Skip to main content

candle_transformers/models/qwen3_vl/
config.rs

1use candle_nn::Activation;
2
3use crate::serde_default_fn;
4
5serde_default_fn!(Activation, default_vision_hidden_act, Activation::Gelu);
6serde_default_fn!(usize, default_in_channels, 3);
7serde_default_fn!(usize, default_depth, 32);
8serde_default_fn!(usize, default_hidden_size, 3584);
9serde_default_fn!(usize, default_out_hidden_size, 3584);
10serde_default_fn!(usize, default_intermediate_size, 3420);
11serde_default_fn!(usize, default_num_heads, 16);
12serde_default_fn!(usize, default_patch_size, 14);
13serde_default_fn!(usize, default_spatial_merge_size, 2);
14serde_default_fn!(usize, default_temporal_patch_size, 2);
15serde_default_fn!(usize, default_num_position_embeddings, 576);
16serde_default_fn!(Vec<usize>, default_deepstack_visual_indexes, Vec::new());
17
18#[derive(Debug, Clone, serde::Deserialize)]
19pub struct VisionConfig {
20    #[serde(default = "default_depth")]
21    pub depth: usize,
22    #[serde(default = "default_hidden_size")]
23    pub hidden_size: usize,
24    #[serde(default = "default_out_hidden_size")]
25    pub out_hidden_size: usize,
26    #[serde(default = "default_vision_hidden_act")]
27    pub hidden_act: Activation,
28    #[serde(default = "default_intermediate_size")]
29    pub intermediate_size: usize,
30    #[serde(default = "default_num_heads")]
31    pub num_heads: usize,
32    #[serde(default = "default_in_channels")]
33    pub in_chans: usize,
34    #[serde(default = "default_patch_size")]
35    pub patch_size: usize,
36    #[serde(default = "default_spatial_merge_size")]
37    pub spatial_merge_size: usize,
38    #[serde(default = "default_temporal_patch_size")]
39    pub temporal_patch_size: usize,
40    #[serde(default = "default_num_position_embeddings")]
41    pub num_position_embeddings: usize,
42    #[serde(default = "default_deepstack_visual_indexes")]
43    pub deepstack_visual_indexes: Vec<usize>,
44}
45
46#[derive(Debug, Clone, serde::Deserialize)]
47pub struct TextConfig {
48    pub head_dim: usize,
49    pub vocab_size: usize,
50    pub hidden_size: usize,
51    pub intermediate_size: usize,
52    pub num_hidden_layers: usize,
53    pub num_attention_heads: usize,
54    pub num_key_value_heads: usize,
55    pub hidden_act: Activation,
56    pub max_position_embeddings: usize,
57    pub rms_norm_eps: f64,
58    pub tie_word_embeddings: bool,
59    pub rope_theta: f64,
60    pub sliding_window: Option<usize>,
61}
62
63#[derive(Debug, Clone, serde::Deserialize)]
64pub struct Config {
65    pub text_config: TextConfig,
66    pub vision_config: VisionConfig,
67    pub image_token_id: u32,
68    pub video_token_id: u32,
69    pub vision_start_token_id: u32,
70    pub vision_end_token_id: u32,
71}