Microsoft Phi-2模型使用

!pip uninstall -y transformers && pip install git+https://github.com/huggingface/transformers
!pip install accelerate PyTorch DeepSpeed Flash-Attention

(mypy) [www@ecs-1c82 phi]$ cat test.py
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
#Set default device for torch
torch.set_default_device("cuda" if torch.cuda.is_available() else "cpu")

# Load Phi-2 model and tokenizer
tokenizer = AutoTokenizer.from_pretrained("microsoft/phi-2", trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained("microsoft/phi-2", device_map="auto", trust_remote_code=True)

# Ask for input prompt from the user
input_text =input("Enter your prompt : ")
# Tokenize input
inputs = tokenizer(input_text, return_tensors="pt",return_attention_mask=False)
# Generate output from the model
outputs = model.generate(**inputs, max_length=300,  pad_token_id=tokenizer.eos_token_id)
text = tokenizer.batch_decode(outputs)[0]

print(text)

import streamlit as st
from transformers import AutoTokenizer, AutoModelForCausalLM
import torch

# Load the Phi 2 model and tokenizer
tokenizer = AutoTokenizer.from_pretrained(
    "microsoft/phi-2",
    trust_remote_code=True
)

model = AutoModelForCausalLM.from_pretrained(
    "microsoft/phi-2",
    device_map="auto",
    trust_remote_code=True
)

# Streamlit UI
st.title("Microsoft Phi 2 Streamlit App")

# User input prompt
prompt = st.text_area("Enter your prompt:", """Write a story about Nasa""")

# Generate output based on user input
if st.button("Generate Output"):
    with torch.no_grad():
        token_ids = tokenizer.encode(prompt, add_special_tokens=False, return_tensors="pt")
        output_ids = model.generate(
            token_ids.to(model.device),
            max_new_tokens=512,
            do_sample=True,
            temperature=0.3
        )

    output = tokenizer.decode(output_ids[0][token_ids.size(1):])
    st.text("Generated Output:")
    st.write(output)