本地大模型LocalAI使用教程指南

## Set number of threads.
## Note: 优先选择物理内核数量。过多的 CPU 会明显降低性能。
THREADS=12
 
## Specify a different bind address (defaults to ":8080")
# ADDRESS=127.0.0.1:8080
 
## Default models context size
# CONTEXT_SIZE=512
#
## Define galleries.
## 要安装的模型将在 `/models/available` 中显示
GALLERIES=[{"name":"model-gallery", "url":"github:go-skynet/model-gallery/index.yaml"}, {"url": "github:go-skynet/model-gallery/huggingface.yaml","name":"huggingface"}]

name: lunademo
parameters:
  model: luna-ai-llama2-uncensored.Q5_K_M.gguf
  top_k: 80
  temperature: 0.2
  top_p: 0.7
context_size: 1024
threads: 12
backend: llama
roles:
  assistant: 'ASSISTANT:'
  system: 'SYSTEM:'
  user: 'USER:'
template:
  chat: lunademo-chat
  completion: lunademo-completion

USER: {{.Input}}
 
ASSISTANT:

Complete the following sentence: {{.Input}}

$ curl http://localhost:8080/v1/chat/completions -H "Content-Type: application/json" -d '{
     "model": "lunademo",
     "messages": [{"role": "user", "content": "How are you?"}],
     "temperature": 0.9 
   }'
{
   "created":1700993538,
   "object":"chat.completion",
   "id":"2fe33052-f4be-4724-8b53-fdade80b49de",
   "model":"lunademo",
   "choices":[
      {
         "index":0,
         "finish_reason":"stop",
         "message":{
            "role":"assistant",
            "content":"I'm doing well, thank you. How about yourself?"
         }
      }
   ],
   "usage":{
      "prompt_tokens":0,
      "completion_tokens":0,
      "total_tokens":0
   }
}

<dependency>
  <groupId>dev.langchain4j</groupId>
  <artifactId>langchain4j-local-ai</artifactId>
  <version>0.24.0</version>
</dependency>

ChatLanguageModel model = LocalAiChatModel.builder()
        .baseUrl("http://localhost:8080")
        .modelName("lunademo")
        .temperature(0.9)
        .build();
 
String answer = model.generate("How are you?");
System.out.println(answer);

private static void askQuestion(String question) {
    EmbeddingModel embeddingModel = new AllMiniLmL6V2EmbeddingModel();
 
    EmbeddingStore<TextSegment> embeddingStore = new InMemoryEmbeddingStore<>();
 
    //读取文件并将其分割成 500 块的片段
    Document springsteenDiscography = loadDocument(toPath("example-files/Bruce_Springsteen_discography.pdf"));
    Document springsteenSongList = loadDocument(toPath("example-files/List_of_songs_recorded_by_Bruce_Springsteen.pdf"));
    ArrayList<Document> documents = new ArrayList<>();
    documents.add(springsteenDiscography);
    documents.add(springsteenSongList);
 
    DocumentSplitter documentSplitter = DocumentSplitters.recursive(500, 0);
    List<TextSegment> documentSegments = documentSplitter.splitAll(documents);
 
    // 嵌入分段
    Response<List<Embedding>> embeddings = embeddingModel.embedAll(documentSegments);
    embeddingStore.addAll(embeddings.content(), documentSegments);
 
    //嵌入问题并查找相关片段
    Embedding queryEmbedding = embeddingModel.embed(question).content();
    List<EmbeddingMatch<TextSegment>> embeddingMatch = embeddingStore.findRelevant(queryEmbedding,1);
    System.out.println(embeddingMatch.get(0).score());
    System.out.println(embeddingMatch.get(0).embedded().text());
    System.out.println(embeddingMatch.get(0).embedded().metadata());
}

public static void main(String args) {
    askQuestion("on which album was \"adam raised a cain\" originally released?");
    askQuestion("what is the highest chart position of \"Greetings from Asbury Park, N.J.\" in the US?");
    askQuestion("what is the highest chart position of the album \"tracks\" in canada?");
    askQuestion("in which year was \"Highway Patrolman\" released?");
    askQuestion("who produced \"all or nothin' at all?\"");
}

List<Document> documents = loadDocuments(toPath("markdown-files"));
 
List<TextSegment> segments = new ArrayList<>();
for (Document document : documents) {
    String splittedDocument = document.text().split("\n");
    String dataOnly = Arrays.copyOfRange(splittedDocument, 2, splittedDocument.length);
    String header = splittedDocument[0] + "\n" + splittedDocument[1] + "\n";
 
    for (String splittedLine : dataOnly) {
        segments.add(TextSegment.from(header + splittedLine, document.metadata()));
    }
}