Delete files m*.py with huggingface_hub

modeling_minicpmv.py

@@ -1,447 +0,0 @@
-import math
-from typing import List, Optional
-import json
-import torch
-import torchvision
-
-from threading import Thread
-from copy import deepcopy
-from PIL import Image
-from transformers import AutoProcessor, TextIteratorStreamer
-
-from .configuration_minicpm import MiniCPMVConfig
-from transformers import LlamaForCausalLM, LlamaPreTrainedModel
-from .modeling_navit_siglip import SiglipVisionTransformer
-from .resampler import Resampler
-
-
-
-class MiniCPMVPreTrainedModel(LlamaPreTrainedModel):
-    config_class = MiniCPMVConfig
-
-
-class MiniCPMV(MiniCPMVPreTrainedModel):
-    def __init__(self, config):
-        super().__init__(config)
-        self.llm = LlamaForCausalLM(config)
-        self.vpm = self.init_vision_module()
-        self.vision_dim = self.vpm.embed_dim
-        self.embed_dim = self.llm.config.hidden_size
-        self.resampler = self.init_resampler(self.embed_dim, self.vision_dim)
-        self.processor = None
-
-        self.terminators = ['<|im_end|>', '</s>']
-
-    def init_vision_module(self):
-        # same as HuggingFaceM4/siglip-so400m-14-980-flash-attn2-navit add tgt_sizes
-        if self.config._attn_implementation == 'flash_attention_2':
-            self.config.vision_config._attn_implementation = 'flash_attention_2'
-        else:
-            # not suport sdpa
-            self.config.vision_config._attn_implementation = 'eager'
-        model = SiglipVisionTransformer(self.config.vision_config)
-        if self.config.drop_vision_last_layer:
-            model.encoder.layers = model.encoder.layers[:-1]
-
-        setattr(model, 'embed_dim', model.embeddings.embed_dim)
-        setattr(model, 'patch_size', model.embeddings.patch_size)
-
-        return model
-
-    def init_resampler(self, embed_dim, vision_dim):
-        return Resampler(
-            num_queries=self.config.query_num,
-            embed_dim=embed_dim,
-            num_heads=embed_dim // 128,
-            kv_dim=vision_dim,
-            adaptive=True
-        )
-
-    def get_input_embeddings(self):
-        return self.llm.get_input_embeddings()
-
-    def set_input_embeddings(self, value):
-        self.llm.embed_tokens = value
-
-    def get_output_embeddings(self):
-        return self.llm.lm_head
-
-    def set_output_embeddings(self, new_embeddings):
-        self.llm.lm_head = new_embeddings
-
-    def set_decoder(self, decoder):
-        self.llm = decoder
-
-    def get_decoder(self):
-        return self.llm
-
-    def get_vllm_embedding(self, data):
-        if 'vision_hidden_states' not in data:
-            dtype = self.llm.model.embed_tokens.weight.dtype
-            device = self.llm.model.embed_tokens.weight.device
-            tgt_sizes = data['tgt_sizes']
-            pixel_values_list = data['pixel_values']
-            vision_hidden_states = []
-            all_pixel_values = []
-            img_cnt = []
-            for pixel_values in pixel_values_list:
-                img_cnt.append(len(pixel_values))
-                all_pixel_values.extend([i.flatten(end_dim=1).permute(1, 0) for i in pixel_values])
-
-            # exist image
-            if all_pixel_values:
-                tgt_sizes = [tgt_size for tgt_size in tgt_sizes if isinstance(tgt_size, torch.Tensor)]
-                tgt_sizes = torch.vstack(tgt_sizes).type(torch.int32)
-
-                max_patches = torch.max(tgt_sizes[:, 0] * tgt_sizes[:, 1])
-
-                all_pixel_values = torch.nn.utils.rnn.pad_sequence(all_pixel_values, batch_first=True,
-                                                                   padding_value=0.0)
-                B, L, _ = all_pixel_values.shape
-                all_pixel_values = all_pixel_values.permute(0, 2, 1).reshape(B, 3, -1, L)
-
-                patch_attn_mask = torch.zeros((B, 1, max_patches), dtype=torch.bool, device=device)
-                for i in range(B):
-                    patch_attn_mask[i, 0, :tgt_sizes[i][0] * tgt_sizes[i][1]] = True
-
-                vision_batch_size = self.config.vision_batch_size
-                all_pixel_values = all_pixel_values.type(dtype).to(device=device)
-                if B > vision_batch_size:
-                    hs = []
-                    for i in range(0, B, vision_batch_size):
-                        start_idx = i
-                        end_idx = i + vision_batch_size
-                        tmp_hs = self.vpm(all_pixel_values[start_idx:end_idx], patch_attention_mask=patch_attn_mask[start_idx:end_idx], tgt_sizes=tgt_sizes[start_idx:end_idx]).last_hidden_state
-                        hs.append(tmp_hs)
-                    vision_embedding = torch.cat(hs, dim=0)
-                else:
-                    vision_embedding = self.vpm(all_pixel_values, patch_attention_mask=patch_attn_mask, tgt_sizes=tgt_sizes).last_hidden_state
-                vision_embedding = self.resampler(vision_embedding, tgt_sizes)
-
-                start = 0
-                for pixel_values in pixel_values_list:
-                    img_cnt = len(pixel_values)
-                    if img_cnt > 0:
-                        vision_hidden_states.append(vision_embedding[start: start + img_cnt])
-                        start += img_cnt
-                    else:
-                        vision_hidden_states.append([])
-            else: # no image
-                if self.training:
-                    dummy_image = torch.zeros(
-                        (1, 3, 224, 224),
-                        device=device, dtype=dtype
-                    )
-                    tgt_sizes = torch.Tensor([[(224 // self.config.patch_size), math.ceil(224 / self.config.patch_size)]]).type(torch.int32)
-                    dummy_feature = self.resampler(self.vpm(dummy_image).last_hidden_state, tgt_sizes)
-                else:
-                    dummy_feature = []
-                for _ in range(len(pixel_values_list)):
-                    vision_hidden_states.append(dummy_feature)
-
-        else:
-            vision_hidden_states = data['vision_hidden_states']
-
-        if hasattr(self.llm.config, 'scale_emb'):
-            vllm_embedding = self.llm.model.embed_tokens(data['input_ids']) * self.llm.config.scale_emb
-        else:
-            vllm_embedding = self.llm.model.embed_tokens(data['input_ids'])
-
-        vision_hidden_states = [i.type(vllm_embedding.dtype) if isinstance(
-            i, torch.Tensor) else i for i in vision_hidden_states]
-
-        bs = len(data['input_ids'])
-        device = vllm_embedding.device
-        embed_dim = vllm_embedding.shape[-1]
-
-        new_vllm_embeddings = []
-        
-        for i in range(bs):
-            cur_vs_hs = vision_hidden_states[i]
-            cur_vllm_emb = vllm_embedding[i]
-
-            if len(cur_vs_hs) == 0:
-                new_vllm_embeddings.append(cur_vllm_emb)
-                continue
-                
-            cur_image_bound = data['image_bound'][i]
-
-            if len(cur_image_bound) > 0:
-                image_indices = torch.stack([
-                    torch.arange(r[0], r[1], dtype=torch.long) 
-                    for r in cur_image_bound
-                ], dim=0).flatten().to(device)
-
-                indices_expanded = image_indices.view(-1, 1).expand(-1, embed_dim)
-                vision_features = cur_vs_hs.view(-1, embed_dim)
-                
-                updated_emb = cur_vllm_emb.scatter(0, indices_expanded, vision_features)
-                new_vllm_embeddings.append(updated_emb)
-            elif self.training:
-                dummy_term = cur_vs_hs[0].sum() * 0 
-                new_vllm_embeddings.append(cur_vllm_emb + dummy_term)
-            else:
-                new_vllm_embeddings.append(cur_vllm_emb)
-
-        vllm_embedding = torch.stack(new_vllm_embeddings, dim=0)
-
-        return vllm_embedding, vision_hidden_states
-
-    def forward(self, data=None, **kwargs):
-        if isinstance(data, torch.Tensor):
-            attention_mask = torch.ones_like(data, dtype=torch.bool)
-            kwargs = {'attention_mask': attention_mask}
-            return self.llm(
-                input_ids=data,
-                **kwargs
-            )
-
-        if data is None:
-            data = {
-                "input_ids": kwargs.pop("input_ids", None),
-                "pixel_values": kwargs.pop("pixel_values", None),
-                "image_bound": kwargs.pop("image_bound", None),
-                "tgt_sizes": kwargs.pop("tgt_sizes", None),
-                "position_ids": kwargs.pop("position_ids", None),
-            }
-        else:
-            kwargs.pop("input_ids", None)
-            kwargs.pop("pixel_values", None)
-            kwargs.pop("image_bound", None)
-            kwargs.pop("tgt_sizes", None)
-            kwargs.pop("position_ids", None)
-        kwargs.pop("inputs_embeds", None)
-        
-        vllm_embedding, vision_hidden_states = self.get_vllm_embedding(data)
-        position_ids = data["position_ids"]
-        if position_ids.dtype != torch.int64:
-            position_ids = position_ids.long()
-
-        return self.llm(
-            input_ids=None,
-            position_ids=position_ids,
-            inputs_embeds=vllm_embedding,
-            **kwargs
-        )
-    
-    def _decode(self, inputs_embeds, tokenizer, attention_mask, decode_text=False, **kwargs):
-        terminators = [tokenizer.convert_tokens_to_ids(i) for i in self.terminators]
-        output = self.llm.generate(
-            inputs_embeds=inputs_embeds,
-            pad_token_id=0,
-            eos_token_id=terminators,
-            attention_mask=attention_mask,
-            **kwargs
-        )
-        if decode_text:
-            return self._decode_text(output, tokenizer)
-        return output
-
-    def _decode_stream(self, inputs_embeds, tokenizer, **kwargs):
-        terminators = [tokenizer.convert_tokens_to_ids(i) for i in self.terminators]
-        streamer = TextIteratorStreamer(tokenizer=tokenizer)
-        generation_kwargs = {
-            'inputs_embeds': inputs_embeds,
-            'pad_token_id': 0,
-            'eos_token_id': terminators,
-            'streamer': streamer
-        }
-        generation_kwargs.update(kwargs)
-
-        thread = Thread(target=self.llm.generate, kwargs=generation_kwargs)
-        thread.start()
-    
-        return streamer
-
-    def _decode_text(self, result_ids, tokenizer):
-        terminators = [tokenizer.convert_tokens_to_ids(i) for i in self.terminators]
-        result_text = []
-        for result in result_ids:
-            result = result[result != 0]
-            if result[0] == tokenizer.bos_id:
-                result = result[1:]
-            if result[-1] in terminators:
-                result = result[:-1]
-            result_text.append(tokenizer.decode(result).strip())
-        return result_text
-
-    def generate(
-        self,
-        input_ids=None,
-        pixel_values=None,
-        tgt_sizes=None,
-        image_bound=None,
-        attention_mask=None,
-        tokenizer=None,
-        vision_hidden_states=None,
-        return_vision_hidden_states=False,
-        stream=False,
-        decode_text=False,
-        **kwargs
-    ):
-        assert input_ids is not None
-        assert len(input_ids) == len(pixel_values)
-
-        model_inputs = {
-            "input_ids": input_ids,
-            "image_bound": image_bound,
-        }
-
-        if vision_hidden_states is None:
-            model_inputs["pixel_values"] = pixel_values
-            model_inputs['tgt_sizes'] = tgt_sizes
-        else:
-            model_inputs["vision_hidden_states"] = vision_hidden_states
-
-        with torch.inference_mode():
-            (
-                model_inputs["inputs_embeds"],
-                vision_hidden_states,
-            ) = self.get_vllm_embedding(model_inputs)
-
-            if stream:
-                result = self._decode_stream(model_inputs["inputs_embeds"], tokenizer, **kwargs)
-            else:
-                result = self._decode(model_inputs["inputs_embeds"], tokenizer, attention_mask, decode_text=decode_text, **kwargs)
-
-        if return_vision_hidden_states:
-            return result, vision_hidden_states
-        
-        return result
-
-    def chat(
-        self,
-        image=None,
-        msgs=None,
-        tokenizer=None,
-        processor=None,
-        vision_hidden_states=None,
-        max_new_tokens=2048,
-        min_new_tokens=0,
-        sampling=True,
-        max_inp_length=32768,
-        system_prompt='',
-        stream=False,
-        max_slice_nums=None,
-        use_image_id=None,
-        **kwargs
-    ):
-        if isinstance(msgs[0], list):
-            batched = True
-        else:
-            batched = False
-        msgs_list = msgs
-        images_list = image
-        
-        if batched is False:
-            images_list, msgs_list = [images_list], [msgs_list]
-        else:
-            assert images_list is None, "Please integrate image to msgs when using batch inference."
-            images_list = [None] * len(msgs_list)
-        assert len(images_list) == len(msgs_list), "The batch dim of images_list and msgs_list should be the same."
-
-        if processor is None:
-            if self.processor is None:
-                self.processor = AutoProcessor.from_pretrained(self.config._name_or_path, trust_remote_code=True)
-            processor = self.processor
-        
-        assert self.config.query_num == processor.image_processor.image_feature_size, "These two values should be the same. Check `config.json` and `preprocessor_config.json`."
-        assert self.config.patch_size == processor.image_processor.patch_size, "These two values should be the same. Check `config.json` and `preprocessor_config.json`."
-        assert self.config.use_image_id == processor.image_processor.use_image_id, "These two values should be the same. Check `config.json` and `preprocessor_config.json`."
-        assert self.config.slice_config.max_slice_nums == processor.image_processor.max_slice_nums, "These two values should be the same. Check `config.json` and `preprocessor_config.json`."
-        assert self.config.slice_mode == processor.image_processor.slice_mode, "These two values should be the same. Check `config.json` and `preprocessor_config.json`."
-
-        prompts_lists = []
-        input_images_lists = []
-        for image, msgs in zip(images_list, msgs_list):
-            if isinstance(msgs, str):
-                msgs = json.loads(msgs)
-            copy_msgs = deepcopy(msgs)
-
-            assert len(msgs) > 0, "msgs is empty"
-            assert sampling or not stream, "if use stream mode, make sure sampling=True"
-
-            if image is not None and isinstance(copy_msgs[0]["content"], str):
-                copy_msgs[0]["content"] = [image, copy_msgs[0]["content"]]
-
-            images = []
-            for i, msg in enumerate(copy_msgs):
-                role = msg["role"]
-                content = msg["content"]
-                assert role in ["user", "assistant"]
-                if i == 0:
-                    assert role == "user", "The role of first msg should be user"
-                if isinstance(content, str):
-                    content = [content]
-                cur_msgs = []
-                for c in content:
-                    if isinstance(c, Image.Image):
-                        images.append(c)
-                        cur_msgs.append("(<image>./</image>)")
-                    elif isinstance(c, str):
-                        cur_msgs.append(c)
-                msg["content"] = "\n".join(cur_msgs)
-
-            if system_prompt:
-                sys_msg = {'role': 'system', 'content': system_prompt}
-                copy_msgs = [sys_msg] + copy_msgs        
-
-            prompts_lists.append(processor.tokenizer.apply_chat_template(copy_msgs, tokenize=False, add_generation_prompt=True))
-            input_images_lists.append(images)
-
-        inputs = processor(
-            prompts_lists, 
-            input_images_lists, 
-            max_slice_nums=max_slice_nums,
-            use_image_id=use_image_id,
-            return_tensors="pt", 
-            max_length=max_inp_length
-        ).to(self.device)
-
-        if sampling:
-            generation_config = {
-                "top_p": 0.8,
-                "top_k": 100,
-                "temperature": 0.7,
-                "do_sample": True,
-                "repetition_penalty": 1.05
-            }
-        else:
-            generation_config = {
-                "num_beams": 3,
-                "repetition_penalty": 1.2,
-            }
-            
-        if min_new_tokens > 0:
-            generation_config['min_new_tokens'] = min_new_tokens
-
-        generation_config.update(
-            (k, kwargs[k]) for k in generation_config.keys() & kwargs.keys()
-        )
-
-        inputs.pop("image_sizes")
-        with torch.inference_mode():
-            res = self.generate(
-                **inputs,
-                tokenizer=tokenizer,
-                max_new_tokens=max_new_tokens,
-                vision_hidden_states=vision_hidden_states,
-                stream=stream,
-                decode_text=True,
-                **generation_config
-            )
-        
-        if stream:
-            def stream_gen():
-                for text in res:
-                    for term in self.terminators:
-                        text = text.replace(term, '')
-                    yield text
-            return stream_gen()
-
-        else:
-            if batched:
-                answer = res
-            else:
-                answer = res[0]
-            return answer

modeling_navit_siglip.py

@@ -1,937 +0,0 @@
-# coding=utf-8
-# Copyright 2024 Google AI and The HuggingFace Team. All rights reserved.
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-#     http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-""" PyTorch Siglip model. """
-# Copied from  HuggingFaceM4/siglip-so400m-14-980-flash-attn2-navit and add tgt_sizes
-
-
-import os
-import math
-import warnings
-from dataclasses import dataclass
-from typing import Any, Optional, Tuple, Union
-
-import numpy as np
-import torch
-import torch.nn.functional as F
-import torch.utils.checkpoint
-from torch import nn
-from torch.nn.init import _calculate_fan_in_and_fan_out
-
-from transformers.activations import ACT2FN
-from transformers.modeling_attn_mask_utils import _prepare_4d_attention_mask
-from transformers.modeling_outputs import BaseModelOutput, BaseModelOutputWithPooling
-from transformers.modeling_utils import PreTrainedModel
-from transformers.configuration_utils import PretrainedConfig
-from transformers.utils import (
-    ModelOutput,
-    add_start_docstrings,
-    add_start_docstrings_to_model_forward,
-    is_flash_attn_2_available,
-    logging,
-    replace_return_docstrings,
-)
-from transformers.utils import logging
-
-logger = logging.get_logger(__name__)
-
-class SiglipVisionConfig(PretrainedConfig):
-    r"""
-    This is the configuration class to store the configuration of a [`SiglipVisionModel`]. It is used to instantiate a
-    Siglip vision encoder according to the specified arguments, defining the model architecture. Instantiating a
-    configuration with the defaults will yield a similar configuration to that of the vision encoder of the Siglip
-    [google/siglip-base-patch16-224](https://huggingface.co/google/siglip-base-patch16-224) architecture.
-    Configuration objects inherit from [`PretrainedConfig`] and can be used to control the model outputs. Read the
-    documentation from [`PretrainedConfig`] for more information.
-    Args:
-        hidden_size (`int`, *optional*, defaults to 768):
-            Dimensionality of the encoder layers and the pooler layer.
-        intermediate_size (`int`, *optional*, defaults to 3072):
-            Dimensionality of the "intermediate" (i.e., feed-forward) layer in the Transformer encoder.
-        num_hidden_layers (`int`, *optional*, defaults to 12):
-            Number of hidden layers in the Transformer encoder.
-        num_attention_heads (`int`, *optional*, defaults to 12):
-            Number of attention heads for each attention layer in the Transformer encoder.
-        num_channels (`int`, *optional*, defaults to 3):
-            Number of channels in the input images.
-        image_size (`int`, *optional*, defaults to 224):
-            The size (resolution) of each image.
-        patch_size (`int`, *optional*, defaults to 16):
-            The size (resolution) of each patch.
-        hidden_act (`str` or `function`, *optional*, defaults to `"gelu_pytorch_tanh"`):
-            The non-linear activation function (function or string) in the encoder and pooler. If string, `"gelu"`,
-            `"relu"`, `"selu"` and `"gelu_new"` ``"quick_gelu"` are supported.
-        layer_norm_eps (`float`, *optional*, defaults to 1e-06):
-            The epsilon used by the layer normalization layers.
-        attention_dropout (`float`, *optional*, defaults to 0.0):
-            The dropout ratio for the attention probabilities.
-    Example:
-    ```python
-    >>> from transformers import SiglipVisionConfig, SiglipVisionModel
-    >>> # Initializing a SiglipVisionConfig with google/siglip-base-patch16-224 style configuration
-    >>> configuration = SiglipVisionConfig()
-    >>> # Initializing a SiglipVisionModel (with random weights) from the google/siglip-base-patch16-224 style configuration
-    >>> model = SiglipVisionModel(configuration)
-    >>> # Accessing the model configuration
-    >>> configuration = model.config
-    ```"""
-
-    model_type = "siglip_vision_model"
-
-    def __init__(
-        self,
-        hidden_size=768,
-        intermediate_size=3072,
-        num_hidden_layers=12,
-        num_attention_heads=12,
-        num_channels=3,
-        image_size=224,
-        patch_size=16,
-        hidden_act="gelu_pytorch_tanh",
-        layer_norm_eps=1e-6,
-        attention_dropout=0.0,
-        **kwargs,
-    ):
-        super().__init__(**kwargs)
-
-        self.hidden_size = hidden_size
-        self.intermediate_size = intermediate_size
-        self.num_hidden_layers = num_hidden_layers
-        self.num_attention_heads = num_attention_heads
-        self.num_channels = num_channels
-        self.patch_size = patch_size
-        self.image_size = image_size
-        self.attention_dropout = attention_dropout
-        self.layer_norm_eps = layer_norm_eps
-        self.hidden_act = hidden_act
-
-    @classmethod
-    def from_pretrained(cls, pretrained_model_name_or_path: Union[str, os.PathLike], **kwargs) -> "PretrainedConfig":
-        cls._set_token_in_kwargs(kwargs)
-
-        config_dict, kwargs = cls.get_config_dict(pretrained_model_name_or_path, **kwargs)
-
-        # get the vision config dict if we are loading from SiglipConfig
-        if config_dict.get("model_type") == "siglip":
-            config_dict = config_dict["vision_config"]
-
-        if "model_type" in config_dict and hasattr(cls, "model_type") and config_dict["model_type"] != cls.model_type:
-            logger.warning(
-                f"You are using a model of type {config_dict['model_type']} to instantiate a model of type "
-                f"{cls.model_type}. This is not supported for all configurations of models and can yield errors."
-            )
-
-        return cls.from_dict(config_dict, **kwargs)
-        
-
-_CHECKPOINT_FOR_DOC = "google/siglip-base-patch16-224"
-
-SIGLIP_PRETRAINED_MODEL_ARCHIVE_LIST = [
-    "google/siglip-base-patch16-224",
-    # See all SigLIP models at https://huggingface.co/models?filter=siglip
-]
-
-if is_flash_attn_2_available():
-    from flash_attn import flash_attn_func, flash_attn_varlen_func
-    from flash_attn.bert_padding import index_first_axis, pad_input, unpad_input  # noqa
-
-
-# Copied from transformers.models.llama.modeling_llama._get_unpad_data
-def _get_unpad_data(attention_mask):
-    seqlens_in_batch = attention_mask.sum(dim=-1, dtype=torch.int32)
-    indices = torch.nonzero(attention_mask.flatten(), as_tuple=False).flatten()
-    max_seqlen_in_batch = seqlens_in_batch.max().item()
-    cu_seqlens = F.pad(torch.cumsum(seqlens_in_batch, dim=0, dtype=torch.torch.int32), (1, 0))
-    return (
-        indices,
-        cu_seqlens,
-        max_seqlen_in_batch,
-    )
-
-
-def _trunc_normal_(tensor, mean, std, a, b):
-    # Cut & paste from PyTorch official master until it's in a few official releases - RW
-    # Method based on https://people.sc.fsu.edu/~jburkardt/presentations/truncated_normal.pdf
-    def norm_cdf(x):
-        # Computes standard normal cumulative distribution function
-        return (1.0 + math.erf(x / math.sqrt(2.0))) / 2.0
-
-    if (mean < a - 2 * std) or (mean > b + 2 * std):
-        warnings.warn(
-            "mean is more than 2 std from [a, b] in nn.init.trunc_normal_. "
-            "The distribution of values may be incorrect.",
-            stacklevel=2,
-        )
-
-    # Values are generated by using a truncated uniform distribution and
-    # then using the inverse CDF for the normal distribution.
-    # Get upper and lower cdf values
-    l = norm_cdf((a - mean) / std)
-    u = norm_cdf((b - mean) / std)
-
-    # Uniformly fill tensor with values from [l, u], then translate to
-    # [2l-1, 2u-1].
-    tensor.uniform_(2 * l - 1, 2 * u - 1)
-
-    # Use inverse cdf transform for normal distribution to get truncated
-    # standard normal
-    if tensor.dtype in [torch.float16, torch.bfloat16]:
-        # The `erfinv_` op is not (yet?) defined in float16+cpu, bfloat16+gpu
-        og_dtype = tensor.dtype
-        tensor = tensor.to(torch.float32)
-        tensor.erfinv_()
-        tensor = tensor.to(og_dtype)
-    else:
-        tensor.erfinv_()
-
-    # Transform to proper mean, std
-    tensor.mul_(std * math.sqrt(2.0))
-    tensor.add_(mean)
-
-    # Clamp to ensure it's in the proper range
-    if tensor.dtype == torch.float16:
-        # The `clamp_` op is not (yet?) defined in float16+cpu
-        tensor = tensor.to(torch.float32)
-        tensor.clamp_(min=a, max=b)
-        tensor = tensor.to(torch.float16)
-    else:
-        tensor.clamp_(min=a, max=b)
-
-
-def trunc_normal_tf_(
-    tensor: torch.Tensor, mean: float = 0.0, std: float = 1.0, a: float = -2.0, b: float = 2.0
-) -> torch.Tensor:
-    """Fills the input Tensor with values drawn from a truncated
-    normal distribution. The values are effectively drawn from the
-    normal distribution :math:`\\mathcal{N}(\text{mean}, \text{std}^2)`
-    with values outside :math:`[a, b]` redrawn until they are within
-    the bounds. The method used for generating the random values works
-    best when :math:`a \\leq \text{mean} \\leq b`.
-    NOTE: this 'tf' variant behaves closer to Tensorflow / JAX impl where the
-    bounds [a, b] are applied when sampling the normal distribution with mean=0, std=1.0
-    and the result is subsquently scaled and shifted by the mean and std args.
-    Args:
-        tensor: an n-dimensional `torch.Tensor`
-        mean: the mean of the normal distribution
-        std: the standard deviation of the normal distribution
-        a: the minimum cutoff value
-        b: the maximum cutoff value
-    """
-    with torch.no_grad():
-        _trunc_normal_(tensor, 0, 1.0, a, b)
-        tensor.mul_(std).add_(mean)
-
-
-def variance_scaling_(tensor, scale=1.0, mode="fan_in", distribution="normal"):
-    fan_in, fan_out = _calculate_fan_in_and_fan_out(tensor)
-    if mode == "fan_in":
-        denom = fan_in
-    elif mode == "fan_out":
-        denom = fan_out
-    elif mode == "fan_avg":
-        denom = (fan_in + fan_out) / 2
-
-    variance = scale / denom
-
-    if distribution == "truncated_normal":
-        # constant is stddev of standard normal truncated to (-2, 2)
-        trunc_normal_tf_(tensor, std=math.sqrt(variance) / 0.87962566103423978)
-    elif distribution == "normal":
-        with torch.no_grad():
-            tensor.normal_(std=math.sqrt(variance))
-    elif distribution == "uniform":
-        bound = math.sqrt(3 * variance)
-        with torch.no_grad():
-            tensor.uniform_(-bound, bound)
-    else:
-        raise ValueError(f"invalid distribution {distribution}")
-
-
-def lecun_normal_(tensor):
-    variance_scaling_(tensor, mode="fan_in", distribution="truncated_normal")
-
-
-def default_flax_embed_init(tensor):
-    variance_scaling_(tensor, mode="fan_in", distribution="normal")
-
-
-@dataclass
-# Copied from transformers.models.clip.modeling_clip.CLIPVisionModelOutput with CLIP->Siglip
-class SiglipVisionModelOutput(ModelOutput):
-    """
-    Base class for vision model's outputs that also contains image embeddings of the pooling of the last hidden states.
-    Args:
-        image_embeds (`torch.FloatTensor` of shape `(batch_size, output_dim)` *optional* returned when model is initialized with `with_projection=True`):
-            The image embeddings obtained by applying the projection layer to the pooler_output.
-        last_hidden_state (`torch.FloatTensor` of shape `(batch_size, sequence_length, hidden_size)`):
-            Sequence of hidden-states at the output of the last layer of the model.
-        hidden_states (`tuple(torch.FloatTensor)`, *optional*, returned when `output_hidden_states=True` is passed or when `config.output_hidden_states=True`):
-            Tuple of `torch.FloatTensor` (one for the output of the embeddings, if the model has an embedding layer, +
-            one for the output of each layer) of shape `(batch_size, sequence_length, hidden_size)`.
-            Hidden-states of the model at the output of each layer plus the optional initial embedding outputs.
-        attentions (`tuple(torch.FloatTensor)`, *optional*, returned when `output_attentions=True` is passed or when `config.output_attentions=True`):
-            Tuple of `torch.FloatTensor` (one for each layer) of shape `(batch_size, num_heads, sequence_length,
-            sequence_length)`.
-            Attentions weights after the attention softmax, used to compute the weighted average in the self-attention
-            heads.
-    """
-
-    image_embeds: Optional[torch.FloatTensor] = None
-    last_hidden_state: torch.FloatTensor = None
-    hidden_states: Optional[Tuple[torch.FloatTensor]] = None
-    attentions: Optional[Tuple[torch.FloatTensor]] = None
-
-
-class SiglipVisionEmbeddings(nn.Module):
-    def __init__(self, config: SiglipVisionConfig):
-        super().__init__()
-        self.config = config
-        self.embed_dim = config.hidden_size
-        self.image_size = config.image_size
-        self.patch_size = config.patch_size
-
-        self.patch_embedding = nn.Conv2d(
-            in_channels=config.num_channels,
-            out_channels=self.embed_dim,
-            kernel_size=self.patch_size,
-            stride=self.patch_size,
-            padding="valid",
-        )
-
-        self.num_patches_per_side = self.image_size // self.patch_size
-        self.num_patches = self.num_patches_per_side**2
-        self.num_positions = self.num_patches
-        self.position_embedding = nn.Embedding(self.num_positions, self.embed_dim)
-
-    def forward(self, pixel_values: torch.FloatTensor, patch_attention_mask: torch.BoolTensor, tgt_sizes: Optional[torch.IntTensor]=None) -> torch.Tensor:
-        batch_size = pixel_values.size(0)
-
-        patch_embeds = self.patch_embedding(pixel_values)
-        embeddings = patch_embeds.flatten(2).transpose(1, 2)
-
-        max_im_h, max_im_w = pixel_values.size(2), pixel_values.size(3)
-        max_nb_patches_h, max_nb_patches_w = max_im_h // self.patch_size, max_im_w // self.patch_size
-        boundaries = torch.arange(1 / self.num_patches_per_side, 1.0, 1 / self.num_patches_per_side)
-        position_ids = torch.full(
-            size=(
-                batch_size,
-                max_nb_patches_h * max_nb_patches_w,
-            ),
-            fill_value=0,
-        )
-
-        for batch_idx, p_attn_mask in enumerate(patch_attention_mask):
-            if tgt_sizes is not None:
-                nb_patches_h = tgt_sizes[batch_idx][0]
-                nb_patches_w = tgt_sizes[batch_idx][1]
-            else:
-                nb_patches_h = p_attn_mask[:, 0].sum()
-                nb_patches_w = p_attn_mask[0].sum()
-
-            fractional_coords_h = torch.arange(0, 1 - 1e-6, 1 / nb_patches_h)
-            fractional_coords_w = torch.arange(0, 1 - 1e-6, 1 / nb_patches_w)
-
-            bucket_coords_h = torch.bucketize(fractional_coords_h, boundaries, right=True)
-            bucket_coords_w = torch.bucketize(fractional_coords_w, boundaries, right=True)
-
-            pos_ids = (bucket_coords_h[:, None] * self.num_patches_per_side + bucket_coords_w).flatten()
-            position_ids[batch_idx][p_attn_mask.view(-1).cpu()] = pos_ids
-
-        position_ids = position_ids.to(self.position_embedding.weight.device)
-
-        embeddings = embeddings + self.position_embedding(position_ids)
-        return embeddings
-
-
-class SiglipAttention(nn.Module):
-    """Multi-headed attention from 'Attention Is All You Need' paper"""
-
-    # Copied from transformers.models.clip.modeling_clip.CLIPAttention.__init__
-    def __init__(self, config):
-        super().__init__()
-        self.config = config
-        self.embed_dim = config.hidden_size
-        self.num_heads = config.num_attention_heads
-        self.head_dim = self.embed_dim // self.num_heads
-        if self.head_dim * self.num_heads != self.embed_dim:
-            raise ValueError(
-                f"embed_dim must be divisible by num_heads (got `embed_dim`: {self.embed_dim} and `num_heads`:"
-                f" {self.num_heads})."
-            )
-        self.scale = self.head_dim**-0.5
-        self.dropout = config.attention_dropout
-
-        self.k_proj = nn.Linear(self.embed_dim, self.embed_dim)
-        self.v_proj = nn.Linear(self.embed_dim, self.embed_dim)
-        self.q_proj = nn.Linear(self.embed_dim, self.embed_dim)
-        self.out_proj = nn.Linear(self.embed_dim, self.embed_dim)
-
-    def forward(
-        self,
-        hidden_states: torch.Tensor,
-        attention_mask: Optional[torch.Tensor] = None,
-        output_attentions: Optional[bool] = False,
-    ) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor]]]:
-        """Input shape: Batch x Time x Channel"""
-
-        batch_size, q_len, _ = hidden_states.size()
-
-        query_states = self.q_proj(hidden_states)
-        key_states = self.k_proj(hidden_states)
-        value_states = self.v_proj(hidden_states)
-
-        query_states = query_states.view(batch_size, q_len, self.num_heads, self.head_dim).transpose(1, 2)
-        key_states = key_states.view(batch_size, q_len, self.num_heads, self.head_dim).transpose(1, 2)
-        value_states = value_states.view(batch_size, q_len, self.num_heads, self.head_dim).transpose(1, 2)
-
-        k_v_seq_len = key_states.shape[-2]
-        attn_weights = torch.matmul(query_states, key_states.transpose(2, 3)) * self.scale
-
-        if attn_weights.size() != (batch_size, self.num_heads, q_len, k_v_seq_len):
-            raise ValueError(
-                f"Attention weights should be of size {(batch_size, self.num_heads, q_len, k_v_seq_len)}, but is"
-                f" {attn_weights.size()}"
-            )
-
-        if attention_mask is not None:
-            if attention_mask.size() != (batch_size, 1, q_len, k_v_seq_len):
-                raise ValueError(
-                    f"Attention mask should be of size {(batch_size, 1, q_len, k_v_seq_len)}, but is {attention_mask.size()}"
-                )
-            attn_weights = attn_weights + attention_mask
-
-        # upcast attention to fp32
-        attn_weights = nn.functional.softmax(attn_weights, dim=-1, dtype=torch.float32).to(query_states.dtype)
-        attn_weights = nn.functional.dropout(attn_weights, p=self.dropout, training=self.training)
-        attn_output = torch.matmul(attn_weights, value_states)
-
-        if attn_output.size() != (batch_size, self.num_heads, q_len, self.head_dim):
-            raise ValueError(
-                f"`attn_output` should be of size {(batch_size, self.num_heads, q_len, self.head_dim)}, but is"
-                f" {attn_output.size()}"
-            )
-
-        attn_output = attn_output.transpose(1, 2).contiguous()
-        attn_output = attn_output.reshape(batch_size, q_len, self.embed_dim)
-
-        attn_output = self.out_proj(attn_output)
-
-        return attn_output, attn_weights
-
-
-class SiglipFlashAttention2(SiglipAttention):
-    """
-    Llama flash attention module. This module inherits from `LlamaAttention` as the weights of the module stays
-    untouched. The only required change would be on the forward pass where it needs to correctly call the public API of
-    flash attention and deal with padding tokens in case the input contains any of them.
-    """
-
-    def __init__(self, *args, **kwargs):
-        super().__init__(*args, **kwargs)
-        self.is_causal = False  # Hack to make sure we don't use a causal mask
-
-    def forward(
-        self,
-        hidden_states: torch.Tensor,
-        attention_mask: Optional[torch.LongTensor] = None,
-        position_ids: Optional[torch.LongTensor] = None,
-        past_key_value: Optional[Tuple[torch.Tensor]] = None,
-        output_attentions: bool = False,
-        use_cache: bool = False,
-        **kwargs,
-    ) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor]]]:
-        output_attentions = False
-
-        bsz, q_len, _ = hidden_states.size()
-
-        query_states = self.q_proj(hidden_states)
-        key_states = self.k_proj(hidden_states)
-        value_states = self.v_proj(hidden_states)
-
-        # Flash attention requires the input to have the shape
-        # batch_size x seq_length x head_dim x hidden_dim
-        # therefore we just need to keep the original shape
-        query_states = query_states.view(bsz, q_len, self.num_heads, self.head_dim).transpose(1, 2)
-        key_states = key_states.view(bsz, q_len, self.num_heads, self.head_dim).transpose(1, 2)
-        value_states = value_states.view(bsz, q_len, self.num_heads, self.head_dim).transpose(1, 2)
-
-        kv_seq_len = key_states.shape[-2]
-        if past_key_value is not None:
-            kv_seq_len += past_key_value.get_usable_length(kv_seq_len, self.layer_idx)
-        # cos, sin = self.rotary_emb(value_states, seq_len=kv_seq_len)
-        # query_states, key_states = apply_rotary_pos_emb(query_states, key_states, cos, sin, position_ids)
-
-        # if past_key_value is not None:
-        #     cache_kwargs = {"sin": sin, "cos": cos}  # Specific to RoPE models
-        #     key_states, value_states = past_key_value.update(key_states, value_states, self.layer_idx, cache_kwargs)
-
-        # TODO: These transpose are quite inefficient but Flash Attention requires the layout [batch_size, sequence_length, num_heads, head_dim]. We would need to refactor the KV cache
-        # to be able to avoid many of these transpose/reshape/view.
-        query_states = query_states.transpose(1, 2)
-        key_states = key_states.transpose(1, 2)
-        value_states = value_states.transpose(1, 2)
-
-        dropout_rate = self.dropout if self.training else 0.0
-
-        # In PEFT, usually we cast the layer norms in float32 for training stability reasons
-        # therefore the input hidden states gets silently casted in float32. Hence, we need
-        # cast them back in the correct dtype just to be sure everything works as expected.
-        # This might slowdown training & inference so it is recommended to not cast the LayerNorms
-        # in fp32. (LlamaRMSNorm handles it correctly)
-
-        input_dtype = query_states.dtype
-        if input_dtype == torch.float32:
-            if torch.is_autocast_enabled():
-                target_dtype = torch.get_autocast_gpu_dtype()
-            # Handle the case where the model is quantized
-            elif hasattr(self.config, "_pre_quantization_dtype"):
-                target_dtype = self.config._pre_quantization_dtype
-            else:
-                target_dtype = self.q_proj.weight.dtype
-
-            logger.warning_once(
-                "The input hidden states seems to be silently casted in float32, this might be related to the fact"
-                " you have upcasted embedding or layer norm layers in float32. We will cast back the input in"
-                f" {target_dtype}."
-            )
-
-            query_states = query_states.to(target_dtype)
-            key_states = key_states.to(target_dtype)
-            value_states = value_states.to(target_dtype)
-
-        attn_output = self._flash_attention_forward(
-            query_states, key_states, value_states, attention_mask, q_len, dropout=dropout_rate
-        )
-
-        attn_output = attn_output.reshape(bsz, q_len, self.embed_dim).contiguous()
-        attn_output = self.out_proj(attn_output)
-
-        if not output_attentions:
-            attn_weights = None
-
-        return attn_output, attn_weights
-
-    def _flash_attention_forward(
-        self, query_states, key_states, value_states, attention_mask, query_length, dropout=0.0, softmax_scale=None
-    ):
-        """
-        Calls the forward method of Flash Attention - if the input hidden states contain at least one padding token
-        first unpad the input, then computes the attention scores and pad the final attention scores.
-        Args:
-            query_states (`torch.Tensor`):
-                Input query states to be passed to Flash Attention API
-            key_states (`torch.Tensor`):
-                Input key states to be passed to Flash Attention API
-            value_states (`torch.Tensor`):
-                Input value states to be passed to Flash Attention API
-            attention_mask (`torch.Tensor`):
-                The padding mask - corresponds to a tensor of size `(batch_size, seq_len)` where 0 stands for the
-                position of padding tokens and 1 for the position of non-padding tokens.
-            dropout (`int`, *optional*):
-                Attention dropout
-            softmax_scale (`float`, *optional*):
-                The scaling of QK^T before applying softmax. Default to 1 / sqrt(head_dim)
-        """
-
-        # TODO: Remove the `query_length != 1` check once Flash Attention for RoCm is bumped to 2.1. For details, please see the comment in LlamaFlashAttention2 __init__.
-        causal = self.is_causal and query_length != 1
-
-        # Contains at least one padding token in the sequence
-        if attention_mask is not None:
-            batch_size = query_states.shape[0]
-            query_states, key_states, value_states, indices_q, cu_seq_lens, max_seq_lens = self._upad_input(
-                query_states, key_states, value_states, attention_mask, query_length
-            )
-
-            cu_seqlens_q, cu_seqlens_k = cu_seq_lens
-            max_seqlen_in_batch_q, max_seqlen_in_batch_k = max_seq_lens
-
-            attn_output_unpad = flash_attn_varlen_func(
-                query_states,
-                key_states,
-                value_states,
-                cu_seqlens_q=cu_seqlens_q,
-                cu_seqlens_k=cu_seqlens_k,
-                max_seqlen_q=max_seqlen_in_batch_q,
-                max_seqlen_k=max_seqlen_in_batch_k,
-                dropout_p=dropout,
-                softmax_scale=softmax_scale,
-                causal=causal,
-            )
-
-            attn_output = pad_input(attn_output_unpad, indices_q, batch_size, query_length)
-        else:
-            attn_output = flash_attn_func(
-                query_states, key_states, value_states, dropout, softmax_scale=softmax_scale, causal=causal
-            )
-
-        return attn_output
-
-    def _upad_input(self, query_layer, key_layer, value_layer, attention_mask, query_length):
-        indices_k, cu_seqlens_k, max_seqlen_in_batch_k = _get_unpad_data(attention_mask)
-        batch_size, kv_seq_len, num_key_value_heads, head_dim = key_layer.shape
-
-        key_layer = index_first_axis(
-            key_layer.reshape(batch_size * kv_seq_len, num_key_value_heads, head_dim), indices_k
-        )
-        value_layer = index_first_axis(
-            value_layer.reshape(batch_size * kv_seq_len, num_key_value_heads, head_dim), indices_k
-        )
-        if query_length == kv_seq_len:
-            query_layer = index_first_axis(
-                query_layer.reshape(batch_size * kv_seq_len, self.num_heads, head_dim), indices_k
-            )
-            cu_seqlens_q = cu_seqlens_k
-            max_seqlen_in_batch_q = max_seqlen_in_batch_k
-            indices_q = indices_k
-        elif query_length == 1:
-            max_seqlen_in_batch_q = 1
-            cu_seqlens_q = torch.arange(
-                batch_size + 1, dtype=torch.int32, device=query_layer.device
-            )  # There is a memcpy here, that is very bad.
-            indices_q = cu_seqlens_q[:-1]
-            query_layer = query_layer.squeeze(1)
-        else:
-            # The -q_len: slice assumes left padding.
-            attention_mask = attention_mask[:, -query_length:]
-            query_layer, indices_q, cu_seqlens_q, max_seqlen_in_batch_q = unpad_input(query_layer, attention_mask)
-
-        return (
-            query_layer,
-            key_layer,
-            value_layer,
-            indices_q,
-            (cu_seqlens_q, cu_seqlens_k),
-            (max_seqlen_in_batch_q, max_seqlen_in_batch_k),
-        )
-
-
-# Copied from transformers.models.clip.modeling_clip.CLIPMLP with CLIP->Siglip
-class SiglipMLP(nn.Module):
-    def __init__(self, config):
-        super().__init__()
-        self.config = config
-        self.activation_fn = ACT2FN[config.hidden_act]
-        self.fc1 = nn.Linear(config.hidden_size, config.intermediate_size)
-        self.fc2 = nn.Linear(config.intermediate_size, config.hidden_size)
-
-    def forward(self, hidden_states: torch.Tensor) -> torch.Tensor:
-        hidden_states = self.fc1(hidden_states)
-        hidden_states = self.activation_fn(hidden_states)
-        hidden_states = self.fc2(hidden_states)
-        return hidden_states
-
-
-# Copied from transformers.models.clip.modeling_clip.CLIPEncoderLayer with CLIP->Siglip
-class SiglipEncoderLayer(nn.Module):
-    def __init__(self, config: SiglipVisionConfig):
-        super().__init__()
-        self.embed_dim = config.hidden_size
-        self._use_flash_attention_2 = config._attn_implementation == "flash_attention_2"
-        self.self_attn = (
-            SiglipAttention(config)
-            if not self._use_flash_attention_2
-            else SiglipFlashAttention2(config)
-        )
-        self.layer_norm1 = nn.LayerNorm(self.embed_dim, eps=config.layer_norm_eps)
-        self.mlp = SiglipMLP(config)
-        self.layer_norm2 = nn.LayerNorm(self.embed_dim, eps=config.layer_norm_eps)
-
-    def forward(
-        self,
-        hidden_states: torch.Tensor,
-        attention_mask: torch.Tensor,
-        output_attentions: Optional[bool] = False,
-    ) -> Tuple[torch.FloatTensor]:
-        """
-        Args:
-            hidden_states (`torch.FloatTensor`):
-                Input to the layer of shape `(batch, seq_len, embed_dim)`.
-            attention_mask (`torch.FloatTensor`):
-                Attention mask of shape `(batch, 1, q_len, k_v_seq_len)` where padding elements are indicated by very large negative values.
-            output_attentions (`bool`, *optional*, defaults to `False`):
-                Whether or not to return the attentions tensors of all attention layers. See `attentions` under
-                returned tensors for more detail.
-        """
-        residual = hidden_states
-
-        hidden_states = self.layer_norm1(hidden_states)
-        hidden_states, attn_weights = self.self_attn(
-            hidden_states=hidden_states,
-            attention_mask=attention_mask,
-            output_attentions=output_attentions,
-        )
-        hidden_states = residual + hidden_states
-
-        residual = hidden_states
-        hidden_states = self.layer_norm2(hidden_states)
-        hidden_states = self.mlp(hidden_states)
-        hidden_states = residual + hidden_states
-
-        outputs = (hidden_states,)
-
-        if output_attentions:
-            outputs += (attn_weights,)
-
-        return outputs
-
-
-class SiglipPreTrainedModel(PreTrainedModel):
-    """
-    An abstract class to handle weights initialization and a simple interface for downloading and loading pretrained
-    models.
-    """
-
-    config_class = SiglipVisionConfig
-    base_model_prefix = "siglip"
-    supports_gradient_checkpointing = True
-
-    def _init_weights(self, module):
-        """Initialize the weights"""
-
-        if isinstance(module, SiglipVisionEmbeddings):
-            width = self.config.hidden_size
-            nn.init.normal_(module.position_embedding.weight, std=1 / np.sqrt(width))
-        elif isinstance(module, nn.Embedding):
-            default_flax_embed_init(module.weight)
-        elif isinstance(module, SiglipAttention):
-            nn.init.normal_(module.q_proj.weight)
-            nn.init.normal_(module.k_proj.weight)
-            nn.init.normal_(module.v_proj.weight)
-            nn.init.normal_(module.out_proj.weight)
-            nn.init.zeros_(module.q_proj.bias)
-            nn.init.zeros_(module.k_proj.bias)
-            nn.init.zeros_(module.v_proj.bias)
-            nn.init.zeros_(module.out_proj.bias)
-        elif isinstance(module, SiglipMLP):
-            nn.init.normal_(module.fc1.weight)
-            nn.init.normal_(module.fc2.weight)
-            nn.init.normal_(module.fc1.bias, std=1e-6)
-            nn.init.normal_(module.fc2.bias, std=1e-6)
-        elif isinstance(module, (nn.Linear, nn.Conv2d)):
-            lecun_normal_(module.weight)
-            if module.bias is not None:
-                nn.init.zeros_(module.bias)
-        elif isinstance(module, nn.LayerNorm):
-            module.bias.data.zero_()
-            module.weight.data.fill_(1.0)
-
-
-SIGLIP_START_DOCSTRING = r"""
-    This model inherits from [`PreTrainedModel`]. Check the superclass documentation for the generic methods the
-    library implements for all its model (such as downloading or saving, resizing the input embeddings, pruning heads
-    etc.)
-    This model is also a PyTorch [torch.nn.Module](https://pytorch.org/docs/stable/nn.html#torch.nn.Module) subclass.
-    Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage
-    and behavior.
-    Parameters:
-        config ([`SiglipVisionConfig`]): Model configuration class with all the parameters of the model.
-            Initializing with a config file does not load the weights associated with the model, only the
-            configuration. Check out the [`~PreTrainedModel.from_pretrained`] method to load the model weights.
-"""
-
-
-SIGLIP_VISION_INPUTS_DOCSTRING = r"""
-    Args:
-        pixel_values (`torch.FloatTensor` of shape `(batch_size, num_channels, height, width)`):
-            Pixel values. Padding will be ignored by default should you provide it. Pixel values can be obtained using
-            [`AutoImageProcessor`]. See [`CLIPImageProcessor.__call__`] for details.
-        output_attentions (`bool`, *optional*):
-            Whether or not to return the attentions tensors of all attention layers. See `attentions` under returned
-            tensors for more detail.
-        output_hidden_states (`bool`, *optional*):
-            Whether or not to return the hidden states of all layers. See `hidden_states` under returned tensors for
-            more detail.
-        return_dict (`bool`, *optional*):
-            Whether or not to return a [`~utils.ModelOutput`] instead of a plain tuple.
-"""
-
-
-# Copied from transformers.models.clip.modeling_clip.CLIPEncoder with CLIP->Siglip
-class SiglipEncoder(nn.Module):
-    """
-    Transformer encoder consisting of `config.num_hidden_layers` self attention layers. Each layer is a
-    [`SiglipEncoderLayer`].
-    Args:
-        config: SiglipConfig
-    """
-
-    def __init__(self, config: SiglipVisionConfig):
-        super().__init__()
-        self.config = config
-        self.layers = nn.ModuleList([SiglipEncoderLayer(config) for _ in range(config.num_hidden_layers)])
-        self.gradient_checkpointing = False
-
-    # Ignore copy
-    def forward(
-        self,
-        inputs_embeds,
-        attention_mask: Optional[torch.Tensor] = None,
-        output_attentions: Optional[bool] = None,
-        output_hidden_states: Optional[bool] = None,
-        return_dict: Optional[bool] = None,
-    ) -> Union[Tuple, BaseModelOutput]:
-        r"""
-        Args:
-            inputs_embeds (`torch.FloatTensor` of shape `(batch_size, sequence_length, hidden_size)`):
-                Optionally, instead of passing `input_ids` you can choose to directly pass an embedded representation.
-                This is useful if you want more control over how to convert `input_ids` indices into associated vectors
-                than the model's internal embedding lookup matrix.
-            attention_mask (`torch.Tensor` of shape `(batch_size, sequence_length)`, *optional*):
-                Mask to avoid performing attention on padding token indices. Mask values selected in `[0, 1]`:
-                - 1 for tokens that are **not masked**,
-                - 0 for tokens that are **masked**.
-                [What are attention masks?](../glossary#attention-mask)
-            output_attentions (`bool`, *optional*):
-                Whether or not to return the attentions tensors of all attention layers. See `attentions` under
-                returned tensors for more detail.
-            output_hidden_states (`bool`, *optional*):
-                Whether or not to return the hidden states of all layers. See `hidden_states` under returned tensors
-                for more detail.
-            return_dict (`bool`, *optional*):
-                Whether or not to return a [`~utils.ModelOutput`] instead of a plain tuple.
-        """
-        output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
-        output_hidden_states = (
-            output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
-        )
-        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
-
-        encoder_states = () if output_hidden_states else None
-        all_attentions = () if output_attentions else None
-
-        hidden_states = inputs_embeds
-        for encoder_layer in self.layers:
-            if output_hidden_states:
-                encoder_states = encoder_states + (hidden_states,)
-            if self.gradient_checkpointing and self.training:
-                layer_outputs = self._gradient_checkpointing_func(
-                    encoder_layer.__call__,
-                    hidden_states,
-                    attention_mask,
-                    output_attentions,
-                )
-            else:
-                layer_outputs = encoder_layer(
-                    hidden_states,
-                    attention_mask,
-                    output_attentions=output_attentions,
-                )
-
-            hidden_states = layer_outputs[0]
-
-            if output_attentions:
-                all_attentions = all_attentions + (layer_outputs[1],)
-
-        if output_hidden_states:
-            encoder_states = encoder_states + (hidden_states,)
-
-        if not return_dict:
-            return tuple(v for v in [hidden_states, encoder_states, all_attentions] if v is not None)
-        return BaseModelOutput(
-            last_hidden_state=hidden_states, hidden_states=encoder_states, attentions=all_attentions
-        )
-
-@add_start_docstrings(
-    """The vision model from SigLIP without any head or projection on top.""",
-    SIGLIP_START_DOCSTRING
-)
-class SiglipVisionTransformer(SiglipPreTrainedModel):
-    config_class = SiglipVisionConfig
-    main_input_name = "pixel_values"
-    _supports_flash_attn_2 = True
-
-    def __init__(self, config: SiglipVisionConfig):
-        super().__init__(config)
-        self.config = config
-        embed_dim = config.hidden_size
-
-        self.embeddings = SiglipVisionEmbeddings(config)
-        self.encoder = SiglipEncoder(config)
-        self.post_layernorm = nn.LayerNorm(embed_dim, eps=config.layer_norm_eps)
-        self._use_flash_attention_2 = config._attn_implementation == "flash_attention_2"
-
-        # Initialize weights and apply final processing
-        self.post_init()
-
-    def get_input_embeddings(self) -> nn.Module:
-        return self.embeddings.patch_embedding
-
-    @add_start_docstrings_to_model_forward(SIGLIP_VISION_INPUTS_DOCSTRING)
-    @replace_return_docstrings(output_type=BaseModelOutputWithPooling, config_class=SiglipVisionConfig)
-    def forward(
-        self,
-        pixel_values,
-        patch_attention_mask: Optional[torch.BoolTensor] = None,
-        tgt_sizes: Optional[torch.IntTensor] = None,
-        output_attentions: Optional[bool] = None,
-        output_hidden_states: Optional[bool] = None,
-        return_dict: Optional[bool] = None,
-    ) -> Union[Tuple, BaseModelOutputWithPooling]:
-        r"""
-        Returns:
-        """
-        output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
-        output_hidden_states = (
-            output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
-        )
-        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
-
-        batch_size = pixel_values.size(0)
-        if patch_attention_mask is None:
-            patch_attention_mask = torch.ones(
-                size=(
-                    batch_size,
-                    pixel_values.size(2) // self.config.patch_size,
-                    pixel_values.size(3) // self.config.patch_size,
-                ),
-                dtype=torch.bool,
-                device=pixel_values.device,
-            )
-
-        hidden_states = self.embeddings(pixel_values=pixel_values, patch_attention_mask=patch_attention_mask, tgt_sizes=tgt_sizes)
-
-        patch_attention_mask = patch_attention_mask.view(batch_size, -1)
-        # The call to `_upad_input` in `_flash_attention_forward` is expensive
-        # So when the `patch_attention_mask` is full of 1s (i.e. attending to the whole sequence),
-        # avoiding passing the attention_mask, which is equivalent to attending to the full sequence
-        if not torch.any(~patch_attention_mask):
-            attention_mask=None
-        else:
-            attention_mask = (
-                _prepare_4d_attention_mask(patch_attention_mask, hidden_states.dtype)
-                if not self._use_flash_attention_2
-                else patch_attention_mask
-            )
-
-        encoder_outputs = self.encoder(
-            inputs_embeds=hidden_states,
-            attention_mask=attention_mask,
-            output_attentions=output_attentions,
-            output_hidden_states=output_hidden_states,
-            return_dict=return_dict,
-        )
-
-        last_hidden_state = encoder_outputs[0]
-        last_hidden_state = self.post_layernorm(last_hidden_state)
-
-        if not return_dict:
-            return (last_hidden_state, None) + encoder_outputs[1:]
-
-        return BaseModelOutputWithPooling(
-            last_hidden_state=last_hidden_state,
-            pooler_output=None,
-            hidden_states=encoder_outputs.hidden_states,
-            attentions=encoder_outputs.attentions,
-        )