Llama 3.3 (70B) Finetuning - now with 90K context length and fits on <41GB VRAM.
What changed in the latest unsloth (2025.2.15) vs 2025.1.5 version that can cause training failure?
Day 6: One More Thing, DeepSeek-V3/R1 Inference System Overview
Phi-4-mini Bug Fixes + GGUFs
DeepSeek Realse 5th Bomb! Cluster Bomb Again! 3FS (distributed file system) & smallpond (A lightweight data processing framework)
Microsoft announces Phi-4-multimodal and Phi-4-mini
[P] Train your own Reasoning model - GRPO works on just 5GB VRAM
DeepSeek Realse 4th Bomb! DualPipe an innovative bidirectional pipeline parallism algorithm
What would you like to see in Unsloth for 2025?
You can now train your own Reasoning model with just 5GB VRAM
DeepSeek Realse 3th Bomb! DeepGEMM a library for efficient FP8 General Matrix
DeepSeek 2nd OSS package - DeepEP - Expert parallel FP8 MOE kernels
DeepSeek Realse 2nd Bomb, DeepEP a communication library tailored for MoE model
FlashMLA - Day 1 of OpenSourceWeek
OpenAI Triton Course/Tutorial Recommendations
Perplexity R1 Llama 70B Uncensored GGUFs & Dynamic 4bit quant
10x longer contexts for reasoning training - 90% less memory GRPO in Unsloth
We GRPO-ed a 1.5B model to test LLM Spatial Reasoning by solving MAZE
R1-1776 Dynamic GGUFs by Unsloth
PerplexityAI releases R1-1776, a DeepSeek-R1 finetune that removes Chinese censorship while maintaining reasoning capabilities
Unsloth is the #1 trending repo on GitHub!
[P] GRPO fits in 8GB VRAM - DeepSeek R1's Zero's recipe
You can now train your own DeepSeek-R1 model on your local device!
Train your own Reasoning model - 80% less VRAM - GRPO now in Unsloth (7GB VRAM min.)
