Wiki.Quizz.vn

User Tools

Site Tools

Trace:
ai:ai-open-sources

Differences

This shows you the differences between two versions of the page.

Link to this comparison view

Next revision		Previous revision
ai:ai-open-sources [2026/01/22 09:30] – created phong2018ai:ai-open-sources [2026/02/17 20:38] (current) – phong2018
Line 1: Line 1:
 ====== AI CMS / Framework / Model ====== ====== AI CMS / Framework / Model ======
  
-===== Tổng quan =====+===== Overview =====
  
-**Có.** Trong AI cũng tồn tại khái niệm tương đương **CMS / Framework / Library** giống lập trình truyền thống, nhưng được chia thành **nhiều tầng đóng gói** khác nhau.+Yes. In AI, there are equivalent concepts to CMS / Framework / Library similar to traditional programming, but they are divided into multiple packaging layers.
  
-**Mục tiêu trang này:** Giải thích AI theo đúng tư duy **OpenCart / Laravel / SDK** để dev hiểu ngay. +Goal of this page: Explain AI using the mindset of OpenCart / Laravel / SDK so developers can understand it immediately. 
-**What it is:** Trong AI, cũng tồn tại khái niệm tương đương **CMS / Framework / Library** giống lập trình truyền thống.+What it is: In AI, there are also equivalent concepts to CMS / Framework / Library similar to traditional programming.
  
-**Key idea:** AI không phải một khối duy nhất, mà được chia thành **nhiều tầng đóng gói**, từ "lấy về dùng ngay" đến "tự build từ lõi".+Key idea: AI is not a single block, but is divided into multiple packaging layers, from “ready to use” to “build from the core.”
  
---- +===== 1) Direct Mapping: Traditional Programming ↔ AI =====
- +
-===== 1) Mapping trực tiếp: Lập trình truyền thống ↔ AI =====+
 ^ Traditional Programming ^ AI Equivalent ^ Meaning ^ ^ Traditional Programming ^ AI Equivalent ^ Meaning ^
-| CMS (OpenCart, WordPress) | AI CMS / Prebuilt AI App | Dùng ngay |+| CMS (OpenCart, WordPress) | AI CMS / Prebuilt AI App | Ready to use |
 | Framework (Laravel, Django) | AI Framework / LLM Framework | Build AI app | | Framework (Laravel, Django) | AI Framework / LLM Framework | Build AI app |
-| Library / SDK | Model / Inference Library | Ghép từng phần | +| Library / SDK | Model / Inference Library | Assemble components | 
-| Custom low-level code | Train model from scratch | Đắt & hiếm | +| Custom low-level code | Train model from scratch | Expensive & rare |
- +
----+
  
-===== 2) Tầng AI CMS – Dùng ngay (No-code / Low-code) ===== +===== 2) AI CMS Layer – Ready to Use (No-code / Low-code) ===== 
-**What it is:** Ứng dụng AI đã đóng gói sẵn UI + workflow + model integration.+What it is: Prepackaged AI applications with built-in UI + workflow + model integration.
  
-**Equivalent:** OpenCart / WordPress+Equivalent: OpenCart / WordPress
  
 ==== Examples ==== ==== Examples ====
  
-* [[[https://anythingllm.com|AnythingLLM](https://anythingllm.com|AnythingLLM)]]+[[https://anythingllm.com|AnythingLLM 
 +]]
  
-* [[[https://flowiseai.com|Flowise](https://flowiseai.com|Flowise)]]+[[https://flowiseai.com|Flowise 
 +]]
  
-* [[[https://openwebui.com|Open](https://openwebui.com|Open) WebUI]]+[[https://openwebui.com|Open 
 + WebUI]]
  
-* [[[https://botpress.com|Botpress](https://botpress.com|Botpress)]]+[[https://botpress.com|Botpress 
 +]]
  
-* [[[https://rasa.com|Rasa](https://rasa.com|Rasa)]]+[[https://rasa.com|Rasa 
 +]]
  
 ==== Key Characteristics ==== ==== Key Characteristics ====
  
-* Có sẵn UI +Built-in UI 
-* Workflow dựng bằng form / node + 
-* Config bằng YAML / JSON +Workflow built using forms / nodes 
-* Gắn LLM là chạy+ 
 +Configured via YAML / JSON 
 + 
 +Plug in an LLM and run
  
 ==== Use When ==== ==== Use When ====
  
-* Internal chatbot +Internal chatbot 
-* Knowledge assistant + 
-* PoC nhanh +Knowledge assistant 
-* Team không có ML skill+ 
 +Fast PoC 
 + 
 +Team without ML skills
  
 ==== Limitations ==== ==== Limitations ====
  
-* Khó custom sâu +Hard to deeply customize
-* Performance không tối ưu +
-* Governance yếu nếu không chỉnh+
  
-👉 **Đây là AI CMS đúng nghĩa.**+Performance not optimized
  
----+Weak governance if not configured properly
  
-===== 3) Tầng AI Framework – Giống Laravel / Django ===== +👉 This is true AI CMS.
-**What it is:** Framework để lập trình ứng dụng AI bằng code.+
  
-**Equivalent:** Laravel / Django+===== 3) AI Framework Layer – Like Laravel / Django ===== 
 +What it is: Frameworks for building AI applications using code. 
 + 
 +Equivalent: Laravel / Django
  
 ==== Popular Frameworks ==== ==== Popular Frameworks ====
  
-* [[[https://www.langchain.com|LangChain](https://www.langchain.com|LangChain)]] ⭐+[[https://www.langchain.com|LangChain 
 +]] ⭐
  
-* [[[https://www.llamaindex.ai|LlamaIndex](https://www.llamaindex.ai|LlamaIndex)]]+[[https://www.llamaindex.ai|LlamaIndex 
 +]]
  
-* [[[https://haystack.deepset.ai|Haystack](https://haystack.deepset.ai|Haystack)]]+[[https://haystack.deepset.ai|Haystack 
 +]]
  
-* [[[https://learn.microsoft.com/semantic-kernel|Semantic](https://learn.microsoft.com/semantic-kernel|Semantic) Kernel]]+[[https://learn.microsoft.com/semantic-kernel|Semantic 
 + Kernel]]
  
 ==== What They Do ==== ==== What They Do ====
  
-* Prompt management +Prompt management 
-* RAG (Retrieval Augmented Generation) + 
-* Tool calling +RAG (Retrieval Augmented Generation) 
-* Memory & agent flow+ 
 +Tool calling 
 + 
 +Memory & agent flow
  
 ==== Mapping ==== ==== Mapping ====
  
-* Laravel ≈ LangChain +Laravel ≈ LangChain 
-* Django ≈ LlamaIndex+ 
 +Django ≈ LlamaIndex
  
 ==== Use When ==== ==== Use When ====
  
-* Build AI product thực sự +Building a real AI product
-* Có backend team +
-* Cần kiểm soát flow & test+
  
----+Have a backend team
  
-===== 4) Tầng Pretrained Model – Model train sẵn ===== +Need flow control & testing
-**What it is:** Model AI đã được train sẵn trên dữ liệu lớn.+
  
-**Equivalent:** Database engine / Search engine+===== 4) Pretrained Model Layer – Pretrained Models ===== 
 +What it is: AI models trained on large datasets. 
 + 
 +Equivalent: Database engine / Search engine
  
 ==== Examples ==== ==== Examples ====
  
-* [[[https://ai.meta.com/llama|LLaMA](https://ai.meta.com/llama|LLaMA) / LLaMA 3]]+[[https://ai.meta.com/llama|LLaMA 
 + / LLaMA 3]]
  
-* [[[https://mistral.ai|Mistral](https://mistral.ai|Mistral) / Mixtral]]+[[https://mistral.ai|Mistral 
 + / Mixtral]]
  
-* [[[https://huggingface.co/Qwen|Qwen](https://huggingface.co/Qwen|Qwen)]]+[[https://huggingface.co/Qwen|Qwen 
 +]]
  
-* [[[https://deepseek.com|DeepSeek](https://deepseek.com|DeepSeek)]]+[[https://deepseek.com|DeepSeek 
 +]]
  
-* [[[https://ai.google.dev/gemma|Gemma](https://ai.google.dev/gemma|Gemma)]]+[[https://ai.google.dev/gemma|Gemma 
 +]]
  
 ==== Characteristics ==== ==== Characteristics ====
  
-* Không có UI +No UI
-* Không có workflow +
-* Có thể self-host +
-* Có thể fine-tune+
  
-🚫 **Không phải CMS** – chỉ là "bộ não".+No workflow
  
----+Can be self-hosted
  
-===== 5) Tầng AI Inference Engine ===== +Can be fine-tuned
-**What it is:** Runtime để chạy model hiệu quả.+
  
-**Equivalent:** JVM / PHP runtime / DB engine core+🚫 Not a CMS – just the “brain.” 
 + 
 +===== 5) AI Inference Engine Layer ===== 
 +What it is: Runtime for executing models efficiently. 
 + 
 +Equivalent: JVM / PHP runtime / DB engine core
  
 ==== Examples ==== ==== Examples ====
  
-* [[[https://github.com/vllm-project/vllm|vLLM](https://github.com/vllm-project/vllm|vLLM)]]+[[https://github.com/vllm-project/vllm|vLLM 
 +]]
  
-* [[[https://github.com/ggerganov/llama.cpp|llama.cpp](https://github.com/ggerganov/llama.cpp|llama.cpp)]]+[[https://github.com/ggerganov/llama.cpp|llama.cpp 
 +]]
  
-* [[[https://developer.nvidia.com/tensorrt-llm|TensorRT-LLM](https://developer.nvidia.com/tensorrt-llm|TensorRT-LLM)]]+[[https://developer.nvidia.com/tensorrt-llm|TensorRT-LLM 
 +]]
  
-* [[[https://github.com/huggingface/text-generation-inference|TGI](https://github.com/huggingface/text-generation-inference|TGI)]]+[[https://github.com/huggingface/text-generation-inference|TGI 
 +]]
  
 ==== Use When ==== ==== Use When ====
  
-* High throughput +High throughput
-* Tối ưu cost +
-* On-prem / air-gapped+
  
----+Cost optimization
  
-===== 6) Tầng Managed AI Platform (AI SaaS) ===== +On-prem / air-gapped
-**What it is:** Nền tảng AI được vận hành hoàn toàn bởi vendor.+
  
-**Equivalent:** Shopify / Salesforce+===== 6) Managed AI Platform (AI SaaS) ===== 
 +What it is: AI platforms fully operated by vendors. 
 + 
 +Equivalent: Shopify / Salesforce
  
 ==== Examples ==== ==== Examples ====
  
-* [[[https://openai.com|OpenAI](https://openai.com|OpenAI) / Azure OpenAI]]+[[https://openai.com|OpenAI 
 + / Azure OpenAI]]
  
-* [[[https://aws.amazon.com/bedrock|AWS](https://aws.amazon.com/bedrock|AWS) Bedrock]]+[[https://aws.amazon.com/bedrock|AWS 
 + Bedrock]]
  
-* [[[https://cloud.google.com/vertex-ai|Google](https://cloud.google.com/vertex-ai|Google) Vertex AI]]+[[https://cloud.google.com/vertex-ai|Google 
 + Vertex AI]]
  
 ==== Pros / Cons ==== ==== Pros / Cons ====
  
-* ✅ Nhanh, không lo infra +✅ Fast, no infrastructure management
-* ❌ Chi phí cao, lock-in+
  
----+❌ Higher cost, vendor lock-in
  
-===== 7) Sơ đồ tổng thể các tầng ===== +===== 7) Overall Layer Diagram =====
 <code> <code>
 AI CMS (No-code) AI CMS (No-code)
-│  Flowise / AnythingLLM+│ Flowise / AnythingLLM
 │ │
 AI Framework (Code) AI Framework (Code)
-│  LangChain / LlamaIndex+│ LangChain / LlamaIndex
 │ │
 Pretrained Model Pretrained Model
-│  LLaMA / Mistral+│ LLaMA / Mistral
 │ │
 Inference Engine Inference Engine
-│  vLLM / llama.cpp+│ vLLM / llama.cpp
 │ │
 Infrastructure Infrastructure
-│  Cloud / On-prem GPU </code>+│ Cloud / On-prem GPU </code> 
 + 
 +===== 8) What to Choose in Practice? ===== 
 + 
 +"Like OpenCart, download and run" → AnythingLLM / Flowise 
 + 
 +"Like Laravel" → LangChain / LlamaIndex 
 + 
 +"Enterprise, data control" → Open-source model + vLLM + RAG 
 + 
 +====== AI Service Classification – Company & Use Case ====== 
 + 
 +===== Goal ===== 
 +This page classifies popular AI services today using a senior developer mindset (CMS / Framework / Runtime), including: 
 + 
 +Company behind it 
 + 
 +Real-world use case 
 + 
 +Position in system architecture 
 + 
 +===== Layer Overview ===== 
 +AI is not a single block, but consists of 6 layers, from “ready to use” to “infrastructure core.” 
 + 
 +<code> AI App / Tool ↓ AI Platform (API) ↓ AI Framework ↓ Pretrained Model ↓ Inference Engine ↓ Infrastructure (GPU) </code> 
 + 
 +===== 1) AI App / AI Tool (Ready to Use) ===== 
 +Equivalent: CMS (WordPress / OpenCart) 
 + 
 +^ Service ^ Company ^ Main Use ^ 
 +| ChatGPT | OpenAI | General assistant, Q&A, coding, spec writing | 
 +| Claude Chat | Anthropic | Long document analysis, reasoning | 
 +| GitHub Copilot | GitHub / Microsoft | Code completion, pair programming | 
 +| AnythingLLM | Mintplex Labs | Internal document chatbot | 
 +| Flowise | FlowiseAI (OSS) | Rapid AI workflow building | 
 +| Open WebUI | Open-source | Chat UI for self-hosted models | 
 +| Botpress | Botpress Inc. | Customer service chatbot | 
 +| Rasa | Rasa Technologies | Conversation engine | 
 + 
 +Use when: Fast PoC, internal use, small team 
 + 
 +===== 2) AI Platform / AI SaaS ===== 
 +Equivalent: Shopify / Firebase 
 + 
 +^ Platform ^ Company ^ Main Use ^ 
 +| OpenAI API | OpenAI | LLM API: chat, embedding, tool calling | 
 +| Azure OpenAI | Microsoft | OpenAI + enterprise security | 
 +| AWS Bedrock | Amazon | Multi-model AI for enterprise | 
 +| Google Vertex AI | Google | End-to-end AI platform | 
 +| IBM watsonx | IBM | AI + data governance | 
 + 
 +Use when: Calling AI via API, no GPU management 
 + 
 +===== 3) AI Framework (Build AI Logic) ===== 
 +Equivalent: Laravel / Django 
 + 
 +^ Framework ^ Company / Organization ^ Main Use ^ 
 +| LangChain | LangChain Inc. | Orchestrate prompt, agent, tools | 
 +| LlamaIndex | LlamaIndex Inc. | RAG (document → answer) | 
 +| Haystack | deepset | Search + QA pipeline | 
 +| Semantic Kernel | Microsoft | Enterprise AI orchestration | 
 +| CrewAI | CrewAI Inc. | Multi-agent workflow | 
 +| AutoGen | Microsoft Research | Agent collaboration | 
 + 
 +Use when: Building real AI features, need testing & CI/CD 
 + 
 +===== 4) Pretrained Model (AI Brain) ===== 
 +Equivalent: Database / Search Engine 
 + 
 +^ Model ^ Company ^ Main Use ^ 
 +| GPT-4 / GPT-5 | OpenAI | Reasoning, general-purpose | 
 +| Claude 3 / 4 | Anthropic | Long context, safety | 
 +| Gemini | Google | Multimodal | 
 +| LLaMA 3 | Meta | Open-source, self-host | 
 +| Mistral / Mixtral | Mistral AI | Lightweight, fast, low cost | 
 +| Qwen | Alibaba | Multilingual | 
 +| DeepSeek | DeepSeek AI | Reasoning, open | 
 + 
 +Note: Models have no UI and no workflow. 
 + 
 +===== 5) Inference Engine (Runtime) ===== 
 +Equivalent: JVM / PHP-FPM 
 + 
 +^ Engine ^ Company / Organization ^ Main Use ^ 
 +| vLLM | UC Berkeley | High-throughput LLM serving | 
 +| llama.cpp | Open-source | Run LLM on CPU / edge | 
 +| TensorRT-LLM | NVIDIA | GPU optimization, low latency | 
 +| TGI | Hugging Face | Production LLM endpoint | 
 +| ONNX Runtime | Microsoft | Cross-platform inference | 
 + 
 +Use when: Self-host models, high traffic, cost optimization
  
----+===== 6) Infrastructure (GPU / Cloud) ===== 
 +Equivalent: Server / Datacenter
  
-===== 8) Chọn gì trong thực tế? =====+^ Infra ^ Company ^ Main Use ^ 
 +| AWS GPU | Amazon | AI cloud | 
 +| Azure GPU | Microsoft | AI cloud | 
 +| GCP GPU | Google | AI cloud | 
 +| NVIDIA A100 / H100 | NVIDIA | Large-scale training / inference | 
 +| On-prem GPU | Enterprise | Full privacy |
  
-* "Giống OpenCart, kéo về chạy" → **AnythingLLM / Flowise** +===== Real Architecture Example ===== <code> 
-* "Giống Laravel" → **LangChain / LlamaIndex** +React / Vue 
-* "Enterprise, kiểm soát data" → **Open-source model + vLLM + RAG**+↓ 
 +Laravel / Go API 
 +↓ 
 +LangChain / LlamaIndex 
 +↓ 
 +OpenAI API or LLaMA 
 +↓ 
 +(vLLM if self-host) </code>
  
----+===== Important Notes ===== 
 +⚠️ AI is not deterministic like traditional code.
  
-===== 9) Sự thật quan trọng ===== +Always require:
-⚠️ **AI ≠ CMS truyền thống**+
  
-* AI không deterministic +Guardrails
-* Output không đảm bảo đúng +
-* Luôn cần guardrail + evaluation+
  
-👉 AI CMS **không bao giờ là "cài xong là chạy"**, luôn cần con người kiểm soát.+Evaluation
  
 +Human review
ai/ai-open-sources.1769074234.txt.gz · Last modified: by phong2018