题意：怎样增加 max_new_tokens 的值

问题背景：

I'm facing this error while running my code:        运行代码时遇到如下错误：

ValueError: Input length of input_ids is 1495, but max_length is set to 20. This can lead to unexpected behavior. You should consider increasing max_length or, better yet, setting max_new_tokens.

ValueError：input_ids 的输入长度为 1495，但 max_length 被设置为 20。这可能导致意外的行为。你应该考虑增加 max_length 的值，或者更好的是，设置 max_new_tokens。

I wanted the code to generate the query instead it says about the max length issue as basically I am using 8 bit quantized llama using vector embedding to develop a RAG chat bot

我希望代码能够生成查询，但它却提示了最大长度问题。实际上，我正在使用8位量化的llama模型，并利用向量嵌入来开发一个基于检索的增强型生成（RAG）聊天机器人

import os
from langchain.document_loaders import PyPDFLoader
from langchain.text_splitter import CharacterTextSplitter
from transformers import BitsAndBytesConfig, AutoTokenizer
from langchain_huggingface import ChatHuggingFace, HuggingFacePipeline
from time import time
from langchain.embeddings import HuggingFaceEmbeddings
from langchain.vectorstores import Chroma
from langchain.chains import RetrievalQA

# Load PDF and split into chunks
def split_doc(file, chunk_size, chunk_overlap):
    text_splitter = CharacterTextSplitter(
        separator="\n\n",
        chunk_size=chunk_size,
        chunk_overlap=chunk_overlap,
        length_function=len
    )
    return text_splitter.split_documents(file)

loader = PyPDFLoader("/kaggle/input/report/report.pdf")
pages = loader.load()
docs = split_doc(pages, 700, 450)

# Configure and load the model
model_name = "NousResearch/Llama-2-7b-chat-hf"
bnb_config = BitsAndBytesConfig(load_in_8bit=True)
llm = HuggingFacePipeline.from_model_id(
    model_id=model_name,
    task="text-generation",
    model_kwargs={"trust_remote_code": True, "quantization_config": bnb_config}
)
chat_model = ChatHuggingFace(llm=llm)

# Set up embeddings and vector store
embedding_model_name = "sentence-transformers/all-mpnet-base-v2"
embedding_model_kwargs = {"device": "cuda"}
embeddings = HuggingFaceEmbeddings(model_name=embedding_model_name, model_kwargs=embedding_model_kwargs)
vectordb = Chroma.from_documents(documents=docs, embedding=embeddings, persist_directory="chroma_db")
retriever = vectordb.as_retriever()

# Set up the QA system
qa = RetrievalQA.from_chain_type(
    llm=llm,
    chain_type="stuff",
    retriever=retriever,
    verbose=True
)

# Define the testing function
def test_rag(qa, query):
    print(f"Query: {query}\n")
    result = qa.run(query)
    print("\nResult: ", result)

query = "What were the main topics in this "
test_rag(qa, query)# get stuck here

问题解决：

thanks it helped i added the following details: using the pipeline_kwargs in huggingface.py file i was able to find the variable i could use although using this method will render the quantization method a bit useless as you will consume more memory upon increasing the tokens

“谢谢，这很有帮助。我添加了以下细节：通过在huggingface.py文件中使用pipeline_kwargs，我能够找到我可以使用的变量。然而，使用这种方法会使量化方法变得有点无用，因为当你增加tokens时，会消耗更多的内存。

llm = HuggingFacePipeline.from_model_id(
    model_id=model_name,
    task="text-generation",
    model_kwargs={
        "trust_remote_code": True,
        "quantization_config": bnb_config,
        "use_auth_token": auth_token
    },
    pipeline_kwargs={"max_new_tokens": 8096}# this part is how i reconfigured the tokens

)