ferrum-models 0.7.6

Model architectures (LLaMA, Qwen, BERT) for Ferrum inference
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
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69

//! Ferrum 模型层
//!
//! 该 crate 负责围绕 `ferrum-interfaces`/`ferrum-types` 定义的核心抽象
//! 提供模型定义解析、构建器与权重加载占位实现，确保上层可以在
//! 重构阶段编译。

// TTS / Whisper架构里存在大量在不同调用路径下"当下未使用"的字段/方法
// (e.g. 推理期不走的 layers、只在特定 feature 下用的导入)。抑制到 crate 级，
// 新写代码的真实警告由 engine / sampler / scheduler 等下游 crate 的 CI 门把守。
#![allow(
    dead_code,
    unused_imports,
    unused_variables,
    unused_mut,
    unused_parens,
    unused_assignments
)]

pub mod audio_processor;
pub mod common;
pub mod definition;
pub mod executor;
pub mod gguf_config;
pub mod gguf_engine_loader;
pub mod gguf_runtime;
pub mod hf_download;
pub mod image_processor;
pub mod loader;
pub mod lora;
pub mod mel;
pub mod models;
pub mod moe;
pub mod moe_config;
pub mod multimodal;
pub mod registry;
pub mod source;
pub mod tensor_wrapper;
pub mod tokenizer;
pub mod weight_format;

pub use common::{DecoderOnlyLLM, LlmRuntimeConfig};
pub use definition::{ConfigManager, ModelDefinition};
pub use executor::{
    BertModelExecutor, ClipModelExecutor, LlmExecutor, StubModelExecutor, TtsModelExecutor,
    WhisperModelExecutor,
};
pub use hf_download::HfDownloader;
pub use image_processor::ClipImageProcessor;
pub use loader::SafeTensorsLoader;
pub use lora::{
    default_lora_model_id, load_runtime_lora_adapter, load_startup_lora_adapter,
    load_startup_lora_adapters, render_lora_model_id, ActiveLoraAdapter, LoraAdapterConfig,
    RuntimeLoraAdapter, StartupLoraAdapter, StartupLoraSpec,
};
pub use multimodal::{BertModelWrapper, ClipModelWrapper, WhisperModelWrapper};
pub use registry::{
    Architecture, DefaultModelRegistry, ModelAlias, ModelDiscoveryEntry, ModelFormatType,
};
pub use source::{
    DefaultModelSourceResolver, ModelFormat, ModelSourceConfig, ModelSourceResolver,
    ResolvedModelSource,
};
pub use tensor_wrapper::CandleTensorWrapper;
pub use tokenizer::{TokenizerFactory, TokenizerHandle};

pub use ferrum_interfaces::ModelExecutor;
pub use ferrum_types::{
    Activation, AttentionConfig, ModelConfig, ModelInfo, ModelType, NormType, Result, RopeScaling,
};