使用LlamaIndex，Qdrant，Ollama和FastAPI构建本地RAG api_llamaindex ollama

docker run -p 6333:6333 qdrant/qdrant

def ingest(self, embedder, llm):
    print("Indexing data...")
    # Loading
    documents = SimpleDirectoryReader(self.config["data_path"]).load_data()
 
    client = qdrant_client.QdrantClient(url=self.config["qdrant_url"])
    qdrant_vector_store = QdrantVectorStore(
        client=client, collection_name=self.config["collection_name"]
    )
    storage_context = StorageContext.from_defaults(vector_store=qdrant_vector_store)
    service_context = ServiceContext.from_defaults(
        llm=llm, embed_model=embedder, chunk_size=self.config["chunk_size"]
    )
 
    # Chunking + Embedding + Storing
    index = VectorStoreIndex.from_documents(
        documents, storage_context=storage_context, service_context=service_context
    )
    print(
        f"Data indexed successfully to Qdrant. Collection: {self.config['collection_name']}"
    )
    return index

def qdrant_index(self):
    client = qdrant_client.QdrantClient(url=self.config["qdrant_url"])
    qdrant_vector_store = QdrantVectorStore(
        client=client, collection_name=self.config['collection_name']
    )
    service_context = ServiceContext.from_defaults(
        llm=self.llm, embed_model=self.load_embedder(), chunk_size=self.config["chunk_size"]
    )
 
    index = VectorStoreIndex.from_vector_store(
        vector_store=qdrant_vector_store, service_context=service_context
    )
    return index

FROM models/zephyr-models/zephyr-7b-beta.Q5_K_M.gguf
PARAMETER num_ctx 3900
PARAMETER temperature 0.7
PARAMETER top_k 50
PARAMETER top_p 0.95
PARAMETER stop "<|system|>"
PARAMETER stop "<|user|>"
PARAMETER stop "<|assistant|>"
PARAMETER stop "</s>"
TEMPLATE """
<|system|>\n system
{{ .System }}</s>
<|user|>\n user
{{ .Prompt }}</s>
<|assistant|>\n 
"""
 
SYSTEM """
As a personal assistant for researchers, your task is to analyze the provided research papers and extract pertinent information on using the provided keywords. Summarize key findings, methodologies, and any notable insights. This assistant plays a crucial role in facilitating researchers' understanding of the current state of knowledge on the provided keywords.
Your Motive:
Give the learner "aha" moment on every Topic he needs to understand. You can do this with the art of explaining things.
Focus on Conciseness and Clarity: Ensure that the output is concise yet comprehensive. Focus on clarity and readability to provide researchers with easily digestible insights.
IMPORTANT:
If the user query cannot be answered using the provided context, do not improvise, you should only answer using the provided context from the research papers.
If the user asks something that does not exist within the provided context, Answer ONLY with: 'Sorry, the provided query is not clear enough for me to answer from the provided research papers'.
"""

ollama create research_assistant -f Modelfile

ollama run research_assistant

class Query(BaseModel):
    query: str
    similarity_top_k: Optional[int] = Field(default=1, ge=1, le=5)
 
 
class Response(BaseModel):
    search_result: str 
    source: str

llm = Ollama(model=config["llm_name"], url=config["llm_url"])
rag = RAG(config_file=config, llm=llm)
index = rag.qdrant_index()
 
 
app = FastAPI()

a = "You can only answer based on the provided context. If a response cannot be formed strictly using the context, politely say you don’t have knowledge about that topic"
 
@app.post("/api/search", response_model=Response, status_code=200)
def search(query: Query):
 
    query_engine = index.as_query_engine(similarity_top_k=query.similarity_top_k, output=Response, response_mode="tree_summarize", verbose=True)
    response = query_engine.query(query.query + a)
    response_object = Response(
        search_result=str(response).strip(), source=[response.metadata[k]["file_path"] for k in response.metadata.keys()][0]
    )
    return response_object

data_path: "data/"
llm_url: "http://localhost:11434"
llm_name: "research_assistant"
embedding_model: "sentence-transformers/all-mpnet-base-v2"
qdrant_url: "http://localhost:6333"
collection_name: "researchpapers"
chunk_size: 1024

{
  "query": "How can robots imitate human actions?",
  "similarity_top_k": 3
}