import json import os from glob import glob from datetime import datetime, timezone import numpy as np import pandas as pd from src.display.formatting import styled_error, styled_message, styled_warning from src.envs import API, EVAL_REQUESTS_PATH, TOKEN, QUEUE_REPO, RESULTS_REPO from src.submission.check_validity import ( already_submitted_models, check_model_card, get_model_size, is_model_on_hub, ) from src.display.utils import ( BENCHMARK_COLS, BENCHMARK_COL_IDS, COLS ) REQUESTED_MODELS = None USERS_TO_SUBMISSION_DATES = None def add_new_eval( eval_name: str, upload: object, precision: str, hf_model_id: str, contact_email: str, captcha_ok: bool, ): try: if not eval_name: return styled_error("Please provide a model name."), captcha_ok if not precision: return styled_error("Please select precision."), captcha_ok if not contact_email: return styled_error("Please provide your contact email."), captcha_ok if not upload: return styled_error("Please upload a results file."), captcha_ok if not captcha_ok: return styled_error("Please prove you are a human!"), captcha_ok with open(upload, mode="r") as f: data = json.load(f) results = data['results'] acc_keys = ['exact_match,none', 'exact_match,flexible-extract', 'exact_match,strict-match'] ret = { 'eval_name': eval_name, 'precision': precision, 'hf_model_id': hf_model_id, 'contact_email': contact_email } for k, v in results.items(): for acc_k in acc_keys: if acc_k in v and k in BENCHMARK_COL_IDS: ret[k] = v[acc_k] #validation for k,v in ret.items(): if k in ['eval_name', 'precision', 'hf_model_id', 'contact_email']: continue if k not in BENCHMARK_COL_IDS: print(f"Missing: {k}") return styled_error(f'Missing: {k}'), captcha_ok if len(BENCHMARK_COL_IDS) != len(ret) - 4: print(f"Missing columns") return styled_error(f'Missing result entries'), captcha_ok # TODO add complex validation #print(results.keys()) #print(BENCHMARK_COLS) #for input_col in results.keys(): # if input_col not in BENCHMARK_COLS: # print(input_col) # return styled_error(f'Missing: {input_col}') #ret.update({i:j['acc,none'] for i,j in results.items()}) # fake data for testing... #ret.update({i:round(np.random.normal(1, 0.5, 1)[0], 2) for i,j in results.items()}) user_name = "czechbench_leaderboard" OUT_DIR = f"{EVAL_REQUESTS_PATH}/{user_name}" existing_eval_names = [] for fname in glob(f"{OUT_DIR}/*.json"): with open(fname, mode="r") as f: existing_eval = json.load(f) existing_eval_names.append(existing_eval['eval_name']) if ret['eval_name'] in existing_eval_names: print(f"Model name {ret['eval_name']} is used!") return styled_error(f"Model name {ret['eval_name']} is used!"), captcha_ok out_path = f"{OUT_DIR}/{eval_name}_eval_request.json" with open(out_path, "w") as f: f.write(json.dumps(ret)) print("Uploading eval file") print("path_or_fileobj: ", out_path) print("path_in_repo: ",out_path.split("eval-queue/")[1]) print("repo_id: ", RESULTS_REPO) print("repo_type: ", "dataset") response = API.upload_file( path_or_fileobj=out_path, path_in_repo=out_path.split("eval-queue/")[1], repo_id=RESULTS_REPO, repo_type="dataset", commit_message=f"Add {eval_name} to eval queue", ) """ 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") 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, 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.") 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 new eval") eval_entry = { "model": model, "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, } # Check for duplicate submission if f"{model}_{revision}_{precision}" in REQUESTED_MODELS: return styled_warning("This model has been already submitted.") 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_False_{precision}_{weight_type}.json" with open(out_path, "w") as f: f.write(json.dumps(eval_entry)) print("Uploading eval file") API.upload_file( path_or_fileobj=out_path, path_in_repo=out_path.split("eval-queue/")[1], repo_id=QUEUE_REPO, repo_type="dataset", commit_message=f"Add {model} to eval queue", ) # Remove the local file os.remove(out_path) """ return styled_message( "Your results have been successfully submitted. They will be added to the leaderboard upon verification." ), False except Exception as e: return styled_error(f"An error occurred: {e}"), captcha_ok