Duplicate from vikhyatk/moondream2

Co-authored-by: Vik Korrapati <[email protected]>

.gitattributes

@@ -0,0 +1,37 @@
+*.7z filter=lfs diff=lfs merge=lfs -text
+*.arrow filter=lfs diff=lfs merge=lfs -text
+*.bin filter=lfs diff=lfs merge=lfs -text
+*.bz2 filter=lfs diff=lfs merge=lfs -text
+*.ckpt filter=lfs diff=lfs merge=lfs -text
+*.ftz filter=lfs diff=lfs merge=lfs -text
+*.gz filter=lfs diff=lfs merge=lfs -text
+*.h5 filter=lfs diff=lfs merge=lfs -text
+*.joblib filter=lfs diff=lfs merge=lfs -text
+*.lfs.* filter=lfs diff=lfs merge=lfs -text
+*.mlmodel filter=lfs diff=lfs merge=lfs -text
+*.model filter=lfs diff=lfs merge=lfs -text
+*.msgpack filter=lfs diff=lfs merge=lfs -text
+*.npy filter=lfs diff=lfs merge=lfs -text
+*.npz filter=lfs diff=lfs merge=lfs -text
+*.onnx filter=lfs diff=lfs merge=lfs -text
+*.ot filter=lfs diff=lfs merge=lfs -text
+*.parquet filter=lfs diff=lfs merge=lfs -text
+*.pb filter=lfs diff=lfs merge=lfs -text
+*.pickle filter=lfs diff=lfs merge=lfs -text
+*.pkl filter=lfs diff=lfs merge=lfs -text
+*.pt filter=lfs diff=lfs merge=lfs -text
+*.pth filter=lfs diff=lfs merge=lfs -text
+*.rar filter=lfs diff=lfs merge=lfs -text
+*.safetensors filter=lfs diff=lfs merge=lfs -text
+saved_model/**/* filter=lfs diff=lfs merge=lfs -text
+*.tar.* filter=lfs diff=lfs merge=lfs -text
+*.tar filter=lfs diff=lfs merge=lfs -text
+*.tflite filter=lfs diff=lfs merge=lfs -text
+*.tgz filter=lfs diff=lfs merge=lfs -text
+*.wasm filter=lfs diff=lfs merge=lfs -text
+*.xz filter=lfs diff=lfs merge=lfs -text
+*.zip filter=lfs diff=lfs merge=lfs -text
+*.zst filter=lfs diff=lfs merge=lfs -text
+*tfevents* filter=lfs diff=lfs merge=lfs -text
+moondream2-mmproj-f16.gguf filter=lfs diff=lfs merge=lfs -text
+moondream2-text-model-f16.gguf filter=lfs diff=lfs merge=lfs -text

README.md

@@ -0,0 +1,88 @@
+---
+license: apache-2.0
+pipeline_tag: image-text-to-text
+new_version: moondream/moondream3-preview
+---
+
+⚠️ This repository contains the latest version of Moondream 2, our previous generation model. The latest version of Moondream is [Moondream 3 (Preview)](https://huggingface.co/moondream/moondream3-preview).
+
+---
+
+Moondream is a small vision language model designed to run efficiently everywhere. 
+
+[Website](https://moondream.ai/) / [Demo](https://moondream.ai/playground) / [GitHub](https://github.com/vikhyat/moondream)
+
+This repository contains the latest (**2025-06-21**) release of Moondream 2, as well as [historical releases](https://huggingface.co/vikhyatk/moondream2/blob/main/versions.txt). The model is updated frequently, so we recommend specifying a revision as shown below if you're using it in a production application.
+
+
+### Usage
+
+```python
+from transformers import AutoModelForCausalLM, AutoTokenizer
+from PIL import Image
+
+model = AutoModelForCausalLM.from_pretrained(
+    "vikhyatk/moondream2",
+    revision="2025-06-21",
+    trust_remote_code=True,
+    device_map={"": "cuda"}  # ...or 'mps', on Apple Silicon
+)
+
+# Captioning
+print("Short caption:")
+print(model.caption(image, length="short")["caption"])
+
+print("\nNormal caption:")
+for t in model.caption(image, length="normal", stream=True)["caption"]:
+    # Streaming generation example, supported for caption() and detect()
+    print(t, end="", flush=True)
+print(model.caption(image, length="normal"))
+
+# Visual Querying
+print("\nVisual query: 'How many people are in the image?'")
+print(model.query(image, "How many people are in the image?")["answer"])
+
+# Object Detection
+print("\nObject detection: 'face'")
+objects = model.detect(image, "face")["objects"]
+print(f"Found {len(objects)} face(s)")
+
+# Pointing
+print("\nPointing: 'person'")
+points = model.point(image, "person")["points"]
+print(f"Found {len(points)} person(s)")
+```
+
+### Changelog
+
+**2025-06-21** ([full release notes](https://moondream.ai/blog/moondream-2025-06-21-release))
+
+* **Grounded Reasoning**
+  Introduces a new step-by-step reasoning mode that explicitly grounds reasoning in spatial positions within the image before answering, leading to more precise visual interpretation (e.g., chart median calculations, accurate counting). Enable with `reasoning=True` in the `query` skill to trade off speed vs. accuracy.
+* **Sharper Object Detection**
+  Uses reinforcement learning on higher-quality bounding-box annotations to reduce object clumping and improve fine-grained detections (e.g., distinguishing “blue bottle” vs. “bottle”).
+* **Faster Text Generation**
+  Yields 20–40 % faster response generation via a new “superword” tokenizer and lightweight tokenizer transfer hypernetwork, which reduces the number of tokens emitted without loss in accuracy and eases future multilingual extensions.
+* **Improved UI Understanding**
+  Boosts ScreenSpot (UI element localization) performance from an F1\@0.5 of 60.3 to 80.4, making Moondream more effective for UI-focused applications.
+* **Reinforcement Learning Enhancements**
+  RL fine-tuning applied across 55 vision-language tasks to reinforce grounded reasoning and detection capabilities, with a roadmap to expand to \~120 tasks in the next update.
+
+**2025-04-15** ([full release notes](https://moondream.ai/blog/moondream-2025-04-14-release))
+
+1. Improved chart understanding (ChartQA up from 74.8 to 77.5, 82.2 with PoT)
+2. Added temperature and nucleus sampling to reduce repetitive outputs
+3. Better OCR for documents and tables (prompt with “Transcribe the text” or “Transcribe the text in natural reading order”)
+4. Object detection supports document layout detection (figure, formula, text, etc)
+5. UI understanding (ScreenSpot F1\@0.5 up from 53.3 to 60.3)
+6. Improved text understanding (DocVQA up from 76.5 to 79.3, TextVQA up from 74.6 to 76.3)
+
+**2025-03-27** ([full release notes](https://moondream.ai/blog/moondream-2025-03-27-release))
+
+1. Added support for long-form captioning
+2. Open vocabulary image tagging
+3. Improved counting accuracy (e.g. CountBenchQA increased from 80 to 86.4)
+4. Improved text understanding (e.g. OCRBench increased from 58.3 to 61.2)
+5. Improved object detection, especially for small objects (e.g. COCO up from 30.5 to 51.2)
+6. Fixed token streaming bug affecting multi-byte unicode characters
+7. gpt-fast style `compile()` now supported in HF Transformers implementation

added_tokens.json

@@ -0,0 +1,40 @@
+{
+  "\t\t": 50294,
+  "\t\t\t": 50293,
+  "\t\t\t\t": 50292,
+  "\t\t\t\t\t": 50291,
+  "\t\t\t\t\t\t": 50290,
+  "\t\t\t\t\t\t\t": 50289,
+  "\t\t\t\t\t\t\t\t": 50288,
+  "\t\t\t\t\t\t\t\t\t": 50287,
+  "  ": 50286,
+  "   ": 50285,
+  "    ": 50284,
+  "     ": 50283,
+  "      ": 50282,
+  "       ": 50281,
+  "        ": 50280,
+  "         ": 50279,
+  "          ": 50278,
+  "           ": 50277,
+  "            ": 50276,
+  "             ": 50275,
+  "              ": 50274,
+  "               ": 50273,
+  "                ": 50272,
+  "                 ": 50271,
+  "                  ": 50270,
+  "                   ": 50269,
+  "                    ": 50268,
+  "                     ": 50267,
+  "                      ": 50266,
+  "                       ": 50265,
+  "                        ": 50264,
+  "                         ": 50263,
+  "                          ": 50262,
+  "                           ": 50261,
+  "                            ": 50260,
+  "                             ": 50259,
+  "                              ": 50258,
+  "                               ": 50257
+}

config.json

@@ -0,0 +1,13 @@
+{
+  "architectures": [
+    "HfMoondream"
+  ],
+  "auto_map": {
+    "AutoConfig": "hf_moondream.HfConfig",
+    "AutoModelForCausalLM": "hf_moondream.HfMoondream"
+  },
+  "config": {},
+  "model_type": "moondream1",
+  "torch_dtype": "bfloat16",
+  "transformers_version": "4.52.4"
+}

config.py

@@ -0,0 +1,94 @@
+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
+    group_size: Optional[int] = None
+
+
+@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
+    group_size: Optional[int] = None
+
+
+@dataclass(frozen=True)
+class TokenizerConfig:
+    bos_id: int = 0
+    eos_id: int = 0
+    answer_id: int = 3
+    thinking_id: int = 4
+    coord_id: int = 5
+    size_id: int = 6
+    start_ground_points_id: int = 7
+    end_ground_id: int = 9
+    templates: Dict[str, Optional[Dict[str, List[int]]]] = field(
+        default_factory=lambda: {
+            "caption": {
+                "short": [1, 32708, 2, 12492, 3],
+                "normal": [1, 32708, 2, 6382, 3],
+                "long": [1, 32708, 2, 4059, 3],
+            },
+            "query": {"prefix": [1, 15381, 2], "suffix": [3]},
+            "detect": {"prefix": [1, 7235, 476, 2], "suffix": [3]},
+            "point": {"prefix": [1, 2581, 2], "suffix": [3]},
+        }
+    )
+
+
+@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__,
+        }

configuration_moondream.py

@@ -0,0 +1,96 @@
+from transformers import PretrainedConfig
+
+
+class PhiConfig(PretrainedConfig):
+    model_type = "phi"
+    keys_to_ignore_at_inference = ["past_key_values"]
+
+    def __init__(
+        self,
+        vocab_size=51200,
+        hidden_size=2048,
+        intermediate_size=8192,
+        num_hidden_layers=24,
+        num_attention_heads=32,
+        num_key_value_heads=None,
+        resid_pdrop=0.0,
+        embd_pdrop=0.0,
+        attention_dropout=0.0,
+        hidden_act="gelu_new",
+        max_position_embeddings=2048,
+        initializer_range=0.02,
+        layer_norm_eps=1e-5,
+        use_cache=True,
+        tie_word_embeddings=False,
+        rope_theta=10000.0,
+        rope_scaling=None,
+        partial_rotary_factor=0.5,
+        bos_token_id=1,
+        eos_token_id=2,
+        **kwargs,
+    ):
+        self.vocab_size = vocab_size
+        self.hidden_size = hidden_size
+        self.intermediate_size = intermediate_size
+        self.num_hidden_layers = num_hidden_layers
+        self.num_attention_heads = num_attention_heads
+
+        if num_key_value_heads is None:
+            num_key_value_heads = num_attention_heads
+
+        self.num_key_value_heads = num_key_value_heads
+        self.resid_pdrop = resid_pdrop
+        self.embd_pdrop = embd_pdrop
+        self.attention_dropout = attention_dropout
+        self.hidden_act = hidden_act
+        self.max_position_embeddings = max_position_embeddings
+        self.initializer_range = initializer_range
+        self.layer_norm_eps = layer_norm_eps
+        self.use_cache = use_cache
+        self.rope_theta = rope_theta
+        self.rope_scaling = rope_scaling
+        self.partial_rotary_factor = partial_rotary_factor
+        self._rope_scaling_validation()
+
+        super().__init__(
+            bos_token_id=bos_token_id,
+            eos_token_id=eos_token_id,
+            tie_word_embeddings=tie_word_embeddings,
+            **kwargs,
+        )
+
+    # Copied from transformers.models.llama.configuration_llama.LlamaConfig._rope_scaling_validation
+    def _rope_scaling_validation(self):
+        """
+        Validate the `rope_scaling` configuration.
+        """
+        if self.rope_scaling is None:
+            return
+
+        if not isinstance(self.rope_scaling, dict) or len(self.rope_scaling) != 2:
+            raise ValueError(
+                "`rope_scaling` must be a dictionary with with two fields, `type` and `factor`, "
+                f"got {self.rope_scaling}"
+            )
+        rope_scaling_type = self.rope_scaling.get("type", None)
+        rope_scaling_factor = self.rope_scaling.get("factor", None)
+        if rope_scaling_type is None or rope_scaling_type not in ["linear", "dynamic"]:
+            raise ValueError(
+                f"`rope_scaling`'s type field must be one of ['linear', 'dynamic'], got {rope_scaling_type}"
+            )
+        if (
+            rope_scaling_factor is None
+            or not isinstance(rope_scaling_factor, float)
+            or rope_scaling_factor <= 1.0
+        ):
+            raise ValueError(
+                f"`rope_scaling`'s factor field must be a float > 1, got {rope_scaling_factor}"
+            )
+
+
+class MoondreamConfig(PretrainedConfig):
+    model_type = "moondream1"
+
+    def __init__(self, **kwargs):
+        self.text_config = PhiConfig(**kwargs.pop("text_config", {}))
+        super().__init__(**kwargs)

fourier_features.py

@@ -0,0 +1,18 @@
+# Adopted from https://github.com/crowsonkb/k-diffusion/blob/transformer-model-v2/k_diffusion/layers.py
+
+import torch
+import torch.nn as nn
+import math
+
+
+class FourierFeatures(nn.Module):
+    def __init__(self, in_features, out_features, std=1.0):
+        super().__init__()
+        assert out_features % 2 == 0
+        self.register_buffer(
+            "weight", torch.randn([out_features // 2, in_features]) * std
+        )
+
+    def forward(self, input):
+        f = 2 * math.pi * input @ self.weight.T
+        return torch.cat([f.cos(), f.sin()], dim=-1)

generation_config.json

@@ -0,0 +1,4 @@
+{
+  "_from_model_config": true,
+  "transformers_version": "4.44.0"
+}

handler.py

@@ -0,0 +1,58 @@
+from transformers import AutoModelForCausalLM, AutoTokenizer
+from PIL import Image
+import torch
+from io import BytesIO
+import base64
+
+class EndpointHandler:
+    def __init__(self, model_dir):
+        self.model_id = "vikhyatk/moondream2"
+        self.model = AutoModelForCausalLM.from_pretrained(self.model_id, trust_remote_code=True)
+        self.tokenizer = AutoTokenizer.from_pretrained("vikhyatk/moondream2", trust_remote_code=True)
+
+        # Check if CUDA (GPU support) is available and then set the device to GPU or CPU
+        self.device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+        self.model.to(self.device)
+
+    def preprocess_image(self, encoded_image):
+        """Decode and preprocess the input image."""
+        decoded_image = base64.b64decode(encoded_image)
+        img = Image.open(BytesIO(decoded_image)).convert("RGB")
+        return img
+
+    def __call__(self, data):
+        """Handle the incoming request."""
+        try:
+            # Extract the inputs from the data
+            inputs = data.pop("inputs", data)
+            input_image = inputs['image']
+            question = inputs.get('question', "move to the red ball")
+
+            # Preprocess the image
+            img = self.preprocess_image(input_image)
+
+            # Perform inference
+            enc_image = self.model.encode_image(img).to(self.device)
+            answer = self.model.answer_question(enc_image, question, self.tokenizer)
+
+            # If the output is a tensor, move it back to CPU and convert to list
+            if isinstance(answer, torch.Tensor):
+                answer = answer.cpu().numpy().tolist()
+
+            # Create the response
+            response = {
+                "statusCode": 200,
+                "body": {
+                    "answer": answer
+                }
+            }
+            return response
+        except Exception as e:
+            # Handle any errors
+            response = {
+                "statusCode": 500,
+                "body": {
+                    "error": str(e)
+                }
+            }
+            return response

hf_moondream.py

@@ -0,0 +1,183 @@
+import torch
+import torch.nn as nn
+
+from transformers import PreTrainedModel, PretrainedConfig
+from typing import Union
+
+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):
+    _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._is_kv_cache_setup = False
+
+    def _setup_caches(self):
+        if not self._is_kv_cache_setup:
+            self.model._setup_caches()
+            self._is_kv_cache_setup = True
+
+    @property
+    def encode_image(self):
+        self._setup_caches()
+        return self.model.encode_image
+
+    @property
+    def query(self):
+        self._setup_caches()
+        return self.model.query
+
+    @property
+    def caption(self):
+        self._setup_caches()
+        return self.model.caption
+
+    @property
+    def detect(self):
+        self._setup_caches()
+        return self.model.detect
+
+    @property
+    def point(self):
+        self._setup_caches()
+        return self.model.point
+
+    @property
+    def detect_gaze(self):
+        self._setup_caches()
+        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_answer(
+                    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) -> nn.Embedding:
+        """
+        Lazily wrap the raw parameter `self.model.text.wte` in a real
+        `nn.Embedding` layer so that HF mix-ins recognise it.  The wrapper
+        **shares** the weight tensor—no copy is made.
+        """
+        if not hasattr(self, "_input_embeddings"):
+            self._input_embeddings = nn.Embedding.from_pretrained(
+                self.model.text.wte,  # tensor created in text.py
+                freeze=True,  # set to False if you need it trainable
+            )
+        return self._input_embeddings
+
+    def set_input_embeddings(self, value: Union[nn.Embedding, nn.Module]) -> None:
+        """
+        Lets HF functions (e.g. `resize_token_embeddings`) replace or resize the
+        embeddings and keeps everything tied to `self.model.text.wte`.
+        """
+        # 1. point the low-level parameter to the new weight matrix
+        self.model.text.wte = value.weight
+        # 2. keep a reference for get_input_embeddings()
+        self._input_embeddings = value
+
+    def input_embeds(
+        self,
+        input_ids: Union[torch.LongTensor, list, tuple],
+        *,
+        device: torch.device | None = None
+    ) -> torch.FloatTensor:
+        """
+        Back-compat wrapper that turns token IDs into embeddings.
+
+        Example:
+            ids = torch.tensor([[1, 2, 3]])
+            embeds = model.input_embeds(ids)      # (1, 3, hidden_dim)
+        """
+        if not torch.is_tensor(input_ids):
+            input_ids = torch.as_tensor(input_ids)
+        if device is not None:
+            input_ids = input_ids.to(device)
+
+        return self.get_input_embeddings()(input_ids)

image_crops.py

@@ -0,0 +1,231 @@
+import math
+import numpy as np
+import torch
+
+from typing import TypedDict
+
+try:
+    import pyvips
+
+    HAS_VIPS = True
+except:
+    from PIL import Image
+
+    HAS_VIPS = False
+
+
+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,
+    max_crops: int,
+    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 = patch_size * overlap_margin
+    total_margin_pixels = margin_pixels * 2  # Both sides
+
+    # Calculate crop parameters
+    crop_patches = base_size[0] // patch_size  # patches per crop dimension
+    crop_window_patches = crop_patches - (2 * overlap_margin)  # usable patches
+    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,
+    )
+
+    if HAS_VIPS:
+        # 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()
+    else:
+        # Fallback to PIL
+        pil_img = Image.fromarray(image)
+        resized = pil_img.resize(
+            (int(target_size[1]), int(target_size[0])),
+            resample=Image.Resampling.LANCZOS,
+        )
+        image = np.asarray(resized)
+
+        # Create global crop
+        global_pil = pil_img.resize(
+            (int(base_size[1]), int(base_size[0])), resample=Image.Resampling.LANCZOS
+        )
+        crops[0] = np.asarray(global_pil)
+
+    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

layers.py

@@ -0,0 +1,166 @@
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+
+from dataclasses import dataclass
+from typing import Literal, Optional
+
+try:
+    from torchao import quantize_
+    from torchao.quantization import int4_weight_only
+except ImportError:
+
+    def quantize_(model, quant_mode):
+        raise ImportError(
+            "torchao is not installed. Please install it with `pip install torchao`."
+        )
+
+    def int4_weight_only(group_size):
+        raise ImportError(
+            "torchao is not installed. Please install it with `pip install torchao`."
+        )
+
+
+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)
+
+
+def dequantize_tensor(W_q, scale, zero, orig_shape, dtype=torch.bfloat16):
+    _step = W_q.shape[0]
+    W_r = torch.empty([2 * _step, W_q.shape[1]], dtype=dtype, device=W_q.device)
+    W_r[:_step] = (W_q & 0b11110000) >> 4
+    W_r[_step:] = W_q & 0b00001111
+    W_r.sub_(zero).mul_(scale)
+    return W_r.reshape(orig_shape)
+
+
+class QuantizedLinear(nn.Module):
+    def __init__(
+        self,
+        in_features: int,
+        out_features: int,
+        dtype: torch.dtype,
+    ):
+        # TODO: Take group_size as an input instead of hardcoding it here.
+        super().__init__()
+        self.in_features = in_features
+        self.out_features = out_features
+        self.weight = nn.ParameterDict(
+            {
+                "packed": nn.Parameter(
+                    torch.empty(
+                        out_features * in_features // (128 * 2), 128, dtype=torch.uint8
+                    ),
+                    requires_grad=False,
+                ),
+                "scale": nn.Parameter(
+                    torch.empty(out_features * in_features // 128, 1),
+                    requires_grad=False,
+                ),
+                "zero_point": nn.Parameter(
+                    torch.empty(out_features * in_features // 128, 1),
+                    requires_grad=False,
+                ),
+            }
+        )
+        self.bias = nn.Parameter(torch.empty(out_features), requires_grad=False)
+        self.unpacked = False
+
+    def unpack(self):
+        if self.unpacked:
+            return
+
+        self.weight = nn.Parameter(
+            dequantize_tensor(
+                self.weight["packed"],
+                self.weight["scale"],
+                self.weight["zero_point"],
+                (self.out_features, self.in_features),
+                torch.bfloat16,
+            )
+        )
+        with torch.device("meta"):
+            self.linear = nn.Linear(
+                self.in_features, self.out_features, dtype=torch.bfloat16
+            )
+        self.linear.weight = self.weight
+        self.linear.bias = nn.Parameter(
+            self.bias.to(torch.bfloat16), requires_grad=False
+        )
+
+        del self.weight, self.bias
+        quantize_(self, int4_weight_only(group_size=128))
+        self.unpacked = True
+        torch.cuda.empty_cache()
+
+    def forward(self, x: torch.Tensor) -> torch.Tensor:
+        if not self.unpacked:
+            self.unpack()
+        return self.linear(x)
+
+
+@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, lora: Optional[dict] = None) -> torch.Tensor:
+    x0 = w.fc1(x)
+    if lora is not None:
+        x1 = F.linear(F.linear(x, lora["fc1"]["A"]), lora["fc1"]["B"])
+        x = x0 + x1
+    else:
+        x = x0
+
+    x = gelu_approx(x)
+
+    x0 = w.fc2(x)
+    if lora is not None:
+        x1 = F.linear(F.linear(x, lora["fc2"]["A"]), lora["fc2"]["B"])
+        x = x0 + x1
+    else:
+        x = x0
+
+    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

lora.py

@@ -0,0 +1,82 @@
+import functools
+import os
+import shutil
+import torch
+
+from pathlib import Path
+from urllib.request import Request, urlopen
+from typing import Optional
+
+
+def variant_cache_dir():
+    hf_hub_cache = os.environ.get("HF_HUB_CACHE")
+    if hf_hub_cache is not None:
+        return Path(hf_hub_cache) / "md_variants"
+
+    hf_home = os.environ.get("HF_HOME")
+    if hf_home is not None:
+        return Path(hf_home) / "hub" / "md_variants"
+
+    return Path("~/.cache/huggingface/hub").expanduser() / "md_variants"
+
+
+def cached_variant_path(variant_id: str):
+    variant, *rest = variant_id.split("/", 1)
+    step = rest[0] if rest else "final"
+
+    cache_dir = variant_cache_dir() / variant
+    os.makedirs(cache_dir, exist_ok=True)
+    dest = cache_dir / f"{step}.pt"
+    if dest.exists():
+        return dest
+
+    md_endpoint = os.getenv("MOONDREAM_ENDPOINT", "https://api.moondream.ai")
+
+    headers = {"User-Agent": "moondream-torch"}
+    api_key = os.getenv("MOONDREAM_API_KEY")
+    if api_key is not None:
+        headers["X-Moondream-Auth"] = api_key
+
+    req = Request(f"{md_endpoint}/v1/variants/{variant_id}/download", headers=headers)
+    with urlopen(req) as r, open(dest, "wb") as f:
+        shutil.copyfileobj(r, f)
+    return dest
+
+
+def nest(flat):
+    tree = {}
+    for k, v in flat.items():
+        parts = k.split(".")
+        d = tree
+        for p in parts[:-1]:
+            d = d.setdefault(p, {})
+        d[parts[-1]] = v
+    return tree
+
+
+@functools.lru_cache(maxsize=5)
+def variant_state_dict(variant_id: Optional[str] = None, device: str = "cpu"):
+    if variant_id is None:
+        return None
+
+    state_dict = torch.load(
+        cached_variant_path(variant_id), map_location=device, weights_only=True
+    )
+
+    # TODO: Move these into the training code that saves checkpoints...
+    rename_rules = [
+        ("text_model.transformer.h", "text.blocks"),
+        (".mixer", ".attn"),
+        (".out_proj", ".proj"),
+        (".Wqkv", ".qkv"),
+        (".parametrizations.weight.0", ""),
+    ]
+    new_state_dict = {}
+    for key, tensor in state_dict.items():
+        new_key = key
+        for old, new in rename_rules:
+            if old in new_key:
+                new_key = new_key.replace(old, new)
+        new_state_dict[new_key] = tensor
+
+    return nest(new_state_dict)

model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:70a7d94c0c8349eb58ed2d9e636ef2d0916960f321ecabeac6354b8ba3d7403f
+size 3854538968

moondream.py

@@ -0,0 +1,986 @@
+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,
+    encode_spatial_refs,
+    SpatialRefs,
+)
+from .layers import QuantizedLinear
+from .lora import variant_state_dict
+from .utils import remove_outlier_points
+
+ImageEncodingSettings = TypedDict(
+    "ImageEncodingSettings",
+    {"variant": str},
+    total=False,
+)
+
+TextSamplingSettings = TypedDict(
+    "TextSamplingSettings",
+    {
+        "max_tokens": int,
+        "temperature": float,
+        "top_p": float,
+        "variant": str,
+    },
+    total=False,
+)
+
+ObjectSamplingSettings = TypedDict(
+    "ObjectSamplingSettings",
+    {"max_objects": int, "variant": str},
+    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.bfloat16, setup_caches=True
+    ):
+        super().__init__()
+        self.config = config
+
+        self.tokenizer = Tokenizer.from_pretrained("moondream/starmie-v1")
+        self.vision = build_vision_model(config.vision, dtype)
+        self.text = build_text_model(config.text, dtype)
+
+        # Region Model
+        linear_cls = (
+            QuantizedLinear if config.region.group_size is not None else nn.Linear
+        )
+        self.region = nn.ModuleDict(
+            {
+                "coord_encoder": linear_cls(
+                    config.region.coord_feat_dim, config.region.dim, dtype=dtype
+                ),
+                "coord_decoder": nn.ModuleDict(
+                    {
+                        "fc1": linear_cls(
+                            config.region.dim, config.region.inner_dim, dtype=dtype
+                        ),
+                        "fc2": linear_cls(
+                            config.region.inner_dim,
+                            config.region.coord_out_dim,
+                            dtype=dtype,
+                        ),
+                    }
+                ),
+                "size_encoder": linear_cls(
+                    config.region.size_feat_dim, config.region.dim, dtype=dtype
+                ),
+                "size_decoder": nn.ModuleDict(
+                    {
+                        "fc1": linear_cls(
+                            config.region.dim, config.region.inner_dim, dtype=dtype
+                        ),
+                        "fc2": linear_cls(
+                            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,
+        lora: Optional[torch.Tensor],
+    ):
+        return text_decoder(x, self.text, attn_mask, pos_ids, self.config.text, lora)
+
+    def _decode_one_tok(
+        self,
+        x: torch.Tensor,
+        attn_mask: torch.Tensor,
+        pos_ids: torch.Tensor,
+        lora: Optional[torch.Tensor],
+    ):
+        hidden = text_decoder(x, self.text, attn_mask, pos_ids, self.config.text, lora)
+        logits = lm_head(hidden, self.text)
+        return logits, hidden
+
+    def compile(self):
+        for module in self.modules():
+            if isinstance(module, QuantizedLinear):
+                module.unpack()
+
+        # 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],
+        settings: Optional[ImageEncodingSettings] = None,
+    ) -> EncodedImage:
+        if isinstance(image, EncodedImage):
+            return image
+        elif not isinstance(image, Image.Image):
+            raise ValueError("image must be a PIL Image or EncodedImage")
+
+        lora = (
+            variant_state_dict(settings["variant"], device=self.device)
+            if settings is not None and "variant" in settings
+            else None
+        )
+
+        # 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():
+            img_emb = self._run_vision_encoder(image)
+            bos_emb = text_encoder(
+                torch.tensor([[self.config.tokenizer.bos_id]], device=self.device),
+                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)
+            self._prefill(inputs_embeds, mask, pos_ids, lora)
+
+        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,
+        spatial_refs: Optional[SpatialRefs] = None,
+        attn_mask: Optional[torch.Tensor] = None,
+        lora: Optional[dict] = None,
+    ):
+        with torch.inference_mode():
+            prompt_emb = text_encoder(prompt_tokens, self.text)
+
+            if spatial_refs:
+                encoded_refs = encode_spatial_refs(spatial_refs, self.region)
+                prompt_emb[prompt_tokens == self.config.tokenizer.coord_id] = (
+                    encoded_refs["coords"]
+                )
+                if encoded_refs["sizes"] is not None:
+                    prompt_emb[prompt_tokens == self.config.tokenizer.size_id] = (
+                        encoded_refs["sizes"]
+                    )
+
+            torch._dynamo.mark_dynamic(prompt_emb, 1)
+
+            if attn_mask is None:
+                attn_mask = self.attn_mask
+
+            mask = attn_mask[:, :, pos : pos + prompt_emb.size(1), :]
+            pos_ids = torch.arange(pos, pos + prompt_emb.size(1), dtype=torch.long)
+            hidden_BC = self._prefill(prompt_emb, mask, pos_ids, lora)
+            logits_BV = lm_head(hidden_BC, self.text)
+
+            if temperature == 0:
+                next_token = torch.argmax(logits_BV, dim=-1).unsqueeze(1)
+            else:
+                probs = torch.softmax(logits_BV / 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_BV, hidden_BC, next_token, pos
+
+    def _generate_reasoning(
+        self,
+        prompt_tokens,
+        pos,
+        settings: Optional[TextSamplingSettings] = None,
+        spatial_refs: Optional[SpatialRefs] = None,
+        attn_mask: Optional[torch.Tensor] = None,
+    ) -> Tuple[int, str, List[dict]]:
+        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
+        )
+        lora = (
+            variant_state_dict(settings["variant"], device=self.device)
+            if settings is not None and "variant" in settings
+            else None
+        )
+
+        top_p = settings.get("top_p", DEFAULT_TOP_P) if settings else DEFAULT_TOP_P
+        eos_id = self.config.tokenizer.answer_id
+
+        _, last_hidden_BC, next_token, pos = self._prefill_prompt(
+            prompt_tokens,
+            pos,
+            temperature,
+            top_p,
+            spatial_refs,
+            attn_mask=attn_mask,
+            lora=lora,
+        )
+
+        text_token_chunks = [[]]
+        grounding_chunks = [[]]
+
+        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
+
+        while (
+            next_token_id := next_token.item()
+        ) != eos_id and generated_tokens < max_tokens:
+            if (
+                next_token_id == self.config.tokenizer.start_ground_points_id
+                or next_token_id == self.config.tokenizer.end_ground_id
+            ):
+                text_token_chunks.append([])
+                grounding_chunks.append([])
+
+            text_token_chunks[-1].append(next_token_id)
+
+            with torch.inference_mode():
+                if next_token_id == self.config.tokenizer.coord_id:
+                    coord_logits = decode_coordinate(last_hidden_BC, self.region)
+                    coord = torch.argmax(coord_logits, dim=-1) / coord_logits.size(-1)
+                    grounding_chunks[-1].append(coord.item())
+
+                    next_emb = encode_coordinate(
+                        coord.to(dtype=coord_logits.dtype), self.region
+                    ).unsqueeze(0)
+                else:
+                    next_emb = text_encoder(next_token, self.text)
+
+                mask[:, :, pos], pos_ids[0] = 1, pos
+
+                logits_BV, last_hidden_BC = self._decode_one_tok(
+                    next_emb, mask, pos_ids, lora
+                )
+                logits_BV[:, self.config.tokenizer.eos_id] = float("-inf")
+                logits_BV[:, self.config.tokenizer.size_id] = float("-inf")
+
+                pos += 1
+
+                if temperature == 0:
+                    next_token = torch.argmax(logits_BV, dim=-1).unsqueeze(1)  # (1, 1)
+                else:
+                    probs = torch.softmax(logits_BV / 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
+
+        text_chunks = [
+            self.tokenizer.decode(chunk_tokens) for chunk_tokens in text_token_chunks
+        ]
+        text = "".join(text_chunks)
+
+        start_idx = 0
+        grounding = []
+        for text_chunk, grounding_chunk in zip(text_chunks, grounding_chunks):
+            if len(grounding_chunk) > 1:
+                points = []
+                for i in range(0, len(grounding_chunk) - (len(grounding_chunk) % 2), 2):
+                    points.append((grounding_chunk[i], grounding_chunk[i + 1]))
+                grounding.append(
+                    {
+                        "start_idx": start_idx,
+                        "end_idx": start_idx + len(text_chunk),
+                        "points": points,
+                    }
+                )
+            start_idx += len(text_chunk)
+
+        return pos, text, grounding
+
+    def _generate_answer(
+        self,
+        prompt_tokens: torch.Tensor,
+        pos: int,
+        settings: Optional[TextSamplingSettings] = None,
+        spatial_refs: Optional[SpatialRefs] = None,
+        eos_id: Optional[int] = None,
+        attn_mask: Optional[torch.Tensor] = 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
+        eos_id = eos_id if eos_id is not None else self.config.tokenizer.eos_id
+        lora = (
+            variant_state_dict(settings["variant"], device=self.device)
+            if settings is not None and "variant" in settings
+            else None
+        )
+
+        _, _, next_token, pos = self._prefill_prompt(
+            prompt_tokens,
+            pos,
+            temperature,
+            top_p,
+            spatial_refs,
+            attn_mask=attn_mask,
+            lora=lora,
+        )
+
+        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()
+            ) != 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 yield 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_BV, _ = self._decode_one_tok(next_emb, mask, pos_ids, lora)
+                    logits_BV[:, self.config.tokenizer.answer_id] = float("-inf")
+
+                    pos += 1
+
+                    if temperature == 0:
+                        next_token = torch.argmax(logits_BV, dim=-1).unsqueeze(
+                            1
+                        )  # (1, 1)
+                    else:
+                        probs = torch.softmax(logits_BV / 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 query(
+        self,
+        image: Optional[Union[Image.Image, EncodedImage]] = None,
+        question: str = None,
+        reasoning: bool = False,
+        spatial_refs: Optional[SpatialRefs] = None,
+        stream: bool = False,
+        settings: Optional[TextSamplingSettings] = None,
+    ):
+        if self.config.tokenizer.templates["query"] is None:
+            raise NotImplementedError("Model does not support querying.")
+
+        if question is None:
+            raise ValueError("question must be provided.")
+
+        if spatial_refs and image is None:
+            raise ValueError("spatial_refs can only be used with an image.")
+
+        attn_mask = self.attn_mask
+        if image is not None:
+            image = self.encode_image(image, settings)
+            self.load_encoded_image(image)
+            pos = image.pos
+            prompt_toks = self.config.tokenizer.templates["query"]["prefix"]
+        else:
+            self._setup_caches()
+            pos = 0
+            prompt_toks = [
+                self.config.tokenizer.bos_id
+            ] + self.config.tokenizer.templates["query"]["prefix"]
+            max_context = self.config.text.max_context
+            attn_mask = torch.tril(
+                torch.ones(1, 1, max_context, max_context, dtype=torch.bool)
+            ).to(self.device)
+
+        spatial_toks = []
+        if spatial_refs:
+            for ref in spatial_refs:
+                coord_id = self.config.tokenizer.coord_id
+                size_id = self.config.tokenizer.size_id
+                if len(ref) == 2:
+                    spatial_toks.extend([coord_id, coord_id])
+                else:
+                    spatial_toks.extend([coord_id, coord_id, size_id])
+
+        prompt_tokens = [
+            prompt_toks
+            + spatial_toks
+            + self.tokenizer.encode(question).ids
+            + self.config.tokenizer.templates["query"]["suffix"]
+        ]
+
+        if reasoning:
+            prompt_tokens[0] += [self.config.tokenizer.thinking_id]
+            prompt_tokens = torch.tensor(prompt_tokens, device=self.device)
+            pos, reasoning_text, reasoning_grounding = self._generate_reasoning(
+                prompt_tokens, pos, settings, spatial_refs, attn_mask=attn_mask
+            )
+            prompt_tokens = [self.config.tokenizer.templates["query"]["suffix"]]
+            reasoning_dict = {
+                "reasoning": {"text": reasoning_text, "grounding": reasoning_grounding}
+            }
+        else:
+            prompt_tokens[0] += self.config.tokenizer.templates["query"]["suffix"]
+            reasoning_dict = {}
+
+        prompt_tokens = torch.tensor(prompt_tokens, device=self.device)
+
+        def generator():
+            for token in self._generate_answer(
+                prompt_tokens, pos, settings, spatial_refs, attn_mask=attn_mask
+            ):
+                yield token
+
+        if stream:
+            return {**reasoning_dict, "answer": generator()}
+        else:
+            return {**reasoning_dict, "answer": "".join(list(generator()))}
+
+    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 caption(
+        self,
+        image: Union[Image.Image, EncodedImage],
+        length: Literal["normal", "short", "long"] = "normal",
+        stream: bool = False,
+        settings: Optional[TextSamplingSettings] = None,
+    ):
+        if self.config.tokenizer.templates["caption"] is None:
+            raise NotImplementedError("Model does not support captioning.")
+        if length not in self.config.tokenizer.templates["caption"]:
+            raise ValueError(f"Model does not support caption length '{length}'.")
+
+        image = self.encode_image(image, settings)
+        self.load_encoded_image(image)
+
+        prompt_tokens = torch.tensor(
+            [self.config.tokenizer.templates["caption"][length]], device=self.device
+        )
+
+        def generator():
+            for token in self._generate_answer(prompt_tokens, image.pos, settings):
+                yield token
+
+        if stream:
+            return {"caption": generator()}
+        else:
+            return {"caption": "".join(list(generator()))}
+
+    def _generate_points(
+        self,
+        hidden: torch.Tensor,
+        next_token: torch.Tensor,
+        pos: int,
+        include_size: bool = True,
+        max_objects: int = DEFAULT_MAX_OBJECTS,
+        lora: Optional[dict] = None,
+    ):
+        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, lora)
+                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, lora)
+                    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, lora)
+                pos += 1
+                next_token = torch.argmax(logits, dim=-1)
+
+        return out
+
+    def detect(
+        self,
+        image: Union[Image.Image, EncodedImage],
+        object: str,
+        settings: Optional[ObjectSamplingSettings] = None,
+    ):
+        if self.config.tokenizer.templates["detect"] is None:
+            raise NotImplementedError("Model does not support object detection.")
+
+        image = self.encode_image(image, settings)
+        self.load_encoded_image(image)
+
+        prompt_tokens = torch.tensor(
+            [
+                self.config.tokenizer.templates["detect"]["prefix"]
+                + self.tokenizer.encode(" " + object).ids
+                + self.config.tokenizer.templates["detect"]["suffix"]
+            ],
+            device=self.device,
+        )
+
+        lora = (
+            variant_state_dict(settings["variant"], device=self.device)
+            if settings is not None and "variant" in settings
+            else None
+        )
+
+        _, hidden, next_token, pos = self._prefill_prompt(
+            prompt_tokens, image.pos, temperature=0, top_p=0, lora=lora
+        )
+        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=True,
+            max_objects=max_objects,
+            lora=lora,
+        )
+
+        return {"objects": objects}
+
+    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, settings)
+        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,
+        )
+
+        lora = (
+            variant_state_dict(settings["variant"], device=self.device)
+            if settings is not None and "variant" in settings
+            else None
+        )
+
+        _, hidden, next_token, pos = self._prefill_prompt(
+            prompt_tokens, image.pos, temperature=0, top_p=0, lora=lora
+        )
+        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,
+            lora=lora,
+        )
+
+        return {"points": objects}
+
+    def _detect_gaze(
+        self,
+        image: EncodedImage,
+        source: Tuple[float, float],
+        force_detect: bool = False,
+    ):
+        with torch.inference_mode():
+            before_emb = text_encoder(
+                torch.tensor(
+                    [self.tokenizer.encode("\n\nPoint:").ids], device=self.device
+                ),
+                self.text,
+            )
+            after_emb = text_encoder(
+                torch.tensor(
+                    [self.tokenizer.encode(" gaze\n\n").ids], device=self.device
+                ),
+                self.text,
+            )
+            x_emb = encode_coordinate(
+                torch.tensor([[[source[0]]]], device=self.device, dtype=torch.bfloat16),
+                self.region,
+            )
+            y_emb = encode_coordinate(
+                torch.tensor([[[source[1]]]], device=self.device, dtype=torch.bfloat16),
+                self.region,
+            )
+
+            prompt_emb = torch.cat([before_emb, x_emb, y_emb, after_emb], dim=1)
+
+            self.load_encoded_image(image)
+
+            mask = self.attn_mask[:, :, image.pos : image.pos + prompt_emb.size(1), :]
+            pos_ids = torch.arange(
+                image.pos, image.pos + prompt_emb.size(1), dtype=torch.long
+            )
+            hidden = self._prefill(prompt_emb, mask, pos_ids, lora=None)
+            logits = lm_head(hidden, self.text)
+            next_token = torch.argmax(logits, dim=-1)
+            pos = image.pos + prompt_emb.size(1)
+            hidden = hidden[:, -1:, :]
+
+            if force_detect:
+                next_token = torch.tensor([[0]], device=self.device)
+
+            if next_token.item() == self.config.tokenizer.eos_id:
+                return None
+
+            gaze = self._generate_points(
+                hidden, next_token, pos, include_size=False, max_objects=1
+            )
+            return gaze[0]
+
+    def detect_gaze(
+        self,
+        image: Union[Image.Image, EncodedImage],
+        eye: Optional[Tuple[float, float]] = None,
+        face: Optional[Dict[str, float]] = None,
+        unstable_settings: Dict[str, Any] = {},
+    ):
+        if "force_detect" in unstable_settings:
+            force_detect = unstable_settings["force_detect"]
+        else:
+            force_detect = False
+
+        if "prioritize_accuracy" in unstable_settings:
+            prioritize_accuracy = unstable_settings["prioritize_accuracy"]
+        else:
+            prioritize_accuracy = False
+
+        if not prioritize_accuracy:
+            if eye is None:
+                raise ValueError("eye must be provided when prioritize_accuracy=False")
+            image = self.encode_image(image)
+            return {"gaze": self._detect_gaze(image, eye, force_detect=force_detect)}
+        else:
+            if (
+                not isinstance(image, Image.Image)
+                and "flip_enc_img" not in unstable_settings
+            ):
+                raise ValueError(
+                    "image must be a PIL Image when prioritize_accuracy=True, "
+                    "or flip_enc_img must be provided"
+                )
+            if face is None:
+                raise ValueError("face must be provided when prioritize_accuracy=True")
+
+            encoded_image = self.encode_image(image)
+            if (
+                isinstance(image, Image.Image)
+                and "flip_enc_img" not in unstable_settings
+            ):
+                flipped_pil = image.copy()
+                flipped_pil = flipped_pil.transpose(method=Image.FLIP_LEFT_RIGHT)
+                encoded_flipped_image = self.encode_image(flipped_pil)
+            else:
+                encoded_flipped_image = unstable_settings["flip_enc_img"]
+
+            N = 10
+
+            detections = [
+                self._detect_gaze(
+                    encoded_image,
+                    (
+                        random.uniform(face["x_min"], face["x_max"]),
+                        random.uniform(face["y_min"], face["y_max"]),
+                    ),
+                    force_detect=force_detect,
+                )
+                for _ in range(N)
+            ]
+            detections = [
+                (gaze["x"], gaze["y"]) for gaze in detections if gaze is not None
+            ]
+            flipped_detections = [
+                self._detect_gaze(
+                    encoded_flipped_image,
+                    (
+                        1 - random.uniform(face["x_min"], face["x_max"]),
+                        random.uniform(face["y_min"], face["y_max"]),
+                    ),
+                    force_detect=force_detect,
+                )
+                for _ in range(N)
+            ]
+            detections.extend(
+                [
+                    (1 - gaze["x"], gaze["y"])
+                    for gaze in flipped_detections
+                    if gaze is not None
+                ]
+            )
+
+            if len(detections) < N:
+                return {"gaze": None}
+
+            detections = remove_outlier_points(detections)
+            mean_gaze = (
+                sum(gaze[0] for gaze in detections) / len(detections),
+                sum(gaze[1] for gaze in detections) / len(detections),
+            )
+
+            return {"gaze": {"x": mean_gaze[0], "y": mean_gaze[1]}}
+
+
+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

region.py

@@ -0,0 +1,136 @@
+import torch
+import torch.nn as nn
+import math
+
+from typing import List, Tuple, Union
+
+from .layers import mlp
+
+SpatialRefs = List[Union[Tuple[float, float], Tuple[float, float, float, float]]]
+
+
+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 w.coord_encoder(fourier_features(coord, w.coord_features))
+
+
+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 w.size_encoder(fourier_features(size, w.size_features))
+
+
+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)
+
+
+def encode_spatial_refs(spatial_refs: SpatialRefs, w: nn.Module) -> torch.Tensor:
+    """
+    Takes a list of spatial references (points or regions) and encodes them into
+    hidden states for input to the text model.
+
+    Args:
+        spatial_refs: List of spatial references (points or boxes)
+            - Points are represented as normalized (x, y) tuples
+            - Boxes are represented as normalized (x_min, y_min, x_max, y_max) tuples
+
+    Returns:
+        {"coords": torch.Tensor, "sizes": Optional[torch.Tensor]}
+    """
+    coords, sizes = [], []
+    for ref in spatial_refs:
+        if len(ref) == 2:
+            coords.append(ref[0])
+            coords.append(ref[1])
+        else:
+            x_c = (ref[0] + ref[2]) / 2
+            y_c = (ref[1] + ref[3]) / 2
+            width = ref[2] - ref[0]
+            height = ref[3] - ref[1]
+            coords.append(x_c)
+            coords.append(y_c)
+            sizes.append([width, height])
+
+    coords = torch.tensor(
+        coords, device=w.coord_features.device, dtype=w.coord_features.dtype
+    ).view(-1, 1)
+    coords = encode_coordinate(coords, w)
+
+    if sizes:
+        sizes = torch.tensor(
+            sizes, device=w.size_features.device, dtype=w.size_features.dtype
+        )
+        sizes = encode_size(sizes, w)
+    else:
+        sizes = None
+
+    return {"coords": coords, "sizes": sizes}

region_model.py

@@ -0,0 +1,43 @@
+import torch
+import torch.nn as nn
+from .fourier_features import FourierFeatures
+
+class RegionModel(nn.Module):
+    def __init__(self):
+        super().__init__()
+
+        self.position_features = FourierFeatures(2, 256)
+        self.position_encoder = nn.Linear(256, 2048)
+        self.size_features = FourierFeatures(2, 256)
+        self.size_encoder = nn.Linear(256, 2048)
+
+        self.position_decoder = nn.Linear(2048, 2)
+        self.size_decoder = nn.Linear(2048, 2)
+        self.confidence_decoder = nn.Linear(2048, 1)
+
+    def encode_position(self, position):
+        return self.position_encoder(self.position_features(position))
+
+    def encode_size(self, size):
+        return self.size_encoder(self.size_features(size))
+
+    def decode_position(self, x):
+        return self.position_decoder(x)
+
+    def decode_size(self, x):
+        return self.size_decoder(x)
+
+    def decode_confidence(self, x):
+        return self.confidence_decoder(x)
+
+    def encode(self, position, size):
+        return torch.stack(
+            [self.encode_position(position), self.encode_size(size)], dim=0
+        )
+
+    def decode(self, position_logits, size_logits):
+        return (
+            self.decode_position(position_logits),
+            self.decode_size(size_logits),
+            self.decode_confidence(size_logits),
+        )

requirements.txt

@@ -0,0 +1,3 @@
+einops
+pyvips-binary==8.16.0
+pyvips==2.2.3

rope.py

@@ -0,0 +1,48 @@
+# Ethically sourced from https://github.com/xjdr-alt/entropix
+
+import torch
+
+
+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 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:]
+
+    if interleave:
+        xq_r = x_rot.float().reshape(*x_rot.shape[:-1], -1, 2)[..., 0]
+        xq_i = x_rot.float().reshape(*x_rot.shape[:-1], -1, 2)[..., 1]
+    else:
+        d_q = x_rot.shape[-1] // 2
+        xq_r, xq_i = x_rot[..., :d_q], x_rot[..., d_q:]
+
+    freqs_cos = freqs_cis[..., 0][position_ids, :].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)

special_tokens_map.json

@@ -0,0 +1,5 @@
+{
+  "bos_token": "<|endoftext|>",
+  "eos_token": "<|endoftext|>",
+  "unk_token": "<|endoftext|>"
+}

text.py

@@ -0,0 +1,221 @@
+import torch
+import torch.nn as nn
+
+from torch.nn import functional as F
+from typing import Optional
+
+from .layers import layer_norm, mlp, QuantizedLinear
+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,
+    n_kv_heads: int,
+    position_ids: torch.Tensor,
+    lora: Optional[dict],
+):
+    bsz, q_len, d_model = x.shape
+    head_dim = d_model // n_heads
+
+    qkv_out = w.qkv(x)  # shape: (bsz, q_len, (n_heads + 2*n_kv_heads)*head_dim)
+    if lora is not None:
+        qkv_out += F.linear(F.linear(x, lora["qkv"]["A"]), lora["qkv"]["B"])
+    q_dim = n_heads * head_dim
+    kv_dim = n_kv_heads * head_dim
+    q, k, v = qkv_out.split([q_dim, kv_dim, kv_dim], dim=-1)
+    del qkv_out
+
+    q = q.view(bsz, q_len, n_heads, head_dim).transpose(1, 2)
+    k = k.view(bsz, q_len, n_kv_heads, head_dim).transpose(1, 2)
+    v = v.view(bsz, q_len, n_kv_heads, head_dim).transpose(1, 2)
+
+    q = apply_rotary_emb(q, freqs_cis, position_ids, n_heads)
+    k = apply_rotary_emb(k, freqs_cis, position_ids, n_kv_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, enable_gqa=n_heads != n_kv_heads
+    )
+    out = out.transpose(1, 2).reshape(bsz, q_len, d_model)
+
+    out0 = w.proj(out)
+    if lora is not None:
+        out1 = F.linear(F.linear(x, lora["proj"]["A"]), lora["proj"]["B"])
+        out = out0 + out1
+    else:
+        out = out0
+
+    return out
+
+
+def _attn(
+    x: torch.Tensor,
+    w: torch.Tensor,
+    freqs_cis: torch.Tensor,
+    attn_mask: torch.Tensor,
+    n_heads: int,
+    n_kv_heads: int,
+):
+    bsz, q_len, d_model = x.shape
+    head_dim = d_model // n_heads
+    pos = 0
+
+    qkv_out = w.qkv(x)  # shape: (bsz, q_len, (n_heads + 2*n_kv_heads)*head_dim)
+    q_dim = n_heads * head_dim
+    kv_dim = n_kv_heads * head_dim
+
+    q = qkv_out[..., :q_dim].view(bsz, q_len, n_heads, head_dim).transpose(1, 2)
+    k = (
+        qkv_out[..., q_dim : q_dim + kv_dim]
+        .view(bsz, q_len, n_kv_heads, head_dim)
+        .transpose(1, 2)
+    )
+    v = (
+        qkv_out[..., q_dim + kv_dim :]
+        .view(bsz, q_len, n_kv_heads, head_dim)
+        .transpose(1, 2)
+    )
+
+    position_ids = torch.arange(pos, pos + q_len, dtype=torch.long)
+    q = apply_rotary_emb(q, freqs_cis, position_ids, n_heads)
+    k = apply_rotary_emb(k, freqs_cis, position_ids, n_kv_heads)
+    out = F.scaled_dot_product_attention(
+        q, k, v, attn_mask=attn_mask, enable_gqa=n_heads != n_kv_heads
+    )
+    out = out.transpose(1, 2).reshape(bsz, q_len, d_model)
+    out = w.proj(out)
+    return out
+
+
+def _produce_hidden(inputs_embeds: torch.Tensor, w: nn.Module, config: TextConfig):
+    hidden_BTC = inputs_embeds
+
+    bsz, q_len, d_model = inputs_embeds.shape
+    attn_mask = torch.zeros(q_len, q_len)
+    attn_mask[:730, :730] = 1
+    for i in range(730, q_len):
+        attn_mask[i, : i + 1] = 1
+    attn_mask = attn_mask.to(dtype=torch.bool)
+
+    for i, block in enumerate(w.blocks):
+        l_in = layer_norm(hidden_BTC, block.ln)
+        l_attn = _attn(
+            x=l_in,
+            w=block.attn,
+            freqs_cis=w.freqs_cis,
+            attn_mask=attn_mask,
+            n_heads=config.n_heads,
+            n_kv_heads=config.n_kv_heads,
+        )
+        l_mlp = mlp(l_in, block.mlp)
+        hidden_BTC = hidden_BTC + l_attn + l_mlp
+
+    return hidden_BTC
+
+
+def text_decoder(
+    x: torch.Tensor,
+    w: nn.Module,
+    attn_mask: torch.Tensor,
+    position_ids: torch.Tensor,
+    config: TextConfig,
+    lora: Optional[dict],
+):
+    for i, block in enumerate(w.blocks):
+        if lora is not None:
+            layer_lora = lora["text"]["blocks"][str(i)]
+            mlp_lora = layer_lora["mlp"]
+            attn_lora = layer_lora["attn"]
+        else:
+            mlp_lora = None
+            attn_lora = None
+
+        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,
+            n_kv_heads=config.n_kv_heads,
+            position_ids=position_ids,
+            lora=attn_lora,
+        )
+        l_mlp = mlp(l_in, block.mlp, lora=mlp_lora)
+        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 _lm_head(hidden_BTC: torch.Tensor, w: nn.Module):
+    hidden_BTC = layer_norm(hidden_BTC, w.post_ln)
+    logits = w.lm_head(hidden_BTC)
+    return logits
+
+
+def build_text_model(config: TextConfig, dtype: torch.dtype) -> nn.Module:
+    qkv_dim = int(config.dim * (1 + 2 * config.n_kv_heads / config.n_heads))
+    linear_cls = QuantizedLinear if config.group_size is not None else nn.Linear
+
+    text = nn.ModuleDict(
+        {
+            "blocks": nn.ModuleList(
+                [
+                    nn.ModuleDict(
+                        {
+                            "ln": nn.LayerNorm(config.dim, dtype=dtype),
+                            "attn": nn.ModuleDict(
+                                {
+                                    "qkv": linear_cls(config.dim, qkv_dim, dtype=dtype),
+                                    "proj": linear_cls(
+                                        config.dim, config.dim, dtype=dtype
+                                    ),
+                                }
+                            ),
+                            "mlp": nn.ModuleDict(
+                                {
+                                    "fc1": linear_cls(
+                                        config.dim, config.ff_dim, dtype=dtype
+                                    ),
+                                    "fc2": linear_cls(
+                                        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

tokenizer_config.json

@@ -0,0 +1,323 @@
+{
+  "add_prefix_space": false,
+  "added_tokens_decoder": {
+    "50256": {
+      "content": "<|endoftext|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "50257": {
+      "content": "                               ",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "50258": {
+      "content": "                              ",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "50259": {
+      "content": "                             ",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "50260": {
+      "content": "                            ",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "50261": {
+      "content": "                           ",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "50262": {
+      "content": "                          ",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "50263": {
+      "content": "                         ",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "50264": {
+      "content": "                        ",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "50265": {
+      "content": "                       ",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "50266": {
+      "content": "                      ",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "50267": {
+      "content": "                     ",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "50268": {
+      "content": "                    ",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "50269": {
+      "content": "                   ",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "50270": {
+      "content": "                  ",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "50271": {
+      "content": "                 ",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "50272": {
+      "content": "                ",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "50273": {
+      "content": "               ",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "50274": {
+      "content": "              ",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "50275": {
+      "content": "             ",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "50276": {
+      "content": "            ",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "50277": {
+      "content": "           ",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "50278": {
+      "content": "          ",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "50279": {
+      "content": "         ",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "50280": {
+      "content": "        ",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "50281": {
+      "content": "       ",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "50282": {
+      "content": "      ",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "50283": {
+      "content": "     ",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "50284": {
+      "content": "    ",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "50285": {
+      "content": "   ",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "50286": {
+      "content": "  ",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "50287": {
+      "content": "\t\t\t\t\t\t\t\t\t",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "50288": {
+      "content": "\t\t\t\t\t\t\t\t",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "50289": {
+      "content": "\t\t\t\t\t\t\t",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "50290": {
+      "content": "\t\t\t\t\t\t",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "50291": {
+      "content": "\t\t\t\t\t",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "50292": {
+      "content": "\t\t\t\t",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "50293": {
+      "content": "\t\t\t",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "50294": {
+      "content": "\t\t",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    }
+  },
+  "bos_token": "<|endoftext|>",
+  "clean_up_tokenization_spaces": true,
+  "eos_token": "<|endoftext|>",
+  "model_max_length": 2048,
+  "tokenizer_class": "CodeGenTokenizer",
+  "unk_token": "<|endoftext|>"
+}

utils.py

@@ -0,0 +1,41 @@
+import numpy as np
+
+
+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

versions.txt

@@ -0,0 +1,12 @@
+2024-03-04
+2024-03-06
+2024-03-13
+2024-04-02
+2024-05-08
+2024-05-20
+2024-07-23
+2024-08-26
+2025-01-09
+2025-03-27
+2025-04-14
+2025-06-21

vision.py

@@ -0,0 +1,147 @@
+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, 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]]:
+    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.bfloat16)
+        .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 = w.patch_emb(x)
+    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

vision_encoder.py

@@ -0,0 +1,325 @@
+from typing import Union
+
+import PIL.Image
+import torch
+import torch.nn.functional as F
+from torch import nn
+from einops import rearrange
+import PIL
+from torchvision.transforms.v2 import (
+    Compose,
+    Resize,
+    InterpolationMode,
+    ToImage,
+    ToDtype,
+    Normalize,
+)
+from transformers.utils import is_flash_attn_2_available
+
+try:
+    if is_flash_attn_2_available():
+        from flash_attn.modules.mha import FlashSelfAttention
+    else:
+        FlashSelfAttention = None
+except ImportError:
+    FlashSelfAttention = None
+
+
+class Attention(nn.Module):
+
+    def __init__(self, dim, num_heads=16, use_flash_attn=False):
+        super().__init__()
+        assert dim % num_heads == 0, "dim should be divisible by num_heads"
+
+        self.num_heads = num_heads
+        self.head_dim = dim // num_heads
+
+        self.qkv = nn.Linear(dim, dim * 3)
+        self.proj = nn.Linear(dim, dim)
+
+        if use_flash_attn and FlashSelfAttention is not None:
+            self.flash_attn = FlashSelfAttention()
+        else:
+            self.flash_attn = None
+
+        torch.nn.init.kaiming_normal_(
+            self.qkv.weight, mode="fan_in", nonlinearity="relu"
+        )
+        torch.nn.init.kaiming_normal_(
+            self.proj.weight, mode="fan_in", nonlinearity="relu"
+        )
+
+    def forward(self, x: torch.Tensor) -> torch.Tensor:
+        if self.flash_attn is not None:
+            qkv = self.qkv(x)
+            qkv = rearrange(
+                qkv, "... (three h d) -> ... three h d", three=3, h=self.num_heads
+            )
+            attn_output = self.flash_attn(qkv)
+            output = rearrange(attn_output, "... h d -> ... (h d)")
+            output = self.proj(output)
+            return output
+        else:
+            B, N, C = x.shape
+            qkv = (
+                self.qkv(x)
+                .reshape(B, N, 3, self.num_heads, self.head_dim)
+                .permute(2, 0, 3, 1, 4)
+            )
+            q, k, v = qkv.unbind(0)
+
+            x = F.scaled_dot_product_attention(q, k, v)
+
+            x = x.transpose(1, 2).reshape(B, N, C)
+            x = self.proj(x)
+            return x
+
+
+class VitBlock(nn.Module):
+
+    def __init__(self, embed_dim, use_flash_attn=False):
+        super().__init__()
+        self.attn = Attention(embed_dim, use_flash_attn=use_flash_attn)
+        self.mlp = MLP(embed_dim, 4304)
+        self.norm1 = nn.LayerNorm(embed_dim)
+        self.norm2 = nn.LayerNorm(embed_dim)
+
+    def forward(self, x):
+        x = x + self.attn(self.norm1(x))
+        x = x + self.mlp(self.norm2(x))
+        return x
+
+
+class VisionTransformer(nn.Module):
+
+    def __init__(self, use_flash_attn=False):
+        super().__init__()
+
+        embed_len = 729
+        embed_dim = 1152
+
+        self.patch_embed = LinearPatchEmbedding()
+        self.pos_embed = nn.Parameter(torch.randn(1, embed_len, embed_dim) * 0.02)
+        self.blocks = nn.Sequential(
+            *[VitBlock(embed_dim, use_flash_attn=use_flash_attn) for _ in range(27)]
+        )
+        self.norm = nn.LayerNorm(embed_dim)
+
+    def forward(self, x):
+        x = self.patch_embed(x)
+        x = x + self.pos_embed
+        for block in self.blocks:
+            x = block(x)
+        return self.norm(x)
+
+
+class EncoderWrapper(nn.Module):
+
+    def __init__(self, use_flash_attn=False):
+        super().__init__()
+        self.model = nn.ModuleDict({"visual": VisionTransformer(use_flash_attn)})
+
+    def forward(self, x):
+        return self.model["visual"](x)
+
+
+class LinearPatchEmbedding(nn.Module):
+
+    def __init__(self):
+        super().__init__()
+        self.linear = nn.Linear(588, 1152)
+
+    def forward(self, x):
+        b, c, hp1, wp2 = x.shape
+        p1, p2 = 14, 14
+        h, w = hp1 // p1, wp2 // p2
+        x = x.reshape(b, c, h, p1, w, p2)
+        x = x.permute(0, 2, 4, 1, 3, 5)
+        x = x.reshape(b, h * w, c * p1 * p2)
+
+        return self.linear(x)
+
+
+class MLP(nn.Module):
+    def __init__(
+        self,
+        in_features: int,
+        hidden_features: int = None,
+        out_features: int = None,
+    ) -> None:
+        super().__init__()
+        out_features = out_features or in_features
+        hidden_features = hidden_features or in_features
+        self.fc1 = nn.Linear(in_features, hidden_features)
+        self.act = nn.GELU(approximate="tanh")
+        self.fc2 = nn.Linear(hidden_features, out_features)
+
+        torch.nn.init.kaiming_normal_(
+            self.fc1.weight, mode="fan_in", nonlinearity="relu"
+        )
+        torch.nn.init.kaiming_normal_(
+            self.fc2.weight, mode="fan_in", nonlinearity="relu"
+        )
+
+    def forward(self, x: torch.Tensor) -> torch.Tensor:
+        x = self.fc1(x)
+        x = self.act(x)
+        x = self.fc2(x)
+        return x
+
+
+class VisionProjection(nn.Module):
+    def __init__(self):
+        super().__init__()
+
+        image_embedding_dim = 1152
+        model_dim = 2048
+        hidden_dim = model_dim * 4
+
+        self.mlp = MLP(image_embedding_dim * 2, hidden_dim, model_dim)
+
+    @property
+    def device(self):
+        return self.mlp.fc1.weight.device
+
+    def forward(self, x):
+        return self.mlp(x)
+
+
+def create_patches(image, patch_size=(378, 378)):
+    assert image.dim() == 3, "Image must be in CHW format"
+
+    _, height, width = image.shape  # Channels, Height, Width
+    patch_height, patch_width = patch_size
+
+    if height == patch_height and width == patch_width:
+        return []
+
+    # Iterate over the image and create patches
+    patches = []
+    for i in range(0, height, patch_height):
+        row_patches = []
+        for j in range(0, width, patch_width):
+            patch = image[:, i : i + patch_height, j : j + patch_width]
+            row_patches.append(patch)
+        patches.append(torch.stack(row_patches))
+    return patches
+
+
+class VisionEncoder(nn.Module):
+
+    def __init__(self, use_flash_attn=False):
+        super().__init__()
+
+        self.encoder = EncoderWrapper(use_flash_attn)
+        self.projection = VisionProjection()
+        self.supported_sizes = [(378, 378), (378, 756), (756, 378), (756, 756)]
+
+    @property
+    def device(self):
+        return self.projection.mlp.fc1.weight.device
+
+    @property
+    def dtype(self):
+        return self.projection.mlp.fc1.weight.dtype
+
+    def preprocess(self, image: PIL.Image.Image):
+        width, height = image.size
+        max_dim = max(width, height)
+        if max_dim < 512:
+            im_size = (378, 378)
+        else:
+            aspect_ratio = width / height
+            im_size = min(
+                self.supported_sizes,
+                key=lambda size: (
+                    abs((size[1] / size[0]) - aspect_ratio),
+                    abs(size[0] - width) + abs(size[1] - height),
+                ),
+            )
+
+        return Compose(
+            [
+                Resize(size=im_size, interpolation=InterpolationMode.BICUBIC),
+                ToImage(),
+                ToDtype(torch.float32, scale=True),
+                Normalize(mean=[0.5, 0.5, 0.5], std=[0.5, 0.5, 0.5]),
+            ]
+        )(image)
+
+    def forward(
+        self, images: Union[PIL.Image.Image, list[PIL.Image.Image], torch.Tensor]
+    ) -> torch.Tensor:
+        im_list = None
+        if isinstance(images, torch.Tensor):
+            # Input must have dimensions (B, C, H, W)
+            assert (
+                len(images.shape) == 4
+            ), "Tensor input must have dimensions (B, C, H, W)"
+            im_list = list(images)
+        elif isinstance(images, PIL.Image.Image):
+            im_list = [images]
+        elif isinstance(images, list):
+            im_list = images
+        else:
+            raise ValueError(
+                "Input must be a PIL image, list of PIL images, or a tensor"
+            )
+
+        # Preprocess unless the images are already tensors (indicating that
+        # they have already been preprocessed)
+        if not isinstance(im_list[0], torch.Tensor):
+            im_list = [self.preprocess(im.convert("RGB")) for im in im_list]
+
+        patches = [create_patches(im) for im in im_list]
+        flat_patches = [patch for image_patches in patches for patch in image_patches]
+
+        # Images may be variable size, and need to be resized to a common size after
+        # creating patches.
+        resized_images = [
+            F.interpolate(im.unsqueeze(0), size=(378, 378), mode="bilinear")
+            for im in im_list
+        ]
+
+        combined_images = torch.cat([*resized_images, *flat_patches], dim=0)
+        combined_images = combined_images.to(self.device, dtype=self.dtype)
+
+        combined_features = self.encoder(combined_images)
+
+        full_img_features = combined_features[: len(im_list)]
+        patch_features = (
+            combined_features[len(im_list) :].transpose(1, 2).view(-1, 1152, 27, 27)
+        )
+
+        # Reshape patch features back to their original structure
+        reshaped_patch_features = []
+        patch_idx = 0
+        for i, patch_set in enumerate(patches):
+            if len(patch_set) == 0:
+                reshaped_patch_features.append(
+                    full_img_features[i].transpose(0, 1).view(1152, 27, 27)
+                )
+            else:
+                sample_features = []
+                for row_patches in patch_set:
+                    row_len = len(row_patches)
+                    row_features = patch_features[
+                        patch_idx : patch_idx + row_len
+                    ]  # row_len, T, C
+                    row_features = torch.cat(
+                        list(row_features), dim=2
+                    )  # T, C * row_len
+                    patch_idx += row_len
+                    sample_features.append(row_features)
+                sample_features = torch.cat(sample_features, dim=1)
+                sample_features = F.interpolate(
+                    sample_features.unsqueeze(0), size=(27, 27), mode="bilinear"
+                ).squeeze(0)
+                reshaped_patch_features.append(sample_features)
+        reshaped_patch_features = (
+            torch.stack(reshaped_patch_features).view(-1, 1152, 729).transpose(1, 2)
+        )
+
+        final_features = torch.cat([full_img_features, reshaped_patch_features], dim=2)
+
+        return self.projection(final_features)

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()