Beit: Add the gen_relative_position_index() function (#2306)

Co-authored-by: v-espitalier <>

candle-examples/examples/beit/main.rs

@@ -55,7 +55,7 @@ pub fn main() -> anyhow::Result<()> {
         None => {
             let api = hf_hub::api::sync::Api::new()?;
             let api = api.model("vincent-espitalier/candle-beit".into());
-            api.get("beit_base_patch16_384.in22k_ft_in22k_in1k_adapted.safetensors")?
+            api.get("beit_base_patch16_384.in22k_ft_in22k_in1k.safetensors")?
         }
         Some(model) => model.into(),
     };

candle-transformers/src/models/beit.rs

@@ -1,4 +1,4 @@
-use candle::{DType, IndexOp, Result, Tensor, D};
+use candle::{DType, Device, IndexOp, Result, Tensor, D};
 use candle_nn::{layer_norm, LayerNorm, Linear, Module, VarBuilder};
 
 const IMG_SIZE: usize = 384;
@@ -32,7 +32,6 @@ impl Attention {
         num_heads: usize,
         qkv_bias: bool,
         proj_bias: bool,
-        relative_position_index: &Tensor,
     ) -> Result<Self> {
         let qkv = linear(vb.pp("qkv"), dim, dim * 3, qkv_bias)?;
         let proj = linear(vb.pp("proj"), dim, dim, proj_bias)?;
@@ -42,7 +41,8 @@ impl Attention {
             (num_relative_distance, num_heads),
             "relative_position_bias_table",
         )?;
-        let relative_position_index = relative_position_index.clone();
+        let relative_position_index =
+            Self::gen_relative_position_index(relative_position_bias_table.device())?;
         let scale = 1. / ((dim / num_heads) as f64).sqrt();
         Ok(Self {
             qkv,
@@ -56,6 +56,63 @@ impl Attention {
 }
 
 impl Attention {
+    // See: https://github.com/huggingface/pytorch-image-models/blob/main/timm/models/beit.py#L61
+    fn gen_relative_position_index(device: &Device) -> Result<Tensor> {
+        let num_relative_distance = (2 * WINDOW_SIZE - 1) * (2 * WINDOW_SIZE - 1) + 3;
+        let w_area = WINDOW_SIZE * WINDOW_SIZE;
+
+        let t_arange: Tensor = Tensor::arange(0, WINDOW_SIZE as u32, device)?;
+        let t_ndgrid = Tensor::meshgrid(&[&t_arange, &t_arange], false)?;
+        let coords_flatten = Tensor::stack(&t_ndgrid, 0)?.flatten(1, 2)?;
+
+        let tmp1 = coords_flatten
+            .unsqueeze(2)?
+            .broadcast_as((2, w_area, w_area))?
+            .to_dtype(DType::I64)?;
+        let tmp2 = coords_flatten
+            .unsqueeze(1)?
+            .broadcast_as((2, w_area, w_area))?
+            .to_dtype(DType::I64)?;
+        let relative_coords = (tmp1 - tmp2)?
+            .transpose(0, 1)? // 102
+            .transpose(1, 2)? // 120
+            .contiguous()?;
+
+        let relative_coords = relative_coords.slice_assign(
+            &[0..w_area, 0..w_area, 0..1],
+            &(relative_coords.i((0..w_area, 0..w_area, 0..1))? + (WINDOW_SIZE - 1) as f64)?,
+        )?;
+        let relative_coords = relative_coords.slice_assign(
+            &[0..w_area, 0..w_area, 1..2],
+            &(relative_coords.i((0..w_area, 0..w_area, 1..2))? + (WINDOW_SIZE - 1) as f64)?,
+        )?;
+        let relative_coords = relative_coords.slice_assign(
+            &[0..w_area, 0..w_area, 0..1],
+            &(relative_coords.i((.., .., 0..1))? * (2. * (WINDOW_SIZE as f64) - 1.))?,
+        )?;
+
+        Tensor::zeros((w_area + 1, w_area + 1), DType::I64, device)?
+            .slice_assign(&[1.., 1..], &relative_coords.sum(2)?)?
+            .slice_assign(
+                &[0..1, 0..(w_area + 1)],
+                &(Tensor::ones((1, w_area + 1), DType::I64, device)?
+                    * ((num_relative_distance - 3) as f64))?
+                    .to_dtype(DType::I64)?,
+            )?
+            .slice_assign(
+                &[0..(w_area + 1), 0..1],
+                &(Tensor::ones((w_area + 1, 1), DType::I64, device)?
+                    * ((num_relative_distance - 2) as f64))?
+                    .to_dtype(DType::I64)?,
+            )?
+            .slice_assign(
+                &[0..1, 0..1],
+                &(Tensor::ones((1, 1), DType::I64, device)?
+                    * ((num_relative_distance - 1) as f64))?
+                    .to_dtype(DType::I64)?,
+            )
+    }
+
     fn _get_rel_pos_bias(&self) -> Result<Tensor> {
         self.relative_position_bias_table
             .index_select(
@@ -144,21 +201,9 @@ struct Block {
 }
 
 impl Block {
-    fn new(
-        vb: VarBuilder,
-        dim: usize,
-        num_heads: usize,
-        relative_position_index: &Tensor,
-    ) -> Result<Self> {
+    fn new(vb: VarBuilder, dim: usize, num_heads: usize) -> Result<Self> {
         let norm1 = layer_norm(dim, 1e-6, vb.pp("norm1"))?;
-        let attn = Attention::new(
-            vb.pp("attn"),
-            dim,
-            num_heads,
-            true,
-            true,
-            relative_position_index,
-        )?;
+        let attn = Attention::new(vb.pp("attn"), dim, num_heads, true, true)?;
         let ls1 = LayerScale::new(vb.pp("ls1"), dim)?;
         let norm2 = layer_norm(dim, 1e-6, vb.pp("norm2"))?;
         let mlp = Mlp::new(vb.pp("mlp"), dim, dim * 4, true)?;
@@ -240,18 +285,10 @@ impl BeitVisionTransformer {
         let patch_embed = PatchEmbed::new(vb.pp("patch_embed"), PATCH_SIZE, 3, embed_dim)?;
         let cls_token = vb.get((1, 1, embed_dim), "cls_token")?;
         let head = linear(vb.pp("head"), embed_dim, NUM_CLASSES, true)?;
-        let relative_position_index = vb.get((NB_TOKENS, NB_TOKENS), "relative_position_index")?;
         let norm = layer_norm(embed_dim, 1e-6, vb.pp("norm"))?;
         let vb_b = vb.pp("blocks");
         let blocks = (0..depth)
-            .map(|i| {
-                Block::new(
-                    vb_b.pp(&i.to_string()),
-                    embed_dim,
-                    num_heads,
-                    &relative_position_index,
-                )
-            })
+            .map(|i| Block::new(vb_b.pp(&i.to_string()), embed_dim, num_heads))
             .collect::<Result<Vec<_>>>()?;
         Ok(Self {
             patch_embed,