manuelcozar55's picture
Update app.py
7413e10 verified
raw
history blame
2.12 kB
from huggingface_hub import InferenceClient
import gradio as gr
import json
import PyPDF2
client = InferenceClient(
"mistralai/Mistral-7B-Instruct-v0.3"
)
def format_prompt(mode, message, instructions, history):
prompt = f"<s>[MODE] {mode} [/MODE] "
for user_prompt, bot_response in history:
prompt += f"[INST] {user_prompt} [/INST] {bot_response}</s> "
prompt += f"[INST] {message} {instructions} [/INST]"
return prompt
def process_input(file, file_type):
if file_type == 'pdf':
reader = PyPDF2.PdfFileReader(file.name)
text = ""
for page in range(reader.numPages):
text += reader.getPage(page).extractText()
return text
elif file_type == 'json':
data = json.load(file)
return json.dumps(data, indent=2)
return file
def generate(mode, file, file_type, instructions, history, temperature=0.9, max_new_tokens=256, top_p=0.95, repetition_penalty=1.0):
temperature = max(float(temperature), 1e-2)
top_p = float(top_p)
text_input = process_input(file, file_type)
formatted_prompt = format_prompt(mode, text_input, instructions, history)
generate_kwargs = dict(
temperature=temperature,
max_new_tokens=max_new_tokens,
top_p=top_p,
repetition_penalty=repetition_penalty,
do_sample=True,
seed=42,
)
stream = client.text_generation(formatted_prompt, **generate_kwargs, stream=True, details=True, return_full_text=False)
output = ""
for response in stream:
output += response.token.text
yield output
return output
gr.ChatInterface(
fn=generate,
inputs=[
gr.Dropdown(label="Mode", choices=["translation", "summary", "explanation"], value="translation"),
gr.File(label="Input File", type="file"),
gr.Radio(label="File Type", choices=["pdf", "json"], value="pdf"),
gr.Textbox(label="Additional Instructions", placeholder="Enter any additional instructions here"),
gr.Chatbot()
],
outputs=gr.Chatbot(),
title="Mistral 7B v0.3"
).launch(show_api=False)