AI 开发 教程
系统化的 AI 开发学习路径。从提示词工程到模型微调，从 RAG 到 Agent 开发，覆盖 AI 应用开发全栈知识。共 142 篇教程，18 个分类，入门到高级全覆盖。
找到 17 篇教程
DeepSeek Model Evaluation and Benchmarking
Complete evaluation guide for DeepSeek models: results on mainstream benchmarks such as MMLU, HumanEval, GSM8K, and MT-Bench. Comprehensive comparison of DeepSeek-V3 vs GPT-4o vs Claude 3.5, including evaluation code and model selection recommendations.
阅读教程 →DeepSeek Data Engineering Tutorial
From data collection to quality evaluation, fully master the large model data engineering pipeline. Covers core aspects such as SFT dataset construction, RLHF preference data preparation, data cleaning and deduplication, data augmentation, and large-scale data processing.
阅读教程 →DeepSeek Model Architecture Deep Dive
In-depth analysis of DeepSeek-V3's core technologies such as MoE architecture, MLA attention mechanism, and Multi-Token Prediction. Understand how DeepSeek achieves GPT-4o-level performance at extremely low cost, including architecture diagrams and source code analysis.
阅读教程 →DeepSeek Fine-tuning Hands-on Tutorial
Full practical guide to fine-tuning DeepSeek models: LoRA/QLoRA efficient fine-tuning, full-parameter fine-tuning, data preparation, training configuration, and evaluation validation. Covers fine-tuning solutions for DeepSeek-V3 and DeepSeek-R1, including complete code and best practices.
阅读教程 →Model Distillation in Practice: Training Small Models with Large Models
Knowledge distillation allows small models to approach large model capabilities while significantly reducing inference costs. This article provides an in-depth explanation of distillation principles, data generation strategies, training techniques, and effect evaluation, teaching you to distill cost-effective small models with DeepSeek.
阅读教程 →RLHF Reinforcement Fine-Tuning in Practice: From Preference Data to Reward Model
RLHF aligns model outputs with human preferences. This article walks through the complete RLHF pipeline, from preference data collection and reward model training to PPO reinforcement fine-tuning, and provides engineering insights on data quality and effect evaluation.
阅读教程 →Practical Guide to QLoRA Fine-Tuning
Full-parameter fine-tuning is expensive; QLoRA uses 4-bit quantization and low-rank adaptation to allow ordinary developers to fine-tune large models on consumer-grade GPUs. This article provides a hands-on guide to the complete QLoRA fine-tuning process from data preparation to model deployment, using DeepSeek as the base model.
阅读教程 →The Art of Building Fine-Tuning Datasets
The upper limit of fine-tuning effectiveness is determined by data quality, not the model or algorithm. This article delves into the methodology for building fine-tuning datasets, covering data sources, quality control, diversity assurance, and data mixing strategies, helping you build high-quality fine-tuning datasets.
阅读教程 →Model Evaluation and A/B Testing
After fine-tuning, how can you objectively evaluate model performance? How should A/B tests be designed to obtain statistically significant conclusions? This article establishes a scientific model evaluation system from three dimensions: offline evaluation, online A/B testing, and statistical significance.
阅读教程 →Comprehensive Comparison of LoRA vs QLoRA vs Full Fine-tuning
A comprehensive comparison of LoRA, QLoRA, and full fine-tuning, analyzing from the perspectives of principles, memory usage, training speed, and final performance, to help you make the right technology choice.
阅读教程 →Practical Construction of Instruction Fine-Tuning Data
Data quality determines the upper limit of fine-tuning effectiveness. This article explains the construction methods for instruction fine-tuning data, including data sources, cleaning strategies, quality evaluation, data augmentation, and mixing design.
阅读教程 →Practical Model Quantization and Deployment
Quantization is a key technology to reduce model inference costs. This article systematically explains the principles, toolchains, and deployment practices of model quantization from INT8 to INT4, enabling your model to run efficiently on consumer-grade hardware.
阅读教程 →Detailed Explanation of DPO Preference Alignment Training
DPO (Direct Preference Optimization) is a simplified alternative to RLHF that does not require training a reward model. This article explains the principles, implementation details, and training techniques of DPO in depth.
阅读教程 →Complete Guide to Large Model Evaluation and Benchmarking
How to objectively evaluate the capabilities of large models? This article systematically explains mainstream benchmark tests such as MMLU, HumanEval, and C-Eval, as well as evaluation methods like LLM-as-Judge and arena rankings, to help you establish a scientific model evaluation system.
阅读教程 →Dataset Construction and Quality Evaluation
Data quality determines the upper limit of fine-tuning effectiveness. This article systematically explains the construction methods of fine-tuning datasets, data cleaning strategies, quality evaluation metrics, and data augmentation techniques, helping you build high-quality fine-tuning datasets.
阅读教程 →QLoRA Quantized Fine-Tuning in Practice
QLoRA reduces the memory requirements for fine-tuning large models to the extreme through 4-bit quantization. This article provides an in-depth explanation of QLoRA's technical principles, NF4 quantization, double quantization, and other key technologies, along with complete practical code.
阅读教程 →Complete Guide to LoRA Fine-Tuning
LoRA is currently the most popular parameter-efficient fine-tuning method. This article comprehensively explains the complete LoRA fine-tuning process from principles to practice, including environment configuration, dataset preparation, training parameter settings, and model evaluation.
阅读教程 →