app.py

import gradio as gr
import pandas as pd
import numpy as np
import plotly.express as px

# Pricing data
aws_instances = {
    "g4dn.xlarge": {"vcpus": 4, "memory": 16, "gpu": "1x NVIDIA T4", "hourly_rate": 0.526, "gpu_memory": "16GB"},
    "g4dn.2xlarge": {"vcpus": 8, "memory": 32, "gpu": "1x NVIDIA T4", "hourly_rate": 0.752, "gpu_memory": "16GB"},
    "g5.xlarge": {"vcpus": 4, "memory": 16, "gpu": "1x NVIDIA A10G", "hourly_rate": 0.65, "gpu_memory": "24GB"},
    "g5.2xlarge": {"vcpus": 8, "memory": 32, "gpu": "1x NVIDIA A10G", "hourly_rate": 1.006, "gpu_memory": "24GB"},
    "p3.2xlarge": {"vcpus": 8, "memory": 61, "gpu": "1x NVIDIA V100", "hourly_rate": 3.06, "gpu_memory": "16GB"},
    "p4d.24xlarge": {"vcpus": 96, "memory": 1152, "gpu": "8x NVIDIA A100", "hourly_rate": 32.77, "gpu_memory": "8x40GB"}
}

gcp_instances = {
    "a2-highgpu-1g": {"vcpus": 12, "memory": 85, "gpu": "1x NVIDIA A100", "hourly_rate": 1.46, "gpu_memory": "40GB"},
    "a2-highgpu-2g": {"vcpus": 24, "memory": 170, "gpu": "2x NVIDIA A100", "hourly_rate": 2.93, "gpu_memory": "2x40GB"},
    "a2-highgpu-4g": {"vcpus": 48, "memory": 340, "gpu": "4x NVIDIA A100", "hourly_rate": 5.86, "gpu_memory": "4x40GB"},
    "n1-standard-4-t4": {"vcpus": 4, "memory": 15, "gpu": "1x NVIDIA T4", "hourly_rate": 0.49, "gpu_memory": "16GB"},
    "n1-standard-8-t4": {"vcpus": 8, "memory": 30, "gpu": "1x NVIDIA T4", "hourly_rate": 0.69, "gpu_memory": "16GB"},
    "g2-standard-4": {"vcpus": 4, "memory": 16, "gpu": "1x NVIDIA L4", "hourly_rate": 0.59, "gpu_memory": "24GB"}
}

api_pricing = {
    "OpenAI": {
        "GPT-3.5-Turbo": {"input_per_1M": 0.5, "output_per_1M": 1.5, "token_context": 16385},
        "GPT-4o": {"input_per_1M": 5.0, "output_per_1M": 15.0, "token_context": 32768},
        "GPT-4o-mini": {"input_per_1M": 2.5, "output_per_1M": 7.5, "token_context": 32768},
    },
    "TogetherAI": {
        "Llama-3-8B": {"input_per_1M": 0.15, "output_per_1M": 0.15, "token_context": 8192},
        "Llama-3-70B": {"input_per_1M": 0.9, "output_per_1M": 0.9, "token_context": 8192},
        "Llama-2-13B": {"input_per_1M": 0.6, "output_per_1M": 0.6, "token_context": 4096},
        "Llama-2-70B": {"input_per_1M": 2.5, "output_per_1M": 2.5, "token_context": 4096},
        "DeepSeek-Coder-33B": {"input_per_1M": 2.0, "output_per_1M": 2.0, "token_context": 16384},
    },
    "Anthropic": {
        "Claude-3-Opus": {"input_per_1M": 15.0, "output_per_1M": 75.0, "token_context": 200000},
        "Claude-3-Sonnet": {"input_per_1M": 3.0, "output_per_1M": 15.0, "token_context": 200000},
        "Claude-3-Haiku": {"input_per_1M": 0.25, "output_per_1M": 1.25, "token_context": 200000},
    }
}

model_sizes = {
    "Small (7B parameters)": {"memory_required": 14},
    "Medium (13B parameters)": {"memory_required": 26},
    "Large (70B parameters)": {"memory_required": 140},
    "XL (180B parameters)": {"memory_required": 360},
}

def calculate_aws_cost(instance, hours, storage, reserved=False, spot=False, years=1):
    data = aws_instances[instance]
    rate = data['hourly_rate']
    if spot:
        rate *= 0.3
    elif reserved:
        factors = {1: 0.6, 3: 0.4}
        rate *= factors.get(years, 0.6)
    compute = rate * hours
    storage_cost = storage * 0.10
    return {'total_cost': compute + storage_cost}

def calculate_gcp_cost(instance, hours, storage, reserved=False, spot=False, years=1):
    data = gcp_instances[instance]
    rate = data['hourly_rate']
    if spot:
        rate *= 0.2
    elif reserved:
        factors = {1: 0.7, 3: 0.5}
        rate *= factors.get(years, 0.7)
    compute = rate * hours
    storage_cost = storage * 0.04
    return {'total_cost': compute + storage_cost}

def calculate_api_cost(provider, model, input_tokens, output_tokens, api_calls):
    m = api_pricing[provider][model]
    input_cost = input_tokens * m['input_per_1M']
    output_cost = output_tokens * m['output_per_1M']
    call_cost = api_calls * 0.0001 if provider == 'TogetherAI' else 0
    return {'total_cost': input_cost + output_cost + call_cost}

def filter_compatible(instances, min_mem):
    res = {}
    for name, data in instances.items():
        mem_str = data['gpu_memory']
        if 'x' in mem_str and not mem_str.startswith(('1x','2x','4x','8x')):
            val = int(mem_str.replace('GB',''))
        elif 'x' in mem_str:
            parts = mem_str.split('x')
            val = int(parts[0]) * int(parts[1].replace('GB',''))
        else:
            val = int(mem_str.replace('GB',''))
        if val >= min_mem:
            res[name] = data
    return res

def generate_cost_comparison(
    compute_hours, tokens_per_month, input_ratio, api_calls,
    model_size, storage_gb, reserved_instances, spot_instances, multi_year_commitment
):
    years = int(multi_year_commitment)
    in_tokens = tokens_per_month * (input_ratio / 100)
    out_tokens = tokens_per_month - in_tokens
    min_mem = model_sizes[model_size]['memory_required']

    aws_comp = filter_compatible(aws_instances, min_mem)
    gcp_comp = filter_compatible(gcp_instances, min_mem)

    results = []
    # AWS
    if aws_comp:
        best_aws = min(aws_comp.keys(), key=lambda x: calculate_aws_cost(x, compute_hours, storage_gb, reserved_instances, spot_instances, years)['total_cost'])
        best_aws_cost = calculate_aws_cost(best_aws, compute_hours, storage_gb, reserved_instances, spot_instances, years)['total_cost']
        results.append({'provider': f'AWS ({best_aws})', 'cost': best_aws_cost, 'type': 'Cloud'})
    # GCP
    if gcp_comp:
        best_gcp = min(gcp_comp.keys(), key=lambda x: calculate_gcp_cost(x, compute_hours, storage_gb, reserved_instances, spot_instances, years)['total_cost'])
        best_gcp_cost = calculate_gcp_cost(best_gcp, compute_hours, storage_gb, reserved_instances, spot_instances, years)['total_cost']
        results.append({'provider': f'GCP ({best_gcp})', 'cost': best_gcp_cost, 'type': 'Cloud'})
    # API (TogetherAI only)
    api_opts = { (prov, m): calculate_api_cost(prov, m, in_tokens, out_tokens, api_calls)['total_cost']
                 for prov in api_pricing for m in api_pricing[prov] }
    best_api = min(api_opts, key=api_opts.get)
    results.append({'provider': f'{best_api[0]} ({best_api[1]})', 'cost': api_opts[best_api], 'type': 'API'})

    # Build bar chart
    df_res = pd.DataFrame(results)
    aws_name = df_res[df_res['type']=='Cloud']['provider'].iloc[0]
    gcp_name = df_res[df_res['type']=='Cloud']['provider'].iloc[1]
    api_name = df_res[df_res['type']=='API']['provider'].iloc[0]

    fig = px.bar(
        df_res, x='provider', y='cost', color='provider',
        color_discrete_map={
            aws_name: '#FF9900',  # AWS orange
            gcp_name: '#4285F4',  # GCP blue
            api_name: '#D62828'   # TogetherAI red
        },
        title='Monthly Cost Comparison',
        labels={'provider': 'Provider', 'cost': 'Monthly Cost'}
    )
    fig.update_yaxes(tickprefix='$')
    fig.update_layout(showlegend=False, height=500)

    # HTML summary tables omitted for brevity
    html_tables = '<div>'
    # ... you can reinsert your HTML tables here if needed
    html_tables += '</div>'
    return html_tables, fig

def app_function(
    compute_hours, tokens_per_month, input_ratio, api_calls,
    model_size, storage_gb, reserved_instances, spot_instances, multi_year_commitment
):
    return generate_cost_comparison(
        compute_hours, tokens_per_month, input_ratio, api_calls,
        model_size, storage_gb, reserved_instances, spot_instances, multi_year_commitment
    )

# Gradio UI
def main():
    with gr.Blocks(title="Cloud Cost Estimator", theme=gr.themes.Soft(primary_hue="indigo")) as demo:
        gr.HTML("""
        <div style="text-align:center; margin-bottom:20px;">
            <h1>Cloud Cost Estimator</h1>
            <p>Compare cloud vs API costs</p>
        </div>
        """)
        with gr.Row():
            with gr.Column(scale=1):
                compute_hours = gr.Slider("Compute Hours per Month", 1, 730, 100)
                tokens_per_month = gr.Slider("Tokens per Month (M)", 1, 1000, 10)
                input_ratio = gr.Slider("Input Ratio (%)", 10, 90, 30)
                api_calls = gr.Slider("API Calls per Month", 100, 1_000_000, 10000, step=100)
                model_size = gr.Dropdown(list(model_sizes.keys()), value="Medium (13B parameters)")
                storage_gb = gr.Slider("Storage (GB)", 10, 1000, 100)
                reserved_instances = gr.Checkbox("Reserved Instances", value=False)
                spot_instances = gr.Checkbox("Spot Instances", value=False)
                multi_year_commitment = gr.Radio(["1","3"], value="1")
                submit = gr.Button("Calculate Costs")
            with gr.Column(scale=2):
                out_html = gr.HTML()
                out_plot = gr.Plot()
        submit.click(app_function,
                     inputs=[compute_hours, tokens_per_month, input_ratio, api_calls,
                             model_size, storage_gb, reserved_instances, spot_instances, multi_year_commitment],
                     outputs=[out_html, out_plot])
        demo.launch()

if __name__ == "__main__":
    main()