Notes

!pip install transformers
!pip install accelerate
!pip install peft
!pip install SentencePiece
!pip install langchain
!pip install bitsandbytes



import transformers
from peft import PeftModel
from transformers import pipeline
import torch
from transformers import LlamaTokenizer, LlamaForCausalLM, AutoModel, AutoModelForCausalLM, LlamaForCausalLM

from huggingface_hub import notebook_login

notebook_login()

def load_model_and_tokenizer(base_model):
model = LlamaForCausalLM.from_pretrained(
base_model,
load_in_8bit = True,
torch_dtype=torch.float16,
use_auth_token='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'
)
# Load the tokenizer
tokenizer = LlamaTokenizer.from_pretrained(base_model)
# Set tokenizer parameters
tokenizer.pad_token_id = 0 # unk. we want this to be different from the eos token
tokenizer.padding_side = "left"
return model, tokenizer

BASE_MODEL = "meta-llama/Llama-2-7b-chat-hf"
#DEVICE = "cuda" if torch.cuda.is_available() else "cpu"

model, tokenizer = load_model_and_tokenizer(BASE_MODEL)

pipe = pipeline("text-generation",
model=model,
tokenizer= tokenizer,
torch_dtype=torch.bfloat16,
device_map="auto",
max_new_tokens = 512,
do_sample=True,
top_k=30,
num_return_sequences=1,
eos_token_id=tokenizer.eos_token_id
)


from langchain import HuggingFacePipeline
from langchain import PromptTemplate, LLMChain

llm = HuggingFacePipeline(pipeline = pipe, model_kwargs = {'temperature':1})

template = """

You are an AI assistant whose main function is to extract question and answer from a script. Meaning you would need to identify what is the question(query of customer) and resolution(answer of the same query).
now your main task is to frame the question and respective resolution as first person, not second and third person. After finding the query and reolution, you need to show it like
query : question
resolution : answer

[Additional Guidance]
Be sure to accurately identify and extract the main issue as query and resolution as answer. If the conversation contains multiple issues or resolutions, focus on most prominent issue. Ensure that the extracted text is clear and concise.

Question: {question}
Answer:

"""


prompt = PromptTemplate(template=template, input_variables=["question"])

llm_chain = LLMChain(prompt=prompt, llm=llm)

question = """Ruth asked if Manny available. He was not available. Ruth asked if she can talk to someone who is superior to Manny as she received call regarding the fire alarm issue. Ruth was connected to Manny’s superior. Ruth asked him to share the photo of the device that needs battery replacement and shared it to the number 6463976555"""


x = llm_chain.run(question)
print(x)