Big update for teams pretraining LLMs: the FineWeb dataset has just been updated with a knowledge cut-off at Dec 31th 2024
OPEN SOURCE
-
ComfyUI Startup Journey: AI Engineering for Creative Tools
By
–
pod: AI Engineering for Art! https://
latent.space/p/comfyui First pod of the year, with comfyanonymous face reveal! Going over the origin story of @ComfyUI (now a startup, competing with multiple @ycombinator startups like @comfydeploy
), dishing tea on working at @StabilityAI in -
llama.cpp democratizes edge AI for everyone in 2025
By
–
It’s quite a bit crazy that one fine day @ggerganov released llama.cpp and it set an entire new path for bringing intelligence closer to the edge! 2025 is all about bringing the frontier closer to everyone and beyond
-
PrimeIntellect and Federated Learning Support in PyTorch
By
–
@PrimeIntellect is great and looking forward to federated becoming mainstream, ping if anything feels short in pytorch, we'll work on it! @rice_fry has been working on torchft (fault tolerant), and more to come to try help.
-
Hugging Face Releases smolagents Open Source Library for AI Agents
By
–
When you release a new lib on December 31 and the community gets crazy Check it out if you're interested in agents: smolagents – a barebones library for agents – agents write python code to call tools and orchestrate other agents. Github: https://
github.com/huggingface/sm
olagents
… And the -
Open LLM Leaderboard: Leading LLM Evaluation Proxy
By
–
The closest proxy would be Open LLM Leaderboard
-
PyTorch Core Team Recruitment Opportunity
By
–
cool work! if you end up wanting to work on PyTorch, DM me; would love for you to join the core team.
-
Qwen Model on SambaNova Cloud: 3X Faster LLM Inference
By
–
Introducing another model in the Qwen series on SambaNova Cloud! This open-source test-time compute model from @alibaba_cloud enables LLMs to produce accurate responses in seconds, rather than minutes. Runs 3X faster than GPU providers
-
OLMo 2 introduces reordered norm and QK-norm innovations
By
–
Yo! @allen_ai just dropped OLMo 2 Tech report, some interesting things I found Architecture
> Reordered norm: Normalizing outputs of attention and feedforward layers within transformer blocks instead of inputs
> QK-norm: Normalizing key and query projections with RMSNorm
