first phase

.gitignore

@@ -0,0 +1,16 @@
+model.safetensors
+
+# python and vs code stuff
+__pycache__/
+.vscode/
+.env
+.venv/
+.DS_Store
+.env.local
+.env.development.local
+.env.test.local
+.env.production.local
+# venv
+venv/
+
+*.safetensors

moondream2/config.json

@@ -0,0 +1,13 @@
+{
+  "architectures": [
+    "HfMoondream"
+  ],
+  "auto_map": {
+    "AutoConfig": "hf_moondream.HfConfig",
+    "AutoModelForCausalLM": "hf_moondream.HfMoondream"
+  },
+  "config": {},
+  "model_type": "moondream1",
+  "torch_dtype": "float16",
+  "transformers_version": "4.44.0"
+}

moondream2/config.py

@@ -0,0 +1,86 @@
+from dataclasses import dataclass, field
+from typing import Dict, List, Optional
+
+
+@dataclass(frozen=True)
+class TextConfig:
+    dim: int = 2048
+    ff_dim: int = 8192
+    n_layers: int = 24
+    vocab_size: int = 51200
+    max_context: int = 2048
+    n_heads: int = 32
+    n_kv_heads: int = 32
+    prefix_attn: int = 730
+
+
+@dataclass(frozen=True)
+class VisionConfig:
+    enc_dim: int = 1152
+    enc_patch_size: int = 14
+    enc_n_layers: int = 27
+    enc_ff_dim: int = 4304
+    enc_n_heads: int = 16
+    proj_out_dim: int = 2048
+    crop_size: int = 378
+    in_channels: int = 3
+    max_crops: int = 12
+    overlap_margin: int = 4
+    proj_inner_dim: int = 8192
+
+
+@dataclass(frozen=True)
+class RegionConfig:
+    dim: int = 2048
+    coord_feat_dim: int = 256
+    coord_out_dim: int = 1024
+    size_feat_dim: int = 512
+    size_out_dim: int = 2048
+    inner_dim: int = 8192
+
+
+@dataclass(frozen=True)
+class TokenizerConfig:
+    bos_id: int = 50256
+    eos_id: int = 50256
+    templates: Dict[str, Optional[Dict[str, List[int]]]] = field(
+        default_factory=lambda: {
+            "caption": {
+                "short": [198, 198, 16438, 8305, 25],
+                "normal": [198, 198, 24334, 1159, 25],
+                "long": [198, 198, 14617, 8305, 25],
+            },
+            "query": {"prefix": [198, 198, 24361, 25], "suffix": [198, 198, 33706, 25]},
+            "detect": {"prefix": [198, 198, 47504, 25], "suffix": [628]},
+            "point": {"prefix": [198, 198, 12727, 25], "suffix": [628]},
+        }
+    )
+
+
+@dataclass(frozen=True)
+class MoondreamConfig:
+    text: TextConfig = TextConfig()
+    vision: VisionConfig = VisionConfig()
+    region: RegionConfig = RegionConfig()
+    tokenizer: TokenizerConfig = TokenizerConfig()
+
+    @classmethod
+    def from_dict(cls, config_dict: dict):
+        text_config = TextConfig(**config_dict.get("text", {}))
+        vision_config = VisionConfig(**config_dict.get("vision", {}))
+        region_config = RegionConfig(**config_dict.get("region", {}))
+        tokenizer_config = TokenizerConfig(**config_dict.get("tokenizer", {}))
+        return cls(
+            text=text_config,
+            vision=vision_config,
+            region=region_config,
+            tokenizer=tokenizer_config,
+        )
+
+    def to_dict(self):
+        return {
+            "text": self.text.__dict__,
+            "vision": self.vision.__dict__,
+            "region": self.region.__dict__,
+            "tokenizer": self.tokenizer.__dict__,
+        }

moondream2/hf_moondream.py

@@ -0,0 +1,132 @@
+from transformers import PreTrainedModel, PretrainedConfig
+
+from .config import MoondreamConfig
+from .moondream import MoondreamModel
+
+# Files sometimes don't get loaded without these...
+from .image_crops import *
+from .vision import *
+from .text import *
+from .region import *
+from .utils import *
+
+
+def extract_question(text):
+    prefix = "<image>\n\nQuestion: "
+    suffix = "\n\nAnswer:"
+
+    if text.startswith(prefix) and text.endswith(suffix):
+        return text[len(prefix) : -len(suffix)]
+    else:
+        return None
+
+
+class HfConfig(PretrainedConfig):
+    _auto_class = "AutoConfig"
+    model_type = "moondream1"
+
+    def __init__(self, **kwargs):
+        super().__init__(**kwargs)
+        self.config = {}
+
+
+class HfMoondream(PreTrainedModel):
+    _supports_gradient_checkpointing = True
+    _auto_class = "AutoModelForCausalLM"
+    config_class = HfConfig
+
+    def __init__(self, config):
+        super().__init__(config)
+        self.model = MoondreamModel(
+            MoondreamConfig.from_dict(config.config), setup_caches=False
+        )
+        self.model._setup_caches()
+
+    @property
+    def encode_image(self):
+        return self.model.encode_image
+
+    @property
+    def query(self):
+        return self.model.query
+
+    @property
+    def caption(self):
+        return self.model.caption
+
+    @property
+    def detect(self):
+        return self.model.detect
+
+    @property
+    def point(self):
+        return self.model.point
+
+    @property
+    def detect_gaze(self):
+        return self.model.detect_gaze
+
+    def answer_question(
+        self,
+        image_embeds,
+        question,
+        tokenizer=None,
+        chat_history="",
+        result_queue=None,
+        max_new_tokens=256,
+        **kwargs
+    ):
+        answer = self.query(image_embeds, question)["answer"].strip()
+
+        if result_queue is not None:
+            result_queue.put(answer)
+        return answer
+
+    def batch_answer(self, images, prompts, tokenizer=None, **kwargs):
+        answers = []
+        for image, prompt in zip(images, prompts):
+            answers.append(self.query(image, prompt)["answer"].strip())
+        return answers
+
+    def _unsupported_exception(self):
+        raise NotImplementedError(
+            "This method is not supported in the latest version of moondream. "
+            "Consider upgrading to the updated API spec, or alternately pin "
+            "to 'revision=2024-08-26'."
+        )
+
+    def generate(self, image_embeds, prompt, tokenizer, max_new_tokens=128, **kwargs):
+        """
+        Function definition remains unchanged for backwards compatibility.
+        Be aware that tokenizer, max_new_takens, and kwargs are ignored.
+        """
+        prompt_extracted = extract_question(prompt)
+        if prompt_extracted is not None:
+            answer = self.model.query(
+                image=image_embeds, question=prompt_extracted, stream=False
+            )["answer"]
+        else:
+            image_embeds = self.encode_image(image_embeds)
+            prompt_tokens = torch.tensor(
+                [self.model.tokenizer.encode(prompt).ids],
+                device=self.device,
+            )
+
+            def generator():
+                for token in self.model._generate_text(
+                    prompt_tokens,
+                    image_embeds.kv_cache,
+                    image_embeds.pos,
+                    max_new_tokens,
+                ):
+                    yield token
+
+            answer = "".join(list(generator()))
+
+        return [answer]
+
+    def get_input_embeddings(self):
+        return super().get_input_embeddings()
+
+    def input_embeds(self, *args, **kwargs):
+        self._unsupported_exception()

moondream2/image_crops.py

@@ -0,0 +1,213 @@
+import math
+import numpy as np
+import torch
+import pyvips
+
+from typing import TypedDict
+
+
+def select_tiling(
+    height: int, width: int, crop_size: int, max_crops: int
+) -> tuple[int, int]:
+    """
+    Determine the optimal number of tiles to cover an image with overlapping crops.
+    """
+    if height <= crop_size or width <= crop_size:
+        return (1, 1)
+
+    # Minimum required tiles in each dimension
+    min_h = math.ceil(height / crop_size)
+    min_w = math.ceil(width / crop_size)
+
+    # If minimum required tiles exceed max_crops, return proportional distribution
+    if min_h * min_w > max_crops:
+        ratio = math.sqrt(max_crops / (min_h * min_w))
+        return (max(1, math.floor(min_h * ratio)), max(1, math.floor(min_w * ratio)))
+
+    # Perfect aspect-ratio tiles that satisfy max_crops
+    h_tiles = math.floor(math.sqrt(max_crops * height / width))
+    w_tiles = math.floor(math.sqrt(max_crops * width / height))
+
+    # Ensure we meet minimum tile requirements
+    h_tiles = max(h_tiles, min_h)
+    w_tiles = max(w_tiles, min_w)
+
+    # If we exceeded max_crops, scale down the larger dimension
+    if h_tiles * w_tiles > max_crops:
+        if w_tiles > h_tiles:
+            w_tiles = math.floor(max_crops / h_tiles)
+        else:
+            h_tiles = math.floor(max_crops / w_tiles)
+
+    return (max(1, h_tiles), max(1, w_tiles))
+
+
+class OverlapCropOutput(TypedDict):
+    crops: np.ndarray
+    tiling: tuple[int, int]
+
+
+def overlap_crop_image(
+    image: np.ndarray,
+    overlap_margin: int = 4,
+    max_crops: int = 12,
+    base_size: tuple[int, int] = (378, 378),
+    patch_size: int = 14,
+) -> OverlapCropOutput:
+    """
+    Process an image using an overlap-and-resize cropping strategy with margin handling.
+
+    This function takes an input image and creates multiple overlapping crops with
+    consistent margins. It produces:
+    1. A single global crop resized to base_size
+    2. Multiple overlapping local crops that maintain high resolution details
+    3. A patch ordering matrix that tracks correspondence between crops
+
+    The overlap strategy ensures:
+    - Smooth transitions between adjacent crops
+    - No loss of information at crop boundaries
+    - Proper handling of features that cross crop boundaries
+    - Consistent patch indexing across the full image
+
+    Args:
+        image (np.ndarray): Input image as numpy array with shape (H,W,C)
+        base_size (tuple[int,int]): Target size for crops, default (378,378)
+        patch_size (int): Size of patches in pixels, default 14
+        overlap_margin (int): Margin size in patch units, default 4
+        max_crops (int): Maximum number of crops allowed, default 12
+
+    Returns:
+        OverlapCropOutput: Dictionary containing:
+            - crops: A numpy array containing the global crop of the full image (index 0)
+                followed by the overlapping cropped regions (indices 1+)
+            - tiling: Tuple of (height,width) tile counts
+    """
+    original_h, original_w = image.shape[:2]
+
+    # Convert margin from patch units to pixels
+    # margin_pixels = 56
+    margin_pixels = patch_size * overlap_margin
+    # total_margin_pixels = 112
+    total_margin_pixels = margin_pixels * 2  # Both sides
+
+    # Calculate crop parameters
+    # crop_patches = 27
+    crop_patches = base_size[0] // patch_size  # patches per crop dimension
+    # crop_window_patches = 19
+    crop_window_patches = crop_patches - (2 * overlap_margin)  # usable patches
+    # crop_window_size = 266
+    crop_window_size = crop_window_patches * patch_size  # usable size in pixels
+
+    # Determine tiling
+    tiling = select_tiling(
+        original_h - total_margin_pixels,
+        original_w - total_margin_pixels,
+        crop_window_size,
+        max_crops,
+    )
+
+    # Pre-allocate crops.
+    n_crops = tiling[0] * tiling[1] + 1  # 1 = global crop
+    crops = np.zeros(
+        (n_crops, base_size[0], base_size[1], image.shape[2]), dtype=np.uint8
+    )
+
+    # Resize image to fit tiling
+    target_size = (
+        tiling[0] * crop_window_size + total_margin_pixels,
+        tiling[1] * crop_window_size + total_margin_pixels,
+    )
+
+    # Convert to vips for resizing
+    vips_image = pyvips.Image.new_from_array(image)
+    scale_x = target_size[1] / image.shape[1]
+    scale_y = target_size[0] / image.shape[0]
+    resized = vips_image.resize(scale_x, vscale=scale_y)
+    image = resized.numpy()
+
+    # Create global crop
+    scale_x = base_size[1] / vips_image.width
+    scale_y = base_size[0] / vips_image.height
+    global_vips = vips_image.resize(scale_x, vscale=scale_y)
+    crops[0] = global_vips.numpy()
+
+    for i in range(tiling[0]):
+        for j in range(tiling[1]):
+            # Calculate crop coordinates
+            y0 = i * crop_window_size
+            x0 = j * crop_window_size
+
+            # Extract crop with padding if needed
+            y_end = min(y0 + base_size[0], image.shape[0])
+            x_end = min(x0 + base_size[1], image.shape[1])
+
+            crop_region = image[y0:y_end, x0:x_end]
+            crops[
+                1 + i * tiling[1] + j, : crop_region.shape[0], : crop_region.shape[1]
+            ] = crop_region
+
+    return {"crops": crops, "tiling": tiling}
+
+
+def reconstruct_from_crops(
+    crops: torch.Tensor,
+    tiling: tuple[int, int],
+    overlap_margin: int,
+    patch_size: int = 14,
+) -> torch.Tensor:
+    """
+    Reconstruct the original image from overlapping crops into a single seamless image.
+
+    Takes a list of overlapping image crops along with their positional metadata and
+    reconstructs them into a single coherent image by carefully stitching together
+    non-overlapping regions. Handles both numpy arrays and PyTorch tensors.
+
+    Args:
+        crops: List of image crops as numpy arrays or PyTorch tensors with shape
+            (H,W,C)
+        tiling: Tuple of (height,width) indicating crop grid layout
+        patch_size: Size in pixels of each patch, default 14
+        overlap_margin: Number of overlapping patches on each edge, default 4
+
+    Returns:
+        Reconstructed image as numpy array or PyTorch tensor matching input type,
+        with shape (H,W,C) where H,W are the original image dimensions
+    """
+    tiling_h, tiling_w = tiling
+    crop_height, crop_width = crops[0].shape[:2]
+    margin_pixels = overlap_margin * patch_size
+
+    # Calculate output size (only adding margins once)
+    output_h = (crop_height - 2 * margin_pixels) * tiling_h + 2 * margin_pixels
+    output_w = (crop_width - 2 * margin_pixels) * tiling_w + 2 * margin_pixels
+
+    reconstructed = torch.zeros(
+        (output_h, output_w, crops[0].shape[2]),
+        device=crops[0].device,
+        dtype=crops[0].dtype,
+    )
+
+    for i, crop in enumerate(crops):
+        tile_y = i // tiling_w
+        tile_x = i % tiling_w
+
+        # For each tile, determine which part to keep
+        # Keep left margin only for first column
+        x_start = 0 if tile_x == 0 else margin_pixels
+        # Keep right margin only for last column
+        x_end = crop_width if tile_x == tiling_w - 1 else crop_width - margin_pixels
+        # Keep top margin only for first row
+        y_start = 0 if tile_y == 0 else margin_pixels
+        # Keep bottom margin only for last row
+        y_end = crop_height if tile_y == tiling_h - 1 else crop_height - margin_pixels
+
+        # Calculate where this piece belongs in the output
+        out_x = tile_x * (crop_width - 2 * margin_pixels)
+        out_y = tile_y * (crop_height - 2 * margin_pixels)
+
+        # Place the piece
+        reconstructed[
+            out_y + y_start : out_y + y_end, out_x + x_start : out_x + x_end
+        ] = crop[y_start:y_end, x_start:x_end]
+
+    return reconstructed

moondream2/layers.py

@@ -0,0 +1,63 @@
+from dataclasses import dataclass
+from typing import Literal
+
+import torch
+from torch.nn import functional as F
+
+
+def gelu_approx(x):
+    return F.gelu(x, approximate="tanh")
+
+
+@dataclass
+class LinearWeights:
+    weight: torch.Tensor
+    bias: torch.Tensor
+
+
+def linear(x: torch.Tensor, w: LinearWeights) -> torch.Tensor:
+    return F.linear(x, w.weight, w.bias)
+
+
+@dataclass
+class LayerNormWeights:
+    weight: torch.Tensor
+    bias: torch.Tensor
+
+
+def layer_norm(x: torch.Tensor, w: LayerNormWeights) -> torch.Tensor:
+    return F.layer_norm(x, w.bias.shape, w.weight, w.bias)
+
+
+@dataclass
+class MLPWeights:
+    fc1: LinearWeights
+    fc2: LinearWeights
+    act: Literal["gelu_approx"] = "gelu_approx"
+
+
+def mlp(x: torch.Tensor, w: MLPWeights) -> torch.Tensor:
+    x = w.fc1(x)
+    x = gelu_approx(x)
+    x = w.fc2(x)
+    return x
+
+
+@dataclass
+class AttentionWeights:
+    qkv: LinearWeights
+    proj: LinearWeights
+
+
+def attn(x: torch.Tensor, w: AttentionWeights, n_heads: int) -> torch.Tensor:
+    bsz, q_len, d_model = x.shape
+    head_dim = d_model // n_heads
+
+    q, k, v = [
+        t.view(bsz, q_len, n_heads, head_dim).transpose(1, 2)
+        for t in linear(x, w.qkv).chunk(3, dim=-1)
+    ]
+    out = F.scaled_dot_product_attention(q, k, v)
+    out = out.transpose(1, 2).reshape(bsz, q_len, d_model)
+    out = linear(out, w.proj)
+    return out

moondream2/moondream.py

@@ -0,0 +1,476 @@
+import torch
+import torch.nn as nn
+import random
+
+from typing import Literal, Tuple, TypedDict, Union, Dict, Any, Optional, List
+from PIL import Image
+from dataclasses import dataclass
+from tokenizers import Tokenizer
+
+from .config import MoondreamConfig
+from .image_crops import reconstruct_from_crops
+from .vision import vision_encoder, vision_projection, prepare_crops, build_vision_model
+from .text import build_text_model, text_encoder, lm_head, text_decoder
+from .region import decode_coordinate, encode_coordinate, decode_size, encode_size
+from .utils import remove_outlier_points
+import os
+
+TextSamplingSettings = TypedDict(
+    "TextSamplingSettings",
+    {
+        "max_tokens": int,
+        "temperature": float,
+        "top_p": float,
+    },
+    total=False,
+)
+
+ObjectSamplingSettings = TypedDict(
+    "ObjectSamplingSettings",
+    {"max_objects": int},
+    total=False,
+)
+
+DEFAULT_MAX_TOKENS = 768
+DEFAULT_TEMPERATURE = 0.5
+DEFAULT_TOP_P = 0.3
+DEFAULT_MAX_OBJECTS = 50
+
+
+@dataclass(frozen=True)
+class EncodedImage:
+    pos: int
+    caches: List[Tuple[torch.Tensor, torch.Tensor]]
+
+
+class KVCache(nn.Module):
+
+    def __init__(self, n_heads, n_kv_heads, max_context, dim, device, dtype):
+        super().__init__()
+        cache_shape = (1, n_kv_heads, max_context, dim // n_heads)
+        self.register_buffer(
+            "k_cache", torch.zeros(*cache_shape, device=device, dtype=dtype)
+        )
+        self.register_buffer(
+            "v_cache", torch.zeros(*cache_shape, device=device, dtype=dtype)
+        )
+
+    def update(self, pos_ids, k, v):
+        kout, vout = self.k_cache, self.v_cache
+        kout[:, :, pos_ids, :] = k
+        vout[:, :, pos_ids, :] = v
+        return kout, vout
+
+
+class MoondreamModel(nn.Module):
+    def __init__(self, config: MoondreamConfig, dtype=torch.float16, setup_caches=True):
+        super().__init__()
+        self.config = config
+        current_dir = os.path.dirname(os.path.abspath(__file__))
+        self.tokenizer = Tokenizer.from_file(os.path.join(current_dir, "tokenizer.json"))
+        self.vision = build_vision_model(config.vision, dtype)
+        self.text = build_text_model(config.text, dtype)
+
+        # Region Model
+        self.region = nn.ModuleDict(
+            {
+                "coord_encoder": nn.Linear(
+                    config.region.coord_feat_dim, config.region.dim, dtype=dtype
+                ),
+                "coord_decoder": nn.ModuleDict(
+                    {
+                        "fc1": nn.Linear(
+                            config.region.dim, config.region.inner_dim, dtype=dtype
+                        ),
+                        "fc2": nn.Linear(
+                            config.region.inner_dim,
+                            config.region.coord_out_dim,
+                            dtype=dtype,
+                        ),
+                    }
+                ),
+                "size_encoder": nn.Linear(
+                    config.region.size_feat_dim, config.region.dim, dtype=dtype
+                ),
+                "size_decoder": nn.ModuleDict(
+                    {
+                        "fc1": nn.Linear(
+                            config.region.dim, config.region.inner_dim, dtype=dtype
+                        ),
+                        "fc2": nn.Linear(
+                            config.region.inner_dim,
+                            config.region.size_out_dim,
+                            dtype=dtype,
+                        ),
+                    }
+                ),
+            }
+        )
+        self.region.coord_features = nn.Parameter(
+            torch.empty(config.region.coord_feat_dim // 2, 1, dtype=dtype).T
+        )
+        self.region.size_features = nn.Parameter(
+            torch.empty(config.region.size_feat_dim // 2, 2, dtype=dtype).T
+        )
+
+        attn_mask = torch.tril(
+            torch.ones(
+                1, 1, config.text.max_context, config.text.max_context, dtype=torch.bool
+            )
+        )
+        patch_w = config.vision.crop_size // config.vision.enc_patch_size
+        prefix_attn_len = 1 + patch_w**2
+        attn_mask[..., :prefix_attn_len, :prefix_attn_len] = 1
+        self.register_buffer("attn_mask", attn_mask, persistent=False)
+
+        # Initialize KV caches.
+        if setup_caches:
+            self._setup_caches()
+
+    def _setup_caches(self):
+        c = self.config.text
+        for b in self.text.blocks:
+            b.kv_cache = KVCache(
+                c.n_heads,
+                c.n_kv_heads,
+                c.max_context,
+                c.dim,
+                device=self.device,
+                dtype=self.vision.pos_emb.dtype,
+            )
+    @property
+    def device(self):
+        return self.vision.pos_emb.device
+
+    def _vis_enc(self, x: torch.Tensor):
+        return vision_encoder(x, self.vision, self.config.vision)
+
+    def _vis_proj(self, g: torch.Tensor, r: torch.Tensor):
+        return vision_projection(g, r, self.vision, self.config.vision)
+
+    def _prefill(self, x: torch.Tensor, attn_mask: torch.Tensor, pos_ids: torch.Tensor):
+        return text_decoder(x, self.text, attn_mask, pos_ids, self.config.text)
+
+    def _decode_one_tok(
+        self, x: torch.Tensor, attn_mask: torch.Tensor, pos_ids: torch.Tensor
+    ):
+        hidden = text_decoder(x, self.text, attn_mask, pos_ids, self.config.text)
+        logits = lm_head(hidden, self.text)
+        return logits, hidden
+
+    def compile(self):
+        # TODO: vision_projection is not being compiled
+        self._vis_enc = torch.compile(self._vis_enc, fullgraph=True)
+        self._prefill = torch.compile(self._prefill, fullgraph=True)
+        self._decode_one_tok = torch.compile(
+            self._decode_one_tok, fullgraph=True, mode="reduce-overhead"
+        )
+
+    def _run_vision_encoder(self, image: Image.Image) -> torch.Tensor:
+        all_crops, tiling = prepare_crops(image, self.config.vision, device=self.device)
+        torch._dynamo.mark_dynamic(all_crops, 0)
+
+        outputs = self._vis_enc(all_crops)
+
+        global_features = outputs[0]
+        local_features = outputs[1:].view(
+            -1,
+            self.config.vision.enc_n_layers,
+            self.config.vision.enc_n_layers,
+            self.config.vision.enc_dim,
+        )
+
+        reconstructed = reconstruct_from_crops(
+            local_features,
+            tiling,
+            patch_size=1,
+            overlap_margin=self.config.vision.overlap_margin,
+        )
+
+        return self._vis_proj(global_features, reconstructed)
+
+    def encode_image(self, image: Union[Image.Image, EncodedImage]) -> EncodedImage:
+        if isinstance(image, EncodedImage):
+            return image
+        elif not isinstance(image, Image.Image):
+            raise ValueError("image must be a PIL Image or EncodedImage")
+
+        # Run through text model in addition to the vision encoder, to minimize
+        # re-computation if multiple queries are performed on this image.
+        with torch.inference_mode():
+            
+            bos = torch.tensor([[self.config.tokenizer.bos_id]], device=self.device)
+            
+            img_emb = self._run_vision_encoder(image)
+            bos_emb = text_encoder(
+                bos,
+                self.text,
+            )
+            inputs_embeds = torch.cat([bos_emb, img_emb[None]], dim=1)
+            mask = self.attn_mask[:, :, 0 : inputs_embeds.size(1), :]
+            pos_ids = torch.arange(inputs_embeds.size(1), dtype=torch.long, device=self.device)
+            self._prefill(inputs_embeds, mask, pos_ids)
+
+        return EncodedImage(
+            pos=inputs_embeds.size(1),
+            caches=[
+                (
+                    b.kv_cache.k_cache[:, :, : inputs_embeds.size(1), :].clone(),
+                    b.kv_cache.v_cache[:, :, : inputs_embeds.size(1), :].clone(),
+                )
+                for b in self.text.blocks
+            ],
+        )
+
+    def _apply_top_p(self, probs: torch.Tensor, top_p: float):
+        probs_sort, probs_idx = torch.sort(probs, dim=-1, descending=True)
+        probs_sum = torch.cumsum(probs_sort, dim=-1)
+        mask = probs_sum - probs_sort > top_p
+        probs_sort[mask] = 0.0
+        probs_sort.div_(probs_sort.sum(dim=-1, keepdim=True))
+        next_probs = torch.zeros_like(probs)
+        next_probs.scatter_(dim=-1, index=probs_idx, src=probs_sort)
+        return next_probs
+
+    def _prefill_prompt(
+        self, prompt_tokens: torch.Tensor, pos: int, temperature: float, top_p: float
+    ):
+        with torch.inference_mode():
+            prompt_emb = text_encoder(prompt_tokens, self.text)
+            torch._dynamo.mark_dynamic(prompt_emb, 1)
+            mask = self.attn_mask[:, :, pos : pos + prompt_emb.size(1), :]
+            pos_ids = torch.arange(pos, pos + prompt_emb.size(1), dtype=torch.long, device=self.device)
+            hidden = self._prefill(prompt_emb, mask, pos_ids)
+            logits = lm_head(hidden, self.text)
+
+            if temperature == 0:
+                next_token = torch.argmax(logits, dim=-1).unsqueeze(1)
+            else:
+                probs = torch.softmax(logits / temperature, dim=-1)
+                probs = self._apply_top_p(probs, top_p)
+                next_token = torch.multinomial(probs, num_samples=1)
+
+        pos = pos + prompt_emb.size(1)
+        return logits, hidden, next_token, pos
+
+    def _generate_text(
+        self,
+        prompt_tokens: torch.Tensor,
+        pos: int,
+        settings: Optional[TextSamplingSettings] = None,
+    ):
+        max_tokens = (
+            settings.get("max_tokens", DEFAULT_MAX_TOKENS)
+            if settings
+            else DEFAULT_MAX_TOKENS
+        )
+        temperature = (
+            settings.get("temperature", DEFAULT_TEMPERATURE)
+            if settings
+            else DEFAULT_TEMPERATURE
+        )
+        top_p = settings.get("top_p", DEFAULT_TOP_P) if settings else DEFAULT_TOP_P
+
+        _, _, next_token, pos = self._prefill_prompt(
+            prompt_tokens, pos, temperature, top_p
+        )
+
+        def generator(next_token, pos):
+            mask = torch.zeros(1, 1, 2048, device=self.device, dtype=torch.bool)
+            mask[:, :, :pos] = 1
+            pos_ids = torch.tensor([pos], device=self.device, dtype=torch.long)
+            generated_tokens = 0
+
+            # For properly handling token streaming with Unicode
+            token_cache = []
+            print_len = 0
+
+            while (
+                next_token_id := next_token.item()
+            ) != self.config.tokenizer.eos_id and generated_tokens < max_tokens:
+                # Add token to our cache
+                token_cache.append(next_token_id)
+
+                # Decode all tokens collected so far
+                text = self.tokenizer.decode(token_cache)
+
+                # After a newline, we flush the cache completely
+                if text.endswith("\n"):
+                    printable_text = text[print_len:]
+                    token_cache = []
+                    print_len = 0
+                    if printable_text:
+                        yield printable_text
+                # If the last token is a CJK character, we can safely print it
+                elif len(text) > 0 and _is_cjk_char(ord(text[-1])):
+                    printable_text = text[print_len:]
+                    print_len += len(printable_text)
+                    if printable_text:
+                        yield printable_text
+                # Otherwise, only print up to the last space to avoid cutting words
+                else:
+                    last_space_idx = text.rfind(" ", print_len)
+                    if last_space_idx >= print_len:
+                        printable_text = text[print_len : last_space_idx + 1]
+                        print_len += len(printable_text)
+                        if printable_text:
+                            yield printable_text
+
+                with torch.inference_mode():
+                    next_emb = text_encoder(next_token, self.text)
+                    mask[:, :, pos], pos_ids[0] = 1, pos
+                    logits, _ = self._decode_one_tok(next_emb, mask, pos_ids)
+                    pos += 1
+
+                    if temperature == 0:
+                        next_token = torch.argmax(logits, dim=-1).unsqueeze(1)  # (1, 1)
+                    else:
+                        probs = torch.softmax(logits / temperature, dim=-1)  # (1, V)
+                        probs = self._apply_top_p(probs, top_p)
+                        next_token = torch.multinomial(probs, num_samples=1)  # (1, 1)
+
+                    generated_tokens += 1
+
+            # Flush any remaining text in the cache
+            if token_cache:
+                text = self.tokenizer.decode(token_cache)
+                printable_text = text[print_len:]
+                if printable_text:
+                    yield printable_text
+
+        return generator(next_token, pos)
+
+    def load_encoded_image(self, encoded_image: EncodedImage):
+        for b, (k, v) in zip(self.text.blocks, encoded_image.caches):
+            b.kv_cache.k_cache[:, :, : k.size(2), :] = k
+            b.kv_cache.v_cache[:, :, : v.size(2), :] = v
+
+    def _generate_points(
+        self,
+        hidden: torch.Tensor,
+        next_token: torch.Tensor,
+        pos: int,
+        include_size: bool = True,
+        max_objects: int = DEFAULT_MAX_OBJECTS,
+    ):
+        out = []
+        mask = torch.zeros(1, 1, 2048, device=self.device, dtype=torch.bool)
+        mask[:, :, :pos] = 1
+        pos_ids = torch.tensor([pos], device=self.device, dtype=torch.long)
+
+        with torch.inference_mode():
+            while (
+                next_token.item() != self.config.tokenizer.eos_id
+                and len(out) < max_objects
+            ):
+                x_logits = decode_coordinate(hidden, self.region)
+                x_center = torch.argmax(x_logits, dim=-1) / x_logits.size(-1)
+                next_emb = encode_coordinate(
+                    x_center.to(dtype=x_logits.dtype), self.region
+                ).unsqueeze(0)
+
+                # Decode y-coordinate
+                mask[:, :, pos], pos_ids[0] = 1, pos
+                _, hidden = self._decode_one_tok(next_emb, mask, pos_ids)
+                pos += 1
+                y_logits = decode_coordinate(hidden, self.region)
+                y_center = torch.argmax(y_logits, dim=-1) / y_logits.size(-1)
+                next_emb = encode_coordinate(
+                    y_center.to(dtype=y_logits.dtype), self.region
+                ).unsqueeze(0)
+
+                # Decode size
+                if include_size:
+                    mask[:, :, pos], pos_ids[0] = 1, pos
+                    logits, hidden = self._decode_one_tok(next_emb, mask, pos_ids)
+                    pos += 1
+                    size_logits = decode_size(hidden, self.region)
+
+                    # Get bin indices from the logits
+                    w_bin = torch.argmax(size_logits[0], dim=-1)
+                    h_bin = torch.argmax(size_logits[1], dim=-1)
+
+                    # Convert from bin indices to actual size values using the inverse of the log-scale mapping
+                    # Formula: size = 2^((bin / 1023.0) * 10.0 - 10.0)
+                    w = torch.pow(2.0, (w_bin.float() / 1023.0) * 10.0 - 10.0)
+                    h = torch.pow(2.0, (h_bin.float() / 1023.0) * 10.0 - 10.0)
+
+                    next_emb = (
+                        encode_size(
+                            torch.tensor(
+                                [w, h], device=self.device, dtype=size_logits.dtype
+                            ),
+                            self.region,
+                        )
+                        .unsqueeze(0)
+                        .unsqueeze(0)
+                    )
+
+                    # Add object
+                    out.append(
+                        {
+                            "x_min": x_center.item() - w.item() / 2,
+                            "y_min": y_center.item() - h.item() / 2,
+                            "x_max": x_center.item() + w.item() / 2,
+                            "y_max": y_center.item() + h.item() / 2,
+                        }
+                    )
+                else:
+                    out.append({"x": x_center.item(), "y": y_center.item()})
+
+                # Decode next token (x-coordinate, or eos)
+                mask[:, :, pos], pos_ids[0] = 1, pos
+                logits, hidden = self._decode_one_tok(next_emb, mask, pos_ids)
+                pos += 1
+                next_token = torch.argmax(logits, dim=-1)
+
+        return out
+
+    def point(
+        self,
+        image: Union[Image.Image, EncodedImage],
+        object: str,
+        settings: Optional[ObjectSamplingSettings] = None,
+    ):
+        if self.config.tokenizer.templates["point"] is None:
+            raise NotImplementedError("Model does not support pointing.")
+
+        image = self.encode_image(image)
+        self.load_encoded_image(image)
+
+        prompt_tokens = torch.tensor(
+            [
+                self.config.tokenizer.templates["point"]["prefix"]
+                + self.tokenizer.encode(" " + object).ids
+                + self.config.tokenizer.templates["point"]["suffix"]
+            ],
+            device=self.device,
+        )
+
+        _, hidden, next_token, pos = self._prefill_prompt(
+            prompt_tokens, image.pos, temperature=0, top_p=0
+        )
+        hidden = hidden[:, -1:, :]
+
+        max_objects = (
+            settings.get("max_objects", DEFAULT_MAX_OBJECTS)
+            if settings
+            else DEFAULT_MAX_OBJECTS
+        )
+        objects = self._generate_points(
+            hidden, next_token, pos, include_size=False, max_objects=max_objects
+        )
+
+        return {"points": objects}
+
+def _is_cjk_char(cp):
+    """Checks whether CP is the codepoint of a CJK character."""
+    # This defines a "chinese character" as anything in the CJK Unicode block:
+    # https://en.wikipedia.org/wiki/CJK_Unified_Ideographs_(Unicode_block)
+    if (
+        (cp >= 0x4E00 and cp <= 0x9FFF)
+        or (cp >= 0x3400 and cp <= 0x4DBF)
+        or (cp >= 0x2F800 and cp <= 0x2FA1F)
+    ):
+        return True
+    return False

moondream2/region.py

@@ -0,0 +1,89 @@
+import torch
+import torch.nn as nn
+import math
+
+from .layers import linear, mlp
+
+
+def fourier_features(x: torch.Tensor, w: torch.Tensor) -> torch.Tensor:
+    """
+    Applies Fourier feature mapping to input tensor x using frequency matrix w. This
+    projects inputs through sinusoidal functions to create higher dimensional features
+    that help mitigate spectral bias - the tendency of neural networks to learn
+    low-frequency functions more easily than high-frequency ones. By explicitly
+    mapping inputs to higher frequencies through sin/cos transformations, we enable
+    better learning of fine details and higher frequency patterns.
+
+    Args:
+        x: Input tensor to transform
+        w: Matrix of frequencies for the Fourier features transformation
+
+    Returns:
+        Concatenated cosine and sine transformed features as a tensor
+    """
+    f = 2 * math.pi * x @ w
+    return torch.cat([f.cos(), f.sin()], dim=-1)
+
+
+def encode_coordinate(coord: torch.Tensor, w: nn.Module) -> torch.Tensor:
+    """
+    Takes as input a tensor containing a single float coordinate value (x or y)
+    and encodes it into hidden states for input to the text model.
+
+    Args:
+        coord: Tensor with single float coordinate value
+
+    Returns:
+        Encoded hidden states tensor for input to text model
+    """
+    return linear(fourier_features(coord, w.coord_features), w.coord_encoder)
+
+
+def decode_coordinate(hidden_state: torch.Tensor, w: nn.Module) -> torch.Tensor:
+    """
+    Takes as input the last hidden state from the text model and outputs a single logit
+    representing either an x or y coordinate prediction.
+
+    Args:
+        hidden_state: The final hidden state tensor from the text model.
+
+    Returns:
+        A single logit representing the predicted coordinate value (x or y)
+    """
+    return mlp(hidden_state, w.coord_decoder)
+
+
+def encode_size(size: torch.Tensor, w: nn.Module) -> torch.Tensor:
+    """
+    Takes a tensor containing width and height values and encodes them into
+    hidden states for input to the text model.
+
+    Args:
+        size: Tensor with two floats for width and height
+
+    Returns:
+        Encoded hidden states tensor for input to text model
+    """
+    return linear(fourier_features(size, w.size_features), w.size_encoder)
+
+
+def decode_size(hidden_state: torch.Tensor, w: nn.Module) -> torch.Tensor:
+    """
+    Takes as input the last hidden state from the text model and outputs logits
+    for 1024 bins representing width and height in log-scale.
+
+    The bins are distributed according to the formula:
+    bin = (log2(size) + 10.0) / 10.0 * 1023.0
+    where size values are clamped to be at least 1/1024.
+
+    To convert from bin back to size:
+    size = 2^((bin / 1023.0) * 10.0 - 10.0)
+
+    Args:
+        hidden_state: The final hidden state tensor from the text model.
+
+    Returns:
+        A tensor containing logits for 1024 bins for width and height.
+        Shape is (2, 1024) where the first dimension corresponds to width and height.
+    """
+    return mlp(hidden_state, w.size_decoder).view(2, -1)

moondream2/rope.py

@@ -0,0 +1,90 @@
+# Ethically sourced from https://github.com/xjdr-alt/entropix
+
+import torch
+import time
+
+def precompute_freqs_cis(
+    dim: int,
+    end: int,
+    theta: float = 10000.0,
+    use_scaled: bool = False,
+    dtype: torch.dtype = torch.float32,
+) -> torch.Tensor:
+    freqs = 1.0 / (theta ** (torch.arange(0, dim, 2, dtype=dtype)[: (dim // 2)] / dim))
+    t = torch.arange(end, dtype=dtype).unsqueeze(1)
+    freqs = t * freqs.unsqueeze(0)
+    freqs = torch.exp(1j * freqs)
+    return torch.stack([freqs.real, freqs.imag], dim=-1)
+
+def func1(x):
+    #print(x)
+    pass
+
+def func2(x):
+    #print(x)
+    pass
+
+def func3(x):
+    #print(x)
+    pass
+
+def func4(x):
+    #print(x)
+    pass
+
+def func5(x):
+    #print(x)
+    pass
+
+def func6(x):
+    #print(x)
+    pass
+
+def func7(x):
+    #print(x)
+    pass
+
+def func8(x):
+    #print(x)
+    pass
+
+def func9(x):
+    #print(x)
+    pass
+
+def func10(x):
+    #print(x)
+    pass
+
+def func11(x):
+    #print(x)
+    pass
+
+def apply_rotary_emb(
+    x: torch.Tensor,
+    freqs_cis: torch.Tensor,
+    position_ids: torch.Tensor,
+    num_heads: int,
+    rot_dim: int = 32,
+    interleave: bool = False,
+) -> torch.Tensor:
+    assert rot_dim == freqs_cis.shape[-2] * 2
+    assert num_heads == x.shape[1]
+    x_rot, x_pass = x[..., :rot_dim], x[..., rot_dim:]
+
+    d_q = x_rot.shape[-1] // 2
+    xq_r, xq_i = x_rot[..., :d_q], x_rot[..., d_q:]
+
+    # Get the cosine component from freqs_cis
+    cos_component = freqs_cis[..., 0]
+    # Index with position_ids
+    cos_indexed = cos_component[position_ids, :]
+    # Add two dimensions at the beginning
+    freqs_cos = cos_indexed.unsqueeze(0).unsqueeze(0)
+    freqs_sin = freqs_cis[..., 1][position_ids, :].unsqueeze(0).unsqueeze(0)
+
+    # Complex multiplication: (a + bi) * (c + di) = (ac - bd) + (ad + bc)i
+    xq_out_r = xq_r * freqs_cos - xq_i * freqs_sin
+    xq_out_i = xq_r * freqs_sin + xq_i * freqs_cos
+    xq_out = torch.stack((xq_out_r, xq_out_i), dim=-1).flatten(-2)
+    return torch.cat([xq_out.to(x.dtype), x_pass], dim=-1)

moondream2/text.py

@@ -0,0 +1,129 @@
+import torch
+import torch.nn as nn
+
+from torch.nn import functional as F
+
+from .layers import layer_norm, mlp
+from .rope import apply_rotary_emb, precompute_freqs_cis
+from .config import TextConfig
+
+
+def text_encoder(input_ids: torch.Tensor, w: nn.Module):
+    return F.embedding(input_ids, w.wte)
+
+
+def attn(
+    x: torch.Tensor,
+    w: nn.Module,
+    freqs_cis: torch.Tensor,
+    kv_cache: nn.Module,
+    attn_mask: torch.Tensor,
+    n_heads: int,
+    position_ids: torch.Tensor,
+    do_apply_rotary_emb: bool = True,
+):
+    bsz, q_len, d_model = x.shape
+    head_dim = d_model // n_heads
+
+    qkv_out = w.qkv(x)  # shape: (bsz, q_len, (n_heads * 3)*head_dim)
+
+    qkv_reshaped = qkv_out.view(bsz, q_len, 3, n_heads, head_dim)
+
+    # 2. Permute to bring heads before sequence length and QKV to the front
+    # Current: (bsz, q_len, 3, n_heads, head_dim) -> (0, 1, 2, 3, 4)
+    # Target:  (3, bsz, n_heads, q_len, head_dim) -> (2, 0, 3, 1, 4)
+    qkv_permuted = qkv_reshaped.permute(2, 0, 3, 1, 4)
+
+    # 3. Unpack/Split along the first dimension (which now separates Q, K, V)
+    q, k, v = qkv_permuted[0], qkv_permuted[1], qkv_permuted[2]
+
+    q = apply_rotary_emb(q, freqs_cis, position_ids, n_heads)
+    k = apply_rotary_emb(k, freqs_cis, position_ids, n_heads)
+
+    if kv_cache is not None:
+        k, v = kv_cache.update(position_ids, k, v)
+
+    out = F.scaled_dot_product_attention(
+        q, k, v, attn_mask=attn_mask
+    )
+    out = out.transpose(1, 2).reshape(bsz, q_len, d_model)
+    out = w.proj(out)
+    return out
+
+
+def text_decoder(
+    x: torch.Tensor,
+    w: nn.Module,
+    attn_mask: torch.Tensor,
+    position_ids: torch.Tensor,
+    config: TextConfig,
+):
+    for i, block in enumerate(w.blocks):
+        l_in = layer_norm(x, block.ln)
+        l_attn = attn(
+            l_in,
+            block.attn,
+            freqs_cis=w.freqs_cis,
+            kv_cache=block.kv_cache,
+            attn_mask=attn_mask,
+            n_heads=config.n_heads,
+            position_ids=position_ids,
+        )
+        l_mlp = mlp(l_in, block.mlp)
+        x = x + l_attn + l_mlp
+
+    return x
+
+
+def lm_head(hidden_BTC: torch.Tensor, w: nn.Module):
+    hidden_BC = hidden_BTC[:, -1, :]
+    hidden_BC = layer_norm(hidden_BC, w.post_ln)
+    logits = w.lm_head(hidden_BC)
+    return logits
+
+
+def build_text_model(config: TextConfig, dtype: torch.dtype) -> nn.Module:
+    qkv_dim = int(config.dim * 3)
+
+    text = nn.ModuleDict(
+        {
+            "blocks": nn.ModuleList(
+                [
+                    nn.ModuleDict(
+                        {
+                            "ln": nn.LayerNorm(config.dim, dtype=dtype),
+                            "attn": nn.ModuleDict(
+                                {
+                                    "qkv": nn.Linear(config.dim, qkv_dim, dtype=dtype),
+                                    "proj": nn.Linear(
+                                        config.dim, config.dim, dtype=dtype
+                                    ),
+                                }
+                            ),
+                            "mlp": nn.ModuleDict(
+                                {
+                                    "fc1": nn.Linear(
+                                        config.dim, config.ff_dim, dtype=dtype
+                                    ),
+                                    "fc2": nn.Linear(
+                                        config.ff_dim, config.dim, dtype=dtype
+                                    ),
+                                }
+                            ),
+                        }
+                    )
+                    for _ in range(config.n_layers)
+                ]
+            ),
+            "post_ln": nn.LayerNorm(config.dim, dtype=dtype),
+            "lm_head": nn.Linear(config.dim, config.vocab_size, dtype=dtype),
+        }
+    )
+    text.wte = nn.Parameter(torch.empty(config.vocab_size, config.dim, dtype=dtype))
+    text.register_buffer(
+        "freqs_cis",
+        precompute_freqs_cis(config.dim // (2 * config.n_heads), config.max_context),
+        persistent=False,
+    )
+
+    return text

moondream2/utils.py

@@ -0,0 +1,55 @@
+import numpy as np
+
+import torch.cuda.nvtx
+from functools import wraps
+
+def nvtx_annotate(label=None, color='blue'):
+    def decorator(func):
+        @wraps(func)
+        def wrapper(*args, **kwargs):
+            range_name = label or func.__name__
+            torch.cuda.nvtx.range_push(range_name)
+            result = func(*args, **kwargs)
+            torch.cuda.nvtx.range_pop()
+            return result
+        return wrapper
+    return decorator
+
+def remove_outlier_points(points_tuples, k_nearest=2, threshold=2.0):
+    """
+    Robust outlier detection for list of (x,y) tuples.
+    Only requires numpy.
+
+    Args:
+        points_tuples: list of (x,y) tuples
+        k_nearest: number of neighbors to consider
+        threshold: multiplier for median distance
+
+    Returns:
+        list: filtered list of (x,y) tuples with outliers removed
+        list: list of booleans indicating which points were kept (True = kept)
+    """
+    points = np.array(points_tuples)
+    n_points = len(points)
+
+    # Calculate pairwise distances manually
+    dist_matrix = np.zeros((n_points, n_points))
+    for i in range(n_points):
+        for j in range(i + 1, n_points):
+            # Euclidean distance between points i and j
+            dist = np.sqrt(np.sum((points[i] - points[j]) ** 2))
+            dist_matrix[i, j] = dist
+            dist_matrix[j, i] = dist
+
+    # Get k nearest neighbors' distances
+    k = min(k_nearest, n_points - 1)
+    neighbor_distances = np.partition(dist_matrix, k, axis=1)[:, :k]
+    avg_neighbor_dist = np.mean(neighbor_distances, axis=1)
+
+    # Calculate mask using median distance
+    median_dist = np.median(avg_neighbor_dist)
+    mask = avg_neighbor_dist <= threshold * median_dist
+
+    # Return filtered tuples and mask
+    filtered_tuples = [t for t, m in zip(points_tuples, mask) if m]
+    return filtered_tuples

moondream2/vision.py

@@ -0,0 +1,152 @@
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+import numpy as np
+
+from typing import Union, Tuple
+from PIL import Image
+
+from .layers import attn, layer_norm, linear, mlp
+from .image_crops import overlap_crop_image
+from .config import VisionConfig
+
+if torch.backends.mps.is_available():
+    # Non-divisible input sizes are not implemented on MPS device yet.
+    # https://github.com/pytorch/pytorch/issues/96056
+    def adaptive_avg_pool2d(input, output_size):
+        return F.adaptive_avg_pool2d(input.to("cpu"), output_size).to("mps")
+
+else:
+    adaptive_avg_pool2d = F.adaptive_avg_pool2d
+
+DeviceLike = Union[str, torch.device, int]
+
+def prepare_crops(
+    image: Image.Image, config: VisionConfig, device: DeviceLike
+) -> Tuple[torch.Tensor, Tuple[int, int]]:
+    # config.max_crops = 12
+    # config.overlap_margin = 4
+    
+    # (H, W, C)
+    np_image = np.array(image.convert("RGB"))
+    overlap_crops = overlap_crop_image(
+        np_image, max_crops=config.max_crops, overlap_margin=config.overlap_margin
+    )
+    all_crops = overlap_crops["crops"]
+    all_crops = np.transpose(all_crops, (0, 3, 1, 2))
+    all_crops = (
+        torch.from_numpy(all_crops)
+        .to(device=device, dtype=torch.float16)
+        .div_(255.0)
+        .sub_(0.5)
+        .div_(0.5)
+    )
+    return all_crops, overlap_crops["tiling"]
+
+
+def create_patches(x, patch_size):
+    # Original shape: [B, C, H, W]
+    B, C, H, W = x.shape
+    P1 = P2 = patch_size
+
+    # Step 1: Split H and W dimensions into patches
+    # [B, C, H/P1, P1, W/P2, P2]
+    x = x.reshape(B, C, H // P1, P1, W // P2, P2)
+
+    # Step 2: Rearrange dimensions to match target shape
+    # [B, H/P1, W/P2, C, P1, P2]
+    x = x.permute(0, 2, 4, 1, 3, 5)
+
+    # Step 3: Combine dimensions to get final shape
+    # [B, (H/P1)*(W/P2), C*P1*P2]
+    x = x.reshape(B, (H // P1) * (W // P2), C * P1 * P2)
+
+    return x
+
+
+def vision_encoder(input_BCHW: torch.Tensor, w: nn.Module, config: VisionConfig):
+    x = create_patches(input_BCHW, config.enc_patch_size)
+
+    x = linear(x, w.patch_emb)
+    x = x + w.pos_emb
+    for block in w.blocks:
+        x = x + attn(layer_norm(x, block.ln1), block.attn, n_heads=config.enc_n_heads)
+        x = x + mlp(layer_norm(x, block.ln2), block.mlp)
+    x = layer_norm(x, w.post_ln)
+
+    return x
+
+
+def vision_projection(
+    global_features: torch.Tensor,
+    reconstructed: torch.Tensor,
+    w: nn.Module,
+    config: VisionConfig,
+):
+    reconstructed = reconstructed.permute(2, 0, 1)
+    reconstructed = adaptive_avg_pool2d(
+        reconstructed, output_size=(config.enc_n_layers, config.enc_n_layers)
+    )
+    reconstructed = reconstructed.permute(1, 2, 0).view(729, config.enc_dim)
+    final_features = torch.cat([global_features, reconstructed], dim=-1)
+    return mlp(final_features, w.proj_mlp)
+
+
+def build_vision_model(config: VisionConfig, dtype: torch.dtype):
+    patch_dim = config.enc_patch_size * config.enc_patch_size * config.in_channels
+    grid_size = config.crop_size // config.enc_patch_size
+    num_patches = grid_size * grid_size
+
+    vision = nn.ModuleDict(
+        {
+            "patch_emb": nn.Linear(patch_dim, config.enc_dim, dtype=dtype),
+            "blocks": nn.ModuleList(
+                [
+                    nn.ModuleDict(
+                        {
+                            "ln1": nn.LayerNorm(config.enc_dim, dtype=dtype),
+                            "attn": nn.ModuleDict(
+                                {
+                                    "qkv": nn.Linear(
+                                        config.enc_dim, 3 * config.enc_dim, dtype=dtype
+                                    ),
+                                    "proj": nn.Linear(
+                                        config.enc_dim, config.enc_dim, dtype=dtype
+                                    ),
+                                }
+                            ),
+                            "ln2": nn.LayerNorm(config.enc_dim, dtype=dtype),
+                            "mlp": nn.ModuleDict(
+                                {
+                                    "fc1": nn.Linear(
+                                        config.enc_dim, config.enc_ff_dim, dtype=dtype
+                                    ),
+                                    "fc2": nn.Linear(
+                                        config.enc_ff_dim, config.enc_dim, dtype=dtype
+                                    ),
+                                }
+                            ),
+                        }
+                    )
+                    for _ in range(config.enc_n_layers)
+                ]
+            ),
+            "post_ln": nn.LayerNorm(config.enc_dim, dtype=dtype),
+            "proj_mlp": nn.ModuleDict(
+                {
+                    "fc1": nn.Linear(
+                        config.enc_dim * 2, config.proj_inner_dim, dtype=dtype
+                    ),
+                    "fc2": nn.Linear(
+                        config.proj_inner_dim, config.proj_out_dim, dtype=dtype
+                    ),
+                }
+            ),
+        }
+    )
+    vision.pos_emb = nn.Parameter(
+        torch.zeros(1, num_patches, config.enc_dim, dtype=dtype)
+    )
+    return vision
+
+

moondream2/weights.py

@@ -0,0 +1,292 @@
+import safetensors
+import torch
+import torch.nn as nn
+
+from contextlib import contextmanager
+from dataclasses import dataclass
+from typing import Callable, List
+
+from .layers import AttentionWeights, LayerNormWeights, LinearWeights, MLPWeights
+
+
+@dataclass
+class VisionBlock:
+    ln1: LayerNormWeights
+    attn: AttentionWeights
+    ln2: LayerNormWeights
+    mlp: MLPWeights
+
+
+@dataclass
+class VisionModel:
+    patch_emb: LinearWeights
+    pos_emb: torch.Tensor
+    blocks: List[VisionBlock]
+    post_ln: LayerNormWeights
+    proj_mlp: MLPWeights
+
+
+@dataclass
+class TextBlock:
+    ln: LayerNormWeights
+    attn: AttentionWeights
+    mlp: MLPWeights
+
+
+@dataclass
+class TextModel:
+    wte: torch.Tensor
+    blocks: List[TextBlock]
+    post_ln: LayerNormWeights
+    lm_head: LinearWeights
+
+
+@dataclass
+class RegionModel:
+    coord_features: torch.Tensor
+    coord_encoder: LinearWeights
+    coord_decoder: MLPWeights
+    size_features: torch.Tensor
+    size_encoder: LinearWeights
+    size_decoder: MLPWeights
+
+
+@dataclass
+class MoondreamModel:
+    vision: VisionModel
+    text: TextModel
+    region: RegionModel
+
+
+@contextmanager
+def safetensors_open(safetensors_file: str):
+    """
+    Simplify interfacing with safetensors files. Eliminates the need to ignore
+    type errors when using the `safe_open` function.
+    """
+    with safetensors.safe_open(
+        safetensors_file, framework="pt"
+    ) as st:  # pyright: ignore
+
+        def get_tensor(name: str) -> torch.Tensor:
+            return st.get_tensor(name)
+
+        def get_keys() -> List[str]:
+            return st.keys()
+
+        get_tensor.keys = get_keys
+
+        yield get_tensor
+
+
+def _load_weights(get_tensor: Callable[[str], torch.Tensor], model: nn.Module) -> None:
+    """Internal function to load weights using a tensor getter function."""
+    model = model.to(dtype=torch.float16)
+
+    # Vision Model
+    model.vision["patch_emb"].weight.data.copy_(
+        get_tensor("vision_encoder.encoder.model.visual.patch_embed.linear.weight")
+    )
+    model.vision["patch_emb"].bias.data.copy_(
+        get_tensor("vision_encoder.encoder.model.visual.patch_embed.linear.bias")
+    )
+    model.vision.pos_emb.data.copy_(
+        get_tensor("vision_encoder.encoder.model.visual.pos_embed")
+    )
+
+    for i in range(len(model.vision["blocks"])):
+        prefix = f"vision_encoder.encoder.model.visual.blocks.{i}"
+
+        # Layer norms
+        model.vision["blocks"][i]["ln1"].weight.data.copy_(
+            get_tensor(f"{prefix}.norm1.weight")
+        )
+        model.vision["blocks"][i]["ln1"].bias.data.copy_(
+            get_tensor(f"{prefix}.norm1.bias")
+        )
+        model.vision["blocks"][i]["ln2"].weight.data.copy_(
+            get_tensor(f"{prefix}.norm2.weight")
+        )
+        model.vision["blocks"][i]["ln2"].bias.data.copy_(
+            get_tensor(f"{prefix}.norm2.bias")
+        )
+
+        # Attention
+        model.vision["blocks"][i]["attn"]["qkv"].weight.data.copy_(
+            get_tensor(f"{prefix}.attn.qkv.weight")
+        )
+        model.vision["blocks"][i]["attn"]["qkv"].bias.data.copy_(
+            get_tensor(f"{prefix}.attn.qkv.bias")
+        )
+        model.vision["blocks"][i]["attn"]["proj"].weight.data.copy_(
+            get_tensor(f"{prefix}.attn.proj.weight")
+        )
+        model.vision["blocks"][i]["attn"]["proj"].bias.data.copy_(
+            get_tensor(f"{prefix}.attn.proj.bias")
+        )
+
+        # MLP
+        model.vision["blocks"][i]["mlp"]["fc1"].weight.data.copy_(
+            get_tensor(f"{prefix}.mlp.fc1.weight")
+        )
+        model.vision["blocks"][i]["mlp"]["fc1"].bias.data.copy_(
+            get_tensor(f"{prefix}.mlp.fc1.bias")
+        )
+        model.vision["blocks"][i]["mlp"]["fc2"].weight.data.copy_(
+            get_tensor(f"{prefix}.mlp.fc2.weight")
+        )
+        model.vision["blocks"][i]["mlp"]["fc2"].bias.data.copy_(
+            get_tensor(f"{prefix}.mlp.fc2.bias")
+        )
+
+    model.vision["post_ln"].weight.data.copy_(
+        get_tensor("vision_encoder.encoder.model.visual.norm.weight")
+    )
+    model.vision["post_ln"].bias.data.copy_(
+        get_tensor("vision_encoder.encoder.model.visual.norm.bias")
+    )
+
+    model.vision["proj_mlp"]["fc1"].weight.data.copy_(
+        get_tensor("vision_encoder.projection.mlp.fc1.weight")
+    )
+    model.vision["proj_mlp"]["fc1"].bias.data.copy_(
+        get_tensor("vision_encoder.projection.mlp.fc1.bias")
+    )
+    model.vision["proj_mlp"]["fc2"].weight.data.copy_(
+        get_tensor("vision_encoder.projection.mlp.fc2.weight")
+    )
+    model.vision["proj_mlp"]["fc2"].bias.data.copy_(
+        get_tensor("vision_encoder.projection.mlp.fc2.bias")
+    )
+
+    # Text Model
+    model.text.wte.data.copy_(get_tensor("text_model.transformer.embd.wte.weight"))
+
+    for i in range(len(model.text["blocks"])):
+        prefix = f"text_model.transformer.h.{i}"
+
+        # Layer norm
+        model.text["blocks"][i]["ln"].weight.data.copy_(
+            get_tensor(f"{prefix}.ln.weight")
+        )
+        model.text["blocks"][i]["ln"].bias.data.copy_(get_tensor(f"{prefix}.ln.bias"))
+
+        # Attention
+        model.text["blocks"][i]["attn"]["qkv"].weight.data.copy_(
+            get_tensor(f"{prefix}.mixer.Wqkv.weight")
+        )
+        model.text["blocks"][i]["attn"]["qkv"].bias.data.copy_(
+            get_tensor(f"{prefix}.mixer.Wqkv.bias")
+        )
+        model.text["blocks"][i]["attn"]["proj"].weight.data.copy_(
+            get_tensor(f"{prefix}.mixer.out_proj.weight")
+        )
+        model.text["blocks"][i]["attn"]["proj"].bias.data.copy_(
+            get_tensor(f"{prefix}.mixer.out_proj.bias")
+        )
+
+        # MLP
+        model.text["blocks"][i]["mlp"]["fc1"].weight.data.copy_(
+            get_tensor(f"{prefix}.mlp.fc1.weight")
+        )
+        model.text["blocks"][i]["mlp"]["fc1"].bias.data.copy_(
+            get_tensor(f"{prefix}.mlp.fc1.bias")
+        )
+        model.text["blocks"][i]["mlp"]["fc2"].weight.data.copy_(
+            get_tensor(f"{prefix}.mlp.fc2.weight")
+        )
+        model.text["blocks"][i]["mlp"]["fc2"].bias.data.copy_(
+            get_tensor(f"{prefix}.mlp.fc2.bias")
+        )
+
+    model.text["post_ln"].weight.data.copy_(get_tensor("text_model.lm_head.ln.weight"))
+    model.text["post_ln"].bias.data.copy_(get_tensor("text_model.lm_head.ln.bias"))
+
+    model.text["lm_head"].weight.data.copy_(
+        get_tensor("text_model.lm_head.linear.weight")
+    )
+    model.text["lm_head"].bias.data.copy_(get_tensor("text_model.lm_head.linear.bias"))
+
+    # Region Model
+    model.region.coord_features.data.copy_(
+        get_tensor("region_model.coordinate_features.weight").T
+    )
+    model.region["coord_encoder"].weight.data.copy_(
+        get_tensor("region_model.coordinate_encoder.weight")
+    )
+    model.region["coord_encoder"].bias.data.copy_(
+        get_tensor("region_model.coordinate_encoder.bias")
+    )
+
+    model.region["coord_decoder"]["fc1"].weight.data.copy_(
+        get_tensor("region_model.coordinate_decoder.fc1.weight")
+    )
+    model.region["coord_decoder"]["fc1"].bias.data.copy_(
+        get_tensor("region_model.coordinate_decoder.fc1.bias")
+    )
+    model.region["coord_decoder"]["fc2"].weight.data.copy_(
+        get_tensor("region_model.coordinate_decoder.fc2.weight")
+    )
+    model.region["coord_decoder"]["fc2"].bias.data.copy_(
+        get_tensor("region_model.coordinate_decoder.fc2.bias")
+    )
+
+    model.region.size_features.data.copy_(
+        get_tensor("region_model.size_features.weight").T
+    )
+    model.region["size_encoder"].weight.data.copy_(
+        get_tensor("region_model.size_encoder.weight")
+    )
+    model.region["size_encoder"].bias.data.copy_(
+        get_tensor("region_model.size_encoder.bias")
+    )
+
+    model.region["size_decoder"]["fc1"].weight.data.copy_(
+        get_tensor("region_model.size_decoder.fc1.weight")
+    )
+    model.region["size_decoder"]["fc1"].bias.data.copy_(
+        get_tensor("region_model.size_decoder.fc1.bias")
+    )
+    model.region["size_decoder"]["fc2"].weight.data.copy_(
+        get_tensor("region_model.size_decoder.fc2.weight")
+    )
+    model.region["size_decoder"]["fc2"].bias.data.copy_(
+        get_tensor("region_model.size_decoder.fc2.bias")
+    )
+
+
+def load_weights_from_safetensors(weights_file: str, model: nn.Module) -> None:
+    """Load weights from a safetensors file into a MoondreamModel instance."""
+    with safetensors_open(weights_file) as get_tensor:
+        # Wrap the get_tensor function to handle key normalization
+        name_map = {k.replace("._orig_mod", ""): k for k in get_tensor.keys()}
+        _load_weights(lambda x: get_tensor(name_map[x]).to(dtype=torch.float16), model)
+
+
+def load_weights_from_pt(weights_file: str, model: nn.Module) -> None:
+    """Load weights from a PyTorch file into a MoondreamModel instance."""
+    device = str(torch.empty(0).device)
+    tensors = torch.load(weights_file, map_location=device, weights_only=True)
+    tensors = {
+        k.replace("._orig_mod", ""): v.to(dtype=torch.float16)
+        for k, v in tensors.items()
+    }
+    _load_weights(lambda x: tensors[x], model)
+
+
+def load_weights_into_model(weights_file: str, model: nn.Module) -> None:
+    """
+    Load weights from either a safetensors or PyTorch file directly into a MoondreamModel instance.
+
+    Args:
+        weights_file: Path to weights file (either .safetensors or .pt)
+        model: MoondreamModel instance to load weights into
+    """
+    if weights_file.endswith(".safetensors"):
+        load_weights_from_safetensors(weights_file, model)
+    else:
+        load_weights_from_pt(weights_file, model)
+
+    # Make all parameters contiguous
+    for param in model.parameters():
+        param.data = param.data.contiguous()

notes.ipynb

@@ -0,0 +1,420 @@
+{
+ "cells": [
+  {
+   "cell_type": "code",
+   "execution_count": 2,
+   "metadata": {},
+   "outputs": [
+    {
+     "data": {
+      "text/plain": [
+       "True"
+      ]
+     },
+     "execution_count": 2,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "source": [
+    "import torch\n",
+    "torch.cuda.is_available()"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 1,
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "/home/pixel/Desktop/moondream/venv/lib/python3.13/site-packages/tqdm/auto.py:21: TqdmWarning: IProgress not found. Please update jupyter and ipywidgets. See https://ipywidgets.readthedocs.io/en/stable/user_install.html\n",
+      "  from .autonotebook import tqdm as notebook_tqdm\n"
+     ]
+    },
+    {
+     "data": {
+      "text/plain": [
+       "False"
+      ]
+     },
+     "execution_count": 1,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "source": [
+    "from transformers.utils import is_flash_attn_2_available\n",
+    "is_flash_attn_2_available()\n"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 1,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import torch"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 1,
+   "metadata": {},
+   "outputs": [
+    {
+     "ename": "RuntimeError",
+     "evalue": "Error(s) in loading state_dict for MoondreamModel:\n\tWhile copying the parameter named \"region.size_features\", whose dimensions in the model are torch.Size([2, 256]) and whose dimensions in the checkpoint are torch.Size([2, 256]), an exception occurred : ('CUDA error: out of memory\\nCUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect.\\nFor debugging consider passing CUDA_LAUNCH_BLOCKING=1\\nCompile with `TORCH_USE_CUDA_DSA` to enable device-side assertions.\\n',).",
+     "output_type": "error",
+     "traceback": [
+      "\u001b[31m---------------------------------------------------------------------------\u001b[39m",
+      "\u001b[31mRuntimeError\u001b[39m                              Traceback (most recent call last)",
+      "\u001b[36mCell\u001b[39m\u001b[36m \u001b[39m\u001b[32mIn[1]\u001b[39m\u001b[32m, line 14\u001b[39m\n\u001b[32m     12\u001b[39m \u001b[38;5;28;01mfrom\u001b[39;00m\u001b[38;5;250m \u001b[39m\u001b[34;01msafetensors\u001b[39;00m\u001b[34;01m.\u001b[39;00m\u001b[34;01mtorch\u001b[39;00m\u001b[38;5;250m \u001b[39m\u001b[38;5;28;01mimport\u001b[39;00m load_model\n\u001b[32m     13\u001b[39m weights_path = \u001b[33m\"\u001b[39m\u001b[33mmoondream2/model2.safetensors\u001b[39m\u001b[33m\"\u001b[39m  \u001b[38;5;66;03m# Path to your local weights file\u001b[39;00m\n\u001b[32m---> \u001b[39m\u001b[32m14\u001b[39m state_dict = \u001b[43mload_model\u001b[49m\u001b[43m(\u001b[49m\u001b[43mmodel\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43mweights_path\u001b[49m\u001b[43m)\u001b[49m\n\u001b[32m     15\u001b[39m new_state_dict = {}\n\u001b[32m     16\u001b[39m \u001b[38;5;28;01mfor\u001b[39;00m key, value \u001b[38;5;129;01min\u001b[39;00m state_dict.items():\n\u001b[32m     17\u001b[39m     \u001b[38;5;66;03m# Remove 'model.' prefix if it exists\u001b[39;00m\n",
+      "\u001b[36mFile \u001b[39m\u001b[32m~/Desktop/moondream/venv/lib/python3.13/site-packages/safetensors/torch.py:205\u001b[39m, in \u001b[36mload_model\u001b[39m\u001b[34m(model, filename, strict, device)\u001b[39m\n\u001b[32m    203\u001b[39m model_state_dict = model.state_dict()\n\u001b[32m    204\u001b[39m to_removes = _remove_duplicate_names(model_state_dict, preferred_names=state_dict.keys())\n\u001b[32m--> \u001b[39m\u001b[32m205\u001b[39m missing, unexpected = \u001b[43mmodel\u001b[49m\u001b[43m.\u001b[49m\u001b[43mload_state_dict\u001b[49m\u001b[43m(\u001b[49m\u001b[43mstate_dict\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43mstrict\u001b[49m\u001b[43m=\u001b[49m\u001b[38;5;28;43;01mFalse\u001b[39;49;00m\u001b[43m)\u001b[49m\n\u001b[32m    206\u001b[39m missing = \u001b[38;5;28mset\u001b[39m(missing)\n\u001b[32m    207\u001b[39m \u001b[38;5;28;01mfor\u001b[39;00m to_remove_group \u001b[38;5;129;01min\u001b[39;00m to_removes.values():\n",
+      "\u001b[36mFile \u001b[39m\u001b[32m~/Desktop/moondream/venv/lib/python3.13/site-packages/torch/nn/modules/module.py:2593\u001b[39m, in \u001b[36mModule.load_state_dict\u001b[39m\u001b[34m(self, state_dict, strict, assign)\u001b[39m\n\u001b[32m   2585\u001b[39m         error_msgs.insert(\n\u001b[32m   2586\u001b[39m             \u001b[32m0\u001b[39m,\n\u001b[32m   2587\u001b[39m             \u001b[33m\"\u001b[39m\u001b[33mMissing key(s) in state_dict: \u001b[39m\u001b[38;5;132;01m{}\u001b[39;00m\u001b[33m. \u001b[39m\u001b[33m\"\u001b[39m.format(\n\u001b[32m   2588\u001b[39m                 \u001b[33m\"\u001b[39m\u001b[33m, \u001b[39m\u001b[33m\"\u001b[39m.join(\u001b[33mf\u001b[39m\u001b[33m'\u001b[39m\u001b[33m\"\u001b[39m\u001b[38;5;132;01m{\u001b[39;00mk\u001b[38;5;132;01m}\u001b[39;00m\u001b[33m\"\u001b[39m\u001b[33m'\u001b[39m \u001b[38;5;28;01mfor\u001b[39;00m k \u001b[38;5;129;01min\u001b[39;00m missing_keys)\n\u001b[32m   2589\u001b[39m             ),\n\u001b[32m   2590\u001b[39m         )\n\u001b[32m   2592\u001b[39m \u001b[38;5;28;01mif\u001b[39;00m \u001b[38;5;28mlen\u001b[39m(error_msgs) > \u001b[32m0\u001b[39m:\n\u001b[32m-> \u001b[39m\u001b[32m2593\u001b[39m     \u001b[38;5;28;01mraise\u001b[39;00m \u001b[38;5;167;01mRuntimeError\u001b[39;00m(\n\u001b[32m   2594\u001b[39m         \u001b[33m\"\u001b[39m\u001b[33mError(s) in loading state_dict for \u001b[39m\u001b[38;5;132;01m{}\u001b[39;00m\u001b[33m:\u001b[39m\u001b[38;5;130;01m\\n\u001b[39;00m\u001b[38;5;130;01m\\t\u001b[39;00m\u001b[38;5;132;01m{}\u001b[39;00m\u001b[33m\"\u001b[39m.format(\n\u001b[32m   2595\u001b[39m             \u001b[38;5;28mself\u001b[39m.\u001b[34m__class__\u001b[39m.\u001b[34m__name__\u001b[39m, \u001b[33m\"\u001b[39m\u001b[38;5;130;01m\\n\u001b[39;00m\u001b[38;5;130;01m\\t\u001b[39;00m\u001b[33m\"\u001b[39m.join(error_msgs)\n\u001b[32m   2596\u001b[39m         )\n\u001b[32m   2597\u001b[39m     )\n\u001b[32m   2598\u001b[39m \u001b[38;5;28;01mreturn\u001b[39;00m _IncompatibleKeys(missing_keys, unexpected_keys)\n",
+      "\u001b[31mRuntimeError\u001b[39m: Error(s) in loading state_dict for MoondreamModel:\n\tWhile copying the parameter named \"region.size_features\", whose dimensions in the model are torch.Size([2, 256]) and whose dimensions in the checkpoint are torch.Size([2, 256]), an exception occurred : ('CUDA error: out of memory\\nCUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect.\\nFor debugging consider passing CUDA_LAUNCH_BLOCKING=1\\nCompile with `TORCH_USE_CUDA_DSA` to enable device-side assertions.\\n',)."
+     ]
+    }
+   ],
+   "source": [
+    "import os\n",
+    "os.environ[\"PYTORCH_CUDA_ALLOC_CONF\"] = \"expandable_segments:True\"\n",
+    "\n",
+    "import torch\n",
+    "from moondream2.config import MoondreamConfig\n",
+    "from moondream2.moondream import MoondreamModel\n",
+    "import torch.profiler\n",
+    "\n",
+    "config = MoondreamConfig()\n",
+    "device = \"cuda\" if torch.cuda.is_available() else \"cpu\"\n",
+    "model = MoondreamModel(config, setup_caches=False).to(device)\n",
+    "from safetensors.torch import load_model\n",
+    "weights_path = \"moondream2/model2.safetensors\"  # Path to your local weights file\n",
+    "state_dict = load_model(model, weights_path)\n",
+    "new_state_dict = {}\n",
+    "for key, value in state_dict.items():\n",
+    "    # Remove 'model.' prefix if it exists\n",
+    "    if key.startswith('model.'):\n",
+    "        new_key = key[6:]  # Skip the first 6 characters ('model.')\n",
+    "    else:\n",
+    "        new_key = key\n",
+    "    new_state_dict[new_key] = value\n",
+    "state_dict = new_state_dict\n",
+    "missing_keys, unexpected_keys = model.load_state_dict(state_dict, strict=True)\n",
+    "model._setup_caches()"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": []
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 8,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from PIL import Image\n",
+    "image = Image.open(\"example.png\")\n",
+    "query = \"home icon at the bottom of the screen is visible\"\n",
+    "points = model.point(image, query)[\"points\"]\n"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 2,
+   "metadata": {},
+   "outputs": [
+    {
+     "data": {
+      "text/plain": [
+       "[{'x': 0.0849609375, 'y': 0.9453125}]"
+      ]
+     },
+     "execution_count": 2,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "source": [
+    "\n",
+    "from PIL import Image\n",
+    "image = Image.open(\"example.png\")\n",
+    "query = \"home icon at the bottom\"\n",
+    "with torch.profiler.profile(\n",
+    "    activities=[\n",
+    "        torch.profiler.ProfilerActivity.CPU,\n",
+    "        torch.profiler.ProfilerActivity.CUDA],\n",
+    "    schedule=torch.profiler.schedule(\n",
+    "        wait=1, warmup=1, active=3),\n",
+    "    on_trace_ready=torch.profiler.tensorboard_trace_handler('log'),\n",
+    "    record_shapes=True,\n",
+    "    with_stack=True\n",
+    ") as prof:\n",
+    "    for i in range(3):\n",
+    "        points = model.point(image, query)[\"points\"]\n",
+    "        prof.step()\n",
+    "points"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 7,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "points = model.point(image, query)[\"points\"]\n"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": []
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 17,
+   "metadata": {},
+   "outputs": [
+    {
+     "ename": "IndexError",
+     "evalue": "list index out of range",
+     "output_type": "error",
+     "traceback": [
+      "\u001b[31m---------------------------------------------------------------------------\u001b[39m",
+      "\u001b[31mIndexError\u001b[39m                                Traceback (most recent call last)",
+      "\u001b[36mCell\u001b[39m\u001b[36m \u001b[39m\u001b[32mIn[17]\u001b[39m\u001b[32m, line 29\u001b[39m\n\u001b[32m     27\u001b[39m     plt.axis(\u001b[33m'\u001b[39m\u001b[33moff\u001b[39m\u001b[33m'\u001b[39m)\n\u001b[32m     28\u001b[39m     plt.show()\n\u001b[32m---> \u001b[39m\u001b[32m29\u001b[39m show_point_on_image(\u001b[43mpoints\u001b[49m\u001b[43m[\u001b[49m\u001b[32;43m0\u001b[39;49m\u001b[43m]\u001b[49m[\u001b[33m'\u001b[39m\u001b[33mx\u001b[39m\u001b[33m'\u001b[39m], points[\u001b[32m0\u001b[39m][\u001b[33m'\u001b[39m\u001b[33my\u001b[39m\u001b[33m'\u001b[39m])\n",
+      "\u001b[31mIndexError\u001b[39m: list index out of range"
+     ]
+    }
+   ],
+   "source": [
+    "from PIL import Image\n",
+    "import matplotlib.pyplot as plt\n",
+    "\n",
+    "def show_point_on_image(rel_x, rel_y, point_color='red', point_size=30):\n",
+    "    \"\"\"\n",
+    "    Display 'example.png' with a point marked at the given relative coordinates.\n",
+    "\n",
+    "    Parameters:\n",
+    "    - rel_x: float, relative x-coordinate (0 to 1).\n",
+    "    - rel_y: float, relative y-coordinate (0 to 1).\n",
+    "    - point_color: str, color of the point (default: 'red').\n",
+    "    - point_size: int, size of the point (default: 50).\n",
+    "    \"\"\"\n",
+    "    image_path = 'example.png'  # Hardcoded image path\n",
+    "\n",
+    "    # Load image\n",
+    "    img = Image.open(image_path)\n",
+    "    width, height = img.size\n",
+    "\n",
+    "    # Convert relative coordinates to absolute\n",
+    "    abs_x = rel_x * width\n",
+    "    abs_y = rel_y * height\n",
+    "\n",
+    "    # Plot\n",
+    "    plt.imshow(img)\n",
+    "    plt.scatter([abs_x], [abs_y], color=point_color, s=point_size, alpha=0.7)\n",
+    "    plt.axis('off')\n",
+    "    plt.show()\n",
+    "show_point_on_image(points[0]['x'], points[0]['y'])\n"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 14,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from PIL import Image\n",
+    "image = Image.open(\"example.png\")"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 15,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import time"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 16,
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "Time taken: 0.5298349857330322 seconds\n"
+     ]
+    }
+   ],
+   "source": [
+    "start_time = time.time()\n",
+    "\n",
+    "query = \"the login button\"\n",
+    "points = model.point(image, query)[\"points\"]\n",
+    "end_time = time.time()\n",
+    "print(f\"Time taken: {end_time - start_time} seconds\")\n"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 9,
+   "metadata": {},
+   "outputs": [],
+   "source": []
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 5,
+   "metadata": {},
+   "outputs": [
+    {
+     "data": {
+      "text/plain": [
+       "False"
+      ]
+     },
+     "execution_count": 5,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "source": [
+    "from transformers.utils import is_flash_attn_2_available\n",
+    "is_flash_attn_2_available()\n"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 1,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from transformers import AutoModelForCausalLM, AutoTokenizer\n",
+    "\n",
+    "model = AutoModelForCausalLM.from_pretrained(\n",
+    "    \"vikhyatk/moondream2\",\n",
+    "    revision=\"2025-04-14\",\n",
+    "    trust_remote_code=True,\n",
+    "    cache_dir=\"moondream\",\n",
+    "    # Uncomment to run on GPU.\n",
+    "    device_map={\"\": \"cuda\"}\n",
+    ")"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 7,
+   "metadata": {},
+   "outputs": [
+    {
+     "data": {
+      "text/plain": [
+       "HfMoondream(\n",
+       "  (model): MoondreamModel(\n",
+       "    (vision): ModuleDict(\n",
+       "      (patch_emb): Linear(in_features=588, out_features=1152, bias=True)\n",
+       "      (blocks): ModuleList(\n",
+       "        (0-26): 27 x ModuleDict(\n",
+       "          (ln1): LayerNorm((1152,), eps=1e-05, elementwise_affine=True)\n",
+       "          (attn): ModuleDict(\n",
+       "            (qkv): Linear(in_features=1152, out_features=3456, bias=True)\n",
+       "            (proj): Linear(in_features=1152, out_features=1152, bias=True)\n",
+       "          )\n",
+       "          (ln2): LayerNorm((1152,), eps=1e-05, elementwise_affine=True)\n",
+       "          (mlp): ModuleDict(\n",
+       "            (fc1): Linear(in_features=1152, out_features=4304, bias=True)\n",
+       "            (fc2): Linear(in_features=4304, out_features=1152, bias=True)\n",
+       "          )\n",
+       "        )\n",
+       "      )\n",
+       "      (post_ln): LayerNorm((1152,), eps=1e-05, elementwise_affine=True)\n",
+       "      (proj_mlp): ModuleDict(\n",
+       "        (fc1): Linear(in_features=2304, out_features=8192, bias=True)\n",
+       "        (fc2): Linear(in_features=8192, out_features=2048, bias=True)\n",
+       "      )\n",
+       "    )\n",
+       "    (text): ModuleDict(\n",
+       "      (blocks): ModuleList(\n",
+       "        (0-23): 24 x ModuleDict(\n",
+       "          (ln): LayerNorm((2048,), eps=1e-05, elementwise_affine=True)\n",
+       "          (attn): ModuleDict(\n",
+       "            (qkv): Linear(in_features=2048, out_features=6144, bias=True)\n",
+       "            (proj): Linear(in_features=2048, out_features=2048, bias=True)\n",
+       "          )\n",
+       "          (mlp): ModuleDict(\n",
+       "            (fc1): Linear(in_features=2048, out_features=8192, bias=True)\n",
+       "            (fc2): Linear(in_features=8192, out_features=2048, bias=True)\n",
+       "          )\n",
+       "        )\n",
+       "      )\n",
+       "      (post_ln): LayerNorm((2048,), eps=1e-05, elementwise_affine=True)\n",
+       "      (lm_head): Linear(in_features=2048, out_features=51200, bias=True)\n",
+       "    )\n",
+       "    (region): ModuleDict(\n",
+       "      (coord_encoder): Linear(in_features=256, out_features=2048, bias=True)\n",
+       "      (coord_decoder): ModuleDict(\n",
+       "        (fc1): Linear(in_features=2048, out_features=8192, bias=True)\n",
+       "        (fc2): Linear(in_features=8192, out_features=1024, bias=True)\n",
+       "      )\n",
+       "      (size_encoder): Linear(in_features=512, out_features=2048, bias=True)\n",
+       "      (size_decoder): ModuleDict(\n",
+       "        (fc1): Linear(in_features=2048, out_features=8192, bias=True)\n",
+       "        (fc2): Linear(in_features=8192, out_features=2048, bias=True)\n",
+       "      )\n",
+       "    )\n",
+       "  )\n",
+       ")"
+      ]
+     },
+     "execution_count": 7,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "source": [
+    "model.point()"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": []
+  }
+ ],
+ "metadata": {
+  "kernelspec": {
+   "display_name": "venv",
+   "language": "python",
+   "name": "python3"
+  },
+  "language_info": {
+   "codemirror_mode": {
+    "name": "ipython",
+    "version": 3
+   },
+   "file_extension": ".py",
+   "mimetype": "text/x-python",
+   "name": "python",
+   "nbconvert_exporter": "python",
+   "pygments_lexer": "ipython3",
+   "version": "3.13.3"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 2
+}

ollama.ipynb

@@ -0,0 +1,195 @@
+{
+ "cells": [
+  {
+   "cell_type": "code",
+   "execution_count": 1,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from transformers import AutoConfig\n",
+    "\n",
+    "config = AutoConfig.from_pretrained(\"vikhyatk/moondream2\", trust_remote_code=True)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 2,
+   "metadata": {},
+   "outputs": [
+    {
+     "data": {
+      "text/plain": [
+       "{'return_dict': True,\n",
+       " 'output_hidden_states': False,\n",
+       " 'output_attentions': False,\n",
+       " 'torchscript': False,\n",
+       " 'torch_dtype': 'float16',\n",
+       " 'use_bfloat16': False,\n",
+       " 'tf_legacy_loss': False,\n",
+       " 'pruned_heads': {},\n",
+       " 'tie_word_embeddings': True,\n",
+       " 'chunk_size_feed_forward': 0,\n",
+       " 'is_encoder_decoder': False,\n",
+       " 'is_decoder': False,\n",
+       " 'cross_attention_hidden_size': None,\n",
+       " 'add_cross_attention': False,\n",
+       " 'tie_encoder_decoder': False,\n",
+       " 'max_length': 20,\n",
+       " 'min_length': 0,\n",
+       " 'do_sample': False,\n",
+       " 'early_stopping': False,\n",
+       " 'num_beams': 1,\n",
+       " 'num_beam_groups': 1,\n",
+       " 'diversity_penalty': 0.0,\n",
+       " 'temperature': 1.0,\n",
+       " 'top_k': 50,\n",
+       " 'top_p': 1.0,\n",
+       " 'typical_p': 1.0,\n",
+       " 'repetition_penalty': 1.0,\n",
+       " 'length_penalty': 1.0,\n",
+       " 'no_repeat_ngram_size': 0,\n",
+       " 'encoder_no_repeat_ngram_size': 0,\n",
+       " 'bad_words_ids': None,\n",
+       " 'num_return_sequences': 1,\n",
+       " 'output_scores': False,\n",
+       " 'return_dict_in_generate': False,\n",
+       " 'forced_bos_token_id': None,\n",
+       " 'forced_eos_token_id': None,\n",
+       " 'remove_invalid_values': False,\n",
+       " 'exponential_decay_length_penalty': None,\n",
+       " 'suppress_tokens': None,\n",
+       " 'begin_suppress_tokens': None,\n",
+       " 'architectures': ['HfMoondream'],\n",
+       " 'finetuning_task': None,\n",
+       " 'id2label': {0: 'LABEL_0', 1: 'LABEL_1'},\n",
+       " 'label2id': {'LABEL_0': 0, 'LABEL_1': 1},\n",
+       " 'tokenizer_class': None,\n",
+       " 'prefix': None,\n",
+       " 'bos_token_id': None,\n",
+       " 'pad_token_id': None,\n",
+       " 'eos_token_id': None,\n",
+       " 'sep_token_id': None,\n",
+       " 'decoder_start_token_id': None,\n",
+       " 'task_specific_params': None,\n",
+       " 'problem_type': None,\n",
+       " '_name_or_path': 'vikhyatk/moondream2',\n",
+       " '_attn_implementation_autoset': False,\n",
+       " 'transformers_version': '4.49.0',\n",
+       " 'auto_map': {'AutoConfig': 'vikhyatk/moondream2--hf_moondream.HfConfig',\n",
+       "  'AutoModelForCausalLM': 'vikhyatk/moondream2--hf_moondream.HfMoondream'},\n",
+       " 'config': {},\n",
+       " 'model_type': 'moondream1'}"
+      ]
+     },
+     "execution_count": 2,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "source": [
+    "config.to_dict()\n"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 4,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from moondream2.hf_moondream import HfConfig\n",
+    "# serialize config.json to data variable\n",
+    "import json\n",
+    "with open(\"moondream2/config.json\", \"r\") as f:\n",
+    "    data = json.load(f)\n",
+    "configo = HfConfig(**data)\n"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 5,
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "_name_or_path : vikhyatk/moondream2 != \n",
+      "auto_map : {'AutoConfig': 'vikhyatk/moondream2--hf_moondream.HfConfig', 'AutoModelForCausalLM': 'vikhyatk/moondream2--hf_moondream.HfMoondream'} != {'AutoConfig': 'hf_moondream.HfConfig', 'AutoModelForCausalLM': 'hf_moondream.HfMoondream'}\n"
+     ]
+    }
+   ],
+   "source": [
+    "# compare dicts of config.to_dict() and configo.to_dict()    \n",
+    "\n",
+    "# make it check values of the dicts\n",
+    "\n",
+    "for key, value in config.to_dict().items():\n",
+    "    if key not in configo.to_dict():\n",
+    "        print(key + \" : \" + str(value) + \" not in hf_config\")\n",
+    "    elif value != configo.to_dict()[key]:\n",
+    "        print(key+ \" : \"+str(value)+\" != \"+str(configo.to_dict()[key]))\n",
+    "\n",
+    "for key, value in configo.to_dict().items():\n",
+    "    if key not in config.to_dict():\n",
+    "        print(key + \" : \" + str(value) + \" not in from_pretrained\")\n",
+    "\n",
+    "hparams = config.to_dict()\n"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 24,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "text_config = hparams.get(\"text_config\", {})"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 26,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "text_config.get(\"architectures\")\n"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 27,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "hparams.get(\"vision_config\")"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": []
+  }
+ ],
+ "metadata": {
+  "kernelspec": {
+   "display_name": "venv",
+   "language": "python",
+   "name": "python3"
+  },
+  "language_info": {
+   "codemirror_mode": {
+    "name": "ipython",
+    "version": 3
+   },
+   "file_extension": ".py",
+   "mimetype": "text/x-python",
+   "name": "python",
+   "nbconvert_exporter": "python",
+   "pygments_lexer": "ipython3",
+   "version": "3.12.4"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 2
+}

test.py

@@ -0,0 +1,41 @@
+print("app Started")
+import torch
+from moondream2.config import MoondreamConfig
+from moondream2.moondream import MoondreamModel
+import torch.profiler
+
+config = MoondreamConfig()
+device = "cuda"
+model = MoondreamModel(config, setup_caches=False).to(device)
+from safetensors.torch import load_file
+weights_path = "moondream2/model.safetensors"  # Path to your local weights file
+state_dict = load_file(weights_path, device=device)
+new_state_dict = {}
+for key, value in state_dict.items():
+    # Remove 'model.' prefix if it exists
+    if key.startswith('model.'):
+        new_key = key[6:]  # Skip the first 6 characters ('model.')
+    else:
+        new_key = key
+    new_state_dict[new_key] = value
+state_dict = new_state_dict
+missing_keys, unexpected_keys = model.load_state_dict(state_dict, strict=True)
+model._setup_caches()
+
+
+
+from PIL import Image
+image = Image.open("example.png")
+query = "home icon at the bottom"
+warmup_iters = 2
+
+
+for i in range(3):
+    if i == warmup_iters: torch.cuda.cudart().cudaProfilerStart()
+    if i >= warmup_iters: torch.cuda.nvtx.range_push("iteration{}".format(i))
+    if i >= warmup_iters: torch.cuda.nvtx.range_push("forward")
+    points = model.point(image, query)["points"]
+    if i >= warmup_iters: torch.cuda.nvtx.range_pop()
+    if i >= warmup_iters: torch.cuda.nvtx.range_pop()
+
+torch.cuda.cudart().cudaProfilerStop()