app.py

import gradio as gr
import json
import requests
import pandas as pd  

def update_task_options(framework):
    config = {
        "Custom":["Custom"],
        "Diffusers":["Text To Image"],
        "Transformers":[
            "Text Classification", 
            "Zero Shot Classification", 
            "Token Classifiation", 
            "Question Answering", 
            "Fill Mask", 
            "Summarization", 
            "Translation", 
            "Text to Text Generation", 
            "Text Generation", 
            "Feature Extraction", 
            "Image Classification", 
            "Automatic Speech Recognition", 
            "Audio Classification", 
            "Object Detection", 
            "Image Segmentation", 
            "Table Question Answering", 
            "Conversational", 
            "Visual Question Answering", 
            "Zero Shot Image Classification"]
    }
    return gr.Dropdown.update(
        choices=config[framework],
        value=config[framework][0] if len(config[framework]) > 0 else None
    )

def update_regions(provider):
    available_regions = []
    
    headers = {
        "Content-Type": "application/json",
    }
    endpoint_url = f"https://api.endpoints.huggingface.cloud/provider/{provider}/region"
    response = requests.get(endpoint_url, headers=headers)

    for region in response.json()['items']:
        if region['status'] == 'available':
            available_regions.append(f"{region['region']}/{region['label']}")

    return gr.Dropdown.update(
        choices=available_regions,
        value=available_regions[0] if len(available_regions) > 0 else None
    )

def update_compute_options(provider, region):
    region = region.split("/")[0]
    available_compute_choices = []

    headers = {
        "Content-Type": "application/json",
    }
    endpoint_url = f"https://api.endpoints.huggingface.cloud/provider/{provider}/region/{region}/compute"
    print(endpoint_url)
    response = requests.get(endpoint_url, headers=headers)

    for compute in response.json()['items']:
        if compute['status'] == 'available':
            accelerator = compute['accelerator']
            numAccelerators = compute['numAccelerators']
            memoryGb = compute['memoryGb'].replace("Gi", "GB")
            architecture = compute['architecture']
            instanceType = compute['instanceType']
            
            type = f"{numAccelerators}vCPU {memoryGb} · {architecture}" if accelerator == "cpu" else f"{numAccelerators}x {architecture}"
            
            available_compute_choices.append(
                f"{compute['accelerator'].upper()} [{compute['instanceSize']}] · {type} · {instanceType}"
            )

    return gr.Dropdown.update(
        choices=available_compute_choices,
        value=available_compute_choices[0] if len(available_compute_choices) > 0 else None
    )

def submit(
    hf_token_input,
    endpoint_name_input,
    provider_selector,
    region_selector,
    repository_selector,
    revision_selector,
    task_selector,
    framework_selector,
    compute_selector,
    min_node_selector, 
    max_node_selector, 
    security_selector    
):
    compute_resources = compute_selector.split("·")
    accelerator = compute_resources[0][:3].strip()

    size_l_index = compute_resources[0].index("[") - 1
    size_r_index = compute_resources[0].index("]")
    size = compute_resources[0][size_l_index : size_r_index].strip()

    type = compute_resources[-1].strip()
    
    payload = {
      "accountId": repository_selector.split("/")[0],
      "compute": {
        "accelerator": accelerator.lower(),
        "instanceSize": size[1:],
        "instanceType": type,
        "scaling": {
          "maxReplica": int(max_node_selector),
          "minReplica": int(min_node_selector)
        }
      },
      "model": {
        "framework": "custom",
        "image": {
          "huggingface": {}
        },
        "repository": repository_selector.lower(),
        "revision": revision_selector,
        "task": task_selector.lower()
      },
      "name": endpoint_name_input.strip(),
      "provider": {
        "region": region_selector.split("/")[0].lower(),
        "vendor": provider_selector.lower()
      },
      "type": security_selector.lower()
    }
    

    payload = json.dumps(payload)
    print(f"Payload: {payload}")

    headers = {
        "Authorization": f"Bearer {hf_token_input.strip()}",
        "Content-Type": "application/json",
    }
    endpoint_url = f"https://api.endpoints.huggingface.cloud/endpoint"
    print(f"Endpoint: {endpoint_url}")

    response = requests.post(endpoint_url, headers=headers, data=payload)
    if response.status_code == 400:
        return f"{response.text}. Malformed data in {payload}"
    elif response.status_code == 401:
        return "Invalid token"
    elif response.status_code == 409:
        return f"Error: {response.text}"
    elif response.status_code == 202:
        return f"Endpoint {endpoint_name_input} created successfully on {provider_selector.lower()} using {repository_selector.lower()}@{revision_selector}. \n Please check out the progress at https://ui.endpoints.huggingface.co/endpoints."
    else:
        return f"something went wrong {response.status_code} = {response.text}"

def delete_endpoint(
    hf_token_input,
    endpoint_name_input
):
    response = requests.delete(
        f"https://api.endpoints.huggingface.cloud/endpoint/{endpoint_name_input}",
        headers = {
        "Authorization": f"Bearer {hf_token_input.strip()}",
        "Content-Type": "application/json",
    }
    )
    if response.status_code == 401:
        return "Invalid token"
    elif response.status_code == 404:
        return f"Error: {response.text}"
    elif response.status_code == 202:
        return f"Endpoint {endpoint_name_input} deleted successfully."
    else:
        return f"something went wrong {response.status_code} = {response.text}"

def get_all_endpoints(
    hf_token_input,
):
    response = requests.get(
            f"https://api.endpoints.huggingface.cloud/endpoint",
            headers = {
                "Authorization": f"Bearer {hf_token_input.strip()}",
                "Content-Type": "application/json",
            })
    if response.status_code == 401:
        return "Invalid token"
    elif response.status_code == 200:
        endpoints_json = response.json()
        print(endpoints_json)
        endpoints_df = pd.DataFrame(endpoints_json["items"])
        endpoints_df = endpoints_df[["name", "model", "provider", "compute", "status"]]
        endpoints_df["model"] = endpoints_df["model"].apply(lambda x: x["repository"] + "@" + x["revision"])
        endpoints_df["provider"] = endpoints_df["provider"].apply(lambda x: x["vendor"] + "/" + x["region"])
        endpoints_df["compute"] = endpoints_df["compute"].apply(lambda x: x["instanceType"] + "·" + x["instanceSize"] + " [" + x["accelerator"] + "]")
        endpoints_df["status"] = endpoints_df["status"].apply(lambda x: x["state"])
        endpoints_df["minReplica"] = endpoints_df["compute"].apply(lambda x: x["scaling"]["minReplica"])
        endpoints_df["maxReplica"] = endpoints_df["compute"].apply(lambda x: x["scaling"]["maxReplica"])
        endpoints_df["createdAt"] = endpoints_df["status"].apply(lambda x: x["createdAt"])
        endpoints_df["updatedAt"] = endpoints_df["status"].apply(lambda x: x["updatedAt"])
        endpoints_df = endpoints_df[["name", "model", "provider", "compute", "status", "minReplica", "maxReplica", "createdAt", "updatedAt"]]
        return gr.Dataframe.update(
            value=endpoints_df
        )

def update_endpoint(
    hf_token_input,
    endpoint_name_input,
    min_node_selector,
    max_node_selector,
    instance_type,
):
    payload ={
            "compute": {
                "instanceSize": instance_type.split("·")[0].split("[")[1].split("]")[0],
                "instanceType": instance_type.split("·")[-1].strip(),
                "scaling": {
                    "maxReplica": max_node_selector,
                    "minReplica": min_node_selector
                    }
                }}

    response = requests.put(
            f"https://api.endpoints.huggingface.cloud/endpoint/{endpoint_name_input}",
            headers = {
            "Authorization": f"Bearer {hf_token_input.strip()}",
            "Content-Type": "application/json",
            },
            data=payload,
        )
    if response.status_code == 401:
        return "Invalid token"
    elif response.status_code == 404:
        return f"Error: {response.text}"
    elif response.status_code == 202:
        return f"Endpoint {endpoint_name_input} updated successfully."
    else:
        return f"something went wrong {response.status_code} = {response.text}"

with gr.Blocks() as interface:
    gr.Markdown("""
    #### Your 🤗 Access Token <span style="color:red;">(Required)</span>
    """)
    hf_token_input = gr.Textbox(
        show_label=False,
        type="password"
    )

    # Get All Endpoints Info
    with gr.Tab("Info"):
        gr.Markdown("""
        ### All Deployed Endpoints
        """)
        endpoints_table = gr.Dataframe(
            headers=["Endpoint Name", "Revision", "Provider", "Instance Type", "Status", "Min Replica", "Max Replica", "Created At", "Updated At"],
            col_count=(9, "fixed"),
        )
        endpoint_info_button = gr.Button(value="Get Info")
    
    # Deploy Endpoint
    with gr.Tab("Deploy Endpoint"):
        gr.Markdown(
        """
        ### <br><center style="color:green">(Deploy Your Model on 🤗 Endpoint)</center>
        """)
        gr.Markdown("""    
        #### Endpoint Name
        """)
        endpoint_name_input = gr.Textbox(
            show_label=False
        )

        with gr.Row():
            gr.Markdown("""    
            #### Cloud Provider
            """)

            gr.Markdown("""    
            #### Cloud Region
            """)       
        
        with gr.Row():
            provider_selector = gr.Dropdown(
                choices=["aws", "azure"],
                value="",
                interactive=True,
                show_label=False,
            )
            
            region_selector = gr.Dropdown(
                [],
                value="",
                interactive=True,
                show_label=False,
            )
    
        with gr.Row():
            gr.Markdown("""    
            #### Target Model
            """)

            gr.Markdown("""    
            #### Target Model Version(branch commit hash)
            """)       
        
        with gr.Row():
            repository_selector = gr.Textbox(
                value="",
                interactive=True,
                show_label=False,
            )

            revision_selector = gr.Textbox(
                value="",
                interactive=True,
                show_label=False,
            )        

        with gr.Row():
            gr.Markdown("""    
            #### Task
            """)

            gr.Markdown("""    
            #### Framework
            """)      
        
        with gr.Row():
            framework_selector = gr.Dropdown(
                choices = ["Custom", "Diffusers", "Transformers"],
                value="",
                interactive=True,
                show_label=False,
            )

            task_selector = gr.Dropdown(
                [],
                value="",
                interactive=True,
                show_label=False,
            )

        gr.Markdown("""
        
        #### Select Compute Instance Type
        """)    
        compute_selector = gr.Dropdown(
            [],
            value="",
            interactive=True,
            show_label=False,
        )

        with gr.Row():
            gr.Markdown("""    
            #### Min Number of Nodes
            """)

            gr.Markdown("""    
            #### Max Number of Nodes
            """)

            gr.Markdown("""    
            #### Security Level
            """)        
        
        with gr.Row():
            min_node_selector = gr.Number(
                value=1,
                interactive=True,
                show_label=False,
            )

            max_node_selector = gr.Number(
                value=1,
                interactive=True,
                show_label=False,
            )

            security_selector = gr.Radio(
                choices=["Protected", "Public"],
                value="Protected",
                interactive=True,
                show_label=False,
            )
        submit_button = gr.Button(
            value="Submit",
        )

        status_txt = gr.Textbox(
            value="status",
            interactive=False
        )

    # Update Endpoint
    with gr.Tab("Update Endpoint"):
        gr.Markdown("""
        ### <br><center style="color:green">(Update 🔁 Endpoint)</center>
        """)

        with gr.Row():
            gr.Markdown("""    
            #### Cloud Provider
            """)

            gr.Markdown("""    
            #### Cloud Region
            """)      

        with gr.Row():
            update_provider_selector = gr.Dropdown(
                choices=["aws", "azure"],
                value="",
                interactive=True,
                show_label=False,
            )
            
            update_region_selector = gr.Dropdown(
                [],
                value="",
                interactive=True,
                show_label=False,
            )

        with gr.Row():
            gr.Markdown("""
            #### Endpoint Name
            """)
            gr.Markdown("""
            #### Instance Type
            """)

        with gr.Row():
            update_endpoint_name_input = gr.Textbox(
                show_label=False
            )
            update_compute_selector = gr.Dropdown(
                [],
                value="",
                interactive=True,
                show_label=False,
            )
        with gr.Row():
            gr.Markdown("""
            #### Min Number of Nodes
            """)
            gr.Markdown("""
            #### Max Number of Nodes
            """)
        with gr.Row():
            update_min_node_input = gr.Number(
                value=1,
                interactive=True,
                show_label=False,
            )
            
            update_max_node_input = gr.Number(
                value=1,
                interactive=True,
                show_label=False,
            )
        update_button = gr.Button(
            value="Update",
        )

        update_status_txt = gr.Textbox(
            value="status",
            interactive=False
        )

    # Delete Endpoint
    with gr.Tab("Delete Endpoint"):
        gr.Markdown("""
        ### <br><center style="color:green">(Delete 🗑️ Endpoint)</center>
        """)
        gr.Markdown("""    
        #### Endpoint Name
        """)
        delete_endpoint_name_input = gr.Textbox(
            show_label=False
        )
        delete_button = gr.Button(
            value="Delete",
        )

        delete_status_txt = gr.Textbox(
            value="status",
            interactive=False
        )

    # Pricing Table
    with gr.Tab("Pricing Table"):
        gr.Markdown("""
            ### <br><center style="color:green">(Instance Pricing Table)</center>
            #### Pricing Table(CPU) - 2023/2/22
            """)
        gr.Dataframe(
            headers=["provider", "size", "$/h", "vCPUs", "Memory", "Architecture"],
            datatype=["str", "str", "str", "number", "str", "str"],
            row_count=8,
            col_count=(6, "fixed"),
            value=[
                ["aws", "small", "$0.06", 1, "2GB", "Intel Xeon - Ice Lake"],
                ["aws", "medium", "$0.12", 2, "4GB", "Intel Xeon - Ice Lake"],
                ["aws", "large", "$0.24", 4, "8GB", "Intel Xeon - Ice Lake"],
                ["aws", "xlarge", "$0.48", 8, "16GB", "Intel Xeon - Ice Lake"],
                ["azure", "small", "$0.06", 1, "2GB", "Intel Xeon"],
                ["azure", "medium", "$0.12", 2, "4GB", "Intel Xeon"],
                ["azure", "large", "$0.24", 4, "8GB", "Intel Xeon"],
                ["azure", "xlarge", "$0.48", 8, "16GB", "Intel Xeon"],
            ]
        )

        gr.Markdown("""    
        #### Pricing Table(GPU) - 2023/2/22
        """)    

        gr.Dataframe(
            headers=["provider", "size", "$/h", "GPUs", "Memory", "Architecture"],
            datatype=["str", "str", "str", "number", "str", "str"],
            row_count=6,
            col_count=(6, "fixed"),
            value=[
                ["aws", "small", "$0.60", 1, "14GB", "NVIDIA T4"],
                ["aws", "medium", "$1.30", 1, "24GB", "NVIDIA A10G"],
                ["aws", "large", "$4.50", 4, "56GB", "NVIDIA T4"],
                ["aws", "xlarge", "$6.50", 1, "80GB", "NVIDIA A100"],
                ["aws", "xxlarge", "$7.00", 4, "96GB", "NVIDIA A10G"],
                ["aws", "xxxlarge", "$45.0", 8, "640GB", "NVIDIA A100"],
            ]
        )    

    # Info Tab Events
    endpoint_info_button.click(
            get_all_endpoints, 
            inputs=hf_token_input, 
            outputs=endpoints_table
        )

    # Deploy Tab Events
    framework_selector.change(update_task_options, inputs=framework_selector, outputs=task_selector)
    provider_selector.change(update_regions, inputs=provider_selector, outputs=region_selector)
    region_selector.change(update_compute_options, inputs=[provider_selector, region_selector], outputs=compute_selector)
    submit_button.click(
        submit, 
        inputs=[
            hf_token_input,
            endpoint_name_input,
            provider_selector,
            region_selector,
            repository_selector,
            revision_selector,
            task_selector,
            framework_selector,
            compute_selector,
            min_node_selector, 
            max_node_selector, 
            security_selector],
        outputs=status_txt)

    # Update Tab Events
    update_provider_selector.change(update_regions, inputs=update_provider_selector, outputs=update_region_selector)
    update_region_selector.change(update_compute_options, inputs=[update_provider_selector, update_region_selector], outputs=update_compute_selector)
    update_button.click(
        update_endpoint,
        inputs=[
            hf_token_input,
            update_endpoint_name_input,
            update_min_node_input,
            update_max_node_input,
            update_compute_selector
        ],
        outputs=update_status_txt
    )

    # Delete Tab Events
    delete_button.click(
            delete_endpoint,
            inputs=[
                hf_token_input,
                delete_endpoint_name_input
            ],
            outputs=delete_status_txt
        )

interface.launch()