candding 0.1.0

Pure candle embeddings for Rust with no ONNX: dense, sparse, multi-vector and rerank models today, image models on the roadmap
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45

use crate::models::MultiVectorHead;
use crate::multi_vector::MaskedRowFill;
use crate::multi_vector::ScoreReduction;
use crate::registry::DtypeSupport;
use crate::registry::Family;
use crate::registry::Support;
use crate::registry::WeightsFormat;
use crate::registry::multi_vector::entry::MultiVectorEntry;
use crate::registry::t4::T4_BF16_UNTESTED;

/// `answerdotai/answerai-colbert-small-v1`: colbertv2's layout at a smaller scale, a bias-free `Linear(384, 96)`
/// over a 12-layer BERT; documents are cut at 300 tokens.
pub(super) const ANSWERAI_COLBERT_SMALL_MULTI_VECTOR: MultiVectorEntry = MultiVectorEntry {
  id: "answerdotai/answerai-colbert-small-v1",
  family: Family::Bert,
  head: MultiVectorHead::StanfordColbert,
  default_max_length: 300,
  keeps_cls_row: true,
  projection_in: 384,
  projection_out: 96,
  projection_bias: false,
  normalize: true,
  score_reduction: ScoreReduction::Sum,
  query_marker: Some("[unused0] "),
  document_marker: Some("[unused1] "),
  query_length: Some(32),
  attends_to_mask_tokens: false,
  masked_document_row_fill: MaskedRowFill::DtypeMinimum,
  position_scheme: None,
  cpu: Support::Verified,
  metal: Support::Verified,
  cuda: Support::Verified,
  dtypes: DtypeSupport {
    f32: Support::Verified,
    f16: Support::Verified,
    bf16: Support::Verified,
  },
  license: "Apache-2.0",
  size_gb: 0.13,
  phase: 4,
  weights: WeightsFormat::Safetensors,
  gated: false,
  accel_only_dtypes: &[],
  cell_exceptions: T4_BF16_UNTESTED,
};