Havuz

Tools Are Not Islands: Set-Level Tool Retrieval for LLM Agents via Query-Conditioned Hyperedge Prediction

Daily LLM value rankings - compare 300+ models by intelligence, speed and price. OpenRouter + Artificial Analysis. 大模型性价比排行榜

Run the native 284B-A13B DeepSeek-V4-Flash-0731 LLM locally on one laptop CPU: pure C, 8 GB RAM minimum, no GPU, best TPOT 0.892 s/token, re

面向智能座舱的云边协同 AI Agent：端侧毫秒级车控，云端声明式 Multi-Agent 与 Skill/DAG 编排，支持 S2S 实时语音、声纹多用户、HMI和Android双端；LLM 只负责理解与规划，VAL 负责确定性安全执行。

🧠 One LLM, split across a Mac (Apple MPS) + a Windows PC (NVIDIA CUDA) — heterogeneous pipeline inference over a framework-neutral wire, bit

🚀 One API gateway for GPT, Claude, Gemini, DeepSeek and more. OpenAI-compatible API for developers.

⚽️🤖 Benchmarking LLMs and deep-research agents on real-world football prediction — from the tactical "who scores in minute 67" to the strate

Evaan — a fully local, offline-capable AI companion chatbot built with Python & Hugging Face Transformers, running on Qwen2.5-0.5B-Instruct

Deterministic causality engine for AI roleplay.

A recursive self-improvement (RSI) workflow that evolves both target agents (e.g., Pi and DeepSeek Harness) and the RSI system itself.

Qwen3.8-27B on one laptop CPU: up to 2.52 token/s, 8 GB tested, no runtime accuracy loss. Native OpenAI-compatible API with function tools;

The best model under 200B at near 10 token/s exact batch throughput on a single laptop CPU, with an automatic 8 GB RAM path. Native C, no GP

An experiment-driven laboratory for understanding LLM sampling strategies. Explore how Temperature, Top-k, and Top-p influence token probabi

Official implementation of Latent-GRPO: reinforcement learning for vocabulary-space latent reasoning.

🔥🔥 Accelerating decision inference with exact state forking

Official implementation of “ReSAIL: Mitigating Collapse in Iterative Agent Self-Distillation”

Asking for What Was Never Requested: horizontal and vertical proactivity in LLM agents. Need-graph metrics (no LLM judge) and Q&D, which tra

Caveman prompting, measured. A two-channel evaluation framework for prompt compression and LLM token efficiency: what input vs output compre

Companion code for the Manning book 'LLM Customization and Fine-Tuning.' Adapt open-weights LLMs end to end: prompting and RAG, LoRA/QLoRA,

Deploy Qwen3.6-35B-A3B (Q4_K_XL) + MTP speculative decoding on a single NVIDIA L4 24GB — GCP g2-standard-8 — via the official llama.cpp Dock

An agentic LLM-powered knowledge assistant that enhances RAG capabilities through automated entity extraction, structured data analysis, and

SkillAdaptor: Self-Adapting Skills for LLM Agents from Trajectories

A user's facts become a few rows in a content-addressed memory table — not a rewrite of the model. Paper, code, data, and interactive site f

Automated auditing pipeline for LLM and agent benchmarks — surfaces task ambiguity, environment conflicts, and evaluation bugs.

An end-to-end framework for Agentic Reinforcement Learning.

大模型故障 token（glitch token）资料库：收录各 LLM 分词器中已公开记录的异常 token，附异常表现、来源与复现链接。A curated catalog of LLM glitch tokens.

Live Benchmark for long Horizon tool using Agentic Systems

爻枢 Token 智能网关、Token 智能聚合、大模型评测、排名等。关注“爻枢未来”公众号获得更多全自动编程详情。

Production-grade LLM Evaluation & Benchmarking Framework - GPT-4, Claude, Gemini, Mistral. Accuracy, latency, cost, hallucination, reasonin

Minimal harness-agnostic recursive language model layer — one binary, Python + llm()

A practical Generative AI engineering lab exploring LLM foundations, tokenization, deployment strategies, and real-world AI system design.

Seven Minutes Is All I Can Spare To Play With You.

Benchmark for repository-level code understanding: 500 open-ended questions from real GitHub Discussions across 50 repositories, graded by a

Autonomous web-app pentest engine — a multi-agent pipeline wielding 30+ offensive-security tools as skills, PoC-validated findings. Built on