The article explains how to transform a pretrained speech model into a product‑ready assistant by defining demonstration data, clarifying team debates on persona, safety, and length, contrasting alignment with pretraining, and highlighting common pitfalls to avoid during deployment.
A new study by USTC and Shanghai AI Lab uncovers a power‑law scaling relationship between RL fine‑tuning compute and large‑model reasoning performance, offering a quantitative way to predict and control AI capability growth.
The article introduces mlx‑tune, a community project that wraps the MLX library with Unsloth's API to enable local fine‑tuning of large language, vision, TTS, STT, OCR, and embedding models on Apple Silicon Macs, outlines its workflow from prototype to cloud, provides installation steps, code examples, and discusses its capabilities and limitations.
The article outlines a step‑by‑step learning roadmap for AI Agent development, covering large‑model fundamentals, prompt engineering, retrieval‑augmented generation, agent architecture, production practices, and fine‑tuning concepts to help engineers progress from entry‑level to senior roles.
This article analyzes why multimodal large language models often fail to detect small objects, identifies three core bottlenecks, and presents a four‑tiered optimization roadmap—from zero‑cost inference tricks to data augmentation, model fine‑tuning, and engineering safeguards—backed by three real‑world case studies and actionable guidelines.
This article systematically reviews the main reinforcement‑learning algorithms—PPO, GRPO, DAPO, GSPO, and SAPO—used for fine‑tuning large language models, explaining why supervised fine‑tuning precedes RL, how each method improves training efficiency and stability, and what trade‑offs they entail.
This article details the AI techniques behind Gaode's "Street Ranking" project, explaining the Fancy video concept, the dual training and production pipelines, and the use of SFT, reinforcement learning, MoE‑LoRA, distribution‑matching distillation, and quality‑filtering to achieve 25× faster generation with high aesthetic fidelity.
The article examines why AI product managers repeatedly fall into three traps—over‑relying on prompt engineering, blindly adopting Retrieval‑Augmented Generation, or costly fine‑tuning—by presenting real‑world failures, debunking myths, and offering a five‑layer decision framework with cost, data, resource, and risk analysis to choose the right solution.
This article explains how Xiaomi’s Xiao AI assistant leverages large language models for intent routing, domain‑specific intent understanding, and response generation, detailing the system architecture, challenges such as knowledge requirements and latency constraints, and the shift from prompt engineering to model fine‑tuning.
This article explains how large language models are integrated into Xiaomi’s Xiao AI assistant, covering intent distribution, domain‑specific intent understanding, response generation, architectural design, challenges such as knowledge requirements and latency, and the shift from prompt engineering to model fine‑tuning.
This article explains how Xiaomi’s XiaoAI assistant integrates large language models for intent distribution, vertical intent understanding, and response generation, detailing the architecture, challenges such as knowledge requirements and sub‑200 ms latency, and the shift from prompt engineering to model fine‑tuning that boosted user retention by 10% and query satisfaction by 8%.
This article explores how large language models are integrated into Xiaomi’s XiaoAI assistant, detailing the system’s architecture, intent distribution, domain-specific understanding, and response generation, while sharing practical challenges, prompt engineering solutions, and fine‑tuning strategies that boosted user retention and query satisfaction.
This comprehensive guide walks you through generative AI fundamentals—including transformers, diffusion models, large language models, and multimodal systems—then explores practical API usage with OpenAI, Hugging Face, and Vertex AI, followed by model fine‑tuning, LoRA, knowledge injection, and advanced topics such as model distillation, prompt chaining, AutoML, tool integration, and retrieval‑augmented generation.
This article describes how the Gaode terminal team tackled large‑scale repository upgrades by building a code‑RAG and code‑Agent tool, addressing recall and stability issues, then fine‑tuning a small LLM (Qwen3‑4B) with LoRA and custom datasets to achieve reliable, low‑cost, on‑device code‑query performance.
This article introduces the open‑source Zhi‑Create‑Qwen3‑32B model, detailing its fine‑tuned training on creative‑writing data, the multi‑domain dataset strategy, curriculum‑learning based SFT, evaluation on WritingBench, and practical deployment options across various hardware and inference frameworks.
This article describes how a video review system combines multimodal CLIP models, image‑text feature alignment, and optimized vector‑search databases such as RedisSearch and Elasticsearch to detect prohibited content in real time and perform large‑scale historical recall, while addressing challenges of generalization, storage cost, and inference speed.
This article presents a comprehensive blueprint for constructing a DeepSeek-powered AI Ops platform, detailing the six‑module architecture, data collection stack, AI engine deployment options, application and interaction layers, implementation road‑map, model training, security measures, cost estimates, and risk mitigation strategies.
In a detailed interview, ByteDance AI specialist Cai Conghuai explains how large‑model techniques such as SFT, DPO and RAG address Douyin’s multimodal user‑experience challenges, improve signal detection, root‑cause analysis, and outline future AI‑agent breakthroughs for content platforms.
This paper introduces the Thought Leap Bridge task and the CoT‑Bridge model, which detect and fill missing intermediate steps in chain‑of‑thought reasoning, dramatically improving large language model performance on mathematical and logical benchmarks and enhancing downstream distillation and reinforcement‑learning pipelines.
The article argues that despite the rapid rise of powerful large language models, mastering machine learning remains essential because it underpins these models, offers customized solutions for specialized tasks, and cultivates the mathematical, programming, and analytical skills needed to effectively use and extend AI technologies.
This article explores the application of Large Language Models (LLM) in text information detection and extraction, focusing on blue-collar recruitment data processing. It details the implementation of LLM through prompt engineering, RAG enhancement, and model fine-tuning to improve data cleaning efficiency and accuracy.
This article explains the concept of tool-specific tokens for large language models, detailing how they enable efficient, reliable tool calls, the implementation steps, advantages over JSON, practical advice, comparisons, challenges, and future directions for AI agents.
The article explains why most projects should start with prompt engineering or simple agent workflows, outlines the scenarios where model fine‑tuning adds real value, compares fine‑tuning with Retrieval‑Augmented Generation, and offers practical criteria for deciding which approach to adopt.
This guide walks you through fine‑tuning a large language model without any coding by using LLaMA‑Factory, covering LoRA fundamentals, environment setup, dataset creation, parameter configuration, training, loss monitoring, model export, and a quick evaluation on the Qwen2.5‑0.5B model.
This article explains what constitutes high‑quality LLM training data, why large datasets are essential, outlines the step‑by‑step process for collecting, preprocessing, and fine‑tuning models, and highlights the best data sources—including web content, books, code repositories, and news—while noting available free datasets.
This tutorial shows how to use Alibaba Cloud DataWorks' serverless GPU resource groups together with the open‑source LLaMA‑Factory framework to fine‑tune the Qwen2‑VL‑2B multimodal model for tourism‑domain Q&A, covering environment setup, dataset preparation, parameter configuration, training, and interactive inference.
Experts from Dipu, Ant Financial, iKang, and Zhihu discuss practical strategies for improving large model performance, covering RAG, tool‑using, offline knowledge engineering, multimodal training, evaluation metrics, and future trends, while sharing case studies from manufacturing, healthcare, retail, and C‑end applications.
The article outlines eight distinct enterprise strategies for adopting DeepSeek, categorizing them by model maturity, available data types, and specific business challenges, and maps these approaches onto four capability tiers—from basic compliance requirements to advanced multimodal, low‑cost solutions.
This guide walks you through the hardware and software prerequisites, Docker-based installation, environment configuration, model fine‑tuning, IDE integration, maintenance, and troubleshooting for running the DeepSeek AI programming assistant entirely on your own machine.
This article presents a detailed overview of generative AI lifecycle management, covering practical use cases such as email summarization, the roles of providers, fine‑tuners and consumers, MLOps/LLMOps processes, retrieval‑augmented generation, efficient fine‑tuning methods like PEFT, and Amazon Bedrock services for model deployment and monitoring.
This guide walks you through deploying the GPT‑SoVITS text‑to‑speech model on Alibaba Cloud Function Compute, covering architecture, step‑by‑step deployment, quick voice synthesis, and advanced fine‑tuning using NAS storage, VPC networking, and optional custom domains.
This article examines how programming large models combined with code knowledge graphs can overcome the limited context of traditional code‑completion tools, detailing key techniques, trigger strategies, context acquisition methods, model fine‑tuning practices, current challenges, and future research directions for intelligent, repository‑wide code suggestions.
This article explains how to preprocess diverse error messages from apps, mini‑programs, and browsers, then fine‑tune a small spaCy NLP model with data generated via ChatGPT to automatically split and categorize errors for alerting and workflow handling.
The article introduces Marco‑o1, an open‑source LLM that enhances complex reasoning by fine‑tuning on Chain‑of‑Thought data, integrating Monte‑Carlo Tree Search, introducing mini‑step actions and a reflection mechanism, and evaluates its performance on multilingual math and translation benchmarks.
This article examines the AIGC engineering architecture, detailing its data, model, fine‑tuning, inference, application, and monitoring layers, and explains the distinct responsibilities and challenges of application engineers, algorithm engineers, and “alchemy” specialists, highlighting how this structured approach accelerates generative AI productization.
By leveraging multimodal large models such as GPT‑4o and fine‑tuned Qwen‑VL, the study builds a prompt‑engineered and SFT‑enhanced voucher verification system that classifies product categories, detects diverse defects, and estimates problem counts, achieving up to 90 % accuracy and meeting real‑time business throughput requirements.
The open‑source Yuan 2.0 large model is fully integrated into Baidu’s PaddleNLP, offering quick inference for tasks like code generation, translation, and reasoning, along with efficient distributed training and fine‑tuning features such as Zero Padding optimization, enabling developers to easily deploy and customize the model via simple setup steps and example interactions.
The article describes a multimodal AI video moderation system built on Alibaba’s Chinese‑CLIP model and hybrid RedisSearch/ElasticSearch vector databases, enabling real‑time violation detection and historical recall, with fine‑tuned black‑market ad detection, FP16 quantization, and OpenVINO acceleration to boost speed and cut storage.
This article provides an in‑depth, step‑by‑step guide on how enterprises can choose between open‑source and closed‑source large language models, adapt them through incremental pre‑training, instruction fine‑tuning, and reinforcement learning, and finally deploy them across front‑office, middle‑office, and back‑office scenarios to drive digital transformation.
This guide walks you through using Alibaba Cloud PAI and the open‑source LLaMA Factory framework to fine‑tune a Llama‑3 8B model for Chinese dialogue and role‑playing, create a “Zhuge Liang” chatbot, evaluate its performance, and clean up resources.
This article presents a comprehensive knowledge map for developing large‑model applications, covering a four‑layer technical architecture, the full development lifecycle, core elements such as prompt engineering and model fine‑tuning, evaluation methods, and practical case studies, offering guidance for both enterprises and startups.
This article presents Fudan University's Knowledge Factory Lab research on domain large models, covering background, challenges, data selection, source‑enhanced tagging, capability improvements, self‑correction, collaborative workflows, and retrieval‑augmented generation for practical AI deployment.
Qwen2, Alibaba Cloud's new open‑source LLM series, offers five model sizes with GQA acceleration, and through PAI‑QuickStart developers can zero‑code fine‑tune, evaluate, and deploy these models using cloud resources, SDKs, and OpenAI‑compatible APIs.
This article systematically analyzes popular parameter‑efficient fine‑tuning (PEFT) techniques for large language models—including Adapter Tuning, Prefix Tuning, LoRA, QLoRA, AdaLoRA, and SoRA—detailing their principles, implementation code, experimental results on NLU tasks, and practical acceleration using the Unsloth library.
The guide explains why generic LLMs struggle with enterprise tasks and outlines two remedies—retrieval‑augmented generation and domain‑specific fine‑tuning—detailing dataset creation, training strategies (full‑parameter, LoRA, Q‑LoRA), validation methods, hardware benchmarks, and practical tips such as supervised fine‑tuning, 30% domain data, and a stepwise tuning pipeline.
In his Create 2024 AI Developer Conference keynote, Li Yanhong outlines Baidu’s latest large‑model series, AI‑native development platforms (AgentBuilder, AppBuilder, ModelBuilder), performance breakthroughs, real‑world case studies, and the strategic vision that makes AI development accessible to all developers and enterprises.
This article explains the motivation behind Self‑RAG, describes its core workflow—including conditional retrieval, enhanced generation, and self‑evaluation tokens—details the four evaluation metrics (Retrieve, IsRel, IsSup, IsUse), and provides a Python scoring example using log‑probabilities.
This article outlines a comprehensive knowledge map for building large‑model applications, detailing a four‑layer technical architecture, development lifecycle, core elements such as prompt engineering and fine‑tuning, evaluation methods, and real‑world case studies across various AI use cases.
Baidu’s Qianfan Large Model Platform provides a one‑stop enterprise solution with 54 pre‑installed models, advanced fine‑tuning, comprehensive evaluation metrics, and optimized deployment that cuts costs up to 90% and boosts throughput 3‑5×, enabling rapid, affordable AI application development.
This article explains the principle of LoRA (Low‑Rank Adaptation) for large language models, demonstrates how to decompose weight updates into low‑rank matrices, and provides a complete PyTorch implementation that fine‑tunes a small VGG‑19 network on a custom goldfish dataset.
This article explains how the open‑source AutoDev plugin builds an end‑to‑end AI‑assisted coding solution that fine‑tunes open LLMs, constructs a Unit Eval dataset, engineers prompts for unit‑test generation, and enforces quality through a unified write‑evaluate pipeline.
The article examines how enterprises can adopt domain‑specific large models by balancing demand‑side cost‑reduction needs with supply‑side mature training techniques, discusses team composition, fine‑tuning methods, data governance for unstructured data, and outlines Deepexi’s product ecosystem designed to improve efficiency, performance, and user experience.
The article examines how enterprises can adopt domain‑specific large models by addressing talent and cost challenges, outlining self‑supervised pre‑training, instruction fine‑tuning, data governance for unstructured data, dataset balance, model‑type selection, and integrated product solutions to achieve efficient, high‑performance AI deployments.
The article explores how enterprises can adopt domain‑specific large language models by addressing talent and cost challenges, outlining training pipelines, data governance for unstructured data, dataset balancing, fine‑tuning techniques, and a product ecosystem that lowers deployment barriers while optimizing performance and economics.
This article details how Zhihu's internal Bridge platform integrates large language models for business analysis, knowledge taxonomy, natural‑language‑to‑filter conversion, and ad‑hoc data queries, describing the workflow, technical hurdles, iterative improvements, and future directions.
The article reviews the rapid evolution of open‑source large language models, detailing Meta’s Llama 2 series and Tsinghua’s ChatGLM 2, their enhanced capabilities such as RLHF, larger context windows, safety‑usefulness trade‑offs, performance gains, download and fine‑tuning procedures, and how they increasingly rival proprietary models like GPT‑4.
This article explains the evolution from traditional machine learning to the current large‑model era, introduces LMOps concepts and key technologies, compares them with MLOps, and showcases Baidu Cloud's Qianfan platform as a practical solution for building, deploying, and managing large language models in industry.
The article explains how LoRA and its 4‑bit QLoRA extension dramatically reduce trainable parameters and GPU memory for fine‑tuning large language models, while GPTQ post‑training quantization compresses weights for cheap inference, and shows how KubeAI integrates these techniques into a one‑click workflow for 7 B, 13 B, and 33 B models from data upload to API deployment.
This article introduces the upgraded PAI‑Diffusion Chinese models, highlighting major improvements in image quality and style diversity, detailing lightweight fine‑tuning methods such as LoRA and Textual Inversion, showcasing controllable editing, scenario‑specific customization, and providing step‑by‑step usage instructions on popular platforms.
This article introduces a continuous semantic enhancement approach for neural machine translation that overcomes the limitations of discrete data‑augmentation techniques, details the neighbor risk minimization training objective, presents benchmark improvements on ACL‑2022 datasets, and describes practical deployment and fine‑tuning workflows in the Modu community.
The article details the evolution of the open‑source MOSS series—from OpenChat 001 to MOSS 003—covering data collection, fine‑tuning procedures, multilingual capabilities, plugin architecture, example code for inference, and upcoming releases, providing a comprehensive technical overview for AI practitioners.
This article introduces EasyNLP's Chinese text‑to‑image generation framework, explains the underlying Transformer‑VQGAN architecture, provides model specifications, showcases sample outputs, and offers step‑by‑step code and command‑line instructions for fine‑tuning and inference.
