app.py

import gradio as gr
from transformers import AutoTokenizer, AutoModelForCausalLM

# Load the model and tokenizer
model_name = "migueldeguzmandev/GPT2XL_RLLMv3-Assist-v10"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(model_name)

# Set the pad token ID to the EOS token ID
model.config.pad_token_id = model.config.eos_token_id

# Define the inference function
def generate_response(input_text, temperature):
    # Tokenize the input text
    inputs = tokenizer(input_text, return_tensors="pt")
    input_ids = inputs["input_ids"]
    attention_mask = inputs["attention_mask"]

    # Generate the model's response
    output = model.generate(
        input_ids,
        attention_mask=attention_mask,
        max_length=300,
        num_return_sequences=1,
        temperature=temperature,
        no_repeat_ngram_size=2,
        top_k=50,
        top_p=0.95,
        do_sample=True,  # Set do_sample to True when using temperature
    )

    # Decode the generated response
    response = tokenizer.decode(output[0], skip_special_tokens=True)
    return response.replace(input_text, "").strip()


# Create the Gradio interface
interface = gr.Interface(
    fn=generate_response,
    inputs=[
        gr.Textbox(label="User Input"),
        gr.Slider(minimum=0.00000000000000000000001, maximum=1.0, value=0.7, step=0.1, label="Temperature"),
    ],
    outputs=gr.Textbox(label="Model Response"),
    title="Hello, I'm Assistant AI!",
    description=(
        """
        Unfortunately, Jailbreak attacks destroyed this prototype. All of the almost zero temperature attacks can be found <a href=https://whimsical.com/layer10-q-and-a-EiiYQfKCHivyX3t9t84ukE>here</a>.
        """
    ),
)

# Launch the interface without the share option
interface.launch()