Improvement Technique

LFM2.5-2.6B is an on-device agentic model that plans, calls tools, and runs multi-step tasks at 220 tok/s in under 2.5 GB. Open weights on Hugging Face.

How a fine-tuned 3-billion-parameter model beats today’s frontier models at one job, and how to build one this weekend

How frontier AI labs use multi-teacher on-policy distillation to merge capabilities, recover from RL forgetting, and scale post-training across MiMo-V2-Flash, GLM-5, Nemotron-Cascade 2, and DeepSeek-V4.

How the algorithm that teaches LLMs to reason actually works...

We’re on a journey to advance and democratize artificial intelligence through open source and open science.

LLM quantization explained: accuracy, latency, and memory tradeoffs. Learn GPTQ, AWQ, and GGUF and where each fits production.

We’re on a journey to advance and democratize artificial intelligence through open source and open science.

Learn all the basics and best practices of fine-tuning. Beginner-friendly.

Exploring memory-efficient techniques for LLMs

Hierarchal Agent Loop Optimizer HALO is an RLM-based agent optimization technique capable of recursively self-improving agents by analyzing their execution traces and suggesting changes. This work is inspired by the Mismanaged Genius Hypothesis

A walkthrough of using the bootstrap fewshot and GEPA optimizers in DSPy

Agentic systems often reach a plateau after proof-of-concept because they depend on humans to diagnose edge cases and correct failures. T...

Understanding GRPO and New Insights from Reasoning Model Papers

Discover the key differences between RAG and fine-tuning, what each approach can bring, and how to choose the right AI approach for your business goals.

Learn all about Reinforcement Learning (RL) and how to train your own DeepSeek-R1 reasoning model with Unsloth using GRPO. A complete guide from beginner to advanced.

➤ Check out our updated Reinforcement Learning guide!

After completing our multi-turn tokenization and masking refactoring, we eliminated a critical bottleneck that was preventing us from building a more consistent and flexible rollout system for our Agent RL research. This breakthrough enabled us to implement a LangGraph-based rollout for VeRL in just a few days, which we’ve already successfully deployed in our Agent RL experiments. In this article, I’ll share our journey from VeRL’s native multi-turn implementation to our new LangGraph-based solution, explaining both the motivations driving this evolution and the technical details of our implementation.

A Blog post by Sara Han Díaz on Hugging Face

We’re on a journey to advance and democratize artificial intelligence through open source and open science.

A deep dive into the methods, fine-tuning pipeline, and operational risks of building specialised models.

Best practices for LoRA hyperparameters and how they affect the fine-tuning process.

We’re on a journey to advance and democratize artificial intelligence through open source and open science.

Step-by-step guide to fine-tuning the Mistral reasoning model on a medical MCQs dataset using the Transformers framework.

AI Engineering bootcamp. Contribute to apatti/AIEBootcamp development by creating an account on GitHub.

but wtf not? - skill gap - high fine-tuning costs - lack of standards and unified approaches over the past few weeks I've been working on maestro - streamlined tool for VLM fine-tuning link: — SkalskiP (@skalskip92)