Spaces:
Running
AxionLab
Hi, we are AxionLab-Co, we are interessed on the FMN-GPT archicteture. We are requesting the use of FMN-GPT for a thinking model called DogeAI-v2.5 to be distillated on TeichAI/Claude-Opus-4.5-High-Reasoning-250x dataset, thanks for the attention.
FMN-GPT is currently in experimental/development phase and not yet available for external integration with other projects or services.
Hi, we are AxionLab-Co, we are interessed on the FMN-GPT archicteture. We are requesting the use of FMN-GPT for a thinking model called DogeAI-v2.5 to be distillated on TeichAI/Claude-Opus-4.5-High-Reasoning-250x dataset, thanks for the attention.
Looking back at your profile, I would be happy to collab on discord! You seem pretty low on TFlops.
Thanks for reaching out! Yes, my current compute is limited, so I focus heavily on efficiency, distillation and small-scale architecture experimentation rather than brute-force scaling.
I’m very interested in understanding FMN-GPT’s architectural ideas and exploring lightweight or distilled variants.
If you're open to it, I’d love to discuss possible collaboration paths, benchmarking, architectural tweaks for low-compute environments, or even experimental distilled versions.
Currently I'm in the middle of re-designing some of the architecture (look at latest blog for more details) and cleaning up the code, so It might be a while.
What would some of the use cases of this model be and what is the current goal?
If you have a Discord account I can get back to you later today
Currently I'm in the middle of re-designing some of the architecture (look at latest blog for more details) and cleaning up the code, so It might be a while.
What would some of the use cases of this model be and what is the current goal?
Use cases: product with interface, for a reasoning demo
Current goal: light small model that can think like OpenAI-o1-mini
If you have a Discord account I can get back to you later today
"I will comeback later to give you the discord, i think 5-7 hours later than this
If you have a Discord account I can get back to you later today
"I will comeback later to give you the discord, i think 5-7 hours later than this
I think you forgot lol
I was on school lol, but i can't use discord rn, so i don't know how we are gonna colaborate
Just let me know when you are available and can provide the necessary details.
Ok, i think i'm avaliable on 02/27/26
I’m going to step away from the collab. The communication and timing haven’t really lined up, and I need to focus on projects that are more consistent. Thanks for reaching out.
We could try again when your profile is more populated.
Thanks for the attention, i'm busy working trying to release my from scratch models.
How in the world are you finding my releases so quickly 🤣
bro idk, it just popped up for me in the "for you page" loll, btw your models are fire!! If you want, take a look on my models too!!
Btw TMLM-Haiku-2 have smth wrong, because i can't use any prompt on CompactAI Studio trying the Pretrain and even the Instruct version
Btw TMLM-Haiku-2 have smth wrong, because i can't use any prompt on CompactAI Studio trying the Pretrain and even the Instruct version
Oh yeah, the CompactAI studio was scrapped (may be brought back) and you need to download the python file + requirements
The TMLM-Haiku-2 model has dozens of features that the original(s) didn't have,
Oh i see it! Thats because TMLM Haiku is evolutioning!
Thats not a word but you get it lol :)
mb lol, i'm brazilian and my english is a lil bit broken
CompactAI moved homepages, its an organization now. Yippee :P
I feel like we could talk about the collaboration. With my advancements in dataset generation and model quality, I feel like it may be worth re-visiting this topic now.
Do you have a Discord account that you can actually provide on a date of your choosing?
Alright, i'll add you when I get a chance, I hid your message so nobody else knows you leaked it
Thanks! I gtg now bye!!!