eliebakk

• 200+ pages of Hugging Face secrets on how to train an LLM

  Posted by eliebakk@reddit | LocalLLaMA | View on Reddit | 101 comments

• how to train a tiny model (4B) to prove hard theorems

  Posted by eliebakk@reddit | LocalLLaMA | View on Reddit | 20 comments

• What MoE model sizes and capabilities are currently missing in the open weight ecosystem?

  Posted by eliebakk@reddit | LocalLLaMA | View on Reddit | 47 comments

• AMA with Hugging Face Science, the team behind SmolLM, SmolVLM, Fineweb and more.

  Posted by eliebakk@reddit | LocalLLaMA | View on Reddit | 472 comments

• SmolLM3-3B training logs and intermediate checkpoints

  Posted by eliebakk@reddit | LocalLLaMA | View on Reddit | 23 comments

• Full open source reproduction of R1 in progress ⏳

  Posted by eliebakk@reddit | LocalLLaMA | View on Reddit | 157 comments

• SmolLM3: reasoning, long context and multilinguality for 3B parameter only

  Posted by eliebakk@reddit | LocalLLaMA | View on Reddit | 47 comments

• 350k samples to match distilled R1 on *all* benchmark

  Posted by eliebakk@reddit | LocalLLaMA | View on Reddit | 7 comments

• Qwen3 training recap 🐦‍🔥

  Posted by eliebakk@reddit | LocalLLaMA | View on Reddit | 5 comments

• Gemma3 technical report detailed analysis 💎

  Posted by eliebakk@reddit | LocalLLaMA | View on Reddit | 15 comments

• 7B reasoning model outperforming Claude-3.7 Sonnet on IOI

  Posted by eliebakk@reddit | LocalLLaMA | View on Reddit | 28 comments

• New Reasoning model (Reka Flash 3 - 21B)

  Posted by eliebakk@reddit | LocalLLaMA | View on Reddit | 27 comments

• DCLM dataset but better for smol models

  Posted by eliebakk@reddit | LocalLLaMA | View on Reddit | 5 comments

• Do mechanistic interpretability on your laptop with the new SmolLM2 checkpoints

  Posted by eliebakk@reddit | LocalLLaMA | View on Reddit | 3 comments

• Training LLM on 1000s of GPUs made simple

  Posted by eliebakk@reddit | LocalLLaMA | View on Reddit | 33 comments

• Claude Sonnet 3.7 soon

  Posted by eliebakk@reddit | LocalLLaMA | View on Reddit | 107 comments

• Training on 1000s of GPUs made simple

  Posted by eliebakk@reddit | LocalLLaMA | View on Reddit | 0 comments

• Deepseek R1 GRPO code open sourced 🤯

  Posted by eliebakk@reddit | LocalLLaMA | View on Reddit | 18 comments

• First large scale open source math reasoning dataset with 800k R1 reasoning traces

  Posted by eliebakk@reddit | LocalLLaMA | View on Reddit | 8 comments

• 405B MiniMax MoE technical deepdive

  Posted by eliebakk@reddit | LocalLLaMA | View on Reddit | 10 comments

• Deepseek R1 GRPO code open sourced 🤯

  Posted by eliebakk@reddit | LocalLLaMA | View on Reddit | 0 comments