claude/admiring-lichterman - Branch - vortex-data/vortex

feat: turboquant encoding for vectors

#7167

Comparing

claude/admiring-lichterman

(

de598ad

) with

develop

(

fa6931a

)

Untouched: 1106

New: 14

Skipped: 1522

Benchmarks

2642 total

turboquant_compress_dim128_2bit

vortex/benches/single_encoding_throughput.rs

N/A

N/A15.2 ms

turboquant_compress_dim128_4bit

vortex/benches/single_encoding_throughput.rs

N/A

N/A16.1 ms

turboquant_decompress_dim768_2bit

vortex/benches/single_encoding_throughput.rs

N/A

N/A104.6 ms

turboquant_decompress_dim128_4bit

vortex/benches/single_encoding_throughput.rs

N/A

N/A10.3 ms

turboquant_compress_dim768_2bit

vortex/benches/single_encoding_throughput.rs

N/A

N/A143 ms

turboquant_decompress_dim128_2bit

vortex/benches/single_encoding_throughput.rs

N/A

N/A10.3 ms

turboquant_decompress_dim1536_2bit

vortex/benches/single_encoding_throughput.rs

N/A

N/A226.5 ms

turboquant_compress_dim1024_2bit

vortex/benches/single_encoding_throughput.rs

N/A

N/A144.4 ms

turboquant_decompress_dim1536_4bit

vortex/benches/single_encoding_throughput.rs

N/A

N/A226.8 ms

turboquant_decompress_dim1024_4bit

vortex/benches/single_encoding_throughput.rs

N/A

N/A105.2 ms

turboquant_compress_dim1536_4bit

vortex/benches/single_encoding_throughput.rs

N/A

N/A319.8 ms

turboquant_compress_dim1536_2bit

vortex/benches/single_encoding_throughput.rs

N/A

N/A307.1 ms

turboquant_compress_dim1024_4bit

vortex/benches/single_encoding_throughput.rs

N/A

N/A151.3 ms

turboquant_decompress_dim1024_2bit

vortex/benches/single_encoding_throughput.rs

N/A

N/A105.1 ms

take_map[(0.005, 0.5)]

vortex-array/benches/take_patches.rs

+1%

1.2 ms1.2 ms

runend_compress_u32

vortex/benches/single_encoding_throughput.rs

+1%

7.8 ms7.7 ms

take_struct_sequential_indices

vortex-array/benches/take_struct.rs

+1%

59.6 µs59 µs

bench_compare_primitive[(10000, 4)]

vortex-array/benches/dict_compare.rs

+1%

145.9 µs144.9 µs

chunked_canonicalize_into[(10, 10000, 64, 4)]

encodings/fsst/benches/fsst_compress.rs

+1%

28.3 ms28.1 ms

bench_compare_primitive[(10000, 8)]

vortex-array/benches/dict_compare.rs

+1%

145.8 µs144.9 µs

bench_compare_primitive[(10000, 512)]

vortex-array/benches/dict_compare.rs

+1%

158.1 µs157.2 µs

chunked_bool_canonical_into[(1000, 10)]

vortex-array/benches/chunk_array_builder.rs

+1%

75.5 µs75.1 µs

canonicalize_compare[(1000, 16, 8)]

encodings/fsst/benches/fsst_compress.rs

+1%

257.6 µs256.3 µs

bench_compare_sliced_dict_primitive[(3333, 10000)]

vortex-array/benches/dict_compare.rs

137.7 µs137 µs

take_map[(0.01, 1.0)]

vortex-array/benches/take_patches.rs

2.5 ms2.5 ms

Commits

Click on a commit to change the comparison range

Base

develop

fa6931a

-24.8%

docs[turboquant]: add crate-level docs with compression ratios and error bounds

a9dd626

1 day ago

by lwwmanning

feat[turboquant]: support 9-bit Prod for tensor core int8 GEMM

4e01049

1 day ago

by lwwmanning

+24.79%

Merge remote-tracking branch 'origin/develop' into claude/admiring-lichterman

de598ad

4 hours ago

by lwwmanning

Home Terms Privacy Docs