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

use crate::models::SparseHead;
use crate::registry::DtypeSupport;
use crate::registry::Family;
use crate::registry::Support;
use crate::registry::WeightsFormat;
use crate::registry::sparse::entry::SparseComputation;
use crate::registry::sparse::entry::SparseEntry;
use crate::registry::t4::T4_BF16_UNTESTED;

/// `Bm42` over `sentence-transformers/all-MiniLM-L6-v2`'s weights: its reference repository publishes only ONNX,
/// so nothing is fetched from it, and its English stopword list is compiled in.
pub(super) const BM42: SparseEntry = SparseEntry {
  id: "sentence-transformers/all-MiniLM-L6-v2",
  family: Some(Family::Bert),
  head: SparseHead::Bm42,
  vocab_size: 30522,
  default_max_length: 256,
  cpu: Support::Verified,
  metal: Support::Verified,
  cuda: Support::Verified,
  dtypes: DtypeSupport::verified(),
  license: "Apache-2.0",
  size_gb: 0.09,
  phase: 3,
  weights: WeightsFormat::Safetensors,
  gated: false,
  accel_only_dtypes: &[],
  head_source: None,
  head_revision: None,
  computation: SparseComputation::Encoded,
  cell_exceptions: T4_BF16_UNTESTED,
};