app.py

from smolagents import CodeAgent,HfApiModel,tool
import torch
import yaml
from tools.final_answer import FinalAnswerTool
import numpy as np
import random
from scipy.signal import square, sawtooth

from Gradio_UI import GradioUI

@tool
def generate_random_sound(duration: float = 1.0) -> torch.Tensor:
    """Generates a random sound with varying frequency and waveform and returns a torch tensor.

    Args:
        duration: Length of the sound in seconds (default: 1.0)
    """
    # It seems the playback is about 3x slower, so we'll adjust the duration
    adjusted_duration = duration / 3
    
    # Sample rate
    sample_rate = 44100
    num_samples = int(sample_rate * adjusted_duration)
    t = np.linspace(0, adjusted_duration, num_samples, endpoint=False)

    # Random frequency between 100 and 10000 Hz
    frequency = random.uniform(100, 10000)

    # Random waveform selection
    waveform = random.choice(['sine', 'square', 'sawtooth', 'triangle'])

    if waveform == 'sine':
        signal = np.sin(2 * np.pi * frequency * t)
    elif waveform == 'square':
        signal = square(2 * np.pi * frequency * t)
    elif waveform == 'sawtooth':
        signal = sawtooth(2 * np.pi * frequency * t)
    else:  # triangle
        signal = sawtooth(2 * np.pi * frequency * t, width=0.5)  # Triangle is a symmetric sawtooth

    # Apply fade out
    fade = np.exp(-3 * t / adjusted_duration)  # Normalize fade over duration
    signal = signal * fade

    # Normalize to prevent clipping
    signal = signal / np.max(np.abs(signal))

    # Convert to torch tensor
    tensor_signal = torch.from_numpy(signal.astype(np.float32))
    
    return tensor_signal

final_answer = FinalAnswerTool()

model_id='https://pflgm2locj2t89co.us-east-1.aws.endpoints.huggingface.cloud' 

model = HfApiModel(
max_tokens=2096,
temperature=0.5,
model_id=model_id,
custom_role_conversions=None,
)

with open("prompts.yaml", 'r') as stream:
    prompt_templates = yaml.safe_load(stream)
    
agent = CodeAgent(
    model=model,
    tools=[final_answer, generate_random_sound],
    max_steps=6,
    verbosity_level=1,
    grammar=None,
    planning_interval=None,
    name=None,
    description=None,
    prompt_templates=prompt_templates
)


GradioUI(agent).launch()