AutoRAG 0.3.22__tar.gz → 0.3.24__tar.gz

{autorag-0.3.22 → autorag-0.3.24}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: AutoRAG
-Version: 0.3.22
+Version: 0.3.24
 Summary: Automatically Evaluate RAG pipelines with your own data. Find optimal structure for new RAG product.
 Project-URL: Homepage, https://github.com/Marker-Inc-Korea/AutoRAG
 Author-email: Marker-Inc <vkehfdl1@gmail.com>
@@ -217,8 +217,8 @@ Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
 Classifier: Topic :: Software Development :: Libraries
 Classifier: Topic :: Software Development :: Libraries :: Python Modules
 Requires-Python: <3.13,>=3.10
-Requires-Dist: aiohttp>=3.12.13
-Requires-Dist: banks>=2.1.2
+Requires-Dist: aiohttp<3.14,>=3.12.13
+Requires-Dist: banks>=2.4.2
 Requires-Dist: chromadb>=1.0.0
 Requires-Dist: click>=8.2.1
 Requires-Dist: cohere>=5.18.0
@@ -228,19 +228,19 @@ Requires-Dist: emoji>=2.14.1
 Requires-Dist: evaluate>=0.4.4
 Requires-Dist: fastapi>=0.115.13
 Requires-Dist: fastparquet>=2024.11.0
-Requires-Dist: gradio>=5.34.2
-Requires-Dist: grpcio-health-checking<1.68.0,>=1.66.2
-Requires-Dist: grpcio-status<1.68.0,>=1.66.2
-Requires-Dist: grpcio-tools<1.68.0,>=1.66.2
-Requires-Dist: grpcio<1.68.0,>=1.66.2
+Requires-Dist: gradio>=6.15.0
+Requires-Dist: grpcio-health-checking<1.83.0,>=1.66.2
+Requires-Dist: grpcio-status<1.83.0,>=1.66.2
+Requires-Dist: grpcio-tools<1.83.0,>=1.66.2
+Requires-Dist: grpcio<1.83.0,>=1.66.2
 Requires-Dist: ipykernel>=6.29.5
 Requires-Dist: ipywidgets-bokeh>=1.6.0
 Requires-Dist: ipywidgets>=8.1.7
 Requires-Dist: langchain-community>=0.4.1
-Requires-Dist: langchain-core>=1.2.23
-Requires-Dist: langchain-text-splitters>=1.1.1
-Requires-Dist: langchain>=1.2.14
-Requires-Dist: langsmith>=0.7.23
+Requires-Dist: langchain-core>=1.3.3
+Requires-Dist: langchain-text-splitters>=1.1.2
+Requires-Dist: langchain>=1.3.9
+Requires-Dist: langsmith>=0.8.18
 Requires-Dist: llama-index-core>=0.12.42
 Requires-Dist: llama-index-embeddings-ollama>=0.6.0
 Requires-Dist: llama-index-embeddings-openai-like>=0.1.1
@@ -252,14 +252,14 @@ Requires-Dist: llama-index-readers-file>=0.4.9
 Requires-Dist: llama-index-retrievers-bm25>=0.5.2
 Requires-Dist: llama-index>=0.12.42
 Requires-Dist: mixedbread-ai>=2.2.6
-Requires-Dist: numpy==1.26.4
+Requires-Dist: numpy>=2.2.6
 Requires-Dist: openai>=2.8.0
 Requires-Dist: pandas>=2.2.3
 Requires-Dist: panel>=1.7.1
 Requires-Dist: pinecone[grpc]
-Requires-Dist: pyarrow>=20.0.0
+Requires-Dist: pyarrow>=23.0.1
 Requires-Dist: pydantic>=2.9.2
-Requires-Dist: pymilvus>=2.6.0b0
+Requires-Dist: pymilvus<3,>=2.6.0b0
 Requires-Dist: pyngrok>=7.2.11
 Requires-Dist: pyyaml>=6.0.2
 Requires-Dist: qdrant-client>=1.12.1
@@ -270,7 +270,7 @@ Requires-Dist: rouge-score>=0.1.2
 Requires-Dist: sacrebleu>=2.5.1
 Requires-Dist: scikit-learn>=1.7.0
 Requires-Dist: seaborn>=0.13.2
-Requires-Dist: streamlit>=1.46.0
+Requires-Dist: streamlit>=1.54.0
 Requires-Dist: tiktoken>=0.9.0
 Requires-Dist: tokenlog>=0.0.3
 Requires-Dist: tqdm>=4.67.1
@@ -292,7 +292,7 @@ Requires-Dist: pdfminer-six>=20250506; extra == 'all'
 Requires-Dist: pdfplumber>=0.11.7; extra == 'all'
 Requires-Dist: peft>=0.15.2; extra == 'all'
 Requires-Dist: pymupdf>=1.26.1; extra == 'all'
-Requires-Dist: pypdf2==2.12.1; extra == 'all'
+Requires-Dist: pypdf2>=3.0.1; extra == 'all'
 Requires-Dist: sentence-transformers>=4.1.0; extra == 'all'
 Requires-Dist: sentencepiece>=0.2.0; extra == 'all'
 Requires-Dist: sudachidict-core; extra == 'all'
@@ -327,11 +327,14 @@ Requires-Dist: pdf2image>=1.17.0; extra == 'parse'
 Requires-Dist: pdfminer-six>=20250506; extra == 'parse'
 Requires-Dist: pdfplumber>=0.11.7; extra == 'parse'
 Requires-Dist: pymupdf>=1.26.1; extra == 'parse'
-Requires-Dist: pypdf2==2.12.1; extra == 'parse'
+Requires-Dist: pypdf2>=3.0.1; extra == 'parse'
 Requires-Dist: unstructured[pdf]>=0.17.2; extra == 'parse'
 Description-Content-Type: text/markdown
 
-# AutoRAG
+# AutoRAG (Legacy)
+
+> [!NOTE]
+> This is the original Python-based AutoRAG — the RAG AutoML tool for automatically finding an optimal RAG pipeline for your data. It now lives in the `legacy/` directory of the AutoRAG monorepo and is in **maintenance mode**: it continues to receive bug fixes, dependency updates, and PyPI releases (`pip install AutoRAG`), but new feature development happens in [AutoRAG 2.0](../README.md) at the repository root.
 
 RAG AutoML tool for automatically finding an optimal RAG pipeline for your data.
 

{autorag-0.3.22 → autorag-0.3.24}/README.md

@@ -1,4 +1,7 @@
-# AutoRAG
+# AutoRAG (Legacy)
+
+> [!NOTE]
+> This is the original Python-based AutoRAG — the RAG AutoML tool for automatically finding an optimal RAG pipeline for your data. It now lives in the `legacy/` directory of the AutoRAG monorepo and is in **maintenance mode**: it continues to receive bug fixes, dependency updates, and PyPI releases (`pip install AutoRAG`), but new feature development happens in [AutoRAG 2.0](../README.md) at the repository root.
 
 RAG AutoML tool for automatically finding an optimal RAG pipeline for your data.
 

{autorag-0.3.22 → autorag-0.3.24}/autorag/__init__.py

@@ -10,6 +10,11 @@ from llama_index.llms.openai import OpenAI
 from llama_index.llms.openai_like import OpenAILike
 from rich.logging import RichHandler
 
+try:
+	from llama_index.llms.azure_openai import AzureOpenAI
+except ImportError:
+	AzureOpenAI = None
+
 
 class LazyInit:
 	def __init__(self, factory, *args, **kwargs):
@@ -58,6 +63,9 @@ generator_models = {
 	"bedrock": AutoRAGBedrock,
 }
 
+if AzureOpenAI is not None:
+	generator_models["azure_openai"] = AzureOpenAI
+
 try:
 	from llama_index.llms.huggingface import HuggingFaceLLM
 	from llama_index.llms.ollama import Ollama

{autorag-0.3.22 → autorag-0.3.24}/autorag/data/parse/table_hybrid_parse.py

@@ -3,7 +3,7 @@ import tempfile
 from glob import glob
 from typing import List, Tuple, Dict
 
-from PyPDF2 import PdfFileReader, PdfFileWriter
+from PyPDF2 import PdfReader, PdfWriter
 import pdfplumber
 
 from autorag.support import get_support_modules
@@ -76,8 +76,8 @@ def save_page_by_table(data_path: str, text_dir: str, table_dir: str) -> Dict[st
 	file_name = os.path.basename(data_path).split(".pdf")[0]
 
 	with open(data_path, "rb") as input_data:
-		pdf_reader = PdfFileReader(input_data)
-		num_pages = pdf_reader.getNumPages()
+		pdf_reader = PdfReader(input_data)
+		num_pages = len(pdf_reader.pages)
 
 		path_map_dict = {}
 		for page_num in range(num_pages):
@@ -100,9 +100,9 @@ def _get_output_path(
 		return os.path.join(directory, f"{file_name}_page_{page_num + 1}.pdf")
 
 
-def _save_single_page(pdf_reader: PdfFileReader, page_num: int, output_pdf_path: str):
-	pdf_writer = PdfFileWriter()
-	pdf_writer.addPage(pdf_reader.getPage(page_num))
+def _save_single_page(pdf_reader: PdfReader, page_num: int, output_pdf_path: str):
+	pdf_writer = PdfWriter()
+	pdf_writer.add_page(pdf_reader.pages[page_num])
 
 	with open(output_pdf_path, "wb") as output_file:
 		pdf_writer.write(output_file)

{autorag-0.3.22 → autorag-0.3.24}/autorag/data/qa/evolve/openai_query_evolve.py

@@ -30,12 +30,12 @@ async def query_evolve_openai_base(
 	user_prompt = f"Question: {original_query}\nContext: {context_str}\nOutput: "
 	messages.append(ChatMessage(role=MessageRole.USER, content=user_prompt))
 
-	completion = await client.beta.chat.completions.parse(
+	completion = await client.responses.parse(
 		model=model_name,
-		messages=to_openai_message_dicts(messages),
-		response_format=Response,
+		input=to_openai_message_dicts(messages),
+		text_format=Response,
 	)
-	row["query"] = completion.choices[0].message.parsed.evolved_query
+	row["query"] = completion.output_parsed.evolved_query
 	return row
 
 
@@ -72,10 +72,10 @@ async def compress_ragas(
 	user_prompt = f"Question: {original_query}\nOutput: "
 	messages.append(ChatMessage(role=MessageRole.USER, content=user_prompt))
 
-	completion = await client.beta.chat.completions.parse(
+	completion = await client.responses.parse(
 		model=model_name,
-		messages=to_openai_message_dicts(messages),
-		response_format=Response,
+		input=to_openai_message_dicts(messages),
+		text_format=Response,
 	)
-	row["query"] = completion.choices[0].message.parsed.evolved_query
+	row["query"] = completion.output_parsed.evolved_query
 	return row

{autorag-0.3.22 → autorag-0.3.24}/autorag/data/qa/filter/dontknow.py

@@ -77,14 +77,14 @@ async def dontknow_filter_openai(
 	system_prompt: List[ChatMessage] = FILTER_PROMPT["dontknow_filter"][lang]
 	result = []
 	for gen_gt in row["generation_gt"]:
-		completion = await client.beta.chat.completions.parse(
+		completion = await client.responses.parse(
 			model=model_name,
-			messages=to_openai_message_dicts(
+			input=to_openai_message_dicts(
 				system_prompt + [ChatMessage(role=MessageRole.USER, content=gen_gt)]
 			),
-			response_format=Response,
+			text_format=Response,
 		)
-		result.append(completion.choices[0].message.parsed.is_dont_know)
+		result.append(completion.output_parsed.is_dont_know)
 	return not any(result)
 
 

{autorag-0.3.22 → autorag-0.3.24}/autorag/data/qa/filter/passage_dependency.py

@@ -37,9 +37,9 @@ async def passage_dependency_filter_openai(
 	assert "query" in row.keys(), "query column is not in the row."
 	system_prompt: List[ChatMessage] = FILTER_PROMPT["passage_dependency"][lang]
 	query = row["query"]
-	completion = await client.beta.chat.completions.parse(
+	completion = await client.responses.parse(
 		model=model_name,
-		messages=to_openai_message_dicts(
+		input=to_openai_message_dicts(
 			system_prompt
 			+ [
 				ChatMessage(
@@ -48,9 +48,9 @@ async def passage_dependency_filter_openai(
 				)
 			]
 		),
-		response_format=Response,
+		text_format=Response,
 	)
-	return not completion.choices[0].message.parsed.is_passage_dependent
+	return not completion.output_parsed.is_passage_dependent
 
 
 async def passage_dependency_filter_llama_index(

{autorag-0.3.22 → autorag-0.3.24}/autorag/data/qa/generation_gt/openai_gen_gt.py

@@ -25,16 +25,16 @@ async def make_gen_gt_openai(
 	passage_str = "\n".join(retrieval_gt_contents)
 	user_prompt = f"Text:\n<|text_start|>\n{passage_str}\n<|text_end|>\n\nQuestion:\n{query}\n\nAnswer:"
 
-	completion = await client.beta.chat.completions.parse(
+	completion = await client.responses.parse(
 		model=model_name,
-		messages=[
+		input=[
 			{"role": "system", "content": system_prompt},
 			{"role": "user", "content": user_prompt},
 		],
 		temperature=0.0,
-		response_format=Response,
+		text_format=Response,
 	)
-	response: Response = completion.choices[0].message.parsed
+	response: Response = completion.output_parsed
 	return add_gen_gt(row, response.answer)
 
 

{autorag-0.3.22 → autorag-0.3.24}/autorag/data/qa/query/openai_gen_query.py

@@ -27,12 +27,12 @@ async def query_gen_openai_base(
 	user_prompt = f"{context_str}\n\nGenerated Question from the Text:\n"
 	messages.append(ChatMessage(role=MessageRole.USER, content=user_prompt))
 
-	completion = await client.beta.chat.completions.parse(
+	completion = await client.responses.parse(
 		model=model_name,
-		messages=to_openai_message_dicts(messages),
-		response_format=Response,
+		input=to_openai_message_dicts(messages),
+		text_format=Response,
 	)
-	row["query"] = completion.choices[0].message.parsed.query
+	row["query"] = completion.output_parsed.query
 	return row
 
 
@@ -86,10 +86,10 @@ async def two_hop_incremental(
 	user_prompt = f"{context_str}\n\nGenerated two-hop Question from two Documents:\n"
 	messages.append(ChatMessage(role=MessageRole.USER, content=user_prompt))
 
-	completion = await client.beta.chat.completions.parse(
+	completion = await client.responses.parse(
 		model=model_name,
-		messages=to_openai_message_dicts(messages),
-		response_format=TwoHopIncrementalResponse,
+		input=to_openai_message_dicts(messages),
+		text_format=TwoHopIncrementalResponse,
 	)
-	row["query"] = completion.choices[0].message.parsed.two_hop_question
+	row["query"] = completion.output_parsed.two_hop_question
 	return row

{autorag-0.3.22 → autorag-0.3.24}/autorag/evaluation/metric/generation.py

@@ -19,6 +19,7 @@ from autorag.evaluation.metric.deepeval_prompt import FaithfulnessTemplate
 from autorag.evaluation.metric.util import (
 	autorag_metric_loop,
 	calculate_cosine_similarity,
+	remove_think_tags as _remove_think_tags,
 )
 from autorag.nodes.generator import OpenAILLM
 from autorag.nodes.generator.base import BaseGenerator
@@ -36,7 +37,11 @@ from autorag.utils.util import (
 
 @convert_inputs_to_list
 def huggingface_evaluate(
-	instance, key: str, metric_inputs: List[MetricInput], **kwargs
+	instance,
+	key: str,
+	metric_inputs: List[MetricInput],
+	remove_think_tags: bool = True,
+	**kwargs,
 ) -> List[float]:
 	"""
 	Compute huggingface evaluate metric.
@@ -49,6 +54,8 @@ def huggingface_evaluate(
 	"""
 
 	def compute_score(gt: List[str], pred: str) -> float:
+		if remove_think_tags:
+			pred = _remove_think_tags(pred)
 		return max(
 			list(
 				map(
@@ -189,6 +196,7 @@ def bleu(
 	max_ngram_order: int = 4,
 	trg_lang: str = "",
 	effective_order: bool = True,
+	remove_think_tags: bool = True,
 	**kwargs,
 ) -> List[float]:
 	"""
@@ -213,14 +221,15 @@ def bleu(
 		**kwargs,
 	)
 
-	result = list(
-		map(
-			lambda x: bleu_instance.sentence_score(
-				x.generated_texts, x.generation_gt
-			).score,
-			metric_inputs,
-		)
-	)
+	def compute_score(metric_input: MetricInput) -> float:
+		prediction = metric_input.generated_texts
+		if remove_think_tags:
+			prediction = _remove_think_tags(prediction)
+		return bleu_instance.sentence_score(
+			prediction, metric_input.generation_gt
+		).score
+
+	result = list(map(compute_score, metric_inputs))
 	return result
 
 
@@ -230,6 +239,7 @@ def meteor(
 	alpha: float = 0.9,
 	beta: float = 3.0,
 	gamma: float = 0.5,
+	remove_think_tags: bool = True,
 ) -> List[float]:
 	"""
 	Compute meteor score for generation.
@@ -250,6 +260,7 @@ def meteor(
 		meteor_instance,
 		"meteor",
 		metric_inputs,
+		remove_think_tags=remove_think_tags,
 		alpha=alpha,
 		beta=beta,
 		gamma=gamma,
@@ -265,6 +276,7 @@ def rouge(
 	use_stemmer: bool = False,
 	split_summaries: bool = False,
 	batch: int = os.cpu_count(),
+	remove_think_tags: bool = True,
 ) -> List[float]:
 	"""
 	Compute rouge score for generation.
@@ -295,6 +307,8 @@ def rouge(
 	)
 
 	async def compute(gt: List[str], pred: str) -> float:
+		if remove_think_tags:
+			pred = _remove_think_tags(pred)
 		return rouge_instance.score_multi(targets=gt, prediction=pred)[
 			rouge_type
 		].fmeasure
@@ -476,8 +490,11 @@ def bert_score(
 	lang: str = "en",
 	batch: int = 128,
 	n_threads: int = os.cpu_count(),
+	remove_think_tags: bool = True,
 ) -> List[float]:
 	generations = [metric_input.generated_texts for metric_input in metric_inputs]
+	if remove_think_tags:
+		generations = [_remove_think_tags(generation) for generation in generations]
 	generation_gt = [metric_input.generation_gt for metric_input in metric_inputs]
 	evaluator = evaluate.load("bertscore")
 

{autorag-0.3.22 → autorag-0.3.24}/autorag/evaluation/metric/util.py

@@ -1,4 +1,5 @@
 import functools
+import re
 from typing import List
 
 import numpy as np
@@ -6,6 +7,15 @@ import numpy as np
 from autorag.schema.metricinput import MetricInput
 from autorag.utils.util import convert_inputs_to_list
 
+_THINK_TAG_PATTERN = re.compile(
+	r"<think(?:ing)?>[\s\S]*?</think(?:ing)?>\s*", re.IGNORECASE
+)
+
+
+def remove_think_tags(text: str) -> str:
+	"""Remove complete ``<think>`` or ``<thinking>`` reasoning blocks."""
+	return _THINK_TAG_PATTERN.sub("", text)
+
 
 def calculate_cosine_similarity(a, b):
 	dot_product = np.dot(a, b)

{autorag-0.3.22 → autorag-0.3.24}/autorag/node_line.py

@@ -50,6 +50,10 @@ def run_node_line(
 			os.path.join(node_line_dir, node.node_type, "summary.csv")
 		)
 		best_node_row = node_summary_df.loc[node_summary_df["is_best"]]
+		if best_node_row.empty:
+			raise ValueError(
+				f"No best module found for node type {node.node_type} in summary file."
+			)
 		summary_lst.append(
 			{
 				"node_type": node.node_type,

{autorag-0.3.22 → autorag-0.3.24}/autorag/nodes/generator/minimax_llm.py

@@ -22,9 +22,10 @@ THINK_OPEN_TAG = "<think>"
 THINK_CLOSE_TAG = "</think>"
 
 MAX_TOKEN_DICT = {
-	"MiniMax-M2.7": 1_048_576,
-	"MiniMax-M2.7-highspeed": 1_048_576,
-	"MiniMax-M2.5": 1_048_576,
+	"MiniMax-M3": 1_000_000,
+	"MiniMax-M2.7": 204_800,
+	"MiniMax-M2.7-highspeed": 204_800,
+	"MiniMax-M2.5": 204_800,
 	"MiniMax-M2.5-highspeed": 204_800,
 }
 
@@ -38,7 +39,7 @@ class MiniMaxLLM(BaseGenerator):
 		It returns pseudo token ids and log probs since MiniMax does not support logprobs.
 
 		:param project_dir: The project directory.
-		:param llm: A model name for MiniMax. For example, ``MiniMax-M2.7`` or ``MiniMax-M2.5``.
+		:param llm: A model name for MiniMax. For example, ``MiniMax-M3`` or ``MiniMax-M2.7``.
 		:param batch: Batch size for API calls. Default is 16.
 		:param api_key: MiniMax API key. You can also set this to env variable ``MINIMAX_API_KEY``.
 		:param kwargs: Extra parameters for the MiniMax chat completion API.
@@ -172,7 +173,7 @@ class MiniMaxLLM(BaseGenerator):
 		)
 		answer = response.choices[0].message.content
 
-		# Strip thinking tags if present (MiniMax M2.5+ may include them)
+		# Strip thinking tags if present (MiniMax M2.7+ may include them)
 		if answer and THINK_OPEN_TAG in answer:
 			answer = strip_think_tags(answer)
 

{autorag-0.3.22 → autorag-0.3.24}/autorag/nodes/generator/openai_llm.py

@@ -1,5 +1,6 @@
 import logging
-from typing import List, Tuple, Union, Dict
+import re
+from typing import List, Optional, Tuple, Union, Dict
 
 import pandas as pd
 import tiktoken
@@ -16,16 +17,36 @@ from autorag.utils.util import (
 
 logger = logging.getLogger("AutoRAG")
 
+GPT_5_LONG_CONTEXT = (
+	1_050_000  # gpt-5.4 / gpt-5.5 / gpt-5.6 families share a 1.05M context window
+)
+
+# gpt-5 family prefixes that support the 1.05M context window.
+# Used as a fallback for dated snapshots (e.g. gpt-5.6-sol-2026-07-09)
+# that are not listed explicitly in MAX_TOKEN_DICT.
+GPT_5_LONG_CONTEXT_PREFIXES = ("gpt-5.4", "gpt-5.5", "gpt-5.6")
+
+# Only models still served by the OpenAI API are listed. Models retired per
+# https://developers.openai.com/api/docs/deprecations (gpt-5, gpt-5.1,
+# gpt-5-mini/nano, chatgpt-4o-latest, o1-preview, o1-mini, gpt-4-32k,
+# gpt-4 vision/turbo previews, gpt-3.5-turbo-0613/16k snapshots, ...)
+# are intentionally excluded so they fail fast with a clear error.
 MAX_TOKEN_DICT = {  # model name : token limit
-	"gpt-5.1-2025-11-13": 272_000,
-	"gpt-5.1": 272_000,
-	"gpt-5": 272_000,
-	"gpt-5-pro": 272_000,
-	"gpt-5-2025-08-07": 272_000,
-	"gpt-5-chat-latest": 272_000,
-	"gpt-5-mini-2025-08-07": 272_000,
-	"gpt-5-mini": 272_000,
-	"gpt-5-nano-2025-08-07": 272_000,
+	# gpt-5.6 family (July 2026) - 1.05M context window
+	"gpt-5.6": GPT_5_LONG_CONTEXT,  # alias routing to gpt-5.6-sol
+	"gpt-5.6-sol": GPT_5_LONG_CONTEXT,
+	"gpt-5.6-terra": GPT_5_LONG_CONTEXT,
+	"gpt-5.6-luna": GPT_5_LONG_CONTEXT,
+	"gpt-5.6-pro": GPT_5_LONG_CONTEXT,
+	# gpt-5.5 (April 2026)
+	"gpt-5.5": GPT_5_LONG_CONTEXT,
+	"gpt-5.5-pro": GPT_5_LONG_CONTEXT,
+	"gpt-5.5-chat-latest": GPT_5_LONG_CONTEXT,
+	# gpt-5.4
+	"gpt-5.4": GPT_5_LONG_CONTEXT,
+	"gpt-5.4-pro": GPT_5_LONG_CONTEXT,
+	"gpt-5.4-mini": GPT_5_LONG_CONTEXT,
+	"gpt-5.4-nano": GPT_5_LONG_CONTEXT,
 	"gpt-4.1": 1_000_000,
 	"gpt-4.1-2025-04-14": 1_000_000,
 	"gpt-4.1-mini": 1_047_576,
@@ -33,10 +54,6 @@ MAX_TOKEN_DICT = {  # model name : token limit
 	"gpt-4.1-nano": 1_000_000,
 	"gpt-4.1-nano-2025-04-14": 1_000_000,
 	"o1": 200_000,
-	"o1-preview": 128_000,
-	"o1-preview-2024-09-12": 128_000,
-	"o1-mini": 128_000,
-	"o1-mini-2024-09-12": 128_000,
 	"o1-pro": 200_000,
 	"o1-pro-2025-03-19": 200_000,
 	"o3": 200_000,
@@ -49,28 +66,35 @@ MAX_TOKEN_DICT = {  # model name : token limit
 	"gpt-4o": 128_000,
 	"gpt-4o-2024-08-06": 128_000,
 	"gpt-4o-2024-05-13": 128_000,
-	"chatgpt-4o-latest": 128_000,
 	"gpt-4-turbo": 128_000,
 	"gpt-4-turbo-2024-04-09": 128_000,
-	"gpt-4-turbo-preview": 128_000,
-	"gpt-4-0125-preview": 128_000,
-	"gpt-4-1106-preview": 128_000,
-	"gpt-4-vision-preview": 128_000,
-	"gpt-4-1106-vision-preview": 128_000,
 	"gpt-4": 8_192,
 	"gpt-4-0613": 8_192,
-	"gpt-4-32k": 32_768,
-	"gpt-4-32k-0613": 32_768,
 	"gpt-3.5-turbo-0125": 16_385,
 	"gpt-3.5-turbo": 16_385,
 	"gpt-3.5-turbo-1106": 16_385,
 	"gpt-3.5-turbo-instruct": 4_096,
-	"gpt-3.5-turbo-16k": 16_385,
-	"gpt-3.5-turbo-0613": 4_096,
-	"gpt-3.5-turbo-16k-0613": 16_385,
 }
 
 
+def get_max_token_size(llm: str) -> Optional[int]:
+	"""Resolve the context window size for an OpenAI model name.
+
+	Exact matches in MAX_TOKEN_DICT win first. Dated snapshots
+	(e.g. ``gpt-5.6-sol-2026-07-09``) fall back to their base model entry,
+	and unknown gpt-5 family variants fall back to the family context window
+	so that newly released snapshots keep working without a code change.
+	"""
+	if llm in MAX_TOKEN_DICT:
+		return MAX_TOKEN_DICT[llm]
+	base = re.sub(r"-\d{4}-\d{2}-\d{2}$", "", llm)
+	if base in MAX_TOKEN_DICT:
+		return MAX_TOKEN_DICT[base]
+	if base.startswith(GPT_5_LONG_CONTEXT_PREFIXES):
+		return GPT_5_LONG_CONTEXT
+	return None
+
+
 class OpenAILLM(BaseGenerator):
 	def __init__(self, project_dir, llm: str, batch: int = 16, *args, **kwargs):
 		super().__init__(project_dir, llm, *args, **kwargs)
@@ -84,14 +108,13 @@ class OpenAILLM(BaseGenerator):
 		except KeyError:
 			self.tokenizer = tiktoken.get_encoding("o200k_base")
 
-		self.max_token_size = (
-			MAX_TOKEN_DICT.get(self.llm) - 7
-		)  # because of chat token usage
-		if self.max_token_size is None:
+		max_tokens = get_max_token_size(self.llm)
+		if max_tokens is None:
 			raise ValueError(
 				f"Model {self.llm} does not supported. "
 				f"Please select the model between {list(MAX_TOKEN_DICT.keys())}"
 			)
+		self.max_token_size = max_tokens - 7  # because of chat token usage
 
 	@result_to_dataframe(["generated_texts", "generated_tokens", "generated_log_probs"])
 	def pure(self, previous_result: pd.DataFrame, *args, **kwargs):
@@ -309,12 +332,6 @@ class OpenAILLM(BaseGenerator):
 	async def get_result_gpt_5(self, prompt: Union[str, List[dict]], **kwargs):
 		if not self.llm.startswith("gpt-5"):
 			raise ValueError("get_result_gpt_5 is only for gpt-5 models.")
-		api_key = getattr(self.client, "api_key", None)
-		if isinstance(api_key, str) and api_key.startswith("mock_"):
-			answer = "Why not"
-			tokens = self.tokenizer.encode(answer, allowed_special="all")
-			pseudo_log_probs = [0.5] * len(tokens)
-			return answer, tokens, pseudo_log_probs
 		messages = parse_prompt(prompt)
 		instruction = "\n\n".join(
 			[msg["content"] for msg in messages if msg["role"] == "system"]

{autorag-0.3.22 → autorag-0.3.24}/autorag/nodes/passagereranker/nvidia.py

@@ -85,6 +85,9 @@ class NvidiaReranker(BasePassageReranker):
 				f"len(queries)={len(queries)}, len(contents_list)={len(contents_list)}, len(ids_list)={len(ids_list)}."
 			)
 
+		if not queries:
+			return [], [], []
+
 		tasks = [
 			nvidia_rerank_pure(
 				self.session,

{autorag-0.3.22 → autorag-0.3.24}/autorag/support.py

@@ -131,6 +131,8 @@ def get_support_modules(module_name: str) -> Callable:
 		),
 		"flashrank_reranker": ("autorag.nodes.passagereranker", "FlashRankReranker"),
 		"FlashRankReranker": ("autorag.nodes.passagereranker", "FlashRankReranker"),
+		"nvidia_reranker": ("autorag.nodes.passagereranker", "NvidiaReranker"),
+		"NvidiaReranker": ("autorag.nodes.passagereranker", "NvidiaReranker"),
 		# passage_filter
 		"pass_passage_filter": ("autorag.nodes.passagefilter", "PassPassageFilter"),
 		"similarity_threshold_cutoff": (

{autorag-0.3.22 → autorag-0.3.24}/autorag/utils/util.py

@@ -302,7 +302,23 @@ async def process_batch(tasks, batch_size: int = 64) -> List[Any]:
 
 	for i in range(0, len(tasks), batch_size):
 		batch = tasks[i : i + batch_size]
-		batch_results = await asyncio.gather(*batch)
+		batch_results = await asyncio.gather(*batch, return_exceptions=True)
+
+		first_exception = None
+		for offset, item in enumerate(batch_results):
+			if isinstance(item, BaseException):
+				logger.error(
+					"process_batch task %d (batch offset %d) raised %s: %s",
+					i + offset,
+					offset,
+					type(item).__name__,
+					item,
+				)
+				if first_exception is None:
+					first_exception = item
+		if first_exception is not None:
+			raise first_exception
+
 		results.extend(batch_results)
 
 	return results