Spaces:
Running
Inquiry Regarding "Supra-124M" Development Status and Benchmarks
To: SupraLabs Team
Dear SupraLabs Team,
I hope this message finds you well.
I have been closely following your work on Hugging Face and am incredibly excited about your upcoming model, Supra-124M. Given your track record, the community is highly anticipating this release. I am reaching out today in the hopes of getting a brief sneak peek or status update on the project.
Specifically, I would love to know:
- Release Timeline: Do you have an estimated window for when the weights will be officially hosted on Hugging Face?
- Performance & Benchmarks: What specific benchmarks or capabilities are you prioritizing during training? Given its footprint, do you anticipate it outperforming existing models in the ~125M class, such as AxiomicLabs/GPT-X2-125M?
- Model Insights: Are there any unique architecture adjustments or token-efficiency strategies you are utilizing that you'd be willing to share?
Thank you so much for your time, dedication, and incredible contributions to the open-source AI community. We are all eagerly looking forward to what Supra-124M will achieve.
Warm regards,
Akshit
Hi @GODELEV !
First: I accepted your friend request on discord. Let's continue talking there later on.
Second: We are happy that you're interested in our models ❤️ 🤗
We don't know the exact release date for the 124M model and we do not have a specific plan yet.
For the benchmarks, we will take lm-eval using blimp, arc_easy, arc_challenge, boolq, openbookqa, piqa, lambada, hellaswag, and some more.
We think, the 124M model will definetely outperform the e.g. AxiomicLabs models in the 125/123M class and other models like SmolLM-2-135M or GPT-2-Small.
But first, we will release a second version of our 50M model using better data and improved scripts. 😃
If you want, you can join our HF org, SupraLabs, ...
Best regards,
LH-Tech-AI
From: @GODELEV
To: LH-Tech-AI (SupraLabs)
Hi LH-Tech-AI,
Thank you so much for the detailed and exciting update! It is fantastic to hear about your plans for the benchmarks and the upcoming second version of the 50M model. Hearing that you expect Supra-124M to outperform models like GPT-2-Small and SmolLM-2 is incredibly promising, and I really look forward to seeing it in action.
I also truly appreciate the kind invitation to join the SupraLabs Hugging Face organization. It means a lot! However, due to my current schedule, I am quite inconsistent with my availability right now and wouldn't be able to contribute the way I’d want to. I would absolutely love to revisit this and contribute in the future once things stabilize on my end.
Best regards,
Akshit
Well I think , I don't need to be this much formal ?
yes right 😅