nexaai 1.0.29__cp310-cp310-macosx_14_0_universal2.whl

nexaai/mlx_backend/vlm/modeling/models/qwen3vl_moe/switch_layers.py

@@ -0,0 +1,210 @@
+import math
+import mlx.core as mx
+import mlx.nn as nn
+
+def _gather_sort(x, indices):
+    *_, M = indices.shape
+    indices = indices.flatten()
+    order = mx.argsort(indices)
+    inv_order = mx.argsort(order)
+    return x.flatten(0, -3)[order // M], indices[order], inv_order
+
+
+def _scatter_unsort(x, inv_order, shape=None):
+    x = x[inv_order]
+    if shape is not None:
+        x = mx.unflatten(x, 0, shape)
+    return x
+
+
+class QuantizedSwitchLinear(nn.Module):
+    def __init__(
+        self,
+        input_dims: int,
+        output_dims: int,
+        num_experts: int,
+        bias: bool = True,
+        group_size: int = 64,
+        bits: int = 4,
+    ):
+        super().__init__()
+
+        scale = math.sqrt(1 / input_dims)
+        self.weight, self.scales, self.biases = mx.quantize(
+            mx.random.uniform(
+                low=-scale,
+                high=scale,
+                shape=(num_experts, output_dims, input_dims),
+            ),
+            group_size=group_size,
+            bits=bits,
+        )
+
+        if bias:
+            self.bias = mx.zeros((num_experts, output_dims))
+
+        self.group_size = group_size
+        self.bits = bits
+
+        # Freeze this model's parameters
+        self.freeze()
+
+    def unfreeze(self, *args, **kwargs):
+        """Wrap unfreeze so that we unfreeze any layers we might contain but
+        our parameters will remain frozen."""
+        super().unfreeze(*args, **kwargs)
+        self.freeze(recurse=False)
+
+    @property
+    def input_dims(self):
+        return self.scales.shape[2] * self.group_size
+
+    @property
+    def output_dims(self):
+        return self.weight.shape[1]
+
+    @property
+    def num_experts(self):
+        return self.weight.shape[0]
+
+    def __call__(self, x, indices, sorted_indices=False):
+        x = mx.gather_qmm(
+            x,
+            self["weight"],
+            self["scales"],
+            self["biases"],
+            rhs_indices=indices,
+            transpose=True,
+            group_size=self.group_size,
+            bits=self.bits,
+            sorted_indices=sorted_indices,
+        )
+        if "bias" in self:
+            x = x + mx.expand_dims(self["bias"][indices], -2)
+        return x
+
+
+class SwitchLinear(nn.Module):
+    def __init__(
+        self, input_dims: int, output_dims: int, num_experts: int, bias: bool = True
+    ):
+        super().__init__()
+        scale = math.sqrt(1 / input_dims)
+        self.weight = mx.random.uniform(
+            low=-scale,
+            high=scale,
+            shape=(num_experts, output_dims, input_dims),
+        )
+
+        if bias:
+            self.bias = mx.zeros((num_experts, output_dims))
+
+    @property
+    def input_dims(self):
+        return self.weight.shape[2]
+
+    @property
+    def output_dims(self):
+        return self.weight.shape[1]
+
+    @property
+    def num_experts(self):
+        return self.weight.shape[0]
+
+    def __call__(self, x, indices, sorted_indices=False):
+        x = mx.gather_mm(
+            x,
+            self["weight"].swapaxes(-1, -2),
+            lhs_indices=None,
+            rhs_indices=indices,
+        )
+        if "bias" in self:
+            x = x + mx.expand_dims(self["bias"][indices], -2)
+        return x
+
+    def to_quantized(self, group_size: int = 64, bits: int = 4):
+        num_experts, output_dims, input_dims = self.weight.shape
+        ql = QuantizedSwitchLinear(
+            input_dims, output_dims, num_experts, False, group_size, bits
+        )
+        ql.weight, ql.scales, ql.biases = mx.quantize(self.weight, group_size, bits)
+        if "bias" in self:
+            ql.bias = self.bias
+        return ql
+
+
+class SwitchGLU(nn.Module):
+    def __init__(
+        self,
+        input_dims: int,
+        hidden_dims: int,
+        num_experts: int,
+        activation=nn.SiLU(),
+        bias: bool = False,
+    ):
+        super().__init__()
+
+        self.gate_proj = SwitchLinear(input_dims, hidden_dims, num_experts, bias=bias)
+        self.up_proj = SwitchLinear(input_dims, hidden_dims, num_experts, bias=bias)
+        self.down_proj = SwitchLinear(hidden_dims, input_dims, num_experts, bias=bias)
+        self.activation = activation
+
+    def __call__(self, x, indices) -> mx.array:
+        x = mx.expand_dims(x, (-2, -3))
+
+        # When we have many tokens, then sort them to make sure that the access
+        # of different experts is in order.
+        do_sort = indices.size >= 64
+        idx = indices
+        inv_order = None
+        if do_sort:
+            x, idx, inv_order = _gather_sort(x, indices)
+
+        x_up = self.up_proj(x, idx, sorted_indices=do_sort)
+        x_gate = self.gate_proj(x, idx, sorted_indices=do_sort)
+        x = self.down_proj(
+            self.activation(x_gate) * x_up,
+            idx,
+            sorted_indices=do_sort,
+        )
+
+        if do_sort:
+            x = _scatter_unsort(x, inv_order, indices.shape)
+
+        return x.squeeze(-2)
+
+
+class SwitchMLP(nn.Module):
+    def __init__(
+        self,
+        input_dims: int,
+        hidden_dims: int,
+        num_experts: int,
+        activation=nn.GELU(approx="precise"),
+        bias: bool = False,
+    ):
+        super().__init__()
+
+        self.fc1 = SwitchLinear(input_dims, hidden_dims, num_experts, bias=bias)
+        self.fc2 = SwitchLinear(hidden_dims, input_dims, num_experts, bias=bias)
+        self.activation = activation
+
+    def __call__(self, x, indices) -> mx.array:
+        x = mx.expand_dims(x, (-2, -3))
+
+        # When we have many tokens, then sort them to make sure that the access
+        # of different experts is in order.
+        do_sort = indices.size >= 64
+        idx = indices
+        inv_order = None
+        if do_sort:
+            x, idx, inv_order = _gather_sort(x, indices)
+
+        x = self.fc1(x, idx, sorted_indices=do_sort)
+        x = self.activation(x)
+        x = self.fc2(x, idx, sorted_indices=do_sort)
+
+        if do_sort:
+            x = _scatter_unsort(x, inv_order, indices.shape)
+
+        return x.squeeze(-2)

nexaai/mlx_backend/vlm/modeling/models/smolvlm/__init__.py

@@ -0,0 +1,8 @@
+from .smolvlm import (
+    LanguageModel,
+    Model,
+    ModelConfig,
+    TextConfig,
+    VisionConfig,
+    VisionModel,
+)

nexaai/mlx_backend/vlm/modeling/models/smolvlm/smolvlm.py

@@ -0,0 +1,62 @@
+import mlx.core as mx
+import numpy as np
+
+from ..idefics3 import LanguageModel
+from ..idefics3 import Model as Idefics3Model
+from ..idefics3 import ModelConfig, TextConfig, VisionConfig, VisionModel
+
+
+class Model(Idefics3Model):
+    def _prepare_inputs_for_multimodal(self, image_features, inputs_embeds, input_ids):
+        # Assumes bs == 1
+
+        B, T, D_text = inputs_embeds.shape
+        N, S, D_img = image_features.shape
+
+        image_offset = 0
+        cur_embeds = inputs_embeds[0]
+
+        # Find positions of <image> tokens in the text
+        image_token_index = self.config.image_token_index
+        image_positions = np.where(input_ids == image_token_index)[1].tolist()
+        num_image_tokens = len(image_positions)
+
+        # If no <image> => text-only
+        if num_image_tokens == 0:
+            empty_slice = image_features[0][:0, :]  # shape (0, D)
+            return mx.concatenate([cur_embeds, empty_slice], axis=0)
+
+        # Typically, if each image is S embeddings, we expect the total # of <image> tokens
+        # in this sample to be multiple of S => each group of S tokens = 1 image
+        if num_image_tokens % S != 0:
+            raise ValueError(
+                f"Input has {num_image_tokens} <image> tokens, not a multiple of S={S}. "
+                "Cannot map them to blocks of shape (S, D)."
+            )
+
+        chunks = [image_positions[i : i + S] for i in range(0, num_image_tokens, S)]
+
+        segments = []
+        text_start = 0
+
+        # For each chunk (each chunk => 1 image)
+        for chunk in chunks:
+            cur_block = image_features[image_offset]
+            image_offset += 1
+
+            # We'll iterate over the S positions in ascending order
+            for i_s, pos in enumerate(chunk):
+                if pos > text_start:
+                    segments.append(cur_embeds[text_start:pos])
+                # Then add one row from cur_block => shape (1, D)
+                row_of_block = cur_block[i_s : i_s + 1, :]
+                segments.append(row_of_block)
+                text_start = pos + 1
+
+        # leftover text after the final <image> token
+        if text_start < T:
+            segments.append(cur_embeds[text_start:])
+
+        # cat them into a single (T_b, D) tensor
+        merged_sample = mx.concatenate(segments, axis=0)
+        return mx.expand_dims(merged_sample, axis=0)

nexaai/mlx_backend/vlm/modeling/processing_qwen2_5_vl.py

@@ -0,0 +1,209 @@
+# Copied from transformers. Removed video-related code.
+
+from typing import Optional, Union
+
+import numpy as np
+
+from transformers.feature_extraction_utils import BatchFeature
+from transformers.image_utils import ImageInput
+from transformers.processing_utils import ImagesKwargs, MultiModalData, ProcessingKwargs, ProcessorMixin, Unpack, VideosKwargs
+from transformers.tokenization_utils_base import PreTokenizedInput, TextInput
+
+
+class Qwen2_5_VLImagesKwargs(ImagesKwargs):
+    min_pixels: Optional[int]
+    max_pixels: Optional[int]
+    patch_size: Optional[int]
+    temporal_patch_size: Optional[int]
+    merge_size: Optional[int]
+
+
+class Qwen2_5_VLProcessorKwargs(ProcessingKwargs, total=False):
+    images_kwargs: Qwen2_5_VLImagesKwargs
+    _defaults = {
+        "text_kwargs": {
+            "padding": False,
+            "return_mm_token_type_ids": False,
+        },
+    }
+
+
+class Qwen2_5_VLProcessor(ProcessorMixin):
+    r"""
+    Constructs a Qwen2.5-VL processor which wraps a Qwen2.5-VL image processor and a Qwen2 tokenizer into a single processor.
+    [`Qwen2_5_VLProcessor`] offers all the functionalities of [`Qwen2VLImageProcessor`] and [`Qwen2TokenizerFast`]. See the
+    [`~Qwen2_5_VLProcessor.__call__`] and [`~Qwen2_5_VLProcessor.decode`] for more information.
+    Args:
+        image_processor ([`Qwen2VLImageProcessor`], *optional*):
+            The image processor is a required input.
+        tokenizer ([`Qwen2TokenizerFast`], *optional*):
+            The tokenizer is a required input.
+        chat_template (`str`, *optional*): A Jinja template which will be used to convert lists of messages
+            in a chat into a tokenizable string.
+    """
+
+    attributes = ["image_processor", "tokenizer"]
+
+    image_processor_class = "AutoImageProcessor"
+    tokenizer_class = ("Qwen2Tokenizer", "Qwen2TokenizerFast")
+
+    def __init__(self, image_processor=None, tokenizer=None, chat_template=None, **kwargs):
+        self.image_token = "<|image_pad|>" if not hasattr(tokenizer, "image_token") else tokenizer.image_token
+        self.image_token_id = (
+            tokenizer.image_token_id
+            if getattr(tokenizer, "image_token_id", None)
+            else tokenizer.convert_tokens_to_ids(self.image_token)
+        )
+        super().__init__(image_processor, tokenizer, chat_template=chat_template)
+
+    def __call__(
+        self,
+        images: ImageInput = None,
+        text: Union[TextInput, PreTokenizedInput, list[TextInput], list[PreTokenizedInput]] = None,
+        **kwargs: Unpack[Qwen2_5_VLProcessorKwargs],
+    ) -> BatchFeature:
+        """
+        Main method to prepare for the model one or several sequences(s) and image(s). This method forwards the `text`
+        and `kwargs` arguments to Qwen2TokenizerFast's [`~Qwen2TokenizerFast.__call__`] if `text` is not `None` to encode
+        the text. To prepare the vision inputs, this method forwards the `vision_infos` and `kwrags` arguments to
+        Qwen2VLImageProcessor's [`~Qwen2VLImageProcessor.__call__`] if `vision_infos` is not `None`.
+
+        Args:
+            images (`PIL.Image.Image`, `np.ndarray`, `torch.Tensor`, `list[PIL.Image.Image]`, `list[np.ndarray]`, `list[torch.Tensor]`):
+                The image or batch of images to be prepared. Each image can be a PIL image, NumPy array or PyTorch
+                tensor. Both channels-first and channels-last formats are supported.
+            text (`str`, `list[str]`, `list[list[str]]`):
+                The sequence or batch of sequences to be encoded. Each sequence can be a string or a list of strings
+                (pretokenized string). If the sequences are provided as list of strings (pretokenized), you must set
+                `is_split_into_words=True` (to lift the ambiguity with a batch of sequences).
+            return_tensors (`str` or [`~utils.TensorType`], *optional*):
+                If set, will return tensors of a particular framework. Acceptable values are:
+                - `'tf'`: Return TensorFlow `tf.constant` objects.
+                - `'pt'`: Return PyTorch `torch.Tensor` objects.
+                - `'np'`: Return NumPy `np.ndarray` objects.
+                - `'jax'`: Return JAX `jnp.ndarray` objects.
+
+        Returns:
+            [`BatchFeature`]: A [`BatchFeature`] with the following fields:
+
+            - **input_ids** -- List of token ids to be fed to a model. Returned when `text` is not `None`.
+            - **attention_mask** -- List of indices specifying which tokens should be attended to by the model (when
+              `return_attention_mask=True` or if *"attention_mask"* is in `self.model_input_names` and if `text` is not
+              `None`).
+            - **pixel_values** -- Pixel values to be fed to a model. Returned when `images` is not `None`.
+            - **image_grid_thw** -- List of image 3D grid in LLM. Returned when `images` is not `None`.
+        """
+        output_kwargs = self._merge_kwargs(
+            Qwen2_5_VLProcessorKwargs,
+            tokenizer_init_kwargs=self.tokenizer.init_kwargs,
+            **kwargs,
+        )
+
+        image_inputs = {}
+        if images is not None:
+            image_inputs = self.image_processor(images=images, **output_kwargs["images_kwargs"])
+            image_grid_thw = image_inputs["image_grid_thw"]
+
+        if not isinstance(text, list):
+            text = [text]
+
+        text = text.copy()  # below lines change text in-place
+        if images is not None:
+            merge_length = self.image_processor.merge_size**2
+            index = 0
+            for i in range(len(text)):
+                while self.image_token in text[i]:
+                    num_image_tokens = image_grid_thw[index].prod() // merge_length
+                    text[i] = text[i].replace(self.image_token, "<|placeholder|>" * num_image_tokens, 1)
+                    index += 1
+                text[i] = text[i].replace("<|placeholder|>", self.image_token)
+
+        return_tensors = output_kwargs["text_kwargs"].pop("return_tensors", None)
+        return_mm_token_type_ids = output_kwargs["text_kwargs"].pop("return_mm_token_type_ids", None)
+        text_inputs = self.tokenizer(text, **output_kwargs["text_kwargs"])
+        self._check_special_mm_tokens(text, text_inputs, modalities=["image"])
+
+        if return_mm_token_type_ids:
+            array_ids = np.array(text_inputs["input_ids"])
+            mm_token_type_ids = np.zeros_like(text_inputs["input_ids"])
+            mm_token_type_ids[array_ids == self.image_token_id] = 1
+            text_inputs["mm_token_type_ids"] = mm_token_type_ids.tolist()
+
+        return BatchFeature(data={**text_inputs, **image_inputs}, tensor_type=return_tensors)
+
+    def _get_num_multimodal_tokens(self, image_sizes=None, **kwargs):
+        """
+        Computes the number of placeholder tokens needed for multimodal inputs with the given sizes.
+        Args:
+            image_sizes (`list[list[int]]`, *optional*):
+                The input sizes formatted as (height, width) per each image.
+        Returns:
+            `MultiModalData`: A `MultiModalData` object holding number of tokens per each of the provided
+            input modalities, along with other useful data.
+        """
+
+        vision_data = {}
+        if image_sizes is not None:
+            images_kwargs = Qwen2_5_VLProcessorKwargs._defaults.get("images_kwargs", {})
+            images_kwargs.update(kwargs)
+            merge_size = images_kwargs.get("merge_size", None) or self.image_processor.merge_size
+
+            num_image_patches = [
+                self.image_processor.get_number_of_image_patches(*image_size, images_kwargs)
+                for image_size in image_sizes
+            ]
+            num_image_tokens = [(num_patches // merge_size**2) for num_patches in num_image_patches]
+            vision_data.update({"num_image_tokens": num_image_tokens, "num_image_patches": num_image_patches})
+
+        return MultiModalData(**vision_data)
+
+    def batch_decode(self, *args, **kwargs):
+        """
+        This method forwards all its arguments to Qwen2TokenizerFast's [`~PreTrainedTokenizer.batch_decode`]. Please
+        refer to the docstring of this method for more information.
+        """
+        return self.tokenizer.batch_decode(*args, **kwargs)
+
+    def decode(self, *args, **kwargs):
+        """
+        This method forwards all its arguments to Qwen2TokenizerFast's [`~PreTrainedTokenizer.decode`]. Please refer to
+        the docstring of this method for more information.
+        """
+        return self.tokenizer.decode(*args, **kwargs)
+
+    def post_process_image_text_to_text(
+        self, generated_outputs, skip_special_tokens=True, clean_up_tokenization_spaces=False, **kwargs
+    ):
+        """
+        Post-process the output of the model to decode the text.
+
+        Args:
+            generated_outputs (`torch.Tensor` or `np.ndarray`):
+                The output of the model `generate` function. The output is expected to be a tensor of shape `(batch_size, sequence_length)`
+                or `(sequence_length,)`.
+            skip_special_tokens (`bool`, *optional*, defaults to `True`):
+                Whether or not to remove special tokens in the output. Argument passed to the tokenizer's `batch_decode` method.
+            clean_up_tokenization_spaces (`bool`, *optional*, defaults to `False`):
+                Whether or not to clean up the tokenization spaces. Argument passed to the tokenizer's `batch_decode` method.
+            **kwargs:
+                Additional arguments to be passed to the tokenizer's `batch_decode method`.
+
+        Returns:
+            `list[str]`: The decoded text.
+        """
+        return self.tokenizer.batch_decode(
+            generated_outputs,
+            skip_special_tokens=skip_special_tokens,
+            clean_up_tokenization_spaces=clean_up_tokenization_spaces,
+            **kwargs,
+        )
+
+    @property
+    def model_input_names(self):
+        tokenizer_input_names = self.tokenizer.model_input_names
+        image_processor_input_names = self.image_processor.model_input_names
+        names_from_processor = list(dict.fromkeys(tokenizer_input_names + image_processor_input_names))
+        return names_from_processor
+
+
+__all__ = ["Qwen2_5_VLProcessor"]