import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
def extract_responses(text):
"""
Extracts and returns the responses from the text, excluding the parts
between and including the [INST] tags.
Args:
text (str): The input text containing responses and [INST] tags.
Returns:
str: The extracted responses.
"""
import re
# Split the text by [INST] tags and accumulate non-tag parts
parts = re.split(r'\[INST\].*?\[/INST\]', text, flags=re.DOTALL)
cleaned_text = "".join(parts)
# Return the cleaned and trimmed text
return cleaned_text.strip()
def generate_html():
return(
'''
Your Gradio App
''')
def generate_footer():
return(
'''
Your Gradio App
''')
model = AutoModelForCausalLM.from_pretrained(
"microsoft/phi-2",
torch_dtype=torch.float32,
device_map="cpu",
trust_remote_code=True
)
model.load_adapter('checkpoint-1100')
tokenizer = AutoTokenizer.from_pretrained('checkpoint-1100', trust_remote_code=True)
tokenizer.pad_token = tokenizer.eos_token