This is a Cross Encoder model finetuned from BAAI/bge-reranker-v2-m3 using the sentence-transformers library. It computes scores for pairs of texts, which can be used for text reranking and semantic search.

Model Details

Model Description

Model Type: Cross Encoder
Base model: BAAI/bge-reranker-v2-m3
Maximum Sequence Length: 2048 tokens
Number of Output Labels: 1 label

Model Sources

Documentation: Sentence Transformers Documentation
Documentation: Cross Encoder Documentation
Repository: Sentence Transformers on GitHub
Hugging Face: Cross Encoders on Hugging Face

Usage

Direct Usage (Sentence Transformers)

First install the Sentence Transformers library:

pip install -U sentence-transformers

Then you can load this model and run inference.

python

from sentence_transformers import CrossEncoder

# Download from the 🤗 Hub
model = CrossEncoder("cross_encoder_model_id")
# Get scores for pairs of texts
pairs = [
    ["For the goalie who had the most shutouts in 2010, what's his catching hand? the most shutouts refers to max(SHO); shootCatch = 'L' refers to lefthand; shootCatch = 'R' refers to righthand; shootCatch = 'null' or 'empty' means this player is good at both left and right hand", '/* Table: TeamsPost | tmID → Teams: team ID; year → Teams; lgID: league ID; G: Games; W: wins; L: loses; T: ties; GF: goals for; GA: goals against; PIM: penalty minutes; BenchMinor: Bench minors (minutes); PPG: Power play goals; PPC: Power play chances; SHA: Shorthanded goals against; PKG: Power play goals against; PKC: Penalty kill chances; SHF: Shorthanded goals for; on */\nCREATE TABLE TeamsPost\n(\n    year       INTEGER          not null,\n    lgID       TEXT, -- league ID\n    tmID       TEXT not null, -- team ID\n    G          INTEGER, -- Games\n    W          INTEGER, -- wins\n    L          INTEGER, -- loses\n    T          INTEGER, -- ties\n    GF         INTEGER, -- goals for\n    GA         INTEGER, -- goals against\n    PIM        TEXT, -- penalty minutes\n    BenchMinor TEXT, -- Bench minors (minutes)\n    PPG        TEXT, -- Power play goals\n    PPC        TEXT, -- Power play chances\n    SHA        TEXT, -- Shorthanded goals against\n    PKG        TEXT, -- Power play goals against\n    PKC        TEXT, -- Penalty kill chances\n    SHF        TEXT, -- Shorthanded goals for\n    primary key (year, tmID),\n    foreign key (year, tmID) references Teams (year, tmID)\n        on update cascade on delete cascade\n)\n-- Approximate rows: 927\n-- Sample values for column `year`: 1913, 1962, 2011\n-- Sample values for column `lgID`: NHA, NHL, PCHA\n-- Sample values for column `tmID`: MOC, TBS, MOW\n-- Sample values for column `G`: 2, 14, 26\n-- Sample values for column `W`: 0, 8, 16\n-- Sample values for column `L`: 0, 6, 11\n-- Sample values for column `T`: 0, 2, 4\n-- Sample values for column `GF`: 0, 45, 98\n-- Sample values for column `GA`: 0, 38, 91\n-- Sample values for column `PIM`: 59, 68, 87\n-- Sample values for column `BenchMinor`: 0, 6, 2\n-- Sample values for column `PPG`: 1, 5, 9\n-- Sample values for column `PPC`: 39, 27, 25\n-- Sample values for column `SHA`: 0, 1, 2\n-- Sample values for column `PKG`: 3, 8, 4\n-- Sample values for column `PKC`: 55, 45, 20\n-- Sample values for column `SHF`: 0, 1, 3\n-- Table profile: 927 rows\n-- Profile for `year`: 0% null; 97 distinct (high-cardinality); range 1913..2011, avg 1976.14\n-- Profile for `lgID`: 0% null; 5 distinct (categorical); len 3..4; top: NHL(91%), WHA(6%), PCHA(2%)\n-- Profile for `tmID`: 0% null; 77 distinct (high-cardinality); len 3..3\n-- Profile for `G`: 0% null; 25 distinct (high-cardinality); range 2..26, avg 8.81\n-- Profile for `W`: 0% null; 17 distinct (high-cardinality); range 0..16, avg 4.37\n-- Profile for `L`: 0% null; 12 distinct (high-cardinality); range 0..11, avg 4.37\n-- Profile for `T`: 0% null; 5 distinct (categorical); range 0..4, avg 0.08; top: 0(94%), 1(5%), 2(1%)\n-- Profile for `GF`: 0% null; 91 distinct (high-cardinality); range 0..98, avg 25.6\n-- Profile for `GA`: 0% null; 77 distinct (high-cardinality); range 0..91, avg 25.6\n-- Profile for `PIM`: 9% null; 342 distinct (high-cardinality); len 1..3\n-- Profile for `BenchMinor`: 38% null; 8 distinct (categorical); len 1..2; top: 0(47%), 2(28%), 4(15%)\n-- Profile for `PPG`: 31% null; 31 distinct (high-cardinality); len 1..2\n-- Profile for `PPC`: 37% null; 115 distinct (high-cardinality); len 1..3\n-- Profile for `SHA`: 35% null; 9 distinct (categorical); len 1..1; top: 0(41%), 1(31%), 2(15%)\n-- Profile for `PKG`: 31% null; 27 distinct (high-cardinality); len 1..2\n-- Profile for `PKC`: 37% null; 113 distinct (high-cardinality); len 1..3\n-- Profile for `SHF`: 31% null; 11 distinct (high-cardinality); len 1..2'],
    ["What is the percentage of act number 5 in Titus Andronicus? act number 5 refers to Act = 5; Titus Andronicus refers to Title = 'Titus Andronicus'; percentage = divide(sum(Act = 5), count(Act)) as percentage", '/* Table: state_sector_details | level: Level of institution; stateid: state id; control: Public, Private not-for-profit, Private for-profit; state: state name; counted_pct: counted percentage; awards_per_state_value: awards per state value; awards_per_natl_value: awards per national value; exp_award_state_value: expected award state value; exp_award_natl_value: expected award national value; state_appr_value: state appropriation value; references; state_post; schools_count; state_appr_rank; grad_rate_rank; awards_per_rank */\nCREATE TABLE "state_sector_details"\n(\n    stateid                INTEGER, -- state id\n    state                  TEXT -- state name\n        references institution_details (state),\n    state_post             TEXT,\n    level                  TEXT, -- Level of institution | Values: 4-year: bachelor\'s degree 2-year: associate\'s degree\n    control                TEXT, -- Values: Public, Private not-for-profit, Private for-profit\n    schools_count          INTEGER,\n    counted_pct            TEXT, -- counted percentage\n    awards_per_state_value TEXT, -- awards per state value\n    awards_per_natl_value  REAL, -- awards per national value\n    exp_award_state_value  TEXT, -- expected award state value\n    exp_award_natl_value   INTEGER, -- expected award national value\n    state_appr_value       TEXT, -- state appropriation value\n    state_appr_rank        TEXT,\n    grad_rate_rank         TEXT,\n    awards_per_rank        TEXT,\n    primary key (stateid, level, control)\n)\n-- Approximate rows: 312\n-- Sample values for column `stateid`: 0, 29, 56\n-- Sample values for column `state`: United States, Alabama, Alaska\n-- Sample values for column `state_post`: U.S., Ala., Alaska\n-- Sample values for column `level`: 2-year, 4-year\n-- Sample values for column `control`: Private for-profit, Private not-for-profit, Public\n-- Sample values for column `schools_count`: 0, 32, 1180\n-- Sample values for column `counted_pct`: NULL, 61, 62\n-- Sample values for column `awards_per_state_value`: NULL, 18.8, 17.8\n-- Sample values for column `awards_per_natl_value`: 16.5, 24.6, 32.8\n-- Sample values for column `exp_award_state_value`: NULL, 75743, 92268\n-- Sample values for column `exp_award_natl_value`: 24795, 38763, 101725\n-- Sample values for column `state_appr_value`: NULL, 290.68, 496.78\n-- Sample values for column `state_appr_rank`: NULL, 11, 2\n-- Sample values for column `grad_rate_rank`: 23, 18, 8\n-- Sample values for column `awards_per_rank`: NULL, 42, 49\n-- Table profile: 312 rows\n-- Profile for `stateid`: 0% null; 52 distinct (high-cardinality); range 0..56, avg 28.4\n-- Profile for `state`: 0% null; 52 distinct (high-cardinality); len 4..20\n-- Profile for `state_post`: 0% null; 52 distinct (high-cardinality); len 3..6\n-- Profile for `level`: 0% null; 2 distinct (categorical); len 6..6; top: 4-year(50%), 2-year(50%)\n-- Profile for `control`: 0% null; 3 distinct (categorical); len 6..22; top: Public(33%), Private not-for-profit(33%), Private for-profit(33%)\n-- Profile for `schools_count`: 0% null; 58 distinct (high-cardinality); range 0..1180, avg 24.35\n-- Profile for `counted_pct`: 0% null; 209 distinct (high-cardinality); len 1..4\n-- Profile for `awards_per_state_value`: 0% null; 161 distinct (high-cardinality); len 2..4\n-- Profile for `awards_per_natl_value`: 0% null; 6 distinct (categorical); range 16.5..32.8, avg 23.97; top: 32.8(17%), 25.9(17%), 24.6(17%)\n-- Profile for `exp_award_state_value`: 0% null; 265 distinct (high-cardinality); len 4..6\n-- Profile for `exp_award_natl_value`: 0% null; 6 distinct (categorical); range 24795..101725, avg 50668.17; top: 101725(17%), 66436(17%), 38763(17%)\n-- Profile for `state_appr_value`: 0% null; 50 distinct (high-cardinality); len 4..6\n-- Profile for `state_appr_rank`: 0% null; 50 distinct (high-cardinality); len 1..4\n-- Profile for `grad_rate_rank`: 0% null; 53 distinct (high-cardinality); len 1..4\n-- Profile for `awards_per_rank`: 0% null; 52 distinct (high-cardinality); len 1..4'],
    ['When did user 39115684 rate the movie "A Way of Life"? A Way of Life\' refers to movie_title; user 39115684 refers to userid = 39115684;\xa0 when the user rate refers to rating_timestamp_utc;', '/* Table: person | p_id: person ID; professor: whether the person is a professor; student: whether the person is a student; hasPosition: has position; inPhase: in phase; yearsInProgram: years in program */\nCREATE TABLE person\n(\n    p_id           INTEGER -- person ID\n        constraint person_pk\n            primary key,\n    professor      INTEGER, -- whether the person is a professor | Values: 0: professor 1: student\n    student        INTEGER, -- whether the person is a student | Values: 0: professor 1: student\n    hasPosition    TEXT, -- has position | Values: 0: the person is not a faculty member Common Sense evidence: faculty_aff: affiliated faculty faculty_eme: faculty employee\n    inPhase        TEXT, -- in phase | Values: 0: the person is not undergoing the phase of qualification\n    yearsInProgram TEXT -- years in program | Values: 0: the person is not in any programs Common Sense evidence: yearX means the person is on the Xth year of the program\n)\n-- Approximate rows: 278\n-- Sample values for column `p_id`: 3, 222, 435\n-- Sample values for column `professor`: 0, 1\n-- Sample values for column `student`: 0, 1\n-- Sample values for column `hasPosition`: 0, Faculty, Faculty_adj\n-- Sample values for column `inPhase`: 0, Post_Quals, Post_Generals\n-- Sample values for column `yearsInProgram`: 0, Year_2, Year_5\n-- Table profile: 278 rows\n-- Profile for `p_id`: 0% null; 278 distinct (unique/id); range 3..435, avg 221.06\n-- Profile for `professor`: 0% null; 2 distinct (categorical); range 0..1, avg 0.22; top: 0(78%), 1(22%)\n-- Profile for `student`: 0% null; 2 distinct (categorical); range 0..1, avg 0.78; top: 1(78%), 0(22%)\n-- Profile for `hasPosition`: 0% null; 5 distinct (categorical); len 1..11; top: 0(81%), Faculty(14%), Faculty_adj(2%)\n-- Profile for `inPhase`: 0% null; 4 distinct (categorical); len 1..13; top: 0(50%), Post_Generals(18%), Post_Quals(18%)\n-- Profile for `yearsInProgram`: 0% null; 12 distinct (high-cardinality); len 1..7'],
    ["In complaints received in 2014, how many of them were submitted via call? in 2014 refers to Date received LIKE '2014%'; submitted via call refers to Submitted via = 'Phone'", '/* Table: client | client_id: client id; district_id → district: district id; social: social number; address_2: address 2; sex: sex of client; day: day of the birthday; month: month of the birthday; year: year when is born; first: first name; middle: middle name; last: last name; phone: phone number; address_1: address 1; state: state code; age: teenager: 13-19 adult: 19-65 elder: > 65; email: google email / account: @gamil.com microsoft email / account: xxx@outl...; city; zipcode */\nCREATE TABLE client\n(\n    client_id   TEXT -- client id\n            primary key,\n    sex         TEXT, -- sex of client\n    day         INTEGER, -- day of the birthday\n    month       INTEGER, -- month of the birthday\n    year        INTEGER, -- year when is born\n    age         INTEGER, -- Values: teenager: 13-19 adult: 19-65 elder: > 65\n    social      TEXT, -- social number | Values: ssn: us id number for each person\n    first       TEXT, -- first name\n    middle      TEXT, -- middle name\n    last        TEXT, -- last name\n    phone       TEXT, -- phone number\n    email       TEXT, -- Values: google email / account: @gamil.com microsoft email / account: xxx@outlook.com\n    address_1   TEXT, -- address 1\n    address_2   TEXT, -- address 2 | Values: entire address = (address_1, address_2)\n    city        TEXT,\n    state       TEXT, -- state code\n    zipcode     INTEGER,\n    district_id INTEGER, -- district id\n    foreign key (district_id) references district(district_id)\n)\n-- Approximate rows: 5369\n-- Sample values for column `client_id`: C00000001, C00000002, C00000003\n-- Sample values for column `sex`: Female, Male\n-- Sample values for column `day`: 1, 16, 31\n-- Sample values for column `month`: 1, 7, 12\n-- Sample values for column `year`: 1931, 1969, 2007\n-- Sample values for column `age`: 12, 50, 88\n-- Sample values for column `social`: 926-93-2157, 806-94-5725, 614-70-9100\n-- Sample values for column `first`: Emma, Noah, Olivia\n-- Sample values for column `middle`: Avaya, Everest, Brooklynne\n-- Sample values for column `last`: Smith, Thompson, Johnson\n-- Sample values for column `phone`: 367-171-6840, 212-423-7734, 212-425-6932\n-- Sample values for column `email`: emma.smith@gmail.com, noah.thompson@gmail.com, olivia.johnson@outlook.com\n-- Sample values for column `address_1`: 387 Wellington Ave., 75 W. Berkshire St., 36 Second St.\n-- Sample values for column `address_2`: Unit 1, Unit 6, Unit 80\n-- Sample values for column `city`: Albuquerque, New York City, Indianapolis\n-- Sample values for column `state`: NM, NY, IN\n-- Sample values for column `zipcode`: 1101, 44510, 49998\n-- Sample values for column `district_id`: 1, 39, 77\n-- Table profile: 5.4k rows\n-- Profile for `client_id`: 0% null; 5.4k distinct (unique/id); len 9..9\n-- Profile for `sex`: 0% null; 2 distinct (categorical); len 4..6; top: Male(51%), Female(49%)\n-- Profile for `day`: 0% null; 31 distinct (high-cardinality); range 1..31, avg 15.96\n-- Profile for `month`: 0% null; 12 distinct (high-cardinality); range 1..12, avg 6.52\n-- Profile for `year`: 0% null; 77 distinct (high-cardinality); range 1931..2007, avg 1973.2\n-- Profile for `age`: 0% null; 77 distinct (high-cardinality); range 12..88, avg 45.89\n-- Profile for `social`: 0% null; 5.4k distinct (unique/id); len 11..11\n-- Profile for `first`: 0% null; 2.9k distinct (high-cardinality); len 2..11\n-- Profile for `middle`: 0% null; 3.9k distinct (high-cardinality); len 2..12\n-- Profile for `last`: 0% null; 1.8k distinct (high-cardinality); len 2..13\n-- Profile for `phone`: 0% null; 5.4k distinct (unique/id); len 12..12\n-- Profile for `email`: 0% null; 5.4k distinct (unique/id); len 17..38\n-- Profile for `address_1`: 0% null; 5.4k distinct (unique/id); len 9..30\n-- Profile for `address_2`: 98% null; 24 distinct (high-cardinality); len 6..7\n-- Profile for `city`: 0% null; 75 distinct (high-cardinality); len 4..14\n-- Profile for `state`: 0% null; 51 distinct (high-cardinality); len 2..2\n-- Profile for `zipcode`: 0% null; 3.5k distinct (high-cardinality); range 1101..49998, avg 33506.31\n-- Profile for `district_id`: 0% null; 77 distinct (high-cardinality); range 1..77, avg 37.31'],
    ["Among the active businesses in Arizona, how many businesses work after 12PM? active businesses in Arizona refer to business_id where state = 'Arizona' and active = 'true'; work after 12PM refer to opening_time > '12PM';", '/* Table: position | positionID: position id; educationrequired: education required; positiontitle: position title; minsalary: minimum salary; maxsalary: maximum salary */\nCREATE TABLE position\n(\n    positionID        INTEGER -- position id\n        constraint position_pk\n            primary key,\n    positiontitle     TEXT, -- position title\n    educationrequired TEXT, -- education required | Values: Generally, more complex work requires more education year.\n    minsalary         TEXT, -- minimum salary\n    maxsalary         TEXT -- maximum salary\n)\n-- Approximate rows: 4\n-- Sample values for column `positionID`: 1, 3, 4\n-- Sample values for column `positiontitle`: Account Representative, Manager, Trainee\n-- Sample values for column `educationrequired`: 4 year degree, 2 year degree, 6 year degree\n-- Sample values for column `minsalary`: US$25,000.00, US$50,000.00, US$18,000.00\n-- Sample values for column `maxsalary`: US$75,000.00, US$150,000.00, US$25,000.00\n-- Table profile: 4 rows\n-- Profile for `positionID`: 0% null; 4 distinct (unique/id); range 1..4, avg 2.5; top: 1(25%), 2(25%), 3(25%)\n-- Profile for `positiontitle`: 0% null; 4 distinct (unique/id); len 7..22; top: Trainee(25%), Regional Manager(25%), Manager(25%)\n-- Profile for `educationrequired`: 0% null; 3 distinct (categorical); len 13..13; top: 4 year degree(50%), 6 year degree(25%), 2 year degree(25%)\n-- Profile for `minsalary`: 0% null; 4 distinct (unique/id); len 12..13; top: US$50,000.00(25%), US$25,000.00(25%), US$18,000.00(25%)\n-- Profile for `maxsalary`: 0% null; 4 distinct (unique/id); len 12..13; top: US$75,000.00(25%), US$250,000.00(25%), US$25,000.00(25%)'],
]
scores = model.predict(pairs)
print(scores.shape)
# (5,)

# Or rank different texts based on similarity to a single text
ranks = model.rank(
    "For the goalie who had the most shutouts in 2010, what's his catching hand? the most shutouts refers to max(SHO); shootCatch = 'L' refers to lefthand; shootCatch = 'R' refers to righthand; shootCatch = 'null' or 'empty' means this player is good at both left and right hand",
    [
        '/* Table: TeamsPost | tmID → Teams: team ID; year → Teams; lgID: league ID; G: Games; W: wins; L: loses; T: ties; GF: goals for; GA: goals against; PIM: penalty minutes; BenchMinor: Bench minors (minutes); PPG: Power play goals; PPC: Power play chances; SHA: Shorthanded goals against; PKG: Power play goals against; PKC: Penalty kill chances; SHF: Shorthanded goals for; on */\nCREATE TABLE TeamsPost\n(\n    year       INTEGER          not null,\n    lgID       TEXT, -- league ID\n    tmID       TEXT not null, -- team ID\n    G          INTEGER, -- Games\n    W          INTEGER, -- wins\n    L          INTEGER, -- loses\n    T          INTEGER, -- ties\n    GF         INTEGER, -- goals for\n    GA         INTEGER, -- goals against\n    PIM        TEXT, -- penalty minutes\n    BenchMinor TEXT, -- Bench minors (minutes)\n    PPG        TEXT, -- Power play goals\n    PPC        TEXT, -- Power play chances\n    SHA        TEXT, -- Shorthanded goals against\n    PKG        TEXT, -- Power play goals against\n    PKC        TEXT, -- Penalty kill chances\n    SHF        TEXT, -- Shorthanded goals for\n    primary key (year, tmID),\n    foreign key (year, tmID) references Teams (year, tmID)\n        on update cascade on delete cascade\n)\n-- Approximate rows: 927\n-- Sample values for column `year`: 1913, 1962, 2011\n-- Sample values for column `lgID`: NHA, NHL, PCHA\n-- Sample values for column `tmID`: MOC, TBS, MOW\n-- Sample values for column `G`: 2, 14, 26\n-- Sample values for column `W`: 0, 8, 16\n-- Sample values for column `L`: 0, 6, 11\n-- Sample values for column `T`: 0, 2, 4\n-- Sample values for column `GF`: 0, 45, 98\n-- Sample values for column `GA`: 0, 38, 91\n-- Sample values for column `PIM`: 59, 68, 87\n-- Sample values for column `BenchMinor`: 0, 6, 2\n-- Sample values for column `PPG`: 1, 5, 9\n-- Sample values for column `PPC`: 39, 27, 25\n-- Sample values for column `SHA`: 0, 1, 2\n-- Sample values for column `PKG`: 3, 8, 4\n-- Sample values for column `PKC`: 55, 45, 20\n-- Sample values for column `SHF`: 0, 1, 3\n-- Table profile: 927 rows\n-- Profile for `year`: 0% null; 97 distinct (high-cardinality); range 1913..2011, avg 1976.14\n-- Profile for `lgID`: 0% null; 5 distinct (categorical); len 3..4; top: NHL(91%), WHA(6%), PCHA(2%)\n-- Profile for `tmID`: 0% null; 77 distinct (high-cardinality); len 3..3\n-- Profile for `G`: 0% null; 25 distinct (high-cardinality); range 2..26, avg 8.81\n-- Profile for `W`: 0% null; 17 distinct (high-cardinality); range 0..16, avg 4.37\n-- Profile for `L`: 0% null; 12 distinct (high-cardinality); range 0..11, avg 4.37\n-- Profile for `T`: 0% null; 5 distinct (categorical); range 0..4, avg 0.08; top: 0(94%), 1(5%), 2(1%)\n-- Profile for `GF`: 0% null; 91 distinct (high-cardinality); range 0..98, avg 25.6\n-- Profile for `GA`: 0% null; 77 distinct (high-cardinality); range 0..91, avg 25.6\n-- Profile for `PIM`: 9% null; 342 distinct (high-cardinality); len 1..3\n-- Profile for `BenchMinor`: 38% null; 8 distinct (categorical); len 1..2; top: 0(47%), 2(28%), 4(15%)\n-- Profile for `PPG`: 31% null; 31 distinct (high-cardinality); len 1..2\n-- Profile for `PPC`: 37% null; 115 distinct (high-cardinality); len 1..3\n-- Profile for `SHA`: 35% null; 9 distinct (categorical); len 1..1; top: 0(41%), 1(31%), 2(15%)\n-- Profile for `PKG`: 31% null; 27 distinct (high-cardinality); len 1..2\n-- Profile for `PKC`: 37% null; 113 distinct (high-cardinality); len 1..3\n-- Profile for `SHF`: 31% null; 11 distinct (high-cardinality); len 1..2',
        '/* Table: state_sector_details | level: Level of institution; stateid: state id; control: Public, Private not-for-profit, Private for-profit; state: state name; counted_pct: counted percentage; awards_per_state_value: awards per state value; awards_per_natl_value: awards per national value; exp_award_state_value: expected award state value; exp_award_natl_value: expected award national value; state_appr_value: state appropriation value; references; state_post; schools_count; state_appr_rank; grad_rate_rank; awards_per_rank */\nCREATE TABLE "state_sector_details"\n(\n    stateid                INTEGER, -- state id\n    state                  TEXT -- state name\n        references institution_details (state),\n    state_post             TEXT,\n    level                  TEXT, -- Level of institution | Values: 4-year: bachelor\'s degree 2-year: associate\'s degree\n    control                TEXT, -- Values: Public, Private not-for-profit, Private for-profit\n    schools_count          INTEGER,\n    counted_pct            TEXT, -- counted percentage\n    awards_per_state_value TEXT, -- awards per state value\n    awards_per_natl_value  REAL, -- awards per national value\n    exp_award_state_value  TEXT, -- expected award state value\n    exp_award_natl_value   INTEGER, -- expected award national value\n    state_appr_value       TEXT, -- state appropriation value\n    state_appr_rank        TEXT,\n    grad_rate_rank         TEXT,\n    awards_per_rank        TEXT,\n    primary key (stateid, level, control)\n)\n-- Approximate rows: 312\n-- Sample values for column `stateid`: 0, 29, 56\n-- Sample values for column `state`: United States, Alabama, Alaska\n-- Sample values for column `state_post`: U.S., Ala., Alaska\n-- Sample values for column `level`: 2-year, 4-year\n-- Sample values for column `control`: Private for-profit, Private not-for-profit, Public\n-- Sample values for column `schools_count`: 0, 32, 1180\n-- Sample values for column `counted_pct`: NULL, 61, 62\n-- Sample values for column `awards_per_state_value`: NULL, 18.8, 17.8\n-- Sample values for column `awards_per_natl_value`: 16.5, 24.6, 32.8\n-- Sample values for column `exp_award_state_value`: NULL, 75743, 92268\n-- Sample values for column `exp_award_natl_value`: 24795, 38763, 101725\n-- Sample values for column `state_appr_value`: NULL, 290.68, 496.78\n-- Sample values for column `state_appr_rank`: NULL, 11, 2\n-- Sample values for column `grad_rate_rank`: 23, 18, 8\n-- Sample values for column `awards_per_rank`: NULL, 42, 49\n-- Table profile: 312 rows\n-- Profile for `stateid`: 0% null; 52 distinct (high-cardinality); range 0..56, avg 28.4\n-- Profile for `state`: 0% null; 52 distinct (high-cardinality); len 4..20\n-- Profile for `state_post`: 0% null; 52 distinct (high-cardinality); len 3..6\n-- Profile for `level`: 0% null; 2 distinct (categorical); len 6..6; top: 4-year(50%), 2-year(50%)\n-- Profile for `control`: 0% null; 3 distinct (categorical); len 6..22; top: Public(33%), Private not-for-profit(33%), Private for-profit(33%)\n-- Profile for `schools_count`: 0% null; 58 distinct (high-cardinality); range 0..1180, avg 24.35\n-- Profile for `counted_pct`: 0% null; 209 distinct (high-cardinality); len 1..4\n-- Profile for `awards_per_state_value`: 0% null; 161 distinct (high-cardinality); len 2..4\n-- Profile for `awards_per_natl_value`: 0% null; 6 distinct (categorical); range 16.5..32.8, avg 23.97; top: 32.8(17%), 25.9(17%), 24.6(17%)\n-- Profile for `exp_award_state_value`: 0% null; 265 distinct (high-cardinality); len 4..6\n-- Profile for `exp_award_natl_value`: 0% null; 6 distinct (categorical); range 24795..101725, avg 50668.17; top: 101725(17%), 66436(17%), 38763(17%)\n-- Profile for `state_appr_value`: 0% null; 50 distinct (high-cardinality); len 4..6\n-- Profile for `state_appr_rank`: 0% null; 50 distinct (high-cardinality); len 1..4\n-- Profile for `grad_rate_rank`: 0% null; 53 distinct (high-cardinality); len 1..4\n-- Profile for `awards_per_rank`: 0% null; 52 distinct (high-cardinality); len 1..4',
        '/* Table: person | p_id: person ID; professor: whether the person is a professor; student: whether the person is a student; hasPosition: has position; inPhase: in phase; yearsInProgram: years in program */\nCREATE TABLE person\n(\n    p_id           INTEGER -- person ID\n        constraint person_pk\n            primary key,\n    professor      INTEGER, -- whether the person is a professor | Values: 0: professor 1: student\n    student        INTEGER, -- whether the person is a student | Values: 0: professor 1: student\n    hasPosition    TEXT, -- has position | Values: 0: the person is not a faculty member Common Sense evidence: faculty_aff: affiliated faculty faculty_eme: faculty employee\n    inPhase        TEXT, -- in phase | Values: 0: the person is not undergoing the phase of qualification\n    yearsInProgram TEXT -- years in program | Values: 0: the person is not in any programs Common Sense evidence: yearX means the person is on the Xth year of the program\n)\n-- Approximate rows: 278\n-- Sample values for column `p_id`: 3, 222, 435\n-- Sample values for column `professor`: 0, 1\n-- Sample values for column `student`: 0, 1\n-- Sample values for column `hasPosition`: 0, Faculty, Faculty_adj\n-- Sample values for column `inPhase`: 0, Post_Quals, Post_Generals\n-- Sample values for column `yearsInProgram`: 0, Year_2, Year_5\n-- Table profile: 278 rows\n-- Profile for `p_id`: 0% null; 278 distinct (unique/id); range 3..435, avg 221.06\n-- Profile for `professor`: 0% null; 2 distinct (categorical); range 0..1, avg 0.22; top: 0(78%), 1(22%)\n-- Profile for `student`: 0% null; 2 distinct (categorical); range 0..1, avg 0.78; top: 1(78%), 0(22%)\n-- Profile for `hasPosition`: 0% null; 5 distinct (categorical); len 1..11; top: 0(81%), Faculty(14%), Faculty_adj(2%)\n-- Profile for `inPhase`: 0% null; 4 distinct (categorical); len 1..13; top: 0(50%), Post_Generals(18%), Post_Quals(18%)\n-- Profile for `yearsInProgram`: 0% null; 12 distinct (high-cardinality); len 1..7',
        '/* Table: client | client_id: client id; district_id → district: district id; social: social number; address_2: address 2; sex: sex of client; day: day of the birthday; month: month of the birthday; year: year when is born; first: first name; middle: middle name; last: last name; phone: phone number; address_1: address 1; state: state code; age: teenager: 13-19 adult: 19-65 elder: > 65; email: google email / account: @gamil.com microsoft email / account: xxx@outl...; city; zipcode */\nCREATE TABLE client\n(\n    client_id   TEXT -- client id\n            primary key,\n    sex         TEXT, -- sex of client\n    day         INTEGER, -- day of the birthday\n    month       INTEGER, -- month of the birthday\n    year        INTEGER, -- year when is born\n    age         INTEGER, -- Values: teenager: 13-19 adult: 19-65 elder: > 65\n    social      TEXT, -- social number | Values: ssn: us id number for each person\n    first       TEXT, -- first name\n    middle      TEXT, -- middle name\n    last        TEXT, -- last name\n    phone       TEXT, -- phone number\n    email       TEXT, -- Values: google email / account: @gamil.com microsoft email / account: xxx@outlook.com\n    address_1   TEXT, -- address 1\n    address_2   TEXT, -- address 2 | Values: entire address = (address_1, address_2)\n    city        TEXT,\n    state       TEXT, -- state code\n    zipcode     INTEGER,\n    district_id INTEGER, -- district id\n    foreign key (district_id) references district(district_id)\n)\n-- Approximate rows: 5369\n-- Sample values for column `client_id`: C00000001, C00000002, C00000003\n-- Sample values for column `sex`: Female, Male\n-- Sample values for column `day`: 1, 16, 31\n-- Sample values for column `month`: 1, 7, 12\n-- Sample values for column `year`: 1931, 1969, 2007\n-- Sample values for column `age`: 12, 50, 88\n-- Sample values for column `social`: 926-93-2157, 806-94-5725, 614-70-9100\n-- Sample values for column `first`: Emma, Noah, Olivia\n-- Sample values for column `middle`: Avaya, Everest, Brooklynne\n-- Sample values for column `last`: Smith, Thompson, Johnson\n-- Sample values for column `phone`: 367-171-6840, 212-423-7734, 212-425-6932\n-- Sample values for column `email`: emma.smith@gmail.com, noah.thompson@gmail.com, olivia.johnson@outlook.com\n-- Sample values for column `address_1`: 387 Wellington Ave., 75 W. Berkshire St., 36 Second St.\n-- Sample values for column `address_2`: Unit 1, Unit 6, Unit 80\n-- Sample values for column `city`: Albuquerque, New York City, Indianapolis\n-- Sample values for column `state`: NM, NY, IN\n-- Sample values for column `zipcode`: 1101, 44510, 49998\n-- Sample values for column `district_id`: 1, 39, 77\n-- Table profile: 5.4k rows\n-- Profile for `client_id`: 0% null; 5.4k distinct (unique/id); len 9..9\n-- Profile for `sex`: 0% null; 2 distinct (categorical); len 4..6; top: Male(51%), Female(49%)\n-- Profile for `day`: 0% null; 31 distinct (high-cardinality); range 1..31, avg 15.96\n-- Profile for `month`: 0% null; 12 distinct (high-cardinality); range 1..12, avg 6.52\n-- Profile for `year`: 0% null; 77 distinct (high-cardinality); range 1931..2007, avg 1973.2\n-- Profile for `age`: 0% null; 77 distinct (high-cardinality); range 12..88, avg 45.89\n-- Profile for `social`: 0% null; 5.4k distinct (unique/id); len 11..11\n-- Profile for `first`: 0% null; 2.9k distinct (high-cardinality); len 2..11\n-- Profile for `middle`: 0% null; 3.9k distinct (high-cardinality); len 2..12\n-- Profile for `last`: 0% null; 1.8k distinct (high-cardinality); len 2..13\n-- Profile for `phone`: 0% null; 5.4k distinct (unique/id); len 12..12\n-- Profile for `email`: 0% null; 5.4k distinct (unique/id); len 17..38\n-- Profile for `address_1`: 0% null; 5.4k distinct (unique/id); len 9..30\n-- Profile for `address_2`: 98% null; 24 distinct (high-cardinality); len 6..7\n-- Profile for `city`: 0% null; 75 distinct (high-cardinality); len 4..14\n-- Profile for `state`: 0% null; 51 distinct (high-cardinality); len 2..2\n-- Profile for `zipcode`: 0% null; 3.5k distinct (high-cardinality); range 1101..49998, avg 33506.31\n-- Profile for `district_id`: 0% null; 77 distinct (high-cardinality); range 1..77, avg 37.31',
        '/* Table: position | positionID: position id; educationrequired: education required; positiontitle: position title; minsalary: minimum salary; maxsalary: maximum salary */\nCREATE TABLE position\n(\n    positionID        INTEGER -- position id\n        constraint position_pk\n            primary key,\n    positiontitle     TEXT, -- position title\n    educationrequired TEXT, -- education required | Values: Generally, more complex work requires more education year.\n    minsalary         TEXT, -- minimum salary\n    maxsalary         TEXT -- maximum salary\n)\n-- Approximate rows: 4\n-- Sample values for column `positionID`: 1, 3, 4\n-- Sample values for column `positiontitle`: Account Representative, Manager, Trainee\n-- Sample values for column `educationrequired`: 4 year degree, 2 year degree, 6 year degree\n-- Sample values for column `minsalary`: US$25,000.00, US$50,000.00, US$18,000.00\n-- Sample values for column `maxsalary`: US$75,000.00, US$150,000.00, US$25,000.00\n-- Table profile: 4 rows\n-- Profile for `positionID`: 0% null; 4 distinct (unique/id); range 1..4, avg 2.5; top: 1(25%), 2(25%), 3(25%)\n-- Profile for `positiontitle`: 0% null; 4 distinct (unique/id); len 7..22; top: Trainee(25%), Regional Manager(25%), Manager(25%)\n-- Profile for `educationrequired`: 0% null; 3 distinct (categorical); len 13..13; top: 4 year degree(50%), 6 year degree(25%), 2 year degree(25%)\n-- Profile for `minsalary`: 0% null; 4 distinct (unique/id); len 12..13; top: US$50,000.00(25%), US$25,000.00(25%), US$18,000.00(25%)\n-- Profile for `maxsalary`: 0% null; 4 distinct (unique/id); len 12..13; top: US$75,000.00(25%), US$250,000.00(25%), US$25,000.00(25%)',
    ]
)
# [{'corpus_id': ..., 'score': ...}, {'corpus_id': ..., 'score': ...}, ...]

Training Details

Training Dataset

Unnamed Dataset

Size: 92,768 training samples
Columns: sentence_0, sentence_1, and label
Approximate statistics based on the first 1000 samples:
sentence_0 sentence_1 label
type string string int
details
min: 37 characters
mean: 177.6 characters
max: 609 characters
min: 134 characters
mean: 2211.3 characters
max: 22131 characters
0: ~79.50%
1: ~20.50%

Samples:

sentence_0	sentence_1	label
`For the goalie who had the most shutouts in 2010, what's his catching hand? the most shutouts refers to max(SHO); shootCatch = 'L' refers to lefthand; shootCatch = 'R' refers to righthand; shootCatch = 'null' or 'empty' means this player is good at both left and right hand`	`/* Table: TeamsPost`	tmID → Teams: team ID; year → Teams; lgID: league ID; G: Games; W: wins; L: loses; T: ties; GF: goals for; GA: goals against; PIM: penalty minutes; BenchMinor: Bench minors (minutes); PPG: Power play goals; PPC: Power play chances; SHA: Shorthanded goals against; PKG: Power play goals against; PKC: Penalty kill chances; SHF: Shorthanded goals for; on */ CREATE TABLE TeamsPost ( year INTEGER not null, lgID TEXT, -- league ID tmID TEXT not null, -- team ID G INTEGER, -- Games W INTEGER, -- wins L INTEGER, -- loses T INTEGER, -- ties GF INTEGER, -- goals for GA INTEGER, -- goals against PIM TEXT, -- penalty minutes BenchMinor TEXT, -- Bench minors (minutes) PPG TEXT, -- Power play goals PPC TEXT, -- Power play chances SHA TEXT, -- Shorthanded goals against PKG TEXT, -- Power play goals against ...
`What is the percentage of act number 5 in Titus Andronicus? act number 5 refers to Act = 5; Titus Andronicus refers to Title = 'Titus Andronicus'; percentage = divide(sum(Act = 5), count(Act)) as percentage`	`/* Table: state_sector_details`	level: Level of institution; stateid: state id; control: Public, Private not-for-profit, Private for-profit; state: state name; counted_pct: counted percentage; awards_per_state_value: awards per state value; awards_per_natl_value: awards per national value; exp_award_state_value: expected award state value; exp_award_natl_value: expected award national value; state_appr_value: state appropriation value; references; state_post; schools_count; state_appr_rank; grad_rate_rank; awards_per_rank */ CREATE TABLE "state_sector_details" ( stateid INTEGER, -- state id state TEXT -- state name references institution_details (state), state_post TEXT, level TEXT, -- Level of institution
`When did user 39115684 rate the movie "A Way of Life"? A Way of Life' refers to movie_title; user 39115684 refers to userid = 39115684; when the user rate refers to rating_timestamp_utc;`	`/* Table: person`	p_id: person ID; professor: whether the person is a professor; student: whether the person is a student; hasPosition: has position; inPhase: in phase; yearsInProgram: years in program */ CREATE TABLE person ( p_id INTEGER -- person ID constraint person_pk primary key, professor INTEGER, -- whether the person is a professor

Loss: BinaryCrossEntropyLoss with these parameters:
json
1{ 2 "activation_fn": "torch.nn.modules.linear.Identity", 3 "pos_weight": null 4}

Training Hyperparameters

Non-Default Hyperparameters

per_device_train_batch_size: 32
num_train_epochs: 1
max_grad_norm: 10.0
fp16: True
per_device_eval_batch_size: 32

All Hyperparameters

Click to expand

per_device_train_batch_size: 32
num_train_epochs: 1
max_steps: -1
learning_rate: 5e-05
lr_scheduler_type: linear
lr_scheduler_kwargs: None
warmup_steps: 0
optim: adamw_torch
optim_args: None
weight_decay: 0.0
adam_beta1: 0.9
adam_beta2: 0.999
adam_epsilon: 1e-08
optim_target_modules: None
gradient_accumulation_steps: 1
average_tokens_across_devices: True
max_grad_norm: 10.0
label_smoothing_factor: 0.0
bf16: False
fp16: True
bf16_full_eval: False
fp16_full_eval: False
tf32: None
gradient_checkpointing: False
gradient_checkpointing_kwargs: None
torch_compile: False
torch_compile_backend: None
torch_compile_mode: None
use_liger_kernel: False
liger_kernel_config: None
use_cache: False
neftune_noise_alpha: None
torch_empty_cache_steps: None
auto_find_batch_size: False
log_on_each_node: True
logging_nan_inf_filter: True
include_num_input_tokens_seen: no
log_level: passive
log_level_replica: warning
disable_tqdm: False
project: huggingface
trackio_space_id: trackio
eval_strategy: no
per_device_eval_batch_size: 32
prediction_loss_only: True
eval_on_start: False
eval_do_concat_batches: True
eval_use_gather_object: False
eval_accumulation_steps: None
include_for_metrics: []
batch_eval_metrics: False
save_only_model: False
save_on_each_node: False
enable_jit_checkpoint: False
push_to_hub: False
hub_private_repo: None
hub_model_id: None
hub_strategy: every_save
hub_always_push: False
hub_revision: None
load_best_model_at_end: False
ignore_data_skip: False
restore_callback_states_from_checkpoint: False
full_determinism: False
seed: 42
data_seed: None
use_cpu: False
accelerator_config: {'split_batches': False, 'dispatch_batches': None, 'even_batches': True, 'use_seedable_sampler': True, 'non_blocking': False, 'gradient_accumulation_kwargs': None}
parallelism_config: None
dataloader_drop_last: False
dataloader_num_workers: 0
dataloader_pin_memory: True
dataloader_persistent_workers: False
dataloader_prefetch_factor: None
remove_unused_columns: True
label_names: None
train_sampling_strategy: random
length_column_name: length
ddp_find_unused_parameters: None
ddp_bucket_cap_mb: None
ddp_broadcast_buffers: False
ddp_backend: None
ddp_timeout: 1800
fsdp: []
fsdp_config: {'min_num_params': 0, 'xla': False, 'xla_fsdp_v2': False, 'xla_fsdp_grad_ckpt': False}
deepspeed: None
debug: []
skip_memory_metrics: True
do_predict: False
resume_from_checkpoint: None
warmup_ratio: None
local_rank: -1
prompts: None
batch_sampler: batch_sampler
multi_dataset_batch_sampler: proportional

Training Logs

Epoch	Step	Training Loss
0.1725	500	0.1838
0.3449	1000	0.1190
0.5174	1500	0.0935
0.6899	2000	0.0864
0.8624	2500	0.0746

Framework Versions

Python: 3.10.12
Sentence Transformers: 4.1.0
Transformers: 5.5.0
PyTorch: 2.7.1+cu128
Accelerate: 1.7.0
Datasets: 3.6.0
Tokenizers: 0.22.2

Citation

BibTeX

Sentence Transformers

bibtex

@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",
}

	sentence_0	sentence_1	label
type	string	string	int
details	min: 37 characters mean: 177.6 characters max: 609 characters	min: 134 characters mean: 2211.3 characters max: 22131 characters	0: ~79.50% 1: ~20.50%

treprepr/reranker-bird-evidence-finetuned

CrossEncoder based on BAAI/bge-reranker-v2-m3