A public account focused on deep learning, computer vision, and autonomous driving perception algorithms, covering visual CV, neural networks, pattern recognition, related hardware and software configurations, and open-source projects.
The article provides a detailed analysis of Meta’s PerceptionLM, an open‑source perception language model built on Llama 3, describing its vision encoder, projector, dynamic tiling, three‑stage training pipeline, model variants, and competitive performance on image and video benchmarks.
This article breaks down the core deep‑learning optimizers—gradient descent, Momentum, AdaGrad, RMSProp and Adam—showing why vanilla gradient descent converges slowly, how each method uses exponential moving averages to accelerate training, and why Adam is generally the preferred choice.
This article explains the exponential weighted moving average (EWMA) as a practical time‑series approximation method, detailing its motivation, recursive formula, weight decay behavior, typical beta values, and a bias‑correction technique that improves early‑stage estimates.
GLM‑4.5 and its lightweight Air variant, featuring a deep‑layered MoE design, grouped‑query attention, and dual inference modes, achieve third‑place overall on 12 hard‑core benchmarks, excel in web‑browsing and tool‑calling with a 90.6 % success rate, and introduce novel training tricks such as the Muon optimizer and Slime RL framework.
This article explains the RLHF (Reinforcement Learning from Human Feedback) pipeline that powers ChatGPT and other large language models, covering the limitations of traditional fine‑tuning, the creation of human‑feedback datasets, reward‑model training, loss design, and the final PPO‑based fine‑tuning step.
Alibaba’s Qwen team has released Qwen3-Coder, a 480‑billion‑parameter open‑source LLM specialized for code, featuring a 1‑million‑token context via YaRN, extensive benchmark superiority over most open models, and performance that rivals Claude 4 Sonnet while remaining fully accessible.
This article presents a step‑by‑step agentic object detection pipeline that combines open‑vocabulary detectors such as Grounding‑DINO with visual language models (GPT‑4o, o1) for concept extraction, critique, refinement, and validation, complete with code snippets, design rationale, and real‑world examples.
This article explains how low‑rank adaptation (LoRA) and its quantized variant (QLoRA) compress large language model weights, reduce training cost, and enable flexible adapter switching, while detailing matrix decomposition, training mechanics, and trade‑offs with concrete examples and quantitative analysis.
This tutorial walks through fine‑tuning OpenAI's CLIP ViT‑B/32 on a small image‑text dataset in a Kaggle notebook, covering environment setup, model loading, data preprocessing with CLIPProcessor, training a linear head, and observing loss convergence to align visual and textual embeddings.
Kimi K2, an open‑source trillion‑parameter AI model released by Moonshot AI, offers Base and Instruct variants, achieves leading scores on benchmarks such as SWE‑bench, LiveCodeBench and AceBench, and introduces a novel post‑training autonomous‑exploration stage with MuonClip optimization to enable robust tool use and reinforcement‑learning‑driven self‑improvement.
