|
import streamlit as st |
|
import transformers |
|
from transformers import LLaMATokenizer |
|
import torch |
|
model = "PY007/TinyLlama-1.1B-Chat-v0.1" |
|
tokenizer = transformers.LLaMATokenizer.from_pretrained(model) |
|
pipeline = transformers.pipeline("text-generation",model=model,torch_dtype=torch.bfloat16) |
|
|
|
prompt = "What is 653+2343?" |
|
formatted_prompt = ( |
|
f"### Human: {prompt}### Assistant:" |
|
) |
|
|
|
|
|
sequences = pipeline(formatted_prompt,do_sample=True,top_k=50,top_p = 0.7,num_return_sequences=1,repetition_penalty=1.1,max_new_tokens=500) |
|
for seq in sequences: |
|
st.write(f"Result: {seq['generated_text']}") |
|
|
|
|