Dataset Viewer
Auto-converted to Parquet Duplicate
stem
string
composite_png
image
family
string
standard
string
qa_levels
list
questions
list
gt_code
string
threaded_adapter_000240_s4420
threaded_adapter
ASME B1.20.1
[ "L2", "L3", "L4" ]
[ { "qa_id": "11023b6bf58943d5", "qa_level": "L3", "question": "Is every hole radius in this part less than or equal to 1.0?", "answer": 0, "type": "integer", "source": "tier_a_L3", "qa_json": "{\"question\": \"Is every hole radius in this part less than or equal to 1.0?\", \"answer\": 0, ...
import cadquery as cq result = ( cq.Workplane("XY") .polygon(6, 27.713) .extrude(12.0) .union( cq.Workplane("XY") .transformed(offset=cq.Vector(0, 0, 14.0), rotate=cq.Vector(0, 0, 0)) .cylinder(4.0, 11.7) ) .union( cq.Workplane("XY") .transfor...
motor_end_cap_000500_s4420
motor_end_cap
IEC 60072-1
[ "L2", "L3", "L4" ]
[ { "qa_id": "82cf6cb14d3b451f", "qa_level": "L3", "question": "What is the size of the largest chamfer in this part?", "answer": 1.5, "type": "dim", "source": "auto_dim", "qa_json": "{\"question\": \"What is the size of the largest chamfer in this part?\", \"answer\": 1.5, \"type\": \"dim...
import cadquery as cq result = ( cq.Workplane("XZ") .cylinder(15.6, 91.15) .faces(">Y").workplane() .hole(50.8) .faces(">Y").workplane() .circle(53.25) .extrude(10.4) .edges(">Z") .chamfer(1.5) .edges("<Z") .fillet(1.2) .faces(">Y").workplane() .polarArray(59.4, 0, 3...
keyhole_plate_000358_s4420
keyhole_plate
N/A
[ "L1", "L2", "L3" ]
[ { "qa_id": "340a500eedbc49f4", "qa_level": "L3", "question": "What is the size of the largest chamfer in this part?", "answer": 0.4, "type": "dim", "source": "auto_dim", "qa_json": "{\"question\": \"What is the size of the largest chamfer in this part?\", \"answer\": 0.4, \"type\": \"dim...
import cadquery as cq result = ( cq.Workplane("YZ") .rect(35.9, 59.5) .extrude(1.7) .cut( cq.Workplane("YZ") .transformed(offset=cq.Vector(0, 17.33, 0.85), rotate=cq.Vector(0, 0, 0)) .cylinder(3.7, 6.9) ) .cut( cq.Workplane("YZ") .transformed(...
tee_nut_000282_s4420
tee_nut
N/A
[ "L1", "L2", "L3", "L4" ]
[ { "qa_id": "50ff90f836dc4de7", "qa_level": "L3", "question": "What is the barrel OD to thread diameter ratio?", "answer": 1.5, "type": "ratio", "source": "ORIG", "qa_json": "{\"question\": \"What is the barrel OD to thread diameter ratio?\", \"answer\": 1.5, \"type\": \"ratio\", \"source...
import cadquery as cq result = ( cq.Workplane("XY") .circle(11.0) .extrude(1.5) .union( cq.Workplane("XY") .transformed(offset=cq.Vector(0, 0, -5.5), rotate=cq.Vector(0, 0, 0)) .cylinder(11.0, 4.5) ) .cut( cq.Workplane("XY") .transformed(offse...
twisted_bracket_004225_s4252
twisted_bracket
N/A
[ "L1", "L2", "L3", "L4" ]
[ { "qa_id": "3900f103eeab4e93", "qa_level": "L3", "question": "What is the plate width to thickness ratio?", "answer": 3.6667, "type": "ratio", "source": "ORIG", "qa_json": "{\"question\": \"What is the plate width to thickness ratio?\", \"answer\": 3.6667, \"type\": \"ratio\", \"source\"...
import cadquery as cq result = ( cq.Workplane("XY") .transformed(offset=cq.Vector(15.0, 0, -3.0), rotate=cq.Vector(0, 0, 0)) .rect(30.0, 22.0) .extrude(6.0) .union( cq.Workplane("XY") .transformed(offset=cq.Vector(30.0, 0, 0), rotate=cq.Vector(0, 90, 0)) .rect(6.0, 2...
grease_nipple_000032_s4420
grease_nipple
DIN 71412
[ "L1", "L2", "L3", "L4" ]
[ { "qa_id": "9d620120c7bd44b7", "qa_level": "L3", "question": "What is the sum of all extrude / cutBlind depth values?", "answer": 21, "type": "dim", "source": "tier_a_L3", "qa_json": "{\"question\": \"What is the sum of all extrude / cutBlind depth values?\", \"answer\": 21.0, \"type\": ...
import cadquery as cq result = ( cq.Workplane("XZ") .moveTo(0.0, 0.0) .lineTo(3.3, 0.0) .lineTo(4.0, 0.7) .lineTo(4.0, 5.5) .lineTo(2.561, 5.5) .lineTo(2.561, 8.5) .lineTo(2.561, 11.177) .lineTo(3.272, 13.118) .lineTo(1.788, 16.0) .lineTo(0.0, 16.0) .close() .revolve...
eyebolt_000398_s4420
eyebolt
DIN 580
[ "L1", "L2", "L3", "L4" ]
[ { "qa_id": "0be1232c59f34877", "qa_level": "L3", "question": "What is the sum of all extrude / cutBlind depth values?", "answer": 67, "type": "dim", "source": "tier_a_L3", "qa_json": "{\"question\": \"What is the sum of all extrude / cutBlind depth values?\", \"answer\": 67.0, \"type\": ...
import cadquery as cq result = ( cq.Workplane("XY") .circle(30.0) .extrude(22.0) .union( cq.Workplane("XY") .transformed(offset=cq.Vector(0, 0, 22.0), rotate=cq.Vector(0, 0, 0)) .circle(26.91) .transformed(offset=cq.Vector(0, 0, 11.847000000000001), rotate=cq...
heat_sink_000728_s4420
heat_sink
N/A
[ "L2", "L3", "L4" ]
[ { "qa_id": "83a3277fdd594597", "qa_level": "L3", "question": "What is the largest hole radius/diameter argument value used in this part?", "answer": 2.9, "type": "dim", "source": "tier_a_L3", "qa_json": "{\"question\": \"What is the largest hole radius/diameter argument value used in thi...
import cadquery as cq result = ( cq.Workplane("YZ") .box(89.2, 45.8, 6.5) .faces(">X").workplane() .rarray(1, 5.81, 1, 8) .rect(87.2, 1.1) .extrude(21.0) .faces(">Z") .chamfer(0.4) .faces("<X").workplane() .pushPoints([(-6.5, -16.0), (-6.5, 16.0), (6.5, -16.0), (6.5, 16.0)]) ...
bearing_retainer_cap_001082_s4420
bearing_retainer_cap
N/A
[ "L1", "L2", "L3", "L4" ]
[ { "qa_id": "5d17b7e9f66e4630", "qa_level": "L1", "question": "How many hole-type operations are used to build this part?", "answer": 1, "type": "integer", "source": "auto_count", "qa_json": "{\"question\": \"How many hole-type operations are used to build this part?\", \"answer\": 1, \"t...
import cadquery as cq result = ( cq.Workplane("XZ") .cylinder(4.9, 85.3) .union( cq.Workplane("XZ") .transformed(offset=cq.Vector(0.0, 0.0, 24.1), rotate=cq.Vector(0, 0, 0)) .cylinder(44.3, 40.95) ) .cut( cq.Workplane("XZ") .transformed(offset=cq....
chair_000109_s4420
chair
N/A
[ "L1", "L2", "L3", "L4" ]
[ { "qa_id": "aacbc37ec1cf4ba2", "qa_level": "L3", "question": "What is the leg height to seat thickness ratio?", "answer": 9.5, "type": "ratio", "source": "ORIG", "qa_json": "{\"question\": \"What is the leg height to seat thickness ratio?\", \"answer\": 9.5, \"type\": \"ratio\", \"source...
import cadquery as cq result = ( cq.Workplane("XY") .box(159.6, 69.9, 9.0) .union( cq.Workplane("XY") .transformed(offset=cq.Vector(73.05, 28.2, -47.25), rotate=cq.Vector(0, 0, 0)) .box(13.5, 13.5, 85.5) ) .union( cq.Workplane("XY") .transformed(o...
t_slot_rail_002372_s4420
t_slot_rail
DIN 650
[ "L2", "L3", "L4" ]
[ { "qa_id": "7ecf2f09982f4f97", "qa_level": "L2", "question": "How many distinct sketch workplanes are established (.workplane() calls)?", "answer": 2, "type": "integer", "source": "tier_a_L2", "qa_json": "{\"question\": \"How many distinct sketch workplanes are established (.workplane() ...
import cadquery as cq result = ( cq.Workplane("XZ") .moveTo(-35.0, -35.0) .lineTo(35.0, -35.0) .lineTo(35.0, 35.0) .lineTo(14.0, 35.0) .lineTo(14.0, 14.0) .lineTo(21.0, 14.0) .lineTo(21.0, 7.0) .lineTo(-21.0, 7.0) .lineTo(-21.0, 14.0) .lineTo(-14.0, 14.0) .lineTo(-14.0, ...
dovetail_slide_004636_s4420
dovetail_slide
N/A
[ "L1", "L2", "L3", "L4" ]
[ { "qa_id": "a7f198c65e1344c1", "qa_level": "L2", "question": "Along which axis does the primary extrusion occur? Answer 0=X, 1=Y, 2=Z.", "answer": 1, "type": "integer", "source": "tier_a_L2", "qa_json": "{\"question\": \"Along which axis does the primary extrusion occur? Answer 0=X, 1=Y,...
import cadquery as cq result = ( cq.Workplane("XZ") .polyline([(-14.8, 0.0), (14.8, 0.0), (14.8, 10.8), (1.5, 10.8), (9.8, 2.8), (-9.8, 2.8), (-1.5, 10.8), (-14.8, 10.8)]) .close() .extrude(166.6) .faces(">Y") .chamfer(0.7) .faces("<Z").workplane() .pushPoints([(0.0, -27.767), (0.0, 27....
u_bolt_000272_s4420
u_bolt
DIN 3570
[ "L1", "L2", "L3", "L4" ]
[ { "qa_id": "4910f9f795bf40b3", "qa_level": "L3", "question": "Is the smallest cylinder radius less than half the largest cylinder radius?", "answer": 0, "type": "integer", "source": "tier_a_L3", "qa_json": "{\"question\": \"Is the smallest cylinder radius less than half the largest cylin...
import cadquery as cq result = ( cq.Workplane("XY") .transformed(offset=cq.Vector(-27.5, 0.0, 29.25), rotate=cq.Vector(0, 0, 0)) .cylinder(66.5, 5.0) .union( cq.Workplane("XY") .transformed(offset=cq.Vector(27.5, 0.0, 29.25), rotate=cq.Vector(0, 0, 0)) .cylinder(66.5, 5....
torus_link_013124_s4420
torus_link
N/A
[ "L1", "L2", "L3", "L4" ]
[ { "qa_id": "3394ef572c1140aa", "qa_level": "L3", "question": "What is the smallest cylinder radius / first argument used?", "answer": 7.2, "type": "dim", "source": "tier_a_L3", "qa_json": "{\"question\": \"What is the smallest cylinder radius / first argument used?\", \"answer\": 7.2, \"...
import cadquery as cq result = ( cq.Workplane("XY") .moveTo(25.1, 0) .circle(5.3) .revolve(360, (0, 0, 0), (0, 1, 0)) .union( cq.Workplane("XY") .transformed(offset=cq.Vector(32.5, 0.0, 0.0), rotate=cq.Vector(0.0, -0.0, 0.0)) .box(4.2, 8.2, 6.2) ) .union( ...
bevel_gear_002697_s4420
bevel_gear
ISO 23509
[ "L1", "L2", "L3", "L4" ]
[ { "qa_id": "76106107d6044f8b", "qa_level": "L1", "question": "How many extrusion-style operations (extrude or cutBlind) are used?", "answer": 1, "type": "integer", "source": "auto_count", "qa_json": "{\"question\": \"How many extrusion-style operations (extrude or cutBlind) are used?\", ...
import cadquery as cq result = ( cq.Workplane("YZ") .polyline([(9.34, -2.503), (9.762, -1.434), (9.929, -1.437), (10.424, -1.367), (11.215, -1.105), (12.238, -0.543), (12.25, 0.0), (12.238, 0.543), (11.215, 1.105), (10.424, 1.367), (9.929, 1.437), (9.762, 1.434), (9.34, 2.503), (9.171, 3.64), (9.317, 3.72), (9...
z_bracket_000238_s4420
z_bracket
N/A
[ "L1", "L2", "L3", "L4" ]
[ { "qa_id": "ef8e5ee6bd0b4d99", "qa_level": "L3", "question": "What is the arm offset to arm height ratio?", "answer": 1.2773, "type": "ratio", "source": "ORIG", "qa_json": "{\"question\": \"What is the arm offset to arm height ratio?\", \"answer\": 1.2773, \"type\": \"ratio\", \"source\"...
import cadquery as cq result = ( cq.Workplane("XY") .box(56.4, 57.6, 8.5) .union( cq.Workplane("XY") .transformed(offset=cq.Vector(23.35, 0, 20.1), rotate=cq.Vector(0, 0, 0)) .box(9.7, 57.6, 32.7) ) .faces(">Z").workplane() .pushPoints([(-19.74, -17.28), (-19.74,...
manifold_block_000010_s4420
manifold_block
N/A
[ "L1", "L2", "L3", "L4" ]
[ { "qa_id": "59d3164769534cf0", "qa_level": "L3", "question": "What is the block length to channel diameter ratio?", "answer": 16.9375, "type": "ratio", "source": "ORIG", "qa_json": "{\"question\": \"What is the block length to channel diameter ratio?\", \"answer\": 16.9375, \"type\": \"r...
import cadquery as cq result = ( cq.Workplane("YZ") .box(81.3, 32.2, 30.5) .faces(">X").workplane() .pushPoints([(-13.55, 0.0), (13.55, 0.0)]) .hole(4.8) .faces(">X") .faces(">Y").workplane() .pushPoints([(0.0, 0.0)]) .hole(4.8) ) # Export show_object(result)
clevis_pin_000816_s4420
clevis_pin
ISO 2340
[ "L1", "L2", "L3" ]
[ { "qa_id": "3af012d474bc47bd", "qa_level": "L3", "question": "Is the smallest cylinder radius less than half the largest cylinder radius?", "answer": 1, "type": "integer", "source": "tier_a_L3", "qa_json": "{\"question\": \"Is the smallest cylinder radius less than half the largest cylin...
import cadquery as cq result = ( cq.Workplane("XY") .cylinder(56.0, 5.0) .edges(">Z or <Z") .chamfer(1.0) .cut( cq.Workplane("XY") .transformed(offset=cq.Vector(0, 0, -21.0), rotate=cq.Vector(90, 0, 0)) .cylinder(15.0, 1.25) ) .cut( cq.Workplane("XY")...
cam_001927_s4420
cam
N/A
[ "L2", "L3", "L4" ]
[ { "qa_id": "f644f23a51a747c3", "qa_level": "L2", "question": "On which 1-indexed line of the code is the (first) through-hole/hole operation?", "answer": 18, "type": "integer", "source": "tier_a_L2", "qa_json": "{\"question\": \"On which 1-indexed line of the code is the (first) through-...
import cadquery as cq result = ( cq.Workplane("YZ") .polyline([(27.3, 0.0), (27.169, 2.676), (26.775, 5.326), (26.124, 7.925), (25.222, 10.447), (24.076, 12.869), (22.699, 15.167), (21.103, 17.319), (19.304, 19.304), (17.319, 21.103), (15.167, 22.699), (12.869, 24.076), (10.447, 25.222), (7.925, 26.124), (5.32...
spline_hub_000569_s4420
spline_hub
DIN 5480
[ "L1", "L2", "L3" ]
[ { "qa_id": "6c538cfddcb04d4f", "qa_level": "L2", "question": "How many distinct sketch workplanes are established (.workplane() calls)?", "answer": 0, "type": "integer", "source": "tier_a_L2", "qa_json": "{\"question\": \"How many distinct sketch workplanes are established (.workplane() ...
import cadquery as cq result = ( cq.Workplane("XY") .circle(32.6) .extrude(23.2) .cut( cq.Workplane("XY") .transformed(offset=cq.Vector(0, 0, 24.2), rotate=cq.Vector(0, 0, 0)) .moveTo(14.85542, -2.07762) .threePointArc((16.13152, -1.95872), (17.53856, -1.7434...
coil_spring_001210_s4252
coil_spring
DIN 2095
[ "L1", "L2", "L3" ]
[ { "qa_id": "467de7c5c08d46b7", "qa_level": "L3", "question": "How many active coils does the spring have?", "answer": 3, "type": "integer", "source": "ORIG", "qa_json": "{\"question\": \"How many active coils does the spring have?\", \"answer\": 3.0, \"type\": \"integer\", \"source\": \"...
import cadquery as cq result = ( cq.Workplane("XY") .transformed(offset=cq.Vector(12.5, 0, 0), rotate=cq.Vector(-88.1768, 0, 0)) .circle(1.25) .sweep(cq.Workplane("XY").spline([(12.5, 0.0, 0.0), (12.0741, 3.2352, 0.1042), (10.8253, 6.25, 0.2083), (8.8388, 8.8388, 0.3125), (6.25, 10.8253, 0.4167), (3.23...
grease_nipple_005862_s4252
grease_nipple
DIN 71412
[ "L1", "L2", "L3", "L4" ]
[ { "qa_id": "cdb5a0b004cf4d6d", "qa_level": "L3", "question": "What is the hex across-flats to thread diameter ratio?", "answer": 1.1667, "type": "ratio", "source": "ORIG", "qa_json": "{\"question\": \"What is the hex across-flats to thread diameter ratio?\", \"answer\": 1.1667, \"type\":...
import cadquery as cq result = ( cq.Workplane("XY") .workplane(offset=5.5) .polygon(6, 8.083) .extrude(3.0) .union( cq.Workplane("XZ") .moveTo(0.0, 16.0) .lineTo(1.788, 16.0) .lineTo(3.357, 14.068) .lineTo(2.503, 11.29) .lineTo(2.5...
table_000340_s4420
table
N/A
[ "L1", "L2", "L3", "L4" ]
[ { "qa_id": "97d6b51cfc444ac0", "qa_level": "L3", "question": "What is the tabletop length to width ratio?", "answer": 1.9732, "type": "ratio", "source": "ORIG", "qa_json": "{\"question\": \"What is the tabletop length to width ratio?\", \"answer\": 1.9732, \"type\": \"ratio\", \"source\"...
import cadquery as cq result = ( cq.Workplane("YZ") .box(205.8, 104.3, 10.5) .faces(">Z") .edges(">Z") .chamfer(1.3) .union( cq.Workplane("YZ") .transformed(offset=cq.Vector(96.35, 45.6, -55.0), rotate=cq.Vector(0, 0, 0)) .box(13.1, 13.1, 99.5) ) .union( ...
clevis_003212_s4252
clevis
DIN 71751
[ "L2", "L3", "L4" ]
[ { "qa_id": "f9728918b714446d", "qa_level": "L3", "question": "Does any chamfer dimension exceed the largest hole radius in this part?", "answer": 0, "type": "integer", "source": "tier_a_L3", "qa_json": "{\"question\": \"Does any chamfer dimension exceed the largest hole radius in this pa...
import cadquery as cq result = ( cq.Workplane("XY") .box(31.0, 27.9, 47.6) .faces(">Z").workplane() .rect(11.0, 28.9) .cutBlind(-30.0) .faces(">Z") .edges(">Z") .chamfer(1.2) .faces(">X").workplane() .pushPoints([(0, -10.5)]) .hole(10.0) ) # Export show_object(result)
hinge_001033_s4420
hinge
DIN 7954/7955
[ "L1", "L2", "L3", "L4" ]
[ { "qa_id": "3e92de7a77274928", "qa_level": "L2", "question": "Along which axis does the primary extrusion occur? Answer 0=X, 1=Y, 2=Z.", "answer": 0, "type": "integer", "source": "tier_a_L2", "qa_json": "{\"question\": \"Along which axis does the primary extrusion occur? Answer 0=X, 1=Y,...
import cadquery as cq result = ( cq.Workplane("YZ") .box(40.0, 70.0, 2.5) .faces(">X").workplane() .pushPoints([[-10.0, -11.667], [-10.0, 11.667]]) .cskHole(2.7, 5.4, 82.0) .union( cq.Workplane("YZ") .transformed(offset=cq.Vector(24.5, -22.5, 0.0), rotate=cq.Vector(-90.0, 0....
waffle_plate_000613_s4420
waffle_plate
N/A
[ "L1", "L2", "L3", "L4" ]
[ { "qa_id": "5b8f0231351e4d7d", "qa_level": "L4", "question": "If you delete the last hole-type operation in the code, how many holes would remain?", "answer": 0, "type": "integer", "source": "tier_a_L4", "qa_json": "{\"question\": \"If you delete the last hole-type operation in the code,...
import cadquery as cq result = ( cq.Workplane("YZ") .box(77.7, 87.1, 30.1) .edges(">Z") .chamfer(0.4) .faces(">X").workplane() .rarray(17.54, 27.99, 4, 3) .rect(12.3, 18.2) .cutBlind(-17.7) .faces(">X").workplane() .pushPoints([(33.25, 37.95), (33.25, -37.95), (-33.25, 37.95), (...
connector_faceplate_000980_s4420
connector_faceplate
N/A
[ "L1", "L2", "L3", "L4" ]
[ { "qa_id": "fb8d20c1216a4625", "qa_level": "L1", "question": "How many chamfer operations are applied?", "answer": 1, "type": "integer", "source": "auto_count", "qa_json": "{\"question\": \"How many chamfer operations are applied?\", \"answer\": 1, \"type\": \"integer\", \"source\": \"au...
import cadquery as cq result = ( cq.Workplane("YZ") .box(132.4, 93.7, 3.6) .edges(">Z") .chamfer(0.7) .faces(">X").workplane() .center(-28.7, 0.0) .rect(34.7, 51.5) .cutThruAll() .faces(">X").workplane() .center(12.3, 0.0) .rect(21.0, 32.9) .cutThruAll() .faces(">X")...
bearing_retainer_cap_000082_s4420
bearing_retainer_cap
N/A
[ "L1", "L2", "L3", "L4" ]
[ { "qa_id": "dcc06769b6cb431b", "qa_level": "L2", "question": "Along which axis does the primary extrusion occur? Answer 0=X, 1=Y, 2=Z.", "answer": 2, "type": "integer", "source": "tier_a_L2", "qa_json": "{\"question\": \"Along which axis does the primary extrusion occur? Answer 0=X, 1=Y,...
import cadquery as cq result = ( cq.Workplane("XY") .cylinder(4.6, 71.1) .union( cq.Workplane("XY") .transformed(offset=cq.Vector(0.0, 0.0, 14.3), rotate=cq.Vector(0, 0, 0)) .cylinder(25.0, 30.15) ) .cut( cq.Workplane("XY") .transformed(offset=cq....
piston_002723_s4252
piston
N/A
[ "L1", "L2", "L3", "L4" ]
[ { "qa_id": "8385c957c1354c74", "qa_level": "L2", "question": "How many distinct sketch workplanes are established (.workplane() calls)?", "answer": 0, "type": "integer", "source": "tier_a_L2", "qa_json": "{\"question\": \"How many distinct sketch workplanes are established (.workplane() ...
import cadquery as cq result = ( cq.Workplane("XY") .moveTo(0.0, 107.3) .lineTo(34.2, 107.3) .lineTo(34.2, 0.0) .lineTo(0.0, 0.0) .close() .revolve(360, (0, 0, 0), (0, 1, 0)) .transformed(offset=cq.Vector(0.0, 32.8, 0.0), rotate=cq.Vector(0, 0, 0)) .circle(7.55) .cutThruAll() ...
End of preview. Expand in Data Studio

BenchCAD 1.0 in Lance

Find BenchCAD parts by what their questions ask, then retrieve a matching render, its ground-truth CadQuery, and all questions for that part. This release covers all three BenchCAD 1.0 configurations.

Browse on Hugging Face

Browse every QA part, code-generation part, and edit. The QA and code-generation views show the original composite render for each part. The edit configuration uses the edit_bench split name in the viewer. Use LanceDB below for vector search and the complete image and STEP assets. Open the Dataset Viewer

The Dataset Viewer reads the Parquet browse tables in viewer/. LanceDB opens the indexed tables in data/.

Key features

  • The QA table stores each composite PNG once alongside its 12 original questions.
  • The code generation table preserves all five source PNGs and the CadQuery program for each part.
  • The edit table preserves each instruction, both programs, and both original STEP models.

Splits

Source config Source split Lance table Source rows Lance rows Media
QA QA qa 2,400 questions 200 parts 200 composite PNGs
code_gen code_gen code_gen 17,900 parts 17,900 parts 89,500 PNGs
edit-bench edit_bench edit_bench 748 edits 748 edits 1,496 STEP models

All 18 code-generation shards form one logical table. The QA table groups rows by stem after checking that every question assigned to that part shares identical metadata, code, and render. The auxiliary VLMEvalKit export maps one-to-one to those 2,400 questions. It supplies a separate level label for 2,042 questions, so both level labels are preserved. Release counts and configuration

Schema

Table Columns Contents
qa stem, family, standard, gt_code, composite_png Original part ID, metadata, program, and original PNG blob
qa questions, qa_levels The original question ID, level, prompt, numeric answer, type, source, and raw JSON for each question. Each item also retains the VLMEvalKit index, level, and answer text
qa question_embeddings Ordered list of 12 float32 vectors per part, aligned positionally to questions
code_gen stem, family, variant, difficulty, base_plane, standard, code Original part ID, family and variants, and ground-truth CadQuery
code_gen view_0_png through view_3_png, composite_png, part_embedding Original render blobs and one vector from the descriptive metadata
edit_bench record_id, pos_id, family, edit_type, category, category_label, instruction, iou, orig_code, gt_code Original IDs, edit taxonomy and instruction, IoU, and both programs
edit_bench orig_step, gt_step, instruction_embedding Original STEP model blobs and one vector from the edit instruction and family

Pre-built indices

All three tables ship a cosine IVF_FLAT vector index and scalar indices on the source key and family. The QA table also indexes question levels. The code generation and edit tables index difficulty and edit category, respectively.

The release uses 384-dimensional Multilingual E5 small embeddings. It converts underscores in family and variant names to spaces for encoding. Each QA vector uses the family name and one original question. Each code-generation vector uses the family, variant, difficulty, base plane, and applicable standard. Each edit vector uses the family name and original edit instruction. The vectors describe these text inputs. Pinned model revision

Preprocessing adds passage: to each input and uses the model tokenizer with a 512-token cap, attention-masked mean pooling, and L2 normalization. Query text uses query: . With a single query vector, a QA part is scored by its closest question vector in cosine distance.

Why Lance?

  • Random access fetches the renders for the returned part IDs.
  • Part metadata, questions, image bytes, and vectors share one schema in each table.
  • The index ships with the dataset, so a search can use it as soon as the table opens.
  • A writable copy can gain new features without rewriting its original media.
  • Indexed filters can select parts by family or QA level.
  • Version tags keep the same examples accessible across evaluations.

Load with LanceDB

Open the public QA table directly on Hugging Face to inspect its row count and indices. LanceDB documentation

import lancedb

db = lancedb.connect("hf://datasets/lance-format/benchcad-lance/data")
tbl = db.open_table("qa")
print(tbl.count_rows())
print(tbl.list_indices())

For repeated training or heavy random access, download the dataset once and open ./benchcad-lance/data locally.

hf download lance-format/benchcad-lance --repo-type dataset --local-dir ./benchcad-lance

Search

Which parts have benchmark questions about hole diameter? Search the question index to return one row per part. The example prints the closest question and saves the composite render for each of the three hits.

pip install lancedb==0.41.0 sentence-transformers==6.3.0
from pathlib import Path

import lancedb
import numpy as np
from sentence_transformers import SentenceTransformer

encoder = SentenceTransformer(
    "intfloat/multilingual-e5-small",
    revision="614241f622f53c4eeff9890bdc4f31cfecc418b3",
)
query = encoder.encode(
    "query: what is the diameter of a hole through the part?",
    normalize_embeddings=True,
)
db = lancedb.connect("hf://datasets/lance-format/benchcad-lance/data")
tbl = db.open_table("qa")
hits = (
    tbl.search([query], vector_column_name="question_embeddings")
    .metric("cosine")
    .nprobes(1)
    .select(["stem", "family", "questions", "question_embeddings", "_distance"])
    .with_row_id(True)
    .limit(3)
    .to_list()
)

output = Path("benchcad-matches")
output.mkdir(exist_ok=True)
images = tbl.fetch_blob_files("composite_png", [hit["_rowid"] for hit in hits])
for hit, image in zip(hits, images, strict=True):
    closest = int(np.argmax(np.asarray(hit["question_embeddings"]) @ query))
    question = hit["questions"][closest]
    (output / f"{hit['stem']}.png").write_bytes(image.read())
    print(hit["stem"], question["qa_id"], question["question"])

Curate

Use a QA-level filter to collect parts for an evaluation focused on a publisher-defined difficulty level. The original questions remain attached to each candidate.

examples = (
    tbl.search()
    .where("array_has_all(qa_levels, ['L4'])")
    .select(["stem", "family", "questions"])
    .limit(20)
    .to_list()
)
print([row["stem"] for row in examples])

Evolve

On a writable local subset, add the number of questions as a derived feature. Create qa_samples using the final section first.

local = lancedb.connect("./benchcad-local").open_table("qa_samples")
local.add_columns({"question_count": "array_length(questions)"})

Train

Use column projection to feed part descriptors and cached vectors into a PyTorch loop. Preserve the publisher’s benchmark split when designing an evaluation.

from lancedb.permutation import Permutation
from torch.utils.data import DataLoader

local = lancedb.connect("./benchcad-local").open_table("qa_samples")
features = Permutation.identity(local).select_columns(["stem", "question_embeddings"])
for batch in DataLoader(features, batch_size=16):
    print(batch.keys())
    break

Versioning

Pin the released QA table by version so that comparisons use the same question IDs, renders, and search index.

print(tbl.tags.list())
pinned = db.open_table("qa", version=5)
print(pinned.version, pinned.count_rows())

Materialize a subset

Copy only the required columns and parts into a writable local table for Evolve, Train, or Versioning.

batches = (
    tbl.search()
    .where("array_has_all(qa_levels, ['L4'])")
    .select(["stem", "questions", "question_embeddings"])
    .to_batches()
)
lancedb.connect("./benchcad-local").create_table("qa_samples", batches)

Source, license, and changes

BenchCAD 1.0 is by Haozhe Zhang, Kaichen Liu, Miaomiao Chen, Lei Li, Shaojie Yang, Cheng Peng, and Hanjie Chen. The authors release the benchmark records and their generated renders and geometry under CC BY 4.0. Publisher’s BenchCAD 1.0 page · Paper · License notice

The conversion used Hugging Face revision 5919f578ab09ec283603a082fab07c7639ab56eb. It groups QA rows by original part, folds in the auxiliary export’s fields, and adds text embeddings and Lance indices. Media and ground-truth code retain their original bytes or strings. The Hub reported 118,407 downloads in the preceding 30 days on November 18, 2026. Pinned upstream snapshot

Downloads last month
-

Paper for lance-format/benchcad-lance