walrus_model/
config.rs

1//! Provider configuration.
2//!
3//! `ProviderConfig` and `ApiStandard` are defined in wcore and re-exported here.
4//! `Loader` selects which mistralrs builder to use for local models.
5
6use compact_str::CompactString;
7use serde::{Deserialize, Serialize};
8use std::collections::BTreeMap;
9
10pub use wcore::config::provider::{ApiStandard, ProviderConfig};
11
12/// Selects which mistralrs model builder to use for local inference.
13///
14/// Defaults to `Text` when omitted in config.
15#[derive(Debug, Serialize, Deserialize, Clone, Copy, Default)]
16#[serde(rename_all = "snake_case")]
17pub enum Loader {
18    /// `TextModelBuilder` — standard text models.
19    #[default]
20    Text,
21    /// `GgufModelBuilder` — GGUF quantized models.
22    Gguf,
23    /// `VisionModelBuilder` — vision-language models.
24    Vision,
25}
26
27/// Custom HuggingFace model configuration for local inference.
28///
29/// Allows users to run any HuggingFace model not in the built-in registry.
30/// Memory and quantization are auto-selected at runtime based on system RAM.
31#[derive(Debug, Clone, Serialize, Deserialize)]
32pub struct HfModelConfig {
33    /// HuggingFace repository ID (e.g. `"myorg/MyModel-7B-GGUF"`).
34    pub model_id: String,
35    /// Which mistralrs builder to use.
36    #[serde(default)]
37    pub loader: Loader,
38    /// Explicit GGUF filename (required for GGUF models without a standard
39    /// naming convention).
40    #[serde(default, skip_serializing_if = "Option::is_none")]
41    pub gguf_file: Option<String>,
42    /// Custom chat template override.
43    #[serde(default, skip_serializing_if = "Option::is_none")]
44    pub chat_template: Option<String>,
45}
46
47/// Model provider configuration for the daemon.
48///
49/// Remote providers are configured in `providers`. Local models come from the
50/// built-in registry or from user-defined `models`. The active model name
51/// lives in `[walrus].model`.
52#[derive(Debug, Clone, Serialize, Deserialize, Default)]
53pub struct ModelConfig {
54    /// Optional embedding model
55    #[serde(default, skip_serializing_if = "Option::is_none")]
56    pub embedding: Option<CompactString>,
57    /// Remote providers (local models come from the built-in registry)
58    #[serde(default)]
59    pub providers: BTreeMap<CompactString, ProviderConfig>,
60    /// Custom HuggingFace models for local inference
61    #[serde(default)]
62    pub models: BTreeMap<CompactString, HfModelConfig>,
63}
64
65// MemoryThreshold is generated by build.rs from registry TOML files.
66include!(concat!(env!("OUT_DIR"), "/quantization.rs"));
walrus_model/config.rs

walrus_model/
config.rs
