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
//! # Pretrained Voxtral Realtime models
//!
//! From a model's name to a loaded
//! [`VoxtralConstruct`](super::VoxtralConstruct), through the data layer's
//! pretrained machinery ([`data::pretrained`](crate::data::pretrained)). The
//! whole module needs the `store` and `cache` features.
//!
//! # The index
//!
//! [`default_voxtral_factory`] is the index a caller holds, over two
//! providers. First the catalog row, the built-in catalog's build of
//! [`VOXTRAL_CATALOG_CONTRACT`], stating its geometry, one of
//! [`VoxtralRealtimeConfig`](crate::kits::speech::asr::voxtral::VoxtralRealtimeConfig)'s
//! prefabs:
//!
//! - `mistralai/voxtral-mini-4b-realtime-2602` (`voxtral-mini-4b-realtime`):
//! Mistral's 4B realtime model, its 8.9 GB `bf16` `model.safetensors` and its
//! `tekken.json`. The default.
//!
//! Then `hf:org/repo` names any Hugging Face repo in the published layout:
//! its safetensors checkpoint (one file or shards), its `config.json`, which
//! gives the geometry
//! ([`voxtral_config_from_transformers`](super::weights::voxtral_config_from_transformers)),
//! and its `tekken.json`, which gives the vocabulary and the stream's layout.
//!
//! Every file is pinned to its digest and fetched from a fixed revision of
//! its Hugging Face repo.
//!
//! # Loading
//!
//! [`load_construct`](crate::data::builds::BuildFactory::load_construct)
//! resolves a name to a deferred model carrying [`VoxtralConstructConfig`],
//! brings every file local through the cache, and reads them into the row's
//! prefab through
//! [`load_transformers_tensors`](super::weights::load_transformers_tensors).
pub use *;
pub use *;