Re: Document for my GSoC Tasks and want some advice.

[Mo Zhou <cdluminate@riseup.net>]

Hi Kohei,

Thanks for the update and the investigation!

I sent a ping to the python team list to ask somebody to add you as team 
member.
You can use repositories under your personal namespace as a temporary 
workspace.

In terms of the vllm version, my suggestion is to take the path with 
pyo3 0.22.
Look at here:
https://salsa.debian.org/deeplearning-team/safetensors/-/blob/master/debian/patches/downgrade-to-0.4.5.patch?ref_type=heads

I downgraded the safetensors also becasue pyo3 0.23 is not available.
We are currently in the deep freeze, so I do not really expect pyo3 
transition
to happen very soon.

While that does not lead to the latest version, this way can simplify 
our work
so that we can focus on the core parts. Debian's dependency tree is 
sometimes
complicated due to reasons like this.

And, we can always revisit the upgrade path to new versions afterwards.
Let's proceed with the simplest path without being distracted by pyo3.

On 5/15/25 12:40, 千代航平 wrote:
> Thanks to Mo's advice, I created my repo for tracking my tasks.
> https://salsa.debian.org/k1000dai/gsoc-status
> I also added WORKLOG.md, where I will write some notes and what I did.
>
> Also, I will send my progress bi-weekly to you and discuss next steps.
>
> Now, I want to decide the vllm version to package during this GSoC 
> project.
> I think there are two options.
> First is to follow the latest and catch up on every dependency for that.
>
> Second, and I think this might be much easier, is to follow the v0.8.4.
> v0.8.5 (latest version) requires tokenizers >= 0.21.1 and tokenizers 
> v0.21.1 requires pyo3 24.0 and numpy-rust 0.24.0; however, the rust 
> team says that backporting to pyo3 0.22 (currently in Debian unstable) 
> or upgrading pyo3 to 0.24 after the current Debian release is done in 
> a few months is required.
> In contrast, vllm v0.8.4 requires tokenizers >= 0.19.1 and tokenizers 
> v0.20.3 requires pyo3 22.0 and numpy-rst 0.22.0 which seems to be good 
> for avoiding repackaging pyo3 crate.
>
> I also think it’s important to keep up with the latest developments in 
> LLMs since the field is evolving so quickly. How do you think about this?
>
> Best regards,
> *
> *
> *Kohei Sendai*
> ------------------------------------------------------------------------------------------------------
> kouhei.sendai@gmail.com
>
> University of Tokyo, Faculty of Engineering, system innovation.
>
> ------------------------------------------------------------------------------------------