src/submission/submit.py

import json
import os
from datetime import datetime, timedelta, timezone
import gradio as gr
from src.display.formatting import styled_error, styled_message, styled_warning
from src.envs import API, EVAL_REQUESTS_PATH, TOKEN, QUEUE_REPO
from src.submission.check_validity import (
    already_submitted_models,
    check_model_card,
    get_model_size,
    is_model_on_hub,
)
from huggingface_hub import hf_hub_download

REQUESTED_MODELS = None
USERS_TO_SUBMISSION_DATES = None

def add_new_eval(
    model: str,
    progress=gr.Progress()
    #base_model: str,
    #revision: str,
    #precision: str,
    #weight_type: str,
    #model_type: str,
):
    global REQUESTED_MODELS
    global USERS_TO_SUBMISSION_DATES
    if not REQUESTED_MODELS:
        REQUESTED_MODELS, USERS_TO_SUBMISSION_DATES = already_submitted_models(EVAL_REQUESTS_PATH)

    user_name = ""
    model_path = model
    if "/" in model:
        user_name = model.split("/")[0]
        model_path = model.split("/")[1]

    #precision = precision.split(" ")[0]
    current_time = datetime.now(timezone.utc).strftime("%Y-%m-%dT%H:%M:%SZ")


    progress(0.1, desc=f"Checking model {model} on hub")

    if not is_model_on_hub(model_name=model, token=TOKEN, test_tokenizer=True): #revision=revision
        return styled_error("Model does not exist on HF Hub. Please select a valid model name.")
    

    ##check for org banning
    progress(0.2, desc=f"Checking for banned orgs")
    banned_orgs = [{
        'org_name':'TEMPLATE',
        'banning_reason':'Submitting contaminated models'
    }]

    if user_name in [banned_org['org_name'] for banned_org in banned_orgs]:
        return styled_error(
            f"Your org  \"{user_name}\" is banned from submitting models on ABL. If you think this is a mistake then please contact benchmark@silma.ai"
        )

    """
    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=model)#, revision=revision
    except Exception:
        return styled_error("Could not get your model information. Please fill it up properly.")

    progress(0.3, desc=f"Checking model size")
    model_size = get_model_size(model_info=model_info)#, precision=precision

    if model_size>15:
        return styled_error("We currently accept community-submitted models up to 15 billion parameters only. If you represent an organization then please contact us at benchmark@silma.ai")

    # 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)
    


    ##check if org have submitted in the last 30 days
    progress(0.6, desc=f"Checking last submission date")
    previous_user_submissions = USERS_TO_SUBMISSION_DATES.get(user_name)

    if False and previous_user_submissions:

        previous_user_submission_dates = [datetime.strptime(date.replace("T"," ").split(" ")[0], "%Y-%m-%d") for date in previous_user_submissions]
        previous_user_submission_dates.sort(reverse=True)
        most_recent_submission = previous_user_submission_dates[0]

        time_since_last_submission = datetime.now() - most_recent_submission
        if time_since_last_submission < timedelta(days=30):
            return styled_warning(
                f"Your org \"{user_name}\" have already submitted a model in the last 30 days. Please wait before submitting another model. For exceptions please contact benchmark@silma.ai"
            )
            


    progress(0.8, desc=f"Checking same model submissions")

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

    # Seems good, creating the eval
    print("Preparing a new eval")

    eval_entry = {
        "model": model,
        "model_sha": model_info.sha,
        #"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,
    }

   

    print("Creating eval file")
    OUT_DIR = f"{EVAL_REQUESTS_PATH}/{user_name}"
    os.makedirs(OUT_DIR, exist_ok=True)
    out_path = f"{OUT_DIR}/{model_path}_eval_request.json" #_{precision}_{weight_type}

    with open(out_path, "w") as f:
        f.write(json.dumps(eval_entry))


    ##update queue file
    queue_file_path = "./eval_queue.json"

    ## download queue_file from repo using HuggingFace hub API, update it and upload again
    queue_file = hf_hub_download(
        filename=queue_file_path,
        repo_id=QUEUE_REPO,
        repo_type="space",
        token=TOKEN
    )   

    
    with open(queue_file, "r") as f:
        queue_data = json.load(f)

    queue_len = len(queue_data)
    print(f"Queue length: {queue_len}")

    if queue_len == 0:
        queue_data = []
    elif queue_len >= 1:
        return styled_warning("The evaluation queue is full at the moment. Please try again in one hour")

    queue_data.append(eval_entry)

    print(queue_data)

    #with open(queue_file, "w") as f:
    #    json.dump(queue_data, f)

    print("Updating eval queue file")
    API.upload_file(
        path_or_fileobj=json.dumps(queue_data, indent=2).encode("utf-8"),
        path_in_repo=queue_file_path,
        repo_id=QUEUE_REPO,
        repo_type="space",
        commit_message=f"Add {model} to eval queue"
    )
        

    print("Uploading eval file")
    API.upload_file(
        path_or_fileobj=out_path,
        path_in_repo=out_path,
        repo_id=QUEUE_REPO,
        repo_type="space",
        commit_message=f"Add {model} request file",
    )


    # Remove the local file
    os.remove(out_path)


    return styled_message(
        "✅ Good news! Your model has been added to the evaluation queue.<br>If you do not see the results after 3 hours then please let us know by opening a community discussion."
    )