File size: 5,015 Bytes
14e4843 2a18e0a 14e4843 581fdbd 22ce8a7 14e4843 2a18e0a 14e4843 d6d7ec6 14e4843 581fdbd 14e4843 581fdbd bc48941 14e4843 22ce8a7 14e4843 a549d9d 22ce8a7 14e4843 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 |
import json
import os
from datetime import datetime, timezone
from src.display.formatting import styled_error, styled_message, styled_warning
from src.envs import API, EVAL_REQUESTS_PATH, H4_TOKEN, QUEUE_REPO, RATE_LIMIT_PERIOD, RATE_LIMIT_QUOTA, DEBUG_QUEUE_REPO
from src.leaderboard.filter_models import DO_NOT_SUBMIT_MODELS
from src.submission.check_validity import (
already_submitted_models,
check_model_card,
get_model_size,
is_model_on_hub,
user_submission_permission,
)
REQUESTED_MODELS = None
USERS_TO_SUBMISSION_DATES = None
def add_new_eval(
model: str,
base_model: str,
revision: str,
precision: str,
private: bool,
weight_type: str,
model_type: str,
inference_framework: str,
debug: bool = False,
gpu_type: str = "NVIDIA-A100-PCIe-80GB",
):
global REQUESTED_MODELS
global USERS_TO_SUBMISSION_DATES
if not REQUESTED_MODELS:
REQUESTED_MODELS, USERS_TO_SUBMISSION_DATES = already_submitted_models(EVAL_REQUESTS_PATH)
if debug:
QUEUE_REPO = DEBUG_QUEUE_REPO
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.")
# Is the user rate limited?
if user_name != "":
user_can_submit, error_msg = user_submission_permission(
user_name, USERS_TO_SUBMISSION_DATES, RATE_LIMIT_PERIOD, RATE_LIMIT_QUOTA
)
if not user_can_submit:
return styled_error(error_msg)
# Did the model authors forbid its submission to the leaderboard?
if model in DO_NOT_SUBMIT_MODELS or base_model in DO_NOT_SUBMIT_MODELS:
return styled_warning("Model authors have requested that their model be not submitted on the leaderboard.")
# 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=H4_TOKEN, test_tokenizer=False
)
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=False)
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")
# TODO: Check if the inference framework is valid
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,
"private": private,
"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,
"inference_framework": inference_framework,
"gpu_type": gpu_type
}
# Check for duplicate submission
if f"{model}_{revision}_{precision}_{inference_framework}_{gpu_type}" 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_{private}_{precision}_{weight_type}.json"
out_path = f"{OUT_DIR}/{model_path}_eval_request_{private}_{precision}_{weight_type}_{inference_framework}_{gpu_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 request has been submitted to the evaluation queue!\nPlease wait for up to an hour for the model to show in the PENDING list."
)
|