Master's graduate from Beijing Institute of Technology, published four top‑journal papers, previously worked as a developer at ByteDance and Alibaba. Currently researching large models at a major state‑owned enterprise. Committed to sharing concise, practical AI large‑model development experience, believing that AI large models will become as essential as PCs in the future. Let's start experimenting now!
The article reviews Grok 4’s flashy launch and claimed first‑principles advantage, then presents benchmark results—showing strong reasoning, multimodal and agent scores but disappointing coding performance versus DeepSeek‑R1—concluding that the model’s real‑world capabilities fall short of its hype.
This article demonstrates how to construct a knowledge graph from a 58,099‑character document using GraphRAG, measures the indexing time on an M2 PRO machine, and compares Global, Local, and Drift query modes, highlighting three scenarios where Local mode is unsuitable.
This article walks through the complete GraphRAG workflow—from environment setup and indexing to command‑line and Python API queries—demonstrating how to build a knowledge graph, tune prompts, and retrieve high‑quality answers using Microsoft’s GraphRAG implementation.
This article walks through GraphRAG’s retrieval phase, showing how knowledge‑graph entities, relationships, and community reports are assembled into a query context, comparing local and global modes with traditional RAG, and illustrating the process with a concrete “Age of Big Data” example.
This article walks through the complete GraphRAG knowledge‑graph building pipeline—text splitting, entity extraction, relation mining, community clustering, and report generation—using a concrete example from the book “The Age of Big Data,” and explains why each step improves retrieval and answer quality.
This article explains GraphRAG’s architecture, compares it with traditional RAG, and presents experimental results showing that GraphRAG’s knowledge‑graph‑driven retrieval markedly improves answer accuracy, especially on low‑match, multi‑paragraph queries.
This article explains the GRPO reinforcement‑learning algorithm, shows its core idea of internal group competition without a separate evaluator model, and provides a complete, step‑by‑step code walkthrough—including environment setup, dataset preparation, reward‑function design, training configuration, and evaluation—using the Qwen2.5‑0.5B‑Instruct model on the GSM8K math dataset.
This article introduces EvalScope, an open‑source framework for evaluating large language models, detailing its architecture, built‑in benchmarks, installation steps, and step‑by‑step guides for both performance stress testing and dataset‑based capability assessment, enabling users to independently verify model quality without relying on official documentation.
The DeepSeek‑R1‑0528 model released on May 28 2025 shows major gains in coding, function‑calling and long‑text generation, with benchmark scores that surpass Qwen3‑235B, approach Claude 4 in programming, and include detailed hands‑on prompts and results.
This article walks through the complete installation and configuration of Suna, an open‑source, fully offline AI agent built on Claude 3.7, LangChain, and Daytona, comparing its richer UI and privacy advantages to Manus, and demonstrates its capabilities by automating a Kaggle competition analysis.
