predibase/lorax

jundot/omlx

LLM inference server with continuous batching & SSD caching for Apple Silicon — managed from the macOS menu bar

vllm-project/vllm

A high-throughput and memory-efficient inference and serving engine for LLMs

Sebastian Raschka, PhD (@rasbt)

Is LoRA (Low Rank Adaptation) relevant in 2025 for reasoning models? I recently read "Tina: Tiny Reasoning Models via LoRA (https://arxiv.org/abs/2504.15777)", and it made me pause for a moment: when was the last time I heard someone excitedly talk/write about LoRA? LoRA was one of the most influential fine-tuning methods in the earlier LLM boom (as you may remember, I wrote about it a lot in recent years). The idea is simple but effective: avoid full model updates and instead inject a small number of trainable parameters for downstream tasks. This drastically reduces memory and compute costs. But in the age of ever-larger instruction-tuned models coupled with well-working distillation techniques (like popularized by DeepSeek-R1 etc), LoRA seemed to become more irrelevant recently. What about LoRA work for developing reasoning models? This paper tackles exactly that question. Instead of the usual supervised fine-tuning or instruction distillation pipeline, the authors use LoRA with reinforcement learning (RL) to improve reasoning capabilities. Specifically, they fine-tune a 1.5B base model using LoRA adapters while applying RL on reasoning benchmarks. Their baseline model is DeepSeek-R1-Distill-Qwen-1.5B, which is a model already fine-tuned for reasoning tasks. (I wish they started with the base Qwen-1.5B model; but this way, I guess they have more comparisons with other methods that further trained the DeepSeek-R1-Distill-Qwen-1.5B.) From there, the authors ran experiments across datasets, learning rates, LoRA ranks, and RL algorithms. Their best-performing model was trained on just 7k examples and cost just $9 to train. Even with hyperparameter sweeps and multiple ablations, the entire study cost just $526. So, how well does LoRA work? The top half of the results figure (highlighted in blue) compares models trained with LoRA-based RL versus standard RL (i.e., no LoRA). On every benchmark (AIME24, AIME25, AMC23, MATH500, GPAQ, Minerva), LoRA outperforms the regular RL baseline when applied to the same starting model. Insights from ablations 1) Surprisingly, the best-performing model came from the smallest dataset: just 7k examples from Open-RS. 2) The classic LoRA rank 16 emerged as the sweet spot, but ranks 8 and 32 also worked well. 3) It's nice that they included the recent Dr. GRPO (I recently discussed it in my latest Ahead of AI blog). It substantially reduces training time by length-normalizing rewards and addressing issues in GRPO Bottom line: Reasoning is certainly an interesting use case, and it's interesting (and a bit surprising) that LoRA does so well here. It might also be the first case where I've seen LoRA coupled with RL, which is another interesting aspect. LoRA certainly peaked in popularity 1-2 years ago, and more people now consider (more expensive) full-parameter updates (based on anecdotal perception); there's still a place for LoRA and LoRA-like methods. Let's not forget that one of the key advantages of LoRA is that it doesn't modify the underlying base model. This is key in applications where you either have lots of specialized use cases or lots of customers. For example, instead of storing 100 1B full-parameter tuned models, it would be much cheaper to store a 32B model with 100 sets of LoRA weights.

omlx/docs/oQ_Quantization.md at main · jundot/omlx

LLM inference server with continuous batching & SSD caching for Apple Silicon — managed from the macOS menu bar - jundot/omlx

Overview - GroqDocs

Fast LLM inference, OpenAI-compatible. Simple to integrate, easy to scale. Start building in minutes.

omlx/docs/experimental/dflash_mlx_integration.md at main · jundot/omlx

LLM inference server with continuous batching & SSD caching for Apple Silicon — managed from the macOS menu bar - jundot/omlx

EricLBuehler/mistral.rs

Fast, flexible LLM inference

Luce-Org/lucebox-hub

Fast LLM speculative inference server for consumer hardware.

leobeeson/llm_benchmarks

A collection of benchmarks and datasets for evaluating LLM.

Run LLMs locally on your Mac · mlx-optiq

Quantize, fine-tune and serve LLMs locally on Apple Silicon. MLX-native, no PyTorch, no cloud. On PyPI.

[Tool Release] Finetune & Quantize 1–3B LLMs on 8GB RAM using LoFT CLI (TinyLlama + QLoRA + llama.cpp)

23 votes, 16 comments. Hey folks — I’ve been working on a CLI tool called LoFT (Low-RAM Finetuning Toolkit), and I finally have a working release. 🔧…

Home - NobodyWho

NobodyWho is an inference engine that lets you run LLMs locally on any device

Sakana AI on Twitter / X

We’re excited to introduce Doc-to-LoRA and Text-to-LoRA, two related research exploring how to make LLM customization faster and more accessible.https://t.co/wGKDNhBcJXBy training a Hypernetwork to generate LoRA adapters on the fly, these methods allow models to instantly… pic.twitter.com/gId3J6hgEr— Sakana AI (@SakanaAILabs) February 27, 2026

Instant LLM Updates with Doc-to-LoRA and Text-to-LoRA

Recent LLM agents have shown impressive capabilities on complex computer use and long-horizon tasks. Yet, they still struggle with long-term memory and adaptation--two of the most important cognitive capabilities that still limit LLMs today. Without long-term memory, users have to provide LLMs with relevant content at the start of every new session, creating friction, discontinuity, and longer time-to-response. Additionally, due to the lack of adaptation, they do not learn from mistakes or user preferences from previous sessions, making each interaction as cumbersome as the first. Traditionally, these two problems are tackled by "updating" the model.

Announcing Burn-LM (alpha): LLM Inference Engine

We're happy to announce the Burn-LM, an LLM inference engine based on Burn! The goal is to support any large model, LLM, VLM, and others, for inference but also for training (pre-training, post-training, and fine-tuning).