input_seq stringlengths 2.19k 45k | output_seq stringlengths 105 1.02k | source stringclasses 1
value | difficulty stringclasses 3
values | db_id stringclasses 57
values |
|---|---|---|---|---|
Task Overview:
You are a data science expert. Below, you are provided with a database schema and a natural language question. Your task is to understand the schema and generate a valid SQL query to answer the question.
Database Engine:
SQLite
Database Schema:
CREATE TABLE city (
id integer, -- example: [1, 2]
... | <think>
Tables needed: competitor_event (as T1), event (as T2).
Join keys: INNER join "event" AS "T2" on "T1"."event_id" = "T2"."id".
Filters: "T1"."event_id" = 188. (evidence: event ID 188 refers to competitor_event.event_id = 188)
Aggregation: ordered by "T2"."event_name" ASC; limited to 5.
</think>
<answer>
SELECT "... | datagen | moderate | olympics |
Task Overview:
You are a data science expert. Below, you are provided with a database schema and a natural language question. Your task is to understand the schema and generate a valid SQL query to answer the question.
Database Engine:
SQLite
Database Schema:
CREATE TABLE Author (
Id integer, -- example: [9, 14]
... | <think>
Tables needed: Author (as T1), PaperAuthor (as T2).
Join keys: INNER join "PaperAuthor" AS "T2" on "T1"."Id" = "T2"."AuthorId".
Filters: "T2"."AuthorId" = 894232. (evidence: author ID 894232 refers to PaperAuthor.AuthorId)
Aggregation: aggregates COUNT(*).
</think>
<answer>
SELECT COUNT(*) AS "agg_value" FROM "... | datagen | simple | authors |
Task Overview:
You are a data science expert. Below, you are provided with a database schema and a natural language question. Your task is to understand the schema and generate a valid SQL query to answer the question.
Database Engine:
SQLite
Database Schema:
CREATE TABLE country (
country_id integer, -- example:... | <think>
Tables needed: movie_cast (as T1).
Filters: "T1"."cast_order" >= 110. (evidence: a cast order of 110 or higher refers to cast_order >= 110;)
Aggregation: limited to 5.
</think>
<answer>
SELECT "T1"."character_name" FROM "movie_cast" AS "T1" WHERE "T1"."cast_order" >= 110 LIMIT 5
</answer> | datagen | simple | movies_4 |
Task Overview:
You are a data science expert. Below, you are provided with a database schema and a natural language question. Your task is to understand the schema and generate a valid SQL query to answer the question.
Database Engine:
SQLite
Database Schema:
CREATE TABLE customers (
CustomerID integer, -- exampl... | <think>
Tables needed: location (as T1), geolocation (as T2), rootbeer (as T3).
Join keys: INNER join "geolocation" AS "T2" on "T1"."LocationID" = "T2"."LocationID"; INNER join "rootbeer" AS "T3" on "T1"."LocationID" = "T3"."LocationID".
Filters: "T2"."Latitude" = 38.559615. (evidence: latitude 38.559615 refers to geol... | datagen | moderate | beer_factory |
Task Overview:
You are a data science expert. Below, you are provided with a database schema and a natural language question. Your task is to understand the schema and generate a valid SQL query to answer the question.
Database Engine:
SQLite
Database Schema:
CREATE TABLE Demog (
GEOID integer, -- example: [0, 1]... | <think>
Tables needed: Sales (as T1).
Filters: "T1"."EVENT_DATE" = '2007-01-09 12:00:00.0'. (evidence: January 9, 2007 at 12:00 PM refers to EVENT_DATE = '2007-01-09 12:00:00.0';)
Aggregation: limited to 3.
</think>
<answer>
SELECT "T1"."AMOUNT" FROM "Sales" AS "T1" WHERE "T1"."EVENT_DATE" = '2007-01-09 12:00:00.0' LIM... | datagen | simple | software_company |
Task Overview:
You are a data science expert. Below, you are provided with a database schema and a natural language question. Your task is to understand the schema and generate a valid SQL query to answer the question.
Database Engine:
SQLite
Database Schema:
CREATE TABLE film_text (
film_id integer, -- example: ... | <think>
Tables needed: customer (as T1), address (as T2).
Join keys: INNER join "address" AS "T2" on "T1"."address_id" = "T2"."address_id".
Filters: "T1"."store_id" = 2. (evidence: store 2 refers to customer.store_id = 2;)
</think>
<answer>
SELECT "T1"."last_update" FROM "customer" AS "T1" INNER JOIN "address" AS "T2" ... | datagen | simple | movie_3 |
Task Overview:
You are a data science expert. Below, you are provided with a database schema and a natural language question. Your task is to understand the schema and generate a valid SQL query to answer the question.
Database Engine:
SQLite
Database Schema:
CREATE TABLE all_prevalences (
ITEM text, -- example: ... | <think>
Tables needed: medications (as T1).
Filters: "T1"."CODE" > 978950. (evidence: a code above 978950 refers to CODE > 978950;)
Aggregation: ordered by "T1"."REASONCODE" ASC; limited to 1.
</think>
<answer>
SELECT "T1"."DESCRIPTION" FROM "medications" AS "T1" WHERE "T1"."CODE" > 978950 ORDER BY "T1"."REASONCODE" AS... | datagen | simple | synthea |
Task Overview:
You are a data science expert. Below, you are provided with a database schema and a natural language question. Your task is to understand the schema and generate a valid SQL query to answer the question.
Database Engine:
SQLite
Database Schema:
CREATE TABLE Customers (
CustomerID integer, -- exampl... | <think>
Tables needed: Employees (as T1).
Filters: "T1"."FirstName" = 'Sheryl'. (evidence: 'Sheryl' is an employee's first name;)
Aggregation: ordered by "T1"."FirstName" DESC.
</think>
<answer>
SELECT "T1"."FirstName" FROM "Employees" AS "T1" WHERE "T1"."FirstName" = 'Sheryl' ORDER BY "T1"."FirstName" DESC
</answer> | datagen | simple | sales |
Task Overview:
You are a data science expert. Below, you are provided with a database schema and a natural language question. Your task is to understand the schema and generate a valid SQL query to answer the question.
Database Engine:
SQLite
Database Schema:
CREATE TABLE institution_details (
unitid integer, -- ... | <think>
Tables needed: state_sector_details (as T1).
Filters: "T1"."exp_award_natl_value" > 37780. (evidence: national award expenditure above 37780 refers to exp_award_natl_value > 37780;)
Aggregation: ordered by "T1"."exp_award_natl_value" DESC; limited to 1.
</think>
<answer>
SELECT "T1"."counted_pct" FROM "state_se... | datagen | simple | college_completion |
Task Overview:
You are a data science expert. Below, you are provided with a database schema and a natural language question. Your task is to understand the schema and generate a valid SQL query to answer the question.
Database Engine:
SQLite
Database Schema:
CREATE TABLE current (
ballotpedia_id text, -- example... | <think>
Tables needed: current (as T1), current-terms (as T2).
Join keys: INNER join "current-terms" AS "T2" on "T1"."bioguide_id" = "T2"."bioguide".
Filters: "T1"."votesmart_id" = 542.0. (evidence: votesmart ID 542 refers to votesmart_id = 542.0;)
</think>
<answer>
SELECT "T1"."middle_name" FROM "current" AS "T1" INNE... | datagen | simple | legislator |
Task Overview:
You are a data science expert. Below, you are provided with a database schema and a natural language question. Your task is to understand the schema and generate a valid SQL query to answer the question.
Database Engine:
SQLite
Database Schema:
CREATE TABLE country (
origin integer, -- example: [1,... | <think>
Tables needed: price (as T1), production (as T2).
Join keys: INNER join "production" AS "T2" on "T1"."ID" = "T2"."ID".
Filters: "T2"."model_year" = 1982. (evidence: model year 1982 refers to production.model_year = 1982;)
Aggregation: ordered by "T2"."model_year" DESC; limited to 1.
</think>
<answer>
SELECT "T1... | datagen | moderate | cars |
Task Overview:
You are a data science expert. Below, you are provided with a database schema and a natural language question. Your task is to understand the schema and generate a valid SQL query to answer the question.
Database Engine:
SQLite
Database Schema:
CREATE TABLE ATT_CLASSES (
ATT_CLASS_ID integer, -- ex... | <think>
Tables needed: IMG_REL (as T1), IMG_OBJ (as T2).
Join keys: INNER join "IMG_OBJ" AS "T2" on "T1"."IMG_ID" = "T2"."IMG_ID".
Filters: "T2"."IMG_ID" = 59. (evidence: image 59 refers to IMG_OBJ.IMG_ID = 59;)
Aggregation: ordered by "T2"."H" ASC; limited to 1.
</think>
<answer>
SELECT "T2"."X" FROM "IMG_REL" AS "T1"... | datagen | moderate | image_and_language |
Task Overview:
You are a data science expert. Below, you are provided with a database schema and a natural language question. Your task is to understand the schema and generate a valid SQL query to answer the question.
Database Engine:
SQLite
Database Schema:
CREATE TABLE Customers (
CustomerID integer, -- exampl... | <think>
Tables needed: Employees (as T1).
Filters: "T1"."EmployeeID" = 21. (evidence: employee ID 21 refers to EmployeeID = 21;)
</think>
<answer>
SELECT "T1"."FirstName" FROM "Employees" AS "T1" WHERE "T1"."EmployeeID" = 21
</answer> | datagen | simple | sales |
Task Overview:
You are a data science expert. Below, you are provided with a database schema and a natural language question. Your task is to understand the schema and generate a valid SQL query to answer the question.
Database Engine:
SQLite
Database Schema:
CREATE TABLE current (
ballotpedia_id text, -- example... | <think>
Tables needed: current (as T1), social-media (as T2).
Join keys: INNER join "social-media" AS "T2" on "T1"."bioguide_id" = "T2"."bioguide".
Filters: "T1"."nickname_name" = 'Maggie'. (evidence: 'Maggie' is a legislator's nickname;)
Aggregation: aggregates MAX("T2"."govtrack").
</think>
<answer>
SELECT MAX("T2"."... | datagen | simple | legislator |
Task Overview:
You are a data science expert. Below, you are provided with a database schema and a natural language question. Your task is to understand the schema and generate a valid SQL query to answer the question.
Database Engine:
SQLite
Database Schema:
CREATE TABLE lists (
user_id integer, -- example: [882... | <think>
Tables needed: ratings (as T1).
Aggregation: ordered by "T1"."user_eligible_for_trial" ASC.
</think>
<answer>
SELECT "T1"."user_eligible_for_trial" FROM "ratings" AS "T1" ORDER BY "T1"."user_eligible_for_trial" ASC
</answer> | datagen | simple | movie_platform |
Task Overview:
You are a data science expert. Below, you are provided with a database schema and a natural language question. Your task is to understand the schema and generate a valid SQL query to answer the question.
Database Engine:
SQLite
Database Schema:
CREATE TABLE chapters (
id integer, -- example: [19354... | <think>
Tables needed: paragraphs (as T1), chapters (as T2).
Join keys: INNER join "chapters" AS "T2" on "T1"."chapter_id" = "T2"."id".
Filters: "T2"."id" = 19354. (evidence: chapter 19354 refers to chapters.id = 19354;)
Aggregation: ordered by "T1"."ParagraphNum" ASC; limited to 1.
</think>
<answer>
SELECT "T2"."Scene... | datagen | moderate | shakespeare |
Task Overview:
You are a data science expert. Below, you are provided with a database schema and a natural language question. Your task is to understand the schema and generate a valid SQL query to answer the question.
Database Engine:
SQLite
Database Schema:
CREATE TABLE people (
"Customer ID" text, -- example: ... | <think>
Tables needed: central_superstore (as T1), product (as T2).
Join keys: INNER join "product" AS "T2" on "T1"."Product ID" = "T2"."Product ID".
Filters: "T1"."Order Date" = '2016-06-13'. (evidence: 2016-06-13 refers to central_superstore."Order Date" = '2016-06-13';)
Aggregation: aggregates AVG("T1"."Profit").
</... | datagen | simple | superstore |
Task Overview:
You are a data science expert. Below, you are provided with a database schema and a natural language question. Your task is to understand the schema and generate a valid SQL query to answer the question.
Database Engine:
SQLite
Database Schema:
CREATE TABLE Method (
Id integer, -- example: [1, 2]
... | <think>
Tables needed: Solution (as T1).
Filters: "T1"."Id" = 1293. (evidence: solution ID 1293 refers to Solution.Id = 1293;)
Aggregation: aggregates AVG("T1"."WasCompiled").
</think>
<answer>
SELECT AVG("T1"."WasCompiled") AS "agg_value" FROM "Solution" AS "T1" WHERE "T1"."Id" = 1293
</answer> | datagen | simple | codebase_comments |
Task Overview:
You are a data science expert. Below, you are provided with a database schema and a natural language question. Your task is to understand the schema and generate a valid SQL query to answer the question.
Database Engine:
SQLite
Database Schema:
CREATE TABLE employee (
employee_id integer, -- exampl... | <think>
Tables needed: employee (as T1), inspection (as T2).
Join keys: INNER join "inspection" AS "T2" on "T1"."employee_id" = "T2"."employee_id".
Filters: "T2"."inspection_date" = '2010-06-11'. (evidence: June 11, 2010 refers to inspection.inspection_date = '2010-06-11';)
</think>
<answer>
SELECT "T1"."first_name" FR... | datagen | simple | food_inspection_2 |
Task Overview:
You are a data science expert. Below, you are provided with a database schema and a natural language question. Your task is to understand the schema and generate a valid SQL query to answer the question.
Database Engine:
SQLite
Database Schema:
CREATE TABLE ATT_CLASSES (
ATT_CLASS_ID integer, -- ex... | <think>
Tables needed: ATT_CLASSES (as T1).
Filters: "T1"."ATT_CLASS_ID" = 658. (evidence: ATT_CLASS_ID 658 refers to ATT_CLASSES.ATT_CLASS_ID = 658;)
</think>
<answer>
SELECT "T1"."ATT_CLASS" FROM "ATT_CLASSES" AS "T1" WHERE "T1"."ATT_CLASS_ID" = 658
</answer> | datagen | simple | image_and_language |
Task Overview:
You are a data science expert. Below, you are provided with a database schema and a natural language question. Your task is to understand the schema and generate a valid SQL query to answer the question.
Database Engine:
SQLite
Database Schema:
CREATE TABLE Episode (
episode_id text, -- example: ['... | <think>
Tables needed: Episode (as T1).
Filters: "T1"."votes" = 203. (evidence: 203 votes refers to votes = 203;)
Aggregation: aggregates SUM("T1"."number_in_series").
</think>
<answer>
SELECT SUM("T1"."number_in_series") AS "agg_value" FROM "Episode" AS "T1" WHERE "T1"."votes" = 203
</answer> | datagen | simple | law_episode |
Task Overview:
You are a data science expert. Below, you are provided with a database schema and a natural language question. Your task is to understand the schema and generate a valid SQL query to answer the question.
Database Engine:
SQLite
Database Schema:
CREATE TABLE chapters (
id integer, -- example: [18704... | <think>
Tables needed: chapters (as T1), works (as T2).
Join keys: INNER join "works" AS "T2" on "T1"."work_id" = "T2"."id".
Filters: "T1"."Scene" = 99. (evidence: scene number 99 refers to chapters.Scene = 99;)
</think>
<answer>
SELECT "T1"."Description" FROM "chapters" AS "T1" INNER JOIN "works" AS "T2" ON "T1"."work... | datagen | simple | shakespeare |
Task Overview:
You are a data science expert. Below, you are provided with a database schema and a natural language question. Your task is to understand the schema and generate a valid SQL query to answer the question.
Database Engine:
SQLite
Database Schema:
CREATE TABLE chapters (
id integer, -- example: [18704... | <think>
Tables needed: paragraphs (as T1).
Filters: "T1"."PlainText" = 'I was adored once too.'. (evidence: 'I was adored once too.' is a line of paragraph text;)
</think>
<answer>
SELECT "T1"."ParagraphNum" FROM "paragraphs" AS "T1" WHERE "T1"."PlainText" = 'I was adored once too.'
</answer> | datagen | simple | shakespeare |
Task Overview:
You are a data science expert. Below, you are provided with a database schema and a natural language question. Your task is to understand the schema and generate a valid SQL query to answer the question.
Database Engine:
SQLite
Database Schema:
CREATE TABLE customers (
CustomerID integer, -- exampl... | <think>
Tables needed: rootbeerreview (as T1), customers (as T2).
Join keys: INNER join "customers" AS "T2" on "T1"."CustomerID" = "T2"."CustomerID".
Filters: "T2"."First" = 'Casey'. (evidence: 'Casey' is a customer's first name;)
Aggregation: aggregates AVG("T2"."ZipCode"); limited to 3.
</think>
<answer>
SELECT AVG("... | datagen | moderate | beer_factory |
Task Overview:
You are a data science expert. Below, you are provided with a database schema and a natural language question. Your task is to understand the schema and generate a valid SQL query to answer the question.
Database Engine:
SQLite
Database Schema:
CREATE TABLE playstore (
App text, -- example: ['Photo... | <think>
Tables needed: playstore (as T1).
Filters: "T1"."Price" = '$4.99'. (evidence: '$4.99' is an app's price;)
Aggregation: ordered by "T1"."Reviews" DESC; limited to 1.
</think>
<answer>
SELECT "T1"."Type" FROM "playstore" AS "T1" WHERE "T1"."Price" = '$4.99' ORDER BY "T1"."Reviews" DESC LIMIT 1
</answer> | datagen | simple | app_store |
Task Overview:
You are a data science expert. Below, you are provided with a database schema and a natural language question. Your task is to understand the schema and generate a valid SQL query to answer the question.
Database Engine:
SQLite
Database Schema:
CREATE TABLE "Air Carriers" (
Code integer, -- example... | <think>
Tables needed: Air Carriers (as T1), Airlines (as T2).
Join keys: INNER join "Airlines" AS "T2" on "T1"."Code" = "T2"."OP_CARRIER_AIRLINE_ID".
Filters: "T2"."NAS_DELAY" = 133. (evidence: a NAS delay of 133 refers to NAS_DELAY = 133;)
Aggregation: aggregates SUM("T2"."CRS_DEP_TIME"); limited to 1.
</think>
<answ... | datagen | moderate | airline |
Task Overview:
You are a data science expert. Below, you are provided with a database schema and a natural language question. Your task is to understand the schema and generate a valid SQL query to answer the question.
Database Engine:
SQLite
Database Schema:
CREATE TABLE state (
StateCode text, -- example: ['AL'... | <think>
Tables needed: client (as T1), events (as T2).
Join keys: INNER join "events" AS "T2" on "T1"."client_id" = "T2"."Client_ID".
Filters: "T1"."year" = 1978. (evidence: clients born in 1978 refers to client.year = 1978;)
Aggregation: ordered by "T1"."zipcode" DESC; limited to 1.
</think>
<answer>
SELECT "T1"."day"... | datagen | moderate | retail_complains |
Task Overview:
You are a data science expert. Below, you are provided with a database schema and a natural language question. Your task is to understand the schema and generate a valid SQL query to answer the question.
Database Engine:
SQLite
Database Schema:
CREATE TABLE Customers (
CustomerID integer, -- exampl... | <think>
Tables needed: Employees (as T1), Sales (as T2).
Join keys: INNER join "Sales" AS "T2" on "T1"."EmployeeID" = "T2"."SalesPersonID".
Filters: "T2"."CustomerID" = 9941. (evidence: customer 9941 refers to Sales.CustomerID = 9941;)
</think>
<answer>
SELECT "T1"."FirstName" FROM "Employees" AS "T1" INNER JOIN "Sales... | datagen | simple | sales |
Task Overview:
You are a data science expert. Below, you are provided with a database schema and a natural language question. Your task is to understand the schema and generate a valid SQL query to answer the question.
Database Engine:
SQLite
Database Schema:
CREATE TABLE offices (
officeCode text, -- example: ['... | <think>
Tables needed: offices (as T1).
Filters: "T1"."city" = 'Sydney'. (evidence: 'Sydney' is an office's city;)
Aggregation: aggregates COUNT(*).
</think>
<answer>
SELECT COUNT(*) AS "agg_value" FROM "offices" AS "T1" WHERE "T1"."city" = 'Sydney'
</answer> | datagen | simple | car_retails |
Task Overview:
You are a data science expert. Below, you are provided with a database schema and a natural language question. Your task is to understand the schema and generate a valid SQL query to answer the question.
Database Engine:
SQLite
Database Schema:
CREATE TABLE offices (
officeCode text, -- example: ['... | <think>
Tables needed: products (as T1).
Filters: "T1"."productDescription" = '1:12 scale die-cast about 20" long Hood opens, Rubber wheels'.
</think>
<answer>
SELECT "T1"."productScale" FROM "products" AS "T1" WHERE "T1"."productDescription" = '1:12 scale die-cast about 20" long Hood opens, Rubber wheels'
</answer> | datagen | simple | car_retails |
Task Overview:
You are a data science expert. Below, you are provided with a database schema and a natural language question. Your task is to understand the schema and generate a valid SQL query to answer the question.
Database Engine:
SQLite
Database Schema:
CREATE TABLE customer (
c_custkey integer, -- example:... | <think>
Tables needed: nation (as T1), supplier (as T2).
Join keys: INNER join "supplier" AS "T2" on "T1"."n_nationkey" = "T2"."s_nationkey".
Filters: "T1"."n_comment" = 'instructions detect blithely stealthily pending packages'.
Aggregation: aggregates SUM("T1"."n_nationkey"); limited to 3.
</think>
<answer>
SELECT SU... | datagen | moderate | retails |
Task Overview:
You are a data science expert. Below, you are provided with a database schema and a natural language question. Your task is to understand the schema and generate a valid SQL query to answer the question.
Database Engine:
SQLite
Database Schema:
CREATE TABLE Attributes (
attribute_id integer, -- exa... | <think>
Tables needed: Users (as T1), Users_Compliments (as T2).
Join keys: INNER join "Users_Compliments" AS "T2" on "T1"."user_id" = "T2"."user_id".
Filters: "T1"."user_yelping_since_year" <= 2007. (evidence: joined in 2007 or earlier refers to user_yelping_since_year <= 2007;)
Aggregation: aggregates AVG("T1"."user_... | datagen | simple | public_review_platform |
Task Overview:
You are a data science expert. Below, you are provided with a database schema and a natural language question. Your task is to understand the schema and generate a valid SQL query to answer the question.
Database Engine:
SQLite
Database Schema:
CREATE TABLE awards_players (
playerID text, -- exampl... | <think>
Tables needed: draft (as T1), teams (as T2).
Join keys: INNER join "teams" AS "T2" on "T1"."tmID" = "T2"."tmID".
Filters: "T2"."confRank" = 2. (evidence: a conference rank of 2 refers to teams.confRank = 2;)
</think>
<answer>
SELECT "T1"."draftSelection" FROM "draft" AS "T1" INNER JOIN "teams" AS "T2" ON "T1"."... | datagen | simple | professional_basketball |
Task Overview:
You are a data science expert. Below, you are provided with a database schema and a natural language question. Your task is to understand the schema and generate a valid SQL query to answer the question.
Database Engine:
SQLite
Database Schema:
CREATE TABLE offices (
officeCode text, -- example: ['... | <think>
Tables needed: offices (as T1), employees (as T2).
Join keys: INNER join "employees" AS "T2" on "T1"."officeCode" = "T2"."officeCode".
Filters: "T2"."reportsTo" = 1102. (evidence: manager 1102 refers to employees.reportsTo = 1102;)
</think>
<answer>
SELECT "T1"."territory" FROM "offices" AS "T1" INNER JOIN "emp... | datagen | simple | car_retails |
End of preview. Expand in Data Studio
datagen_5k_v2
Synthetic text-to-SQL SFT mix. SQL is generated by a skeleton sampler against real
SQLite databases, execution-verified, then paired with agent-written questions and a
deterministically-constructed <think> trace.
| split | rows |
|---|---|
train |
4,750 |
validation |
250 |
Columns
| column | description |
|---|---|
input_seq |
Full prompt: task overview, SQLite engine, rendered CREATE TABLE schema with inlined -- example: [...] values, question, instructions, output format. |
output_seq |
SFT target: <think>\n...\n</think>\n<answer>\nSELECT ...\n</answer>. Raw SQL, no fences. |
source |
Always datagen. |
difficulty |
simple / moderate / challenging, from a sqlglot-based scorer. |
db_id |
The source database. Populated here because these rows generate SQL against a known database, unlike the BIRD/Spider trace pools. |
Generation statistics
From 5,700 attempts, 5,000 accepted (87.7%):
rejected: zero-row 256 · oversized 295 · sample-failed 36
difficulty simple 71.3% moderate 26.6% challenging 2.1%
n_tables 1: 32.2% 2: 55.0% 3: 11.3% 4: 1.5%
result shape single-row 72.6% multi-row 27.4% zero-row 0.0%
distinct rate 6.2%
Known limitation — read before training on this alone
The <think> traces are deterministic and templated, not free-form reasoning.
They read as terse structured summaries:
Tables needed: competitor_event (as T1), event (as T2).
Join keys: INNER join "event" AS "T2" on "T1"."event_id" = "T2"."id".
Filters: "T1"."event_id" = 188.
Aggregation: ordered by "T2"."event_name" ASC; limited to 5.
Compare with the verbose, discursive teacher CoT in realcot_11k. These are different
reasoning distributions, and a model trained on this one learns the template.
More importantly, the corpus's construct coverage is narrow — a generator's marginals become the model's, to within about a point. The distribution above is heavily skewed toward simple, 2-table, single-row queries. Every per-row correctness check passes while the corpus as a whole spans very little. Coverage, not per-row correctness, is the gate that matters here.
Length handling
Rows in this build were not filtered through TRL's real
SFTTrainer._prepare_dataset pipeline; a naive length measurement missed 7 rows in a
sibling build that were then silently right-truncated, severing </answer>. Measure
your own length distribution before training and drop rather than truncate.
Provenance
SQL generated against BIRD train databases. Consult that source for licensing and citation.
- Downloads last month
- 34