attempted to install using pip install . from the source without making any code changes, it also failed and returned errors from Cargo.

[LLukas22]

Yeah the tokenizers create had some breaking changes, meaning this PR is needed.

I'll sync the latest main branch of rustformers when i get back home. That means i'll also have to implement custom RoPE scaling and probably Falcon. 🤔

[LLukas22]

Alright this seams to work, keep in mind that langchain doesn't officially support pydantic 2 meaning there still could be some issues. I'm merging it for now.

[andri-jpg]

Alright this seams to work, keep in mind that langchain doesn't officially support pydantic 2 meaning there still could be some issues. I'm merging it for now.

Alright, I'll try reinstalling from the source again. I'll create another GitHub issue if any other problems come up.

[LLukas22]

Yeah let me know if everything works, then i'll create a new release 👍

[andri-jpg]

Nice, everything works well. Both llm_rs.langchain and Langchain are running smoothly without any issues. By the way, I've noticed a slight performance improvement here – the generation time has become slightly faster (from 1.4s to 0.8s), model = gpt2.

[LLukas22]

Yeah there is less copying going on, as i changed all repeat operations into broadcastable mul/add operations, which means less time should be spend copying data around.