app.py

import os
import json
from datetime import datetime, timezone
from dataclasses import dataclass

import gradio as gr
from huggingface_hub import HfApi 
from huggingface_hub.hf_api import ModelInfo
from enum import Enum


OWNER = "EnergyStarAI"
COMPUTE_SPACE = f"{OWNER}/launch-computation-example"
REQUESTS_DATASET_PATH = f"{OWNER}/requests_debug"

TOKEN = os.environ.get("DEBUG")
API = HfApi(token=TOKEN)


## All the model information that we might need
@dataclass
class ModelDetails:
    name: str
    display_name: str = ""
    symbol: str = "" # emoji

class WeightType(Enum):
    Adapter = ModelDetails("Adapter")
    Original = ModelDetails("Original")
    Delta = ModelDetails("Delta")

class Precision(Enum):
    float16 = ModelDetails("float16")
    bfloat16 = ModelDetails("bfloat16")
    float32 = ModelDetails("float32")
    bfloat32 = ModelDetails("bfloat32")
    Unknown = ModelDetails("?")

    def from_str(precision):
        if precision in ["torch.float16", "float16"]:
            return Precision.float16
        if precision in ["torch.bfloat16", "bfloat16"]:
            return Precision.bfloat16
        if precision in ["torch.bfloat32", "bfloat32"]:
            return Precision.bfloat32
        if precision in ["torch.float32", "float32"]:
            return Precision.float32
            
class ModelType(Enum):
    PT = ModelDetails(name="pretrained", symbol="🟢")
    FT = ModelDetails(name="fine-tuned", symbol="🔶")
    IFT = ModelDetails(name="instruction-tuned", symbol="⭕")
    RL = ModelDetails(name="RL-tuned", symbol="🟦")
    Unknown = ModelDetails(name="", symbol="?")

    def to_str(self, separator=" "):
        return f"{self.value.symbol}{separator}{self.value.name}"

    @staticmethod
    def from_str(type):
        if "fine-tuned" in type or "🔶" in type:
            return ModelType.FT
        if "pretrained" in type or "🟢" in type:
            return ModelType.PT
        if "RL-tuned" in type or "🟦" in type:
            return ModelType.RL
        if "instruction-tuned" in type or "⭕" in type:
            return ModelType.IFT
        return ModelType.Unknown

def update(name):
    API.restart_space(COMPUTE_SPACE)  
    return f"Okay! {COMPUTE_SPACE} should be running now!"


def get_model_size(model_info: ModelInfo, precision: str):
    """Gets the model size from the configuration, or the model name if the configuration does not contain the information."""
    try:
        model_size = round(model_info.safetensors["total"] / 1e9, 3)
    except (AttributeError, TypeError):
        return 0  # Unknown model sizes are indicated as 0, see NUMERIC_INTERVALS in app.py
    
    size_factor = 8 if (precision == "GPTQ" or "gptq" in model_info.modelId.lower()) else 1
    model_size = size_factor * model_size
    return model_size


def add_new_eval(
    repo_id: str,
    base_model: str,
    revision: str,
    precision: str,
    weight_type: str,
    model_type: str,
):
    model_owner = repo_id.split("/")[0]
    model_name = repo_id.split("/")[1]
    precision = precision.split(" ")[0]

    out_dir = f"{REQUESTS_DATASET_PATH}/{model_owner}"
    print("Making Dataset directory to output results at %s" % out_dir)
    os.makedirs(out_dir, exist_ok=True)
    out_path = f"{REQUESTS_DATASET_PATH}/{model_owner}/{model_name}_eval_request_{precision}_{weight_type}.json"
    
    current_time = datetime.now(timezone.utc).strftime("%Y-%m-%dT%H:%M:%SZ")

    #if model_type is None or model_type == "":
    #    return styled_error("Please select a model type.")

    # Does the model actually exist?
    #if revision == "":
    revision = "main"

    # Is the model on the hub?
    #if weight_type in ["Delta", "Adapter"]:
    #    base_model_on_hub, error, _ = is_model_on_hub(model_name=base_model, revision=revision, token=TOKEN, test_tokenizer=True)
    #    if not base_model_on_hub:
    #        return styled_error(f'Base model "{base_model}" {error}')

    #if not weight_type == "Adapter":
    #    model_on_hub, error, _ = is_model_on_hub(model_name=model, revision=revision, token=TOKEN, test_tokenizer=True)
    #    if not model_on_hub:
    #        return styled_error(f'Model "{model}" {error}')

    # Is the model info correctly filled?
    try:
        model_info = API.model_info(repo_id=repo_id, revision=revision)
    except Exception:
        print("Could not find information for model %s at revision %s" % (model, revision))
        return
    #    return styled_error("Could not get your model information. Please fill it up properly.")

    model_size = get_model_size(model_info=model_info, precision=precision)

    # Were the model card and license filled?
    #try:
    #    license = model_info.cardData["license"]
    #except Exception:
    #    return styled_error("Please select a license for your model")

    #modelcard_OK, error_msg = check_model_card(model)
    #if not modelcard_OK:
    #    return styled_error(error_msg)

    # Seems good, creating the eval
    print("Adding request")

    request_entry = {
        "model": repo_id,
        "base_model": base_model,
        "revision": revision,
        "precision": precision,
        "weight_type": weight_type,
        "status": "PENDING",
        "submitted_time": current_time,
        "model_type": model_type,
        "likes": model_info.likes,
        "params": model_size}
        #"license": license,
        #"private": False,
    #}

    # Check for duplicate submission
    #if f"{model}_{revision}_{precision}" in REQUESTED_MODELS:
    #    return styled_warning("This model has been already submitted.")

    print("Writing out request file to %s" % out_path)
    with open(out_path, "w") as f:
        f.write(json.dumps(eval_entry))


        

with gr.Blocks() as demo:
    gr.Markdown("This is a super basic example 'frontend'. Fill out below then click **Run** to create the request file and launch the job.")
    gr.Markdown("The request file will be written so %s." % REQUESTS_DATASET_PATH)
    gr.Markdown("The job will be launched at [EnergyStarAI/launch-computation-example](https://huggingface.co/spaces/EnergyStarAI/launch-computation-example).")
    with gr.Row():
            gr.Markdown("# ✉️✨ Submit your model here!", elem_classes="markdown-text")
            with gr.Row():
                with gr.Column():
                    model_name_textbox = gr.Textbox(label="Model name")
                    revision_name_textbox = gr.Textbox(label="Revision commit", placeholder="main")
                    model_type = gr.Dropdown(
                        choices=[t.to_str(" : ") for t in ModelType if t != ModelType.Unknown],
                        label="Model type",
                        multiselect=False,
                        value=None,
                        interactive=True,
                    )

                with gr.Column():
                    precision = gr.Dropdown(
                        choices=[i.value.name for i in Precision if i != Precision.Unknown],
                        label="Precision",
                        multiselect=False,
                        value="float16",
                        interactive=True,
                    )
                    weight_type = gr.Dropdown(
                        choices=[i.value.name for i in WeightType],
                        label="Weights type",
                        multiselect=False,
                        value="Original",
                        interactive=True,
                    )
                    base_model_name_textbox = gr.Textbox(label="Base model (for delta or adapter weights)")

            with gr.Row():
                submit_button = gr.Button("Run Analysis")
                submission_result = gr.Markdown()
                submit_button.click(
                    fn=add_new_eval,
                    inputs=[
                        model_name_textbox,
                        base_model_name_textbox,
                        revision_name_textbox,
                        precision,
                        weight_type,
                        model_type,
                    ],
                    outputs=submission_result,
                )
    
demo.launch()