Gabriele Sarti gsarti

📚

Learning

PhD Student in NLP & XAI @ University of Groningen | @inseq-team core dev | Prev: @awslabs, @aindo-com, ILC-CNR

gsarti / calamita_2024_eureka_rebus_length_hint_config.py

Created September 13, 2024 14:55

Configuration for 2-shot verbalized rebus solving with length hints for the CALAMITA 2024 campaign

	# Huggingface Hub ID of the dataset. If data is not on HF, set the variable to 'other'
	# NOTE: Default config is used
	DATA_HUB_ID = "gsarti/eureka-rebus-calamita-2024"

	# Task type. Values accepted: multiple_choice, open-ended
	OUTPUT_TYPE = "open-ended"

	### Prompting details

	# Template string used to compile the prompt. Use {{}} variables to fill using the dataset columns.

gsarti / calamita_2024_eureka_rebus_config.py

Created September 13, 2024 14:53

Configuration for 2-shot verbalized rebus solving for the CALAMITA 2024 campaign

	# Huggingface Hub ID of the dataset. If data is not on HF, set the variable to 'other'
	# NOTE: Default config is used
	DATA_HUB_ID = "gsarti/eureka-rebus-calamita-2024"

	# Task type. Values accepted: multiple_choice, open-ended
	OUTPUT_TYPE = "open-ended"

	### Prompting details

	# Template string used to compile the prompt. Use {{}} variables to fill using the dataset columns.

gsarti / score_trainslation_qe_comet.py

Created January 6, 2022 13:21

Evaluate the translation of a 🤗 dataset using Comet-based QE models

	import argparse
	import os
	from datasets import load_dataset
	from comet import download_model, load_from_checkpoint

	def replace_missing(scores, sentences, default = -100):
	out = []
	for score, sentence in zip(scores, sentences):
	if not sentence:
	out.append(default)

gsarti / run_marian_translate_flax.py

Last active January 6, 2022 13:17

Use 🤗 transformer FlaxMarianMTModel to translate 🤗 datasets text fields on TPU

	import argparse
	import datasets
	import jax
	from flax.jax_utils import replicate
	from flax.training.common_utils import shard
	from tqdm import tqdm
	from transformers import FlaxMarianMTModel, MarianTokenizer


	def translate(args):