parser.add_argument("--weight_decay", default=1e-2, type=float)
parser.add_argument(
"-l",
"--log",
dest="logLevel",
default="INFO",
choices=["DEBUG", "INFO", "WARNING", "ERROR", "CRITICAL"],
help="Set the logging level (default: 'Info').",
)
main_args = parser.parse_known_args()
if main_args[0].mode == "abstractive":
parser = AbstractiveSummarizer.add_model_specific_args(parser)
else:
parser = ExtractiveSummarizer.add_model_specific_args(parser)
if main_args[0].custom_checkpoint_every_n and (not main_args[0].weights_save_path):
logger.error(
"You must specify the `--weights_save_path` to use `--custom_checkpoint_every_n`."
)
main_args = parser.parse_args()
# Setup logging config
logging.basicConfig(
format="%(asctime)s|%(name)s|%(levelname)s> %(message)s",
level=logging.getLevelName(main_args.logLevel),
)
# Set the `nlp` logging verbosity since its default is not INFO.
!git clone [email protected]:sentian/TransformerSum.git !cd transformersum !conda env create --file environment.yml import os os.chdir('/home/sentian/Projects/code/gentext/TransformerSum/src') from extractive import ExtractiveSummarizer ## the model checkpoint needs to be downloaded from https://github.com/sentian/TransformerSum#extractive ## I use the roberta-base-ext-sum model ## change the path_to_checkpoint below model = ExtractiveSummarizer.load_from_checkpoint("/home/sentian/Projects/model/gentext/transformersum/roberta-base-ext-sum/epoch=3.ckpt", strict=False) text_to_summarize = "I love the extra heat rods which make more even heat. The farther back on the rack, \ the hotter it is. The controls are easy to understand and the pre-heat function is a very nice thing to \ have. It heats up fast. Because the unit is small, the 1800W coils are able to heat it rapidly. \ The heating elements are radiant coils in a quartz tube. I really wish the unit had an extra heating \ element both top and bottom. The Breville is easy to use and heats up quickly. It heats up quickly and \ maintains temperature perfectly. My unit has a 'hot spot' in the right rear corner. It gets a good even \ heat on both sides. It gets hot very quickly so preheating takes no time. The temperature and time display \ is very easy to read. The unit is very light so I can move it to my island if I am worried about the heat. \ It is easy to operate, great dial choices, easy to set or adjust time and temp. Great preheat function. \ The automatic preheat feature is nicer than I expected and the size is perfect. Heats up VERY quickly. \ Way louder than my full size Breville." summary = model.predict(text_to_summarize, num_summary_sentences=5) print(summary)
def summarize_text(text, model_choice):
summarizer = ExtractiveSummarizer.load_from_checkpoint(model_choice)
return summarizer.predict(text)
import time
import json
from datetime import timedelta
from pandas import json_normalize
from extractive import ExtractiveSummarizer
from tqdm import tqdm
tqdm.pandas()
# ======================================================================================================================
# Define summarizer (transformersum)
# ======================================================================================================================
# using extractive model "distilroberta-base-ext-sum"
model = ExtractiveSummarizer.load_from_checkpoint("models\\epoch=3.ckpt")
# ======================================================================================================================
# ACM dataset
# ======================================================================================================================
file = 'datasets\\ACM.json' # TEST data to evaluate the final model
# ======================================================================================================================
# Read data
# ======================================================================================================================
json_data = []
for line in open(file, 'r', encoding="utf8"):
json_data.append(json.loads(line))
# convert json to dataframe
import torch
import time
from extractive import ExtractiveSummarizer
fpath = "./lightning_logs/version_12/checkpoints/epoch=12.ckpt"
model = ExtractiveSummarizer.load_from_checkpoint(fpath)
model.to(torch.device("cuda:0"))
print(f"device: {model.device}")
t = "Do you like games? I want to get a refund for a order I placed. 2 items were damaged. Are you a bot? 123082HSF."
print(f"text: {t}\n\n")
k = 100
start = time.time()
for _ in range(k):
summary_sents = model.predict(t, raw_scores=True)
end = time.time()
print(f"time taken: {(end-start)/k * 1000}ms")
print(summary_sents)
print("\n")
summary = model.predict(t)
print(summary)