The AIGC Era

The main problems being solved right now are text-to-image generation and intelligent Q&A chatbots

Deploying Stable Diffusion for Text-to-Image

1. Servers are typically AWS, Runpod, Tencent, or Alibaba
2. Framework: Stable Diffusion
3. Core: text-to-image models
4. Tuning LoRA: weight 0.3-0.8
5. ControlNet preprocessing <TTP’s RAG>
6. The output is an API for the front end to call

Intelligent Q&A: LLM <---> SLM Configuration

1. Problems faced

2. Ethics

3. Hallucination

4. Timeliness

5. Model selection: ChatGPT, Ollama, Meta, others

6. Basis for reference: HuggingFace popularity and QAnything evaluations

7. Commercial solution

8. User input

9. Model library

10. Knowledge base <---> online/offline

11. Where the three meet, hand it over to RAG

12. RAG (Retrieval-Augmented Generation) ---> GraphRAG, deployed with Langchain

13. User input content

14. Embedding model

15. Rerank model <context reranking, reduces the large model’s memory consumption>

16. RAG processing

17. Knowledge base embedding

18. LLM processes and returns the result