app.py

import gradio as gr
import requests
import json
from transformers import (
    GPT2Tokenizer,
    GPT2LMHeadModel,
    AutoTokenizer,
    AutoModelForCausalLM,
    AutoModelForSeq2SeqLM,
)
import torch

model_specs = [
#    {
#        "name": "distilgpt2",
#        "tokenizer": GPT2Tokenizer.from_pretrained("distilgpt2"),
#        "model": GPT2LMHeadModel.from_pretrained("distilgpt2"),
#    },
#    {
#        "name": "openai-gpt",
#        "tokenizer": AutoTokenizer.from_pretrained("openai-gpt"),
#        "model": AutoModelForCausalLM.from_pretrained("openai-gpt"),
#    },
    {
        "name": "OpenChatKit",
        "tokenizer": AutoTokenizer.from_pretrained("togethercomputer/GPT-NeoXT-Chat-Base-20B"),
        "model": AutoModelForCausalLM.from_pretrained(
            "togethercomputer/GPT-NeoXT-Chat-Base-20B", torch_dtype=torch.float16
        ).to("cuda:0"),
    },
#    {
#        "name": "flan-t5-xl",
#        "tokenizer": AutoTokenizer.from_pretrained("google/flan-t5-xl"),
#        "model": AutoModelForSeq2SeqLM.from_pretrained("google/flan-t5-xl"),
#    },
#    {
#        "name": "LLama",
#        "tokenizer": AutoTokenizer.from_pretrained("decapoda-research/llama-7b-hf"),
#        "model": AutoModelForCausalLM.from_pretrained("decapoda-research/llama-7b-hf"),
#    },
]

def generate_response(model_name, input_text, speak_output):
    selected_model = next(
        (spec for spec in model_specs if spec["name"] == model_name), None
    )
    if not selected_model:
        return "Invalid model selected."

    tokenizer = selected_model["tokenizer"]
    model = selected_model["model"]

    if model_name in ["flan-t5-xl"]:
        # For Seq2Seq models
        input_ids = tokenizer.encode(input_text, return_tensors="pt")
        output_ids = model.generate(input_ids)
        response = tokenizer.decode(output_ids[0], skip_special_tokens=True)
    else:
        # For CausalLM models
        input_ids = tokenizer.encode(input_text, return_tensors="pt")
        output_ids = model.generate(input_ids, max_length=input_ids.shape[1] + 50)
        response = tokenizer.decode(output_ids[0], skip_special_tokens=True)

    if speak_output:
        ELEVENLABS_API_URL = "https://api.elevenlabs.io/v1/text-to-speech/21m00Tcm4TlvDq8ikWAM"
        HEADERS = {
            "accept": "audio/mpeg",
            "xi-api-key": "9360e4509988dc28ff03a5d43cb6941b",
            "Content-Type": "application/json",
        }
        data = {
            "text": response,
            "voice_settings": {"stability": 0, "similarity_boost": 0},
        }
        response_audio = requests.post(ELEVENLABS_API_URL, headers=HEADERS, data=json.dumps(data))

        with open("response_audio.mp3", "wb") as f:
            f.write(response_audio.content)

    return response

if __name__ == "__main__":
    model_names = [spec["name"] for spec in model_specs]
    model_dropdown = gr.inputs.Dropdown(model_names, label="Select Model")
    text_input = gr.inputs.Textbox(lines=5, label="Input Text")
    speak_output = gr.inputs.Checkbox(label="Speak output with ElevenLabs API")

    gr.Interface(
        generate_response,
        inputs=[model_dropdown, text_input, speak_output],
        outputs="text",
        title="Chat Model Selection",
        description="Select a chat model and enter text to generate a response.",
    ).launch()