second sentence giving Astra launch vibes
SSH PRO
ProCreations
AI & ML interests
i make stuff
Recent Activity
liked a model 2 days ago
ProCreations/BetterWright-K2-Horizon-7B-Uno updated a model 2 days ago
ProCreations/BetterWright-K2-Horizon-7B-Uno published a model 2 days ago
ProCreations/BetterWright-K2-Horizon-7B-UnoOrganizations
replied to Banaxi-Tech's post 4 days ago
posted an update 5 days ago
Post
1508
Introducing Bonsai 2 27b GSQ RCO! It applies two newly-popular methods for quants to retain higher accuracy. Bonsai 2 27b GSQ RCO achieves around 6 percent lower perplexity on WikiText-2 compared to Bonsai 2 27b and roughly unchanged benchmark accuracy overall, with small mixed differences. It stays under 7gb, staying small like the original bonsai. Note that this is more of an experiment than a true finished product but the gains we saw are cool! Test it out and let me know what you think!
ProCreations/bonsai-2-27b-gsq-rco-gguf
ProCreations/bonsai-2-27b-gsq-rco-gguf
posted an update 7 days ago
Post
232
Bonsai 2 27b is out! Here's a DFlash2 head for it so you can run it as fast as possible
ProCreations/Ternary-Bonsai-2-27B-DFlash2
ProCreations/Ternary-Bonsai-2-27B-DFlash2
posted an update 9 days ago
Post
96
ICYMI:
Grug 27b v2 released! It brings increased quality, fixes repetitive loop / malformed session title issues seen in grug 27b v1.1, and reasoning efforts now truly work.
ProCreations/grug-27b-v2
Grug 27b v2 released! It brings increased quality, fixes repetitive loop / malformed session title issues seen in grug 27b v1.1, and reasoning efforts now truly work.
ProCreations/grug-27b-v2
- 1 reply
posted an update 14 days ago
Post
111
Grug 27b gguf got around 400,000 downloads. Grug v1.1 gguf got around 1000. Do you guys want new grugs?
posted an update 15 days ago
Post
145
auto 0.4b 2 is out now and it beats out auto 1b at half the size. check it out! https://x.com/SSHCodes/status/2097575612919545869
- 1 reply
replied to their post 20 days ago
nvidia
replied to their post 24 days ago
I didn't know it was based off of ornith, that's pretty neat!
You can also make some more datasets. Every dataset you've made seems to be perfect quality somehow, and all of them do well.While I'm here, how do you generate your datasets?
I thought it'd be cool to do something similar to what you do with grug for a custom finetune, and I have no idea how to generate a dataset anywhere near your 1m for grug v1.1.Do you open source them (or the big one eventually?)
It is pretty much all synthetic data. It is very easy for me to generate data with AI because of my pro 6000 (can get 4500 tokens a sec on qwen 3.8 27b aggregate) but before then i just used gpt 5.6 luna through a codex subscription and that's it.
replied to their post 24 days ago
Grug 35b v1 didn’t get that much demand even though it was based off ornith 1 but I might do this! Thanks for suggestion
replied to their post 24 days ago
Matbe
replied to their post 24 days ago
Random side project
replied to their post about 1 month ago
Its very nicce too
thanks!
replied to their post about 1 month ago
let me know how it goes!
posted an update about 1 month ago
Post
2387
check out ProCreations/booper-chat an extremely tiny LM that is surprisingly coherent at english and hilarious to chat with. space: ProCreations/booper-chat
btw for the compute giveaway, winners have been decided. Booper is apart of this (won, they allowed the model to be posted on my page and they used it for their use case) and the other winners have been contacted.
NOTE!! this model is trained purely as a conversationalist and do NOT depend on it for math, facts, and pretty much everything.
btw for the compute giveaway, winners have been decided. Booper is apart of this (won, they allowed the model to be posted on my page and they used it for their use case) and the other winners have been contacted.
NOTE!! this model is trained purely as a conversationalist and do NOT depend on it for math, facts, and pretty much everything.
- 4 replies
replied to their post about 1 month ago
@ProCreations can you run the thing when you have time? I put it in the dataset link above
I know I can’t do it right away lol