fount-vlm-nell-02 0.3.11__py3-none-any.whl

mlx_vlm/models/deepseekocr/sam.py

@@ -0,0 +1,489 @@
+from typing import Optional, Tuple, Type
+
+import mlx.core as mx
+import mlx.nn as nn
+
+
+def get_abs_pos_sam(abs_pos, tgt_size):
+    """Interpolate absolute positional embeddings to target size."""
+    dtype = abs_pos.dtype
+    src_size = abs_pos.shape[1]
+
+    if src_size != tgt_size:
+        # Transpose to (B, C, H, W) for interpolation
+        old_pos_embed = abs_pos.transpose(0, 3, 1, 2)
+        old_pos_embed = old_pos_embed.astype(mx.float32)
+
+        # Bicubic interpolation
+        from ..kernels import bicubic_interpolate
+
+        new_pos_embed = bicubic_interpolate(
+            old_pos_embed, size=(tgt_size, tgt_size), antialias=True
+        ).astype(dtype)
+
+        # Transpose back to (B, H, W, C)
+        new_pos_embed = new_pos_embed.transpose(0, 2, 3, 1)
+        return new_pos_embed
+    else:
+        return abs_pos
+
+
+class MLPBlock(nn.Module):
+    """MLP block with GELU activation."""
+
+    def __init__(
+        self,
+        embedding_dim: int,
+        mlp_dim: int,
+        act: Type[nn.Module] = nn.GELU,
+    ) -> None:
+        super().__init__()
+        self.lin1 = nn.Linear(embedding_dim, mlp_dim)
+        self.lin2 = nn.Linear(mlp_dim, embedding_dim)
+        self.act = act()
+
+    def __call__(self, x: mx.array) -> mx.array:
+        return self.lin2(self.act(self.lin1(x)))
+
+
+class Attention(nn.Module):
+    """Multi-head Attention block with relative position embeddings."""
+
+    def __init__(
+        self,
+        dim: int,
+        num_heads: int = 8,
+        qkv_bias: bool = True,
+        use_rel_pos: bool = False,
+        input_size: Optional[Tuple[int, int]] = None,
+    ) -> None:
+        """
+        Args:
+            dim (int): Number of input channels.
+            num_heads (int): Number of attention heads.
+            qkv_bias (bool): If True, add a learnable bias to query, key, value.
+            use_rel_pos (bool): If True, add relative positional embeddings to the attention map.
+            input_size (tuple(int, int) or None): Input resolution for calculating the relative
+                positional parameter size.
+        """
+        super().__init__()
+        self.num_heads = num_heads
+        head_dim = dim // num_heads
+        self.scale = head_dim**-0.5
+
+        self.qkv = nn.Linear(dim, dim * 3, bias=qkv_bias)
+        self.proj = nn.Linear(dim, dim)
+
+        self.use_rel_pos = use_rel_pos
+        if self.use_rel_pos:
+            assert (
+                input_size is not None
+            ), "Input size must be provided if using relative positional encoding."
+            # Initialize relative positional embeddings
+            self.rel_pos_h = mx.zeros((2 * input_size[0] - 1, head_dim))
+            self.rel_pos_w = mx.zeros((2 * input_size[1] - 1, head_dim))
+
+    def __call__(self, x: mx.array) -> mx.array:
+        B, H, W, _ = x.shape
+
+        # QKV projection and reshape
+        qkv = (
+            self.qkv(x)
+            .reshape(B, H * W, 3, self.num_heads, -1)
+            .transpose(2, 0, 3, 1, 4)
+        )
+
+        # Separate q, k, v
+        qkv_reshaped = qkv.reshape(3, B * self.num_heads, H * W, -1)
+        q, k, v = qkv_reshaped[0], qkv_reshaped[1], qkv_reshaped[2]
+
+        # Compute relative positional embeddings if needed
+        rel_h, rel_w = None, None
+        if self.use_rel_pos:
+            rel_h, rel_w = add_decomposed_rel_pos(
+                q, self.rel_pos_h, self.rel_pos_w, (H, W), (H, W)
+            )
+
+        # Reshape for attention
+        q = q.reshape(B, self.num_heads, H * W, -1)
+        k = k.reshape(B, self.num_heads, H * W, -1)
+        v = v.reshape(B, self.num_heads, H * W, -1)
+
+        # Apply scaled dot product attention
+        if self.use_rel_pos:
+            rel_h = rel_h.reshape(
+                B, self.num_heads, rel_h.shape[1], rel_h.shape[2], rel_h.shape[3]
+            )
+            rel_w = rel_w.reshape(
+                B, self.num_heads, rel_w.shape[1], rel_w.shape[2], rel_w.shape[3]
+            )
+            attn_bias = (rel_h + rel_w).reshape(
+                B, self.num_heads, rel_h.shape[2], rel_h.shape[3] * rel_w.shape[4]
+            )
+            x = mx.fast.scaled_dot_product_attention(
+                q, k, v, scale=self.scale, mask=attn_bias
+            )
+        else:
+            x = mx.fast.scaled_dot_product_attention(q, k, v, scale=self.scale)
+
+        # Reshape output
+        x = (
+            x.reshape(B, self.num_heads, H, W, -1)
+            .transpose(0, 2, 3, 1, 4)
+            .reshape(B, H, W, -1)
+        )
+
+        x = self.proj(x)
+        return x
+
+
+class Block(nn.Module):
+    """Transformer blocks with support of window attention and residual propagation."""
+
+    def __init__(
+        self,
+        dim: int,
+        num_heads: int,
+        mlp_ratio: float = 4.0,
+        qkv_bias: bool = True,
+        norm_layer: Type[nn.Module] = nn.LayerNorm,
+        act_layer: Type[nn.Module] = nn.GELU,
+        use_rel_pos: bool = False,
+        window_size: int = 0,
+        input_size: Optional[Tuple[int, int]] = None,
+    ) -> None:
+        """
+        Args:
+            dim (int): Number of input channels.
+            num_heads (int): Number of attention heads in each ViT block.
+            mlp_ratio (float): Ratio of mlp hidden dim to embedding dim.
+            qkv_bias (bool): If True, add a learnable bias to query, key, value.
+            norm_layer (nn.Module): Normalization layer.
+            act_layer (nn.Module): Activation layer.
+            use_rel_pos (bool): If True, add relative positional embeddings to the attention map.
+            window_size (int): Window size for window attention blocks. If it equals 0, then
+                use global attention.
+            input_size (tuple(int, int) or None): Input resolution for calculating the relative
+                positional parameter size.
+        """
+        super().__init__()
+        self.norm1 = norm_layer(dim, eps=1e-6)
+        self.attn = Attention(
+            dim,
+            num_heads=num_heads,
+            qkv_bias=qkv_bias,
+            use_rel_pos=use_rel_pos,
+            input_size=input_size if window_size == 0 else (window_size, window_size),
+        )
+
+        self.norm2 = norm_layer(dim, eps=1e-6)
+        self.mlp = MLPBlock(
+            embedding_dim=dim, mlp_dim=int(dim * mlp_ratio), act=act_layer
+        )
+
+        self.window_size = window_size
+
+    def __call__(self, x: mx.array) -> mx.array:
+        shortcut = x
+        x = self.norm1(x)
+
+        # Window partition
+        if self.window_size > 0:
+            H, W = x.shape[1], x.shape[2]
+            x, pad_hw = window_partition(x, self.window_size)
+
+        x = self.attn(x)
+
+        # Reverse window partition
+        if self.window_size > 0:
+            x = window_unpartition(x, self.window_size, pad_hw, (H, W))
+
+        x = shortcut + x
+        x = x + self.mlp(self.norm2(x))
+
+        return x
+
+
+class PatchEmbed(nn.Module):
+    """Image to Patch Embedding."""
+
+    def __init__(
+        self,
+        kernel_size: Tuple[int, int] = (16, 16),
+        stride: Tuple[int, int] = (16, 16),
+        in_chans: int = 3,
+        embed_dim: int = 768,
+    ) -> None:
+        """
+        Args:
+            kernel_size (Tuple): kernel size of the projection layer.
+            stride (Tuple): stride of the projection layer.
+            in_chans (int): Number of input image channels.
+            embed_dim (int): Patch embedding dimension.
+        """
+        super().__init__()
+        self.proj = nn.Conv2d(
+            in_chans, embed_dim, kernel_size=kernel_size, stride=stride
+        )
+
+    def __call__(self, x: mx.array) -> mx.array:
+        x = self.proj(x)
+        return x
+
+
+class SAMEncoder(nn.Module):
+    """Vision Transformer encoder based on SAM architecture."""
+
+    def __init__(
+        self,
+        img_size: int = 1024,
+        patch_size: int = 16,
+        in_chans: int = 3,
+        embed_dim: int = 768,
+        depth: int = 12,
+        num_heads: int = 12,
+        mlp_ratio: float = 4.0,
+        out_chans: int = 256,
+        qkv_bias: bool = True,
+        norm_layer: Type[nn.Module] = nn.LayerNorm,
+        act_layer: Type[nn.Module] = nn.GELU,
+        use_abs_pos: bool = True,
+        use_rel_pos: bool = True,
+        window_size: int = 14,
+        global_attn_indexes: Tuple[int, ...] = (2, 5, 8, 11),
+        final_out_chans: int = 1024,
+    ) -> None:
+        """
+        Args:
+            img_size (int): Input image size.
+            patch_size (int): Patch size.
+            in_chans (int): Number of input image channels.
+            embed_dim (int): Patch embedding dimension.
+            depth (int): Depth of ViT.
+            num_heads (int): Number of attention heads in each ViT block.
+            mlp_ratio (float): Ratio of mlp hidden dim to embedding dim.
+            out_chans (int): Output channels for neck.
+            qkv_bias (bool): If True, add a learnable bias to query, key, value.
+            norm_layer (nn.Module): Normalization layer.
+            act_layer (nn.Module): Activation layer.
+            use_abs_pos (bool): If True, use absolute positional embeddings.
+            use_rel_pos (bool): If True, add relative positional embeddings to the attention map.
+            window_size (int): Window size for window attention blocks.
+            global_attn_indexes (tuple): Indexes for blocks using global attention.
+            final_out_chans (int): Final output channels after net_3 (1024 for OCR, 896 for OCR-2).
+        """
+        super().__init__()
+        self.img_size = img_size
+
+        self.patch_embed = PatchEmbed(
+            kernel_size=(patch_size, patch_size),
+            stride=(patch_size, patch_size),
+            in_chans=in_chans,
+            embed_dim=embed_dim,
+        )
+
+        self.use_abs_pos = use_abs_pos
+        if use_abs_pos:
+            # Initialize absolute positional embedding with pretrain image size
+            self.pos_embed = mx.zeros(
+                (1, img_size // patch_size, img_size // patch_size, embed_dim)
+            )
+
+        self.blocks = []
+        for i in range(depth):
+            block = Block(
+                dim=embed_dim,
+                num_heads=num_heads,
+                mlp_ratio=mlp_ratio,
+                qkv_bias=qkv_bias,
+                norm_layer=norm_layer,
+                act_layer=act_layer,
+                use_rel_pos=use_rel_pos,
+                window_size=window_size if i not in global_attn_indexes else 0,
+                input_size=(img_size // patch_size, img_size // patch_size),
+            )
+            self.blocks.append(block)
+
+        # Neck layers for output processing
+        self.neck = [
+            nn.Conv2d(embed_dim, out_chans, kernel_size=1, bias=False),
+            nn.LayerNorm(out_chans, eps=1e-6),
+            nn.Conv2d(out_chans, out_chans, kernel_size=3, padding=1, bias=False),
+            nn.LayerNorm(out_chans, eps=1e-6),
+        ]
+
+        # Additional downsampling layers
+        self.net_2 = nn.Conv2d(256, 512, kernel_size=3, stride=2, padding=1, bias=False)
+        self.net_3 = nn.Conv2d(
+            512, final_out_chans, kernel_size=3, stride=2, padding=1, bias=False
+        )
+
+    def __call__(self, x: mx.array) -> mx.array:
+        # Patch embedding
+        x = self.patch_embed(x)
+
+        # Add positional embeddings
+        if self.use_abs_pos:
+            x = x + get_abs_pos_sam(self.pos_embed, x.shape[1])
+
+        # Apply transformer blocks
+        for blk in self.blocks:
+            x = blk(x)
+
+        # Apply neck layers
+        for n in self.neck:
+            x = n(x)
+
+        # Additional downsampling
+        x = self.net_2(x)
+        x = self.net_3(x)
+
+        return x
+
+
+# Utility functions
+
+
+def window_partition(x: mx.array, window_size: int) -> Tuple[mx.array, Tuple[int, int]]:
+    """
+    Partition into non-overlapping windows with padding if needed.
+
+    Args:
+        x (mx.array): input tokens with [B, H, W, C].
+        window_size (int): window size.
+
+    Returns:
+        windows: windows after partition with [B * num_windows, window_size, window_size, C].
+        (Hp, Wp): padded height and width before partition
+    """
+    B, H, W, C = x.shape
+
+    pad_h = (window_size - H % window_size) % window_size
+    pad_w = (window_size - W % window_size) % window_size
+
+    if pad_h > 0 or pad_w > 0:
+        x = mx.pad(x, [(0, 0), (0, pad_h), (0, pad_w), (0, 0)])
+
+    Hp, Wp = H + pad_h, W + pad_w
+
+    x = x.reshape(B, Hp // window_size, window_size, Wp // window_size, window_size, C)
+    windows = x.transpose(0, 1, 3, 2, 4, 5).reshape(-1, window_size, window_size, C)
+
+    return windows, (Hp, Wp)
+
+
+def window_unpartition(
+    windows: mx.array,  # FIXED: Changed from np.ndarray to mx.array
+    window_size: int,
+    pad_hw: Tuple[int, int],
+    hw: Tuple[int, int],
+) -> mx.array:  # FIXED: Changed return type from implicit to mx.array
+    """
+    Window unpartition into original sequences and removing padding.
+
+    Args:
+        windows (mx.array): input tokens with [B * num_windows, window_size, window_size, C].
+        window_size (int): window size.
+        pad_hw (Tuple): padded height and width (Hp, Wp).
+        hw (Tuple): original height and width (H, W) before padding.
+
+    Returns:
+        x: unpartitioned sequences with [B, H, W, C].
+    """
+    Hp, Wp = pad_hw
+    H, W = hw
+    B = windows.shape[0] // (Hp * Wp // window_size // window_size)
+
+    x = windows.reshape(
+        B, Hp // window_size, Wp // window_size, window_size, window_size, -1
+    )
+    x = x.transpose(0, 1, 3, 2, 4, 5).reshape(B, Hp, Wp, -1)
+
+    if Hp > H or Wp > W:
+        x = x[:, :H, :W, :]
+
+    return x
+
+
+def get_rel_pos(q_size: int, k_size: int, rel_pos: mx.array) -> mx.array:
+    """
+    Get relative positional embeddings according to the relative positions of
+    query and key sizes.
+
+    Args:
+        q_size (int): size of query q.
+        k_size (int): size of key k.
+        rel_pos (mx.array): relative position embeddings (L, C).
+
+    Returns:
+        Extracted positional embeddings according to relative positions.
+    """
+    max_rel_dist = int(2 * max(q_size, k_size) - 1)
+
+    # Interpolate rel pos if needed
+    if rel_pos.shape[0] != max_rel_dist:
+        dtype = rel_pos.dtype
+        rel_pos = rel_pos.astype(mx.float32)
+        rel_pos_resized = rel_pos.reshape(1, rel_pos.shape[0], -1).transpose(0, 2, 1)
+
+        # Linear interpolation
+        scale = rel_pos_resized.shape[2] / max_rel_dist
+        indices = mx.arange(max_rel_dist, dtype=mx.float32) * scale
+        idx_floor = mx.floor(indices).astype(mx.int32)
+        idx_ceil = mx.minimum(idx_floor + 1, rel_pos_resized.shape[2] - 1)
+        weight = indices - idx_floor.astype(mx.float32)
+
+        rel_pos_resized = (
+            mx.take(rel_pos_resized, idx_floor, axis=2) * (1 - weight)
+            + mx.take(rel_pos_resized, idx_ceil, axis=2) * weight
+        ).astype(dtype)
+
+        rel_pos_resized = rel_pos_resized.reshape(-1, max_rel_dist).transpose(1, 0)
+    else:
+        rel_pos_resized = rel_pos
+
+    # Scale the coords with short length if shapes for q and k are different
+    q_coords = mx.arange(q_size, dtype=mx.float32)[:, None] * max(k_size / q_size, 1.0)
+    k_coords = mx.arange(k_size, dtype=mx.float32)[None, :] * max(q_size / k_size, 1.0)
+    relative_coords = (q_coords - k_coords) + (k_size - 1) * max(q_size / k_size, 1.0)
+
+    return rel_pos_resized[relative_coords.astype(mx.int32)]
+
+
+def add_decomposed_rel_pos(
+    q: mx.array,  # FIXED: Changed from np.ndarray to mx.array
+    rel_pos_h: mx.array,  # FIXED: Changed from np.ndarray to mx.array
+    rel_pos_w: mx.array,  # FIXED: Changed from np.ndarray to mx.array
+    q_size: Tuple[int, int],
+    k_size: Tuple[int, int],
+) -> Tuple[mx.array, mx.array]:  # FIXED: Added explicit return type
+    """
+    Calculate decomposed Relative Positional Embeddings.
+
+    Args:
+        q (mx.array): query q in the attention layer with shape (B, q_h * q_w, C).
+        rel_pos_h (mx.array): relative position embeddings (Lh, C) for height axis.
+        rel_pos_w (mx.array): relative position embeddings (Lw, C) for width axis.
+        q_size (Tuple): spatial sequence size of query q with (q_h, q_w).
+        k_size (Tuple): spatial sequence size of key k with (k_h, k_w).
+
+    Returns:
+        Tuple of (rel_h, rel_w): relative position biases for height and width.
+    """
+    q_h, q_w = q_size
+    k_h, k_w = k_size
+
+    Rh = get_rel_pos(q_h, k_h, rel_pos_h)
+    Rw = get_rel_pos(q_w, k_w, rel_pos_w)
+
+    B, _, dim = q.shape
+    r_q = q.reshape(B, q_h, q_w, dim)
+
+    rel_h = mx.einsum("bhwc,hkc->bhwk", r_q, Rh)
+    rel_w = mx.einsum("bhwc,wkc->bhwk", r_q, Rw)
+    rel_h = rel_h[..., None]
+    rel_w = rel_w[..., None, :]
+    rel_h = rel_h.reshape(B, q_h * q_w, k_h, 1)
+    rel_w = rel_w.reshape(B, q_h * q_w, 1, k_w)
+
+    return rel_h, rel_w