Agent-ModernColBERT

State-of-the-Art ColBERT Model for Agentic Retrieval

Tl;Dr

A few weeks ago, we evaluated Reason-ModernColBERT, a 150M late-interaction model trained on ReasonIR data nearly solved BrowseComp-Plus, reaching 87.56% accuracy with GPT-5 (a +7.59 absolute jump over the previous SOTA) while topping recall and calibration error, while not being trained for agentic retrieval at all (and being one year old). We now present Agent-ModernColBERT, a model specifically fine-tuned for agentic retrieval using the AgentIR dataset released alongside AgentIR. You can find the training boilerplate here. This very lightweight fine-tuning adds increase the performance of Reason-ModernColBERT by another 10%, which allows, when exposing the get_document function and the GPT-OSS-120B model, to beat the original GPT-5 + Qwen3-8B runs, while using a retriever model 54× smaller and an open source LLM.

How it works

Before issuing a query, deep research agents generate explicit reasoning traces describing what they're looking for and why. Conventional retrievers throw all of that away. The AgentIR paper introduced Reasoning-Aware Retrieval: instead of discarding those reasoning traces, concatenate them to the query you send to the retriever. To enable this, the authors generated trajectories from a deep-research agent and released them (the Tevatron/AgentIR-data dataset), then trained AgentIR-4B, a 4B dense model that beats much larger baselines including ReasonIR-8B and rerank pipelines.

BrowseComp-Plus Results

The get_document column indicates whether the LLM was exposed to the possibility of using the get_document function that allows, in addition to the usual search function that returns a snippet of the top-5 documents for a given query (the beginning of the document), to request the full content of a document using its id. It allows the model to get more information when needed, but can also add noise to the process if the model reads full documents that are irrelevant, and thus require a precise retrieval signal.

A few things worth highlighting:

• With the get_document function exposed, Agent-ModernColBERT + GPT-OSS-120B reaches 72.53, higher than the original GPT-5 + Qwen3-8B setup (which doesn't expose get_document).

- Agent-ModernColBERT is **competitive with the single-vector AgentIR-4B** despite being **26× smaller** and trained on the exact same data, direct evidence that late interaction is the right inductive bias for reasoning-trace-augmented queries.

• We use the same prompt as AgentIR to keep the comparison clean (down to a wrongly-escaped line break, for full reproducibility). The prompt has a small but positive effect, as discussed in the next section.

Query format (⚠️ important)

Agent-ModernColBERT was trained on the AgentIR query format, which prepends an instruction and concatenates the agent's reasoning trace with the search query. Documents are encoded as plain text, no prefix.

The exact query layout is:

<INSTRUCT_PREFIX>Reasoning: <reasoning>\n\nQuery: <search_query>

A few non-obvious details worth flagging:

• The instruct prefix itself ends with a literal \nQuery: (a backslash followed by n, 2 characters — not a newline). This is inherited byte-for-byte from the AgentIR training script and the model expects it.
• As a consequence, the word Query: appears twice in every query — once at the end of the prefix, once as a real label before the search query.
• If you don't have a reasoning trace, pass the literal string "Empty" (this matches the AgentIR agent loop).

Helper:

INSTRUCT_PREFIX = (
    "Instruct: Given a user's reasoning followed by a web search query, "
    "retrieve relevant passages that answer the query while incorporating the user's reasoning"
    "\\nQuery:"  # literal backslash-n, NOT a newline — keep as-is
)

def format_query(reasoning: str | None, search_query: str) -> str:
    r = reasoning if reasoning else "Empty"
    return f"{INSTRUCT_PREFIX}Reasoning: {r}\n\nQuery: {search_query}"

Then encode normally with is_query=True for queries and is_query=False for documents:

from pylate import models

model = models.ColBERT(model_name_or_path="lightonai/Agent-ModernColBERT")

queries = [
    format_query(
        reasoning="The user wants a code-level explanation of PLAID, so passages "
                  "with implementation details should be prioritized.",
        search_query="What does the PLAID centroid scoring step do?",
    ),
]
documents = [
    "PLAID first scores documents against quantized centroids of the ColBERT "
    "token embeddings, then re-ranks a short candidate list with the full "
    "late-interaction score.",
]

query_embs = model.encode(queries, is_query=True)
doc_embs   = model.encode(documents, is_query=False)

If you rather want to use a simpler version without the prefix (just reasoning + query), you can use the no-prefix version, at the cost of slighlty lower performance:

Caveats on the get_document setup

get_document is a net improvement in most cases, but the gains depend on both the retrieval model and the agent. The very strong numbers above are not a given for every combination. We tried adding more get_document data points, but were not able to reproduce the original AgentIR results exactly (small deltas remain after fixing environments). We therefore omit those points from the comparison: although they look favorable to Agent-ModernColBERT, we can't yet rule out GPT-OSS-side errors.

Multi-vector embedding model based on lightonai/GTE-ModernColBERT-v1

This is a multi-vector (ColBERT-style late interaction) embedding model finetuned from lightonai/GTE-ModernColBERT-v1 on the agent_ir-data dataset. It maps sentences & paragraphs to sequences of 128-dimensional dense vectors and can be used for semantic textual similarity using the MaxSim operator.

Model Details

Model Description

• Model Type: Multi-vector embedding model
• Base model: lightonai/GTE-ModernColBERT-v1
• Document Length: 4096 tokens
• Query Length: 8192 tokens
• Output Dimensionality: 128 tokens
• Similarity Function: MaxSim
• Training Dataset:
  • agent_ir-data
• Language: en
• License: Apache 2.0

Model Sources

• Documentation: PyLate Documentation
• Repository: PyLate on GitHub
• Hugging Face: PyLate models on Hugging Face

Full Model Architecture

ColBERT(
  (0): Transformer({'max_seq_length': 8191, 'do_lower_case': False, 'architecture': 'ModernBertModel'})
  (1): Dense({'in_features': 768, 'out_features': 128, 'bias': False, 'activation_function': 'torch.nn.modules.linear.Identity', 'use_residual': False})
)

Usage

Sentence Transformers

This model can be used with Sentence Transformers as a multi-vector (ColBERT-style late interaction) retriever via the MultiVectorEncoder:

pip install "sentence-transformers>=6.0.0"

from sentence_transformers import MultiVectorEncoder

model = MultiVectorEncoder("lightonai/Agent-ModernColBERT")

query = "Which planet is known as the Red Planet?"
documents = [
    "Venus is often called Earth's twin because of its similar size and proximity.",
    "Mars, known for its reddish appearance, is often referred to as the Red Planet.",
    "Jupiter, the largest planet in our solar system, has a prominent red spot.",
    "Saturn, famous for its rings, is sometimes mistaken for the Red Planet.",
]

query_embeddings = model.encode_query(query)
document_embeddings = model.encode_document(documents)
print(query_embeddings.shape, document_embeddings[0].shape)
# (12, 128) (18, 128)

# MaxSim late-interaction scoring (higher is more relevant)
scores = model.similarity(query_embeddings, document_embeddings)
print(scores)
# tensor([[10.4239, 11.1039, 10.6760, 10.9058]])

PyLate

First install the PyLate library:

pip install -U pylate

Retrieval

Use this model with PyLate to index and retrieve documents. The index uses FastPLAID for efficient similarity search.

Indexing documents

Load the ColBERT model and initialize the PLAID index, then encode and index your documents:

from pylate import indexes, models, retrieve

# Step 1: Load the ColBERT model
model = models.ColBERT(
    model_name_or_path="pylate_model_id",
)

# Step 2: Initialize the PLAID index
index = indexes.PLAID(
    index_folder="pylate-index",
    index_name="index",
    override=True,  # This overwrites the existing index if any
)

# Step 3: Encode the documents
documents_ids = ["1", "2", "3"]
documents = ["document 1 text", "document 2 text", "document 3 text"]

documents_embeddings = model.encode(
    documents,
    batch_size=32,
    is_query=False,  # Ensure that it is set to False to indicate that these are documents, not queries
    show_progress_bar=True,
)

# Step 4: Add document embeddings to the index by providing embeddings and corresponding ids
index.add_documents(
    documents_ids=documents_ids,
    documents_embeddings=documents_embeddings,
)

Note that you do not have to recreate the index and encode the documents every time. Once you have created an index and added the documents, you can re-use the index later by loading it:

# To load an index, simply instantiate it with the correct folder/name and without overriding it
index = indexes.PLAID(
    index_folder="pylate-index",
    index_name="index",
)

Retrieving top-k documents for queries

Once the documents are indexed, you can retrieve the top-k most relevant documents for a given set of queries. To do so, initialize the ColBERT retriever with the index you want to search in, encode the queries and then retrieve the top-k documents to get the top matches ids and relevance scores.

Note that queries must be wrapped with format_query() — see the Query format section above. Pass the agent's reasoning trace as the first argument, or None if you don't have one.

INSTRUCT_PREFIX = (
    "Instruct: Given a user's reasoning followed by a web search query, "
    "retrieve relevant passages that answer the query while incorporating the user's reasoning"
    "\\nQuery:"  # literal backslash-n, NOT a newline — keep as-is
)

def format_query(reasoning: str | None, search_query: str) -> str:
    r = reasoning if reasoning else "Empty"
    return f"{INSTRUCT_PREFIX}Reasoning: {r}\n\nQuery: {search_query}"
# Step 1: Initialize the ColBERT retriever
retriever = retrieve.ColBERT(index=index)

# Step 2: Build queries in the AgentIR format, then encode them
raw_queries = [
    (
        "The user wants a code-level explanation of PLAID, so passages with "
        "implementation details or pseudo-code should be prioritized.",
        "What does the PLAID centroid scoring step do?",
    ),
    (
        "The user is comparing late-interaction models to dense bi-encoders, "
        "likely with a focus on latency trade-offs.",
        "How does ColBERT compare to dense retrievers in terms of latency?",
    ),
]
queries = [format_query(reasoning, q) for reasoning, q in raw_queries]

queries_embeddings = model.encode(
    queries,
    batch_size=32,
    is_query=True,  # adds the [Q] marker + uses query_length
    show_progress_bar=True,
)

# Step 3: Retrieve top-k documents
scores = retriever.retrieve(
    queries_embeddings=queries_embeddings,
    k=10,  # Retrieve the top 10 matches for each query
)

Reranking

If you only want to use the ColBERT model to perform reranking on top of your first-stage retrieval pipeline without building an index, you can simply use the rank function and pass the queries and documents to rerank.

As with retrieval, queries must be wrapped with format_query() (see the Query format section). Documents are encoded as plain text — no prefix.

from pylate import rank, models
INSTRUCT_PREFIX = (
    "Instruct: Given a user's reasoning followed by a web search query, "
    "retrieve relevant passages that answer the query while incorporating the user's reasoning"
    "\\nQuery:"  # literal backslash-n, NOT a newline — keep as-is
)

def format_query(reasoning: str | None, search_query: str) -> str:
    r = reasoning if reasoning else "Empty"
    return f"{INSTRUCT_PREFIX}Reasoning: {r}\n\nQuery: {search_query}"

raw_queries = [
    (
        "The user wants to understand how MaxSim scoring works at a high level.",
        "What is late interaction in ColBERT?",
    ),
    (
        "The user is benchmarking retrievers and cares about index size.",
        "How large is a PLAID index compared to a dense FAISS index?",
    ),
]
queries = [format_query(reasoning, q) for reasoning, q in raw_queries]

documents = [
    ["document A", "document B"],
    ["document 1", "document C", "document B"],
]

documents_ids = [
    [1, 2],
    [1, 3, 2],
]

model = models.ColBERT(
    model_name_or_path="pylate_model_id",
)

queries_embeddings = model.encode(
    queries,
    is_query=True,  # adds the [Q] marker + uses query_length
)

documents_embeddings = model.encode(
    documents,
    is_query=False,  # adds the [D] marker + uses document_length
)

reranked_documents = rank.rerank(
    documents_ids=documents_ids,
    queries_embeddings=queries_embeddings,
    documents_embeddings=documents_embeddings,
)

Training Details

Training Dataset

agent_ir-data

• Dataset: agent_ir-data at 3f67a6a
• Size: 5,238 training samples
• Columns: query, positive, negative_0, negative_1, negative_2, negative_3, negative_4, negative_5, and negative_6
• Approximate statistics based on the first 1000 samples:

  	query	positive	negative_0	negative_1	negative_2	negative_3	negative_4	negative_5	negative_6
  type	string	string	string	string	string	string	string	string	string
  details	min: 46 tokens mean: 164.15 tokens max: 1179 tokens	min: 77 tokens mean: 1886.55 tokens max: 8192 tokens	min: 15 tokens mean: 1808.24 tokens max: 8192 tokens	min: 10 tokens mean: 1760.88 tokens max: 8192 tokens	min: 38 tokens mean: 1786.12 tokens max: 8192 tokens	min: 28 tokens mean: 1823.68 tokens max: 8192 tokens	min: 21 tokens mean: 1821.01 tokens max: 8192 tokens	min: 24 tokens mean: 1915.61 tokens max: 8192 tokens	min: 13 tokens mean: 1911.1 tokens max: 8192 tokens

• Samples:

  query	positive	negative_0	negative_1	negative_2	negative_3	negative_4	negative_5	negative_6
  Instruct: Given a user's reasoning followed by a web search query, retrieve relevant passages that answer the query while incorporating the user's reasoning\nQuery:Reasoning: Thus we have a clear answer: the manager who was replaced is A.J. Hinch. But to be thorough, we should confirm that his number was retired by the Tigers? Wait, but maybe the question incorrectly states "interim manager who had his number retired by the Detroit Tigers." Actually the interim manager is "Kirk Gibson", whose number might indeed be retired by the Tigers? But I'm not sure. Let's search for "Tigers retired #29" again with maybe a site like MLB.com. Query: "#29" "Detroit Tigers" "retired" "number"	Published Time: 2010-07-03T01:05:31.000Z New Diamondbacks Manager Gibson Brings Intense Attitude - The New York Times Skip to contentSkip to site indexSearch & Section Navigation Section Navigation SEARCH Baseball Subscribe for $1/weekLog in Tuesday, August 12, 2025 Today's Paper Subscribe for $1/week Baseball|Stepping In as Arizona Manager, Gibson Displays His Famously Fiery Attitude Again Share full article Advertisement SKIP ADVERTISEMENT Supported by SKIP ADVERTISEMENT Stepping In as Arizona Manager, Gibson Displays His Famously Fiery Attitude Again Share full article By Thomas Kaplan July 2, 2010 PHOENIX — Kirk Gibson is most famous for a home run he hit while hobbled. Now he will try to turn around a ball club that is also in bad shape. Gibson spent his first day as interim manager of the Arizona Diamondbacks on Friday after Manager A. J. Hinch and General Manager Josh Byrnes were fired the night before. The Diamondbacks were 31-48, last in the National League West, hea...	Tigers retire Hall of Famer Jim Leyland's No. 10 next to World Series winner Sparky Anderson on wall Tigers retire Hall of Famer Jim Leyland's No. 10 next to World Series winner Sparky Anderson on wall DETROIT (AP) — Jim Leyland's No. 10 was retired by the Detroit Tigers, putting the Hall of Fame manager's name and number in white on a brick wall next to World Series winner Sparky Anderson. "When I look out on that wall and see my name with the Tiger greats, it's hard to believe," Leyland said Saturday night during a pregame ceremony before Detroit played the Kansas City Royals. Leyland arrived for the on-field presentation after a slow ride in a white Corvette, giving him a chance to wave to fans from the foul pole in right to Detroit's dugout along the third base line. He was voted into baseball's Hall of Fame last December, two weeks shy of his 79th birthday and last month became the 23rd manager inducted. Leyland won 1,769 regular-season games over 22 seasons, including a 700-597 r...	Below are the retired numbers for every Major League club. On April 15, 1997, every team in MLB retired No. 42 in honor of Jackie Robinson. ATHLETICS 9: Reggie Jackson, RF 24: Rickey Henderson, LF 27: Catfish Hunter, RHP 34: Rollie Fingers, RHP 34: Dave Stewart, RHP 43: Dennis Eckersley, RHP • More on A's retired numbers > ANGELS 11: Jim Fregosi, SS 26: Gene Autry, owner 29: Rod Carew, 1B 30: Nolan Ryan, RHP 50: Jimmie Reese, coach • More on Angels' retired numbers > ASTROS 5: Jeff Bagwell, 1B 7: Craig Biggio, C/2B/OF 24: Jimmy Wynn, OF 25: Jose Cruz, OF 32: Jim Umbricht, RHP 33: Mike Scott, RHP 34: Nolan Ryan, RHP 40: Don Wilson, RHP 49: Larry Dierker, RHP • More on Astros' retired numbers > BLUE JAYS 12: Roberto Alomar, 2B 32: Roy Halladay, RHP • More on Blue Jays' retired numbers > BRAVES 3: Dale Murphy, OF 6: Bobby Cox, manager 10: Chipper Jones, 3B 21: Warren Spahn, LHP 29: John Smoltz, RHP 31: Greg Maddux, RHP 35: Phil Niekro, RHP 41: Eddie Mathews, 3B 44: Hank Aaron, OF 47: Tom ...	[adrotate banner="49″] On August 17, 1980 — At Tiger Stadium, Al Kaline becomes the first player in franchise history to have his uniform number retired. The Hall of Famer, who wore the number 6, roamed the outfield for Detroit from 1953 to 1974. Vintage Baseball HOT ON EBAY Card Collections ENDING SOON ON EBAY MOST WANTED ROOKIE CARDS VINTAGE SPORTS TICKETS Baseball Hall of Famers @ET-DC@eyJkeW5hbWljIjp0cnVlLCJjb250ZW50IjoicG9zdF90YWdzIiwic2V0dGluZ3MiOnsiYmVmb3JlIjoiTGVhcm4gTW9yZSBhYm91dCB0aGUgdGVhbXMsIHBsYXllcnMsIGJhbGwgcGFya3MgYW5kIGV2ZW50cyB0aGF0IGhhcHBlbmVkIG9uIHRoaXMgZGF0ZSBpbiBoaXN0b3J5IC0gLSAtIC0gLSAtIC0gIiwiYWZ0ZXIiOiIiLCJsaW5rX3RvX3Rlcm1fcGFnZSI6Im9uIiwic2VwYXJhdG9yIjoiIHwgIiwiY2F0ZWdvcnlfdHlwZSI6InBvc3RfdGFnIn19@	Detroit Tigers to retire former coach Jim Leyland's number at Comerica Park DETROIT (FOX 2) - The Detroit Tigers plan to retire former coach Jim Leyland's jersey number during a ceremony at Comerica Park this weekend when they play the Kansas City Royals. Leyland, who sported No. 10 during his long tenure with the Tigers, will be only the second manager in team history to have his number retired, with the other being Sparky Anderson. Leyland is one of three managers in Tigers history to win at least 700 games with Detroit and the only one to lead them to four postseason berths in his career. He worked as a manager for 22 seasons, spending the final eight in Detroit. He has the 18-most victories in Major League History. That includes a 700-597 record with the Tigers. The No. 10 will be permanently stenciled alongside other retired numbers at Comerica Park. "We are excited to welcome Jim Leyland and his family back to Detroit, and for his name and number to be forever displayed among the...	Tigers retire Hall of Fame manager Jim Leyland's number Royals @ TigersAugust 3, 2024 | 00:04:45 The Tigers retire their recently inducted Hall of Fame manager Jim Leyland's No. 10 ahead of the team's game against the Royals August 3, 2024 | 00:04:45 The Tigers retire their recently inducted Hall of Fame manager Jim Leyland's No. 10 ahead of the team's game against the Royals	Published Time: 2004-04-01T21:47:42Z Kirk Gibson - Wikipedia Jump to content [x] Main menu Main menu move to sidebar hide Navigation Main page Contents Current events Random article About Wikipedia Contact us Contribute Help Learn to edit Community portal Recent changes Upload file Special pages Search Search [x] Appearance Appearance move to sidebar hide Text Small Standard Large This page always uses small font size Width Standard Wide The content is as wide as possible for your browser window. Color (beta) Automatic Light Dark This page is always in light mode. Donate Create account Log in [x] Personal tools Donate Create account Log in Pages for logged out editors learn more Contributions Talk [x] Toggle the table of contents Contents move to sidebar hide (Top) 1 BiographyToggle Biography subsection 1.1 Early life and collegiate career 1.2 Detroit Tigers 1.3 Los Angeles Dodgers 1.3.1 1988 World Series 1.4 Later career 1.5 Career ...	'A wonderful day': Tigers to retire Hall of Fame manager Jim Leyland's number One of the primary drivers of Detroit's baseball renaissance in the mid-2000s will forever be enshrined on the outfield wall at Comerica Park. The Detroit Tigers on Monday announced former manager Jim Leyland will become the 10th person in franchise history to have his number retired. The Tigers will hold a ceremony to retire Leyland's No. 10 at Comerica Park on Aug. 3. In July, Leyland will be inducted into the Baseball Hall of Fame. "The Hall of Fame is an entire baseball career, and that's one thing, but to have your number retired with one team is something a little bit different," Leyland said on a Zoom call with media members. "Certainly, I would never get into choosing one over the other, but this is the highest honor you can get as an individual that plays on a particular team, so I'm accepting it that way. I'm truly humbled. I've been just in a daze all day, to be honest with you, but it's been a won...
  Instruct: Given a user's reasoning followed by a web search query, retrieve relevant passages that answer the query while incorporating the user's reasoning\nQuery:Reasoning: Scrolling further maybe find South American Championship winner. Let's search "2015 South American U23 Volleyball Championship Brazil champion". Query: "South American" U23 2015 volleyball champion	Jump to content Juma da Silva Add links From Wikipedia, the free encyclopedia Brazilian volleyball player (born 1993) | Juma Silva | | --- | | Juma Silva in 2017 | | | Personal information | | | Full name | Juma Fernandes da Silva | | Nationality | Brazilian | | Born | (1993-01-17) 17 January 1993 (age 32) Belém, Pará | | Height | 1.81 m (5 ft 11 in) | | Weight | 68 kg (150 lb) | | Spike | 295 cm (116 in) | | Block | 280 cm (110 in) | | Volleyball information | | | Position | Setter | | Current club | Sesc Flamengo | | Number | 5 | | National team | | | | | | --- | | 2014–2015 | Brazil U-23 | | | | Honours | | | --- | | | | | Women's Volleyball | | | | Representing Brazil | | | | FIVB U23 World Championship | | | | | 2015 Ankara | Team | | | Juma Fernandes da Silva (born (1993-01-17)17 January 1993) is a Brazilian volleyball player as a setter. Career [edit] She competed at the 2015 FIVB U23 World Championship, 2016 Montreux Volley Masters, and 2018 FIVB Volleyball Women's N...	The 2005 FIFA World Youth Championship was the 15th edition of the FIFA World Youth Championship. It took place in the Netherlands between 10 June and 2 July 2005. == Venues == Doetinchem Emmen Enschede De Vijverberg Univé Stadion Arke Stadion Capacity: 12,600 Capacity: 8,600 Capacity: 13,250 Kerkrade Tilburg Utrecht Parkstad Limburg Stadion Willem II Stadion Stadion Galgenwaard Capacity: 19,979 Capacity: 14,637 Capacity: 24,500 == Qualification == The following 24 teams qualified for the 2005 FIFA World Youth Championship. Host country the Netherlands did not have to qualify for the tournament. Confederation Qualifying Tournament ...	CareerTimeline Most Notable - Made his senior national team debut in 2018 - Has played professionally in France, Poland and Türkiye - Won the Turkish Championship with Halkbank Ankara in 2024 - Won the French Cup with Stade Poitevin Poitiers in 2020 - Won the 2018 NCAA Men's Volleyball National Championship with UCLA; Named to All-Tournament Team - Two-time AVCA First Team All-American at UCLA under U.S. Men's National Team head coach John Speraw - Represented U.S. on Men's Junior and Boys Youth National Teams Events - 2025 Volleyball Nations League Week Two - 2024 Olympic Games - Bronze - 2024 Volleyball Nations League Weeks Two and Three - 2023 Olympic Qualifier - Gold - 2023 NORCECA Continental Championship - Gold - 2023 Volleyball Nations League Final - Silver - 2023 Volleyball Nations League Weeks One and Three - 2019 FIVB Tokyo Men's Volleyball Qualification tournament - Gold - 2019 FIVB World Cup - Bronze - 2019 Volleyball Nationals League Final Round - Silver - 2018 Friendlies ...	Schedule Tabs Featured Athletes Athletes of the Month Bianca Guerrero Nominated for her outstanding performance in the 2025 NJCAA Beach Volleyball Championship. Athletes of the Month Aryanna Spainhower Nominated for her outstanding performance in the 2025 NJCAA Beach Volleyball Championship. Sponsors Recent news SCF Manatees Headlines May. 29 Softball Jun. 01 Women's Volleyball May. 15 Women's Beach Volleyball May. 08 Baseball May. 08 Baseball May. 06 Baseball	The women's doubles competition of the bowling events at the 2015 Pan American Games was held on July 22 and 23 at Planet Bowl (Pan Am Bowling Centre), due to naming rights the venue was known as the latter for the duration of the games. ==Schedule== All times are Eastern Standard Time (UTC-3). Date Time Round July 22, 2015 15:05 1–6 Games July 23, 2015 10:05 7–12 Games ==Qualification== A total of 14 countries qualified two bowlers each through various events. This is summarized below. Event Vacancies Qualified Bowlers per NOC Total Host nation 1 2 2 South American Game 4 2 8 Pan American Sports Festival 3 2 6 PABCON Women's Championship 4 2 8 Central American and Caribbean Games 2 2 4 TOTAL ...	The perennial South American doormats have struggled mightily since opening last year's Copa America with a pair of highly surprising results. Their quarterfinal finish in 2015 marked just the third time in 15 tries that Bolivia has reached the knockouts since the Copa America went to a group stage format back in 1975. It's worth noting they've experienced several rough defeats playing on US soil over the last 20 years. 3 Stars - Yasmani Duk (New York Cosmos/United States) – Some veteran strikers being moved out has made room for the 28-year-old Duk to have his big shot. The forward only grabbed his starting spot over the last few Bolivia games. He also scored on his home debut for the Cosmos to key a win. - Juan Carlos Arce (Bolivar/Bolivia) – The frequent Copa Libertadores menace will try to translate that success to South America's top international tournament. Carlos Arce can wiggle through a crowd on the dribble to drive the attack. - Martin Smedberg-Dalence (IFK Gothenburg/Sweden...	Spain can celebrate tonight after beating Romania 24-3 to be crowned 2015 Rugby Europe Under 19 champions and secure qualification for next year's World Rugby Under 20 Trophy in Zimbabwe. In a tight first half Spain dominated possession but were restricted to one unconverted try by full-back, Guillermo Domingues. Romania replied with a penalty from Ionel Melinte to make the score 3-5 to Spain at the break. Spain’s dominance grew in the second half. Their forwards often on the front foot were able to secure quicker ball in the second half and unleash a set of well drilled and pacy backs, leading to tries from Pacheco, Parry and Zabala to take the game out of the reach of Romania, sparking celebrations. Hosts Portugal finished third with a narrow 22-20 victory over the valiant Russians while, in the day’s early games, Belgium overcame a spirited Polish side and the Netherlands drew 22-22 with Germany, the Germans winning on tries scored (3 to 1). 2017 will see the introduction of a two-t...	2025 Women's Volleyball Roster # | Full Name | Pos. | Ht. | Yr. | Hometown | High School | Previous School | | |---|---|---|---|---|---|---|---|---| | 1 | L/DS | 5' 7'' | R-Fr. | Weston, Fla. | Mater Academy | Tennessee | || | 2 | L/DS | 5' 9'' | Fr. | Puyallup, Wash. | Puyallup | ||| | 3 | OH | 6' 2'' | Fr. | Belgrade, Serbia | Ivanka Gajie HS | ||| | 4 | S | 6' 1'' | Fr. | Fort Lauderdale, Fla. | Westminster Academy | ||| | 5 | S | 5' 10'' | Fr. | Corvallis, Ore. | Crescent Valley HS | ||| | 6 | S | 5' 9'' | Sr. | Independence, Ky. | St. Henry HS | Butler | || | 7 | OPP | 6' 2'' | Fr. | Santa Clarita, Calif. | Bishop Alemany HS | ||| | 8 | OPP/OH | 6' 1'' | Jr. | Cordoba, Argentina | EEM N3 de 10 | Michigan | || | 9 | MB | 6' 4'' | Fr. | Bainbridge Island, Wash. | Bainbridge HS | ||| | 10 | OH | 6' 4'' | Fr. | Cannes, France | Stanislas Private Secondary | ||| | 11 | OPP | 6' 0'' | So. | Glendora, Calif. | Ontario Christian HS | LMU | || | 13 | OH | 5' 11'' | Gr. | Seattle, Wash. | E...
  Instruct: Given a user's reasoning followed by a web search query, retrieve relevant passages that answer the query while incorporating the user's reasoning\nQuery:Reasoning: This is not directly helpful. Let's consider perhaps the novel "The New Nun" (Lady Alice: or, The New Nun) 1837, possibly dealing with religious and social issues? Or "Verschoyle: A Roman Catholic Tale of the Nineteenth Century". Not sure. Let's search for "Victims of Society land inheritance". Query: "The Victims of Society" land inheritance	Long before Jane Austen wrote Northanger Abbey and exposed Frederick Tilney as a profligate heir-at-law, Chief Justice Sir John Popham lashed out at the custom of primogeniture. As J. H. Baker and S.F.C. Milsom record in Sources of English Legal History: Private Law to 1750, Popham objected to laws of landed property that "forced willy-nilly a father to leave his heritage [to his eldest son]." The willy-nilliness of inheritance laws has created a nation of "disobedient and sensual" eldest sons, fumed Popham, "undutiful in demeanour . . . and in manner and conversation dissolute." Furthermore, Popham opined, "such laws are utterly contrary to the providence of God"1 (157). Popham's indictment in 1594 was an objection raised from time to time against primogeniture, the first law of landed property that governed future interests of landowners. Moreover, the system of primogeniture had been further complicated by such institutions as the entail and the strict settlement. All these laws of ...	THE STORY: The play begins by introducing the next generation of Robedauxs in the person of Horace's son, Horace Jr., who comes home from school to find that his maternal grandfather, Henry Vaughn, has died suddenly of a heart attack. From this point, two main storylines are traced in the play. One deals with young Horace's coming to terms with the concepts of life, death and familial relationships. He finds that his mother, Elizabeth, is expecting another baby; he listens to, and asks many questions about, the widely contrasting accounts of his two grandfathers; and he endures the fussing of his paternal grandmother, who's fearful that his love of reading will make him too introspective and "ruin him" for the real world. He must also come to grips with the drowning of Gertrude, a young black girl who was his confidante and friend. The other storyline concerns "Brother" Vaughn, Henry Vaughn's dissolute son, who, in the short time during which he has control, heads his late father's far...	About Project Gutenberg Contact Us History & Philosophy Kindle & eReaders Help Pages Offline Catalogs Donate Frequently Downloaded Main Categories Reading Lists Search Options Frequently Downloaded Main Categories Project Gutenberg 76,495 free eBooks 29 by Harriet Martineau Illustrations of political economy, Volume 2 (of 9) by Harriet Martineau "Illustrations of Political Economy, Volume 2 (of 9)" by Harriet Martineau is a collection of narratives focusing on themes of political economy written in the early 19th century. This volume includes three distinct tales: "Demerara," "Ella of Garveloich," and "Weal and Woe in Garveloich," each exploring complex societal issues and the economic realities of their characters. The likely main focus of the book is the exploration of slavery, its impacts on both the enslaved individuals and society at large, particularly within the context of Demerara. The opening of this volume introduces readers to "Demerara," where the story begins with Alfred...	Warning: Target URL returned error 403: Forbidden JSTOR: Access Check Access Check Our systems have detected unusual traffic activity from your network. Please complete this reCAPTCHA to demonstrate that it's you making the requests and not a robot. If you are having trouble seeing or completing this challenge, this page may help. If you continue to experience issues, you can contact JSTOR support. Block Reference: #446a59e6-89f7-11f0-b21d-fe8a83193c9b VID: # IP: 34.96.49.146 Date and time: Fri, 05 Sep 2025 01:25:54 GMT Go back to JSTOR ©2000-2025 ITHAKA. All Rights Reserved. JSTOR®, the JSTOR logo, JPASS®, and ITHAKA® are registered trademarks of ITHAKA.	Warning: Target URL returned error 403: Forbidden JSTOR: Access Check Access Check Our systems have detected unusual traffic activity from your network. Please complete this reCAPTCHA to demonstrate that it's you making the requests and not a robot. If you are having trouble seeing or completing this challenge, this page may help. If you continue to experience issues, you can contact JSTOR support. Block Reference: #446246ee-89f7-11f0-a854-da3de1a20e81 VID: # IP: 34.34.225.38 Date and time: Fri, 05 Sep 2025 01:25:54 GMT Go back to JSTOR ©2000-2025 ITHAKA. All Rights Reserved. JSTOR®, the JSTOR logo, JPASS®, and ITHAKA® are registered trademarks of ITHAKA.	Click any word in a definition or example to find the entry for that word Society has to be prepared to support its elderly people. The scheme aims to encourage the reintegration of prisoners into society. The protesters were drawn from a broad cross section of society. Good writing still has a place in contemporary society. The novels reflect the values of Victorian society. The question is whether we have sufficient resources to sustain an industrial society. Never forget that we live in a multicultural society. In today's affluent society people are becoming increasingly discontented. This is the British English definition of society. View American English definition of society.	Click any word in a definition or example to find the entry for that word Society has to be prepared to support its elderly people. The program aims to encourage the reintegration of prisoners into society. The protesters were drawn from a broad cross section of society. Good writing still has a place in contemporary society. The novels reflect the values of Victorian society. The question is whether we have sufficient resources to sustain an industrial society. Never forget that we live in a multicultural society. In today's affluent society people are becoming increasingly discontented. She joined the local history society and made some new friends. This is the American English definition of society. View British English definition of society.	NOTE: refers to whole nations of peoples, or sets of people that follow certain norms. We studied several hunting and gathering societies in our anthropology course. These programs are designed to help ex-convicts merge back into society. Her mother said that such things are not done in polite society. (adherents of a certain norm) It's a science fiction story about a futuristic society of intelligent cat-like creatures. (metaphoric extension) WordNet 3.0 Sense Numbers: 1 NOTE: usually implies a formally established club Last year he was president of a prestigious philosophical society. It's a society dedicated to the preservation of the town's parks and conservation lands. There are a number of literary societies based in New York. WordNet 3.0 Sense Numbers: 2 Commentary: SOCIETY[+event][+state][+social][+interaction][+pleasurable] fellowship the state of being with others She needs to get out more in society. (have social relations) He enjoyed the society of his friends. WordNet 3.0 ...

• Loss: pylate.losses.cached_contrastive.CachedContrastive

Training Hyperparameters

Non-Default Hyperparameters

• per_device_train_batch_size: 32
• per_device_eval_batch_size: 32
• learning_rate: 3e-06
• num_train_epochs: 2
• bf16: True
• dataloader_num_workers: 8
• accelerator_config: {'split_batches': True, 'dispatch_batches': None, 'even_batches': True, 'use_seedable_sampler': True, 'non_blocking': False, 'gradient_accumulation_kwargs': None}

All Hyperparameters

Click to expand

• overwrite_output_dir: False
• do_predict: False
• eval_strategy: no
• prediction_loss_only: True
• per_device_train_batch_size: 32
• per_device_eval_batch_size: 32
• per_gpu_train_batch_size: None
• per_gpu_eval_batch_size: None
• gradient_accumulation_steps: 1
• eval_accumulation_steps: None
• torch_empty_cache_steps: None
• learning_rate: 3e-06
• weight_decay: 0.0
• adam_beta1: 0.9
• adam_beta2: 0.999
• adam_epsilon: 1e-08
• max_grad_norm: 1.0
• num_train_epochs: 2
• max_steps: -1
• lr_scheduler_type: linear
• lr_scheduler_kwargs: {}
• warmup_ratio: 0.0
• warmup_steps: 0
• log_level: passive
• log_level_replica: warning
• log_on_each_node: True
• logging_nan_inf_filter: True
• save_safetensors: True
• save_on_each_node: False
• save_only_model: False
• restore_callback_states_from_checkpoint: False
• no_cuda: False
• use_cpu: False
• use_mps_device: False
• seed: 42
• data_seed: None
• jit_mode_eval: False
• bf16: True
• fp16: False
• fp16_opt_level: O1
• half_precision_backend: auto
• bf16_full_eval: False
• fp16_full_eval: False
• tf32: None
• local_rank: 6
• ddp_backend: None
• tpu_num_cores: None
• tpu_metrics_debug: False
• debug: []
• dataloader_drop_last: True
• dataloader_num_workers: 8
• dataloader_prefetch_factor: None
• past_index: -1
• disable_tqdm: False
• remove_unused_columns: True
• label_names: None
• load_best_model_at_end: False
• ignore_data_skip: False
• fsdp: []
• fsdp_min_num_params: 0
• fsdp_config: {'min_num_params': 0, 'xla': False, 'xla_fsdp_v2': False, 'xla_fsdp_grad_ckpt': False}
• fsdp_transformer_layer_cls_to_wrap: None
• accelerator_config: {'split_batches': True, 'dispatch_batches': None, 'even_batches': True, 'use_seedable_sampler': True, 'non_blocking': False, 'gradient_accumulation_kwargs': None}
• parallelism_config: None
• deepspeed: None
• label_smoothing_factor: 0.0
• optim: adamw_torch_fused
• optim_args: None
• adafactor: False
• group_by_length: False
• length_column_name: length
• project: huggingface
• trackio_space_id: trackio
• ddp_find_unused_parameters: None
• ddp_bucket_cap_mb: None
• ddp_broadcast_buffers: False
• dataloader_pin_memory: True
• dataloader_persistent_workers: False
• skip_memory_metrics: True
• use_legacy_prediction_loop: False
• push_to_hub: False
• resume_from_checkpoint: None
• hub_model_id: None
• hub_strategy: every_save
• hub_private_repo: None
• hub_always_push: False
• hub_revision: None
• gradient_checkpointing: False
• gradient_checkpointing_kwargs: None
• include_inputs_for_metrics: False
• include_for_metrics: []
• eval_do_concat_batches: True
• fp16_backend: auto
• push_to_hub_model_id: None
• push_to_hub_organization: None
• mp_parameters:
• auto_find_batch_size: False
• full_determinism: False
• torchdynamo: None
• ray_scope: last
• ddp_timeout: 1800
• torch_compile: False
• torch_compile_backend: None
• torch_compile_mode: None
• include_tokens_per_second: False
• include_num_input_tokens_seen: no
• neftune_noise_alpha: None
• optim_target_modules: None
• batch_eval_metrics: False
• eval_on_start: False
• use_liger_kernel: False
• liger_kernel_config: None
• eval_use_gather_object: False
• average_tokens_across_devices: True
• prompts: None
• batch_sampler: batch_sampler
• multi_dataset_batch_sampler: proportional
• router_mapping: {}
• learning_rate_mapping: {}

Training Logs

Click to expand

Framework Versions

• Python: 3.12.12
• Sentence Transformers: 5.1.1
• PyLate: 1.3.4
• Transformers: 4.57.3
• PyTorch: 2.9.0+cu128
• Accelerate: 1.12.0
• Datasets: 4.6.1
• Tokenizers: 0.22.2

Citation

BibTeX

Agent-ModernColBERT

@misc{Agent-ModernColBERT,
title  = {Agent-ModernColBERT},
author = {Chaffin, Antoine},
url    = {https://huggingface.co/lightonai/Agent-ModernColBERT},
year   = {2026}
}

AgentIR

@article{DBLP:journals/corr/abs-2603-04384,
  author       = {Zijian Chen and
                  Xueguang Ma and
                  Shengyao Zhuang and
                  Jimmy Lin and
                  Akari Asai and
                  Victor Zhong},
  title        = {AgentIR: Reasoning-Aware Retrieval for Deep Research Agents},
  journal      = {CoRR},
  volume       = {abs/2603.04384},
  year         = {2026},
  url          = {https://doi.org/10.48550/arXiv.2603.04384},
  doi          = {10.48550/ARXIV.2603.04384},
  eprinttype   = {arXiv},
  eprint       = {2603.04384},
  timestamp    = {Wed, 08 Apr 2026 10:55:23 +0200},
  biburl       = {https://dblp.org/rec/journals/corr/abs-2603-04384.bib},
  bibsource    = {dblp computer science bibliography, https://dblp.org}
}

Sentence Transformers

@inproceedings{reimers-2019-sentence-bert,
    title = "Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks",
    author = "Reimers, Nils and Gurevych, Iryna",
    booktitle = "Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing",
    month = "11",
    year = "2019",
    publisher = "Association for Computational Linguistics",
    url = "https://arxiv.org/abs/1908.10084"
}

PyLate

@inproceedings{DBLP:conf/cikm/ChaffinS25,
  author       = {Antoine Chaffin and
                  Rapha{"{e}}l Sourty},
  editor       = {Meeyoung Cha and
                  Chanyoung Park and
                  Noseong Park and
                  Carl Yang and
                  Senjuti Basu Roy and
                  Jessie Li and
                  Jaap Kamps and
                  Kijung Shin and
                  Bryan Hooi and
                  Lifang He},
  title        = {PyLate: Flexible Training and Retrieval for Late Interaction Models},
  booktitle    = {Proceedings of the 34th {ACM} International Conference on Information
                  and Knowledge Management, {CIKM} 2025, Seoul, Republic of Korea, November
                  10-14, 2025},
  pages        = {6334--6339},
  publisher    = {{ACM}},
  year         = {2025},
  url          = {https://github.com/lightonai/pylate},
  doi          = {10.1145/3746252.3761608},
}

CachedContrastive

@misc{gao2021scaling,
    title={Scaling Deep Contrastive Learning Batch Size under Memory Limited Setup},
    author={Luyu Gao and Yunyi Zhang and Jiawei Han and Jamie Callan},
    year={2021},
    eprint={2101.06983},
    archivePrefix={arXiv},
    primaryClass={cs.LG}
}