Gemini-Qwen 하이브리드 파이프라인을 통한 PDF 분석 최적화
How I Built an AI Exam App in 8 Months to outsource studying
How I Built an AI Exam App in 8 Months to outsource studying
DSpark: 준자기회귀 구조로 Speculative Decoding 속도 60~85% 가속
DeepSeek's DSpark Brings Speculative Decoding Back Into the Spotlight — Here's What Developers Need to Know
RadixAttention 통한 Agent 처리량 최대 70% 향상 및 PD 분리 아키텍처 구현
The Open-Model Cost Chart Everyone's Sharing Is API Prices. Here's What Self-Hosting Actually Gets You (Measured)
AI API Price War: DeepSeek V4-Pro Cuts 75% & Gemini 3.5 Flash Lands
I spent two weeks optimizing 96GB of VRAM for local LLMs. Paid APIs still won.
RTX 6000 기반 로컬 Qwen 도입을 통한 데이터 주권 확보 및 매출 회수 달성
GLM-5.2: The Most Powerful Open Model yet and the Brutal Reality of Running It
Qwen3.6-35B NVFP4 runs on one H100 — A100 owners are out
Speculative decoding shifted our output distribution and evals missed it
1,175 Redditors Just Told You to Stop Using Ollama — Here's Why Local AI Tooling Got Serious
Open-Weights Long-Horizon Coding LLMs: India's AI Future 2026
GLM-5.2: Built for Long-Horizon Tasks
Commodity GPU로 1T 모델 1000 TPS 달성한 모델-시스템 Codesign
Doubling Qwen3.6-27B on One RTX 3090: ollama llama.cpp + MTP, Lever by Lever (35.7 80.2 tok/s)
MiMo-v2.5-Pro-UltraSpeed: 1T model with 1000 tokens per second
Google LiteRT-LM Speeds Up Local Inference Up to 2.2x With Gemma 4 Multi-Token Prediction
Speculative decoding: when and why it actually speeds up inference
llama.cpp b9455 Finally Caught vLLM: 70t/s on 2x3090 Qwen 27B UQ8