LLMS

→ View original post on X — @rasbt

GRPO-based models and gibberish detection techniques

By

@rasbt

–

28 September 2025 18h18

I think they are all basically based on GPRO++, Dr. GRPO etc. But the one interesting one is the Gibberish detection. Need to borrow this!

→ View original post on X — @datachaz

Anuar Sharafudinov’s ollm Repository: Open Source LLM Project

By

@datachaz

–

28 September 2025 17h33

Awesome work by Anuar Sharafudinov! Here's the repo:
↳ https://
github.com/Mega4alik/ollm

→ View original post on X — @datachaz

oLLM: Lightweight Python Library for LLM Inference

By

@datachaz

–

28 September 2025 17h33

Ace. oLLM is a lightweight Python library for LLM inference built on top of transformers Run qwen3-next-80B, GPT-OSS, Llama3, on consumer hardware. ↳ Handle 100k tokens on an 8GB GPU
↳ Works with contracts, logs, reports
↳ No quantization, just fp16/bf16 Repo in ↓

Embodied AI: From LLMs to World Models Survey

By

–

28 September 2025 16h51

8. Embodied AI: From LLMs to World Models This paper surveys embodied AI through the lens of LLMs and World Models (WMs).

Frontier AI Models Approach Expert Parity in Accuracy

By

–

28 September 2025 16h51

9. GDPval It shows frontier models improve roughly linearly and are nearing expert parity, with Claude Opus 4.1 preferred or tied 47.6% of the time, while GPT-5 leads in accuracy.

Language Models Learn to Think and Chat Better with RL

By

–

28 September 2025 16h51

7. Language Models that Think, Chat Better A simple recipe, RL with Model-rewarded Thinking, makes small open models “plan first, answer second” on regular chat prompts and trains them with online RL against a preference reward.

ARK-V1: Lightweight Agent for Knowledge Graph Navigation

By

–

28 September 2025 16h51

6. ARK-V1 ARK-V1 is a lightweight agent that helps language models answer questions by actively walking through a knowledge graph instead of relying only on memorized text.

LLM-JEPA: Adapting JEPA Training to Language Models

By

–

28 September 2025 16h50

5. LLM-JEPA A JEPA-style training objective is adapted to LLMs by treating paired views of the same underlying content (for example, text and code) as prediction targets in embedding space, added on top of the usual next-token loss.