from itertools import chain
from typing import List, Tuple
import pandas as pd
import torch
from tqdm import tqdm
from transformers import T5Tokenizer, T5ForConditionalGeneration
from autorag.nodes.passagereranker.base import passage_reranker_node
from autorag.utils.util import make_batch, sort_by_scores, flatten_apply, select_top_k
prediction_tokens = {
"castorini/monot5-base-msmarco": ["▁false", "▁true"],
"castorini/monot5-base-msmarco-10k": ["▁false", "▁true"],
"castorini/monot5-large-msmarco": ["▁false", "▁true"],
"castorini/monot5-large-msmarco-10k": ["▁false", "▁true"],
"castorini/monot5-base-med-msmarco": ["▁false", "▁true"],
"castorini/monot5-3b-med-msmarco": ["▁false", "▁true"],
"castorini/monot5-3b-msmarco-10k": ["▁false", "▁true"],
"unicamp-dl/mt5-base-en-msmarco": ["▁no", "▁yes"],
"unicamp-dl/ptt5-base-pt-msmarco-10k-v2": ["▁não", "▁sim"],
"unicamp-dl/ptt5-base-pt-msmarco-100k-v2": ["▁não", "▁sim"],
"unicamp-dl/ptt5-base-en-pt-msmarco-100k-v2": ["▁não", "▁sim"],
"unicamp-dl/mt5-base-en-pt-msmarco-v2": ["▁no", "▁yes"],
"unicamp-dl/mt5-base-mmarco-v2": ["▁no", "▁yes"],
"unicamp-dl/mt5-base-en-pt-msmarco-v1": ["▁no", "▁yes"],
"unicamp-dl/mt5-base-mmarco-v1": ["▁no", "▁yes"],
"unicamp-dl/ptt5-base-pt-msmarco-10k-v1": ["▁não", "▁sim"],
"unicamp-dl/ptt5-base-pt-msmarco-100k-v1": ["▁não", "▁sim"],
"unicamp-dl/ptt5-base-en-pt-msmarco-10k-v1": ["▁não", "▁sim"],
"unicamp-dl/mt5-3B-mmarco-en-pt": ["▁", "▁true"],
"unicamp-dl/mt5-13b-mmarco-100k": ["▁", "▁true"],
}
[docs]
@passage_reranker_node
def monot5(
queries: List[str],
contents_list: List[List[str]],
scores_list: List[List[float]],
ids_list: List[List[str]],
top_k: int,
model_name: str = "castorini/monot5-3b-msmarco-10k",
batch: int = 64,
) -> Tuple[List[List[str]], List[List[str]], List[List[float]]]:
"""
Rerank a list of contents based on their relevance to a query using MonoT5.
:param queries: The list of queries to use for reranking
:param contents_list: The list of lists of contents to rerank
:param scores_list: The list of lists of scores retrieved from the initial ranking
:param ids_list: The list of lists of ids retrieved from the initial ranking
:param top_k: The number of passages to be retrieved
:param model_name: The name of the MonoT5 model to use for reranking
Note: default model name is 'castorini/monot5-3b-msmarco-10k'
If there is a '/' in the model name parameter,
when we create the file to store the results, the path will be twisted because of the '/'.
Therefore, it will be received as '_' instead of '/'.
:param batch: The number of queries to be processed in a batch
:return: tuple of lists containing the reranked contents, ids, and scores
"""
# replace '_' to '/'
if "_" in model_name:
model_name = model_name.replace("_", "/")
# Load the tokenizer and model from the pre-trained MonoT5 model
tokenizer = T5Tokenizer.from_pretrained(model_name)
model = T5ForConditionalGeneration.from_pretrained(model_name).eval()
# Retrieve the tokens used by the model to represent false and true predictions
token_false, token_true = prediction_tokens[model_name]
token_false_id = tokenizer.convert_tokens_to_ids(token_false)
token_true_id = tokenizer.convert_tokens_to_ids(token_true)
# Determine the device to run the model on (GPU if available, otherwise CPU)
device = "cuda" if torch.cuda.is_available() else "cpu"
model.to(device)
nested_list = [
list(map(lambda x: [f"Query: {query} Document: {x}"], content_list))
for query, content_list in zip(queries, contents_list)
]
rerank_scores = flatten_apply(
monot5_run_model,
nested_list,
model=model,
batch_size=batch,
tokenizer=tokenizer,
device=device,
token_false_id=token_false_id,
token_true_id=token_true_id,
)
df = pd.DataFrame(
{
"contents": contents_list,
"ids": ids_list,
"scores": rerank_scores,
}
)
df[["contents", "ids", "scores"]] = df.apply(
sort_by_scores, axis=1, result_type="expand"
)
results = select_top_k(df, ["contents", "ids", "scores"], top_k)
del model
del tokenizer
if torch.cuda.is_available():
torch.cuda.empty_cache()
return (
results["contents"].tolist(),
results["ids"].tolist(),
results["scores"].tolist(),
)
[docs]
def monot5_run_model(
input_texts,
model,
batch_size: int,
tokenizer,
device,
token_false_id,
token_true_id,
):
batch_input_texts = make_batch(input_texts, batch_size)
results = []
for batch_texts in tqdm(batch_input_texts):
flattened_batch_texts = list(chain.from_iterable(batch_texts))
input_encodings = tokenizer(
flattened_batch_texts,
padding=True,
truncation=True,
max_length=512,
return_tensors="pt",
).to(device)
with torch.no_grad():
outputs = model.generate(
input_ids=input_encodings["input_ids"],
attention_mask=input_encodings["attention_mask"],
output_scores=True,
return_dict_in_generate=True,
)
# Extract logits for the 'false' and 'true' tokens from the model's output
logits = outputs.scores[-1][:, [token_false_id, token_true_id]]
# Calculate the softmax probability of the 'true' token
probs = torch.nn.functional.softmax(logits, dim=-1)[:, 1]
results.extend(probs.tolist())
return results