Tag
Release of Ternary-Bonsai-2-27B-gguf, a 27B-class language model using ternary weights for extreme compression (5.9 GB) while retaining 98.2% of FP16 intelligence, optimized for efficient inference on laptops and single GPUs.
Critique of Qwen 3.8 model's conversational output, noting it produces nonsensical sentences while being optimized for coding and agentic usage.
A 230M-parameter chess engine fine-tuned from LFM2.5-230M-Base using Stockfish distillation, capable of predicting chess moves with a language model interface and playing at around 2004 Elo.
The paper introduces NepLEGiT, a specialized small language model pre-trained from scratch on Nepali legal text to enhance legal knowledge accessibility and service delivery in Nepal.
POSPAN introduces a position-constrained span masking framework that unifies existing span-level methods for language model pre-training, showing enhanced performance on NLU benchmarks.
HintMiner is a novel tool that automatically mines hints for user questions from Q&A web posts like Stack Overflow using a language model trained via self-supervised learning, achieving effective performance in evaluations.
The paper proposes Comparison-based Preference Optimization (ComPO), a zeroth-order alignment method for LLMs that uses comparison oracles to avoid likelihood displacement. It includes theoretical guarantees and experimental improvements over existing methods.
This paper investigates how the availability of unnecessary external tools affects large language models' performance on tasks that do not require them. The study reveals a significant drop in correct answering rates and proposes a system instruction as a mitigation strategy.
Atria Dawn Preview is a foundation agentic language model designed for scientific research, achieving competitive benchmark results and demonstrating a shift toward human-AI project-level collaboration.
Yandex has released AliceAI-Foundation-80B-A3B-Base, a foundation language model with a hybrid architecture and MoE layers, featuring 80 billion parameters (3 billion active per token), trained from scratch and excelling in Russian factual knowledge tasks.
This paper investigates portable semantics and negative transfer in latent communication between language-model cells, showing that independently trained societies share semantic interfaces but not raw language, and that inherited interfaces can cause severe negative transfer.
This narrative describes the development and rapid intelligence growth of an AI model named Gordon, surpassing its smaller prototype Algernon, and explores societal impacts and ethical dilemmas.
An article reflecting on the state or impact of ChatGPT two years after its initial release.
GLM 5.3 Flash is an upcoming AI model that showcases high token-per-second performance on 2x DGX Spark systems, with an imminent release.
Mercury 2.5 is announced as the most capable diffusion language model, with a 40% increase in intelligence over Mercury 2, operating at over 1,100 tokens/sec on NVIDIA GPUs, and optimized for production with low latency and cost.
This article likely covers the release of LLM version 0.35, which is an update to a large language model.
MiniCPM5-2B is a 2B-parameter open-source language model that achieves high intelligence density for edge deployment and ranks highly in several AI performance benchmarks.
NeoHorse-1-9B is a 9B parameter language model derived from Qwen3.5-9B, fine-tuned for agent tasks and tool use with benchmark performance improvements.
NeoHorse-1-4B is a 4B parameter language model fine-tuned from Qwen3.5-4B for agentic tasks, using routing-guided post-training to enable recursive self-improvement.
The paper curates and releases open datasets and a model for Armenian, demonstrating that continued pretraining with news and STEM data improves performance and addresses data scarcity in low-resource NLP.