configuration_hf_nomic_bert.py

###################################################################################################
###################################################################################################
###################################################################################################

import collections
import logging

import json
import math
import os
import re
from collections import OrderedDict
from functools import partial
from typing import List, Optional, Tuple, Union

import numpy as np
import torch
import torch.nn as nn
import torch.nn.functional as F
from einops import rearrange, repeat
from safetensors.torch import load_file as safe_load_file
from torch.nn.modules.utils import _pair
from transformers import GPT2Config, PreTrainedModel, ViTConfig, ViTModel
from transformers.models.bert.modeling_bert import (
    BaseModelOutputWithPoolingAndCrossAttentions,
    MaskedLMOutput,
    SequenceClassifierOutput,
)
from transformers.modeling_outputs import (
    MaskedLMOutput,
    MultipleChoiceModelOutput,
    QuestionAnsweringModelOutput,
    SequenceClassifierOutput,
    TokenClassifierOutput,
)
from transformers.utils import SAFE_WEIGHTS_INDEX_NAME, SAFE_WEIGHTS_NAME, WEIGHTS_INDEX_NAME, WEIGHTS_NAME
from transformers.utils.hub import cached_file, get_checkpoint_shard_files


class ContextualNomicBertConfig(GPT2Config):
    model_type = "nomic_bert"

    def __init__(
        self,
        prenorm=False,
        parallel_block=False,
        parallel_block_tied_norm=False,
        rotary_emb_fraction=0.0,
        fused_dropout_add_ln=False,
        fused_bias_fc=False,
        use_flash_attn=False,
        use_xentropy=False,
        qkv_proj_bias=True,
        rotary_emb_base=10_000,
        rotary_emb_scale_base=None,
        rotary_emb_interleaved=False,
        mlp_fc1_bias=True,
        mlp_fc2_bias=True,
        use_rms_norm=False,
        causal=False,
        type_vocab_size=2,
        dense_seq_output=True,
        pad_vocab_size_multiple=1,
        tie_word_embeddings=True,
        rotary_scaling_factor=None,
        max_trained_positions=2048,
        **kwargs,
    ):
        self.prenorm = prenorm
        self.parallel_block = parallel_block
        self.parallel_block_tied_norm = parallel_block_tied_norm
        self.rotary_emb_fraction = rotary_emb_fraction
        self.tie_word_embeddings = tie_word_embeddings
        self.fused_dropout_add_ln = fused_dropout_add_ln
        self.fused_bias_fc = fused_bias_fc
        self.use_flash_attn = use_flash_attn
        self.use_xentropy = use_xentropy
        self.qkv_proj_bias = qkv_proj_bias
        self.rotary_emb_base = rotary_emb_base
        self.rotary_emb_scale_base = rotary_emb_scale_base
        self.rotary_emb_interleaved = rotary_emb_interleaved
        self.mlp_fc1_bias = mlp_fc1_bias
        self.mlp_fc2_bias = mlp_fc2_bias
        self.use_rms_norm = use_rms_norm
        self.causal = causal
        self.type_vocab_size = type_vocab_size
        self.dense_seq_output = dense_seq_output
        self.pad_vocab_size_multiple = pad_vocab_size_multiple
        self.rotary_scaling_factor = rotary_scaling_factor
        self.max_trained_positions = max_trained_positions

        super().__init__(**kwargs)
try:
    from torch.nn.functional import scaled_dot_product_attention
except ImportError:
    scaled_dot_product_attention = None

logger = logging.getLogger(__name__)


# adapted from flash attention, added safe serialization option for hf models
def state_dict_from_pretrained(model_name, safe_serialization=False, device=None, dtype=None):
    # If not fp32, then we don't want to load directly to the GPU
    mapped_device = "cpu" if dtype not in [torch.float32, None] else device
    is_sharded = False
    load_safe = False
    resolved_archive_file = None

    weights_path = os.path.join(model_name, WEIGHTS_NAME)
    weights_index_path = os.path.join(model_name, WEIGHTS_INDEX_NAME)
    safe_weights_path = os.path.join(model_name, SAFE_WEIGHTS_NAME)
    safe_weights_index_path = os.path.join(model_name, SAFE_WEIGHTS_INDEX_NAME)

    if os.path.isfile(weights_path):
        resolved_archive_file = cached_file(model_name, WEIGHTS_NAME, _raise_exceptions_for_missing_entries=False)
    elif os.path.isfile(weights_index_path):
        resolved_archive_file = cached_file(model_name, WEIGHTS_INDEX_NAME, _raise_exceptions_for_missing_entries=False)
        is_sharded = True
    elif os.path.isfile(safe_weights_path):
        resolved_archive_file = cached_file(model_name, SAFE_WEIGHTS_NAME, _raise_exceptions_for_missing_entries=False)
        load_safe = True
    elif os.path.isfile(safe_weights_index_path):
        resolved_archive_file = cached_file(
            model_name, SAFE_WEIGHTS_INDEX_NAME, _raise_exceptions_for_missing_entries=False
        )
        is_sharded = True
        load_safe = True
    else:  # Try loading from HF hub instead of from local files
        resolved_archive_file = None
        for weight_name in [WEIGHTS_NAME, SAFE_WEIGHTS_NAME, WEIGHTS_INDEX_NAME, SAFE_WEIGHTS_INDEX_NAME]:
            resolved_archive_file = cached_file(model_name, weight_name, _raise_exceptions_for_missing_entries=False)
            if resolved_archive_file is not None:
                if weight_name in [SAFE_WEIGHTS_NAME, SAFE_WEIGHTS_INDEX_NAME]:
                    load_safe = True
                if weight_name in [WEIGHTS_INDEX_NAME, SAFE_WEIGHTS_INDEX_NAME]:
                    is_sharded = True
                break

    if resolved_archive_file is None:
        raise EnvironmentError(f"Model name {model_name} was not found.")

    if load_safe:
        loader = partial(safe_load_file, device=mapped_device)
    else:
        loader = partial(torch.load, map_location=mapped_device)

    if is_sharded:
        # resolved_archive_file becomes a list of files that point to the different
        # checkpoint shards in this case.
        resolved_archive_file, sharded_metadata = get_checkpoint_shard_files(model_name, resolved_archive_file)
        state_dict = {}
        for sharded_file in resolved_archive_file:
            state_dict.update(loader(sharded_file))
    else:
        state_dict = loader(resolved_archive_file)
    # Convert dtype before moving to GPU to save memory
    if dtype is not None:
        state_dict = {k: v.to(dtype=dtype) for k, v in state_dict.items()}
    state_dict = {k: v.to(device=device) for k, v in state_dict.items()}
    return state_dict


def filter_shapes(state_dict, model):
    """
    Filters the state dict to match the current model shape.
    """
    filtered_state_dict = {}
    for key, value in state_dict.items():
        if key in model.state_dict():
            if value.shape == model.state_dict()[key].shape:
                filtered_state_dict[key] = value
    return filtered_state_dict


def remap_bert_state_dict(
    state_dict,
    config,
    remove_bert=False,
    remove_cls_weights=False,
    add_pooling_layer=False,
):
    """
    Map the state_dict of a Huggingface BERT model to be flash_attn compatible.
    """

    def add_bert_prefix(key):
        # prepend bert. to the key
        if key.startswith("bert.") or key.startswith("cls."):
            return key
        return f"bert.{key}"

    state_dict = OrderedDict((add_bert_prefix(k), v) for k, v in state_dict.items())

    # LayerNorm
    def key_mapping_ln_gamma_beta(key):
        key = re.sub(r"LayerNorm.gamma$", "LayerNorm.weight", key)
        key = re.sub(r"LayerNorm.beta$", "LayerNorm.bias", key)
        return key

    state_dict = OrderedDict((key_mapping_ln_gamma_beta(k), v) for k, v in state_dict.items())

    # Layers
    def key_mapping_layers(key):
        return re.sub(r"^bert.encoder.layer\.", "bert.encoder.layers.", key)

    state_dict = OrderedDict((key_mapping_layers(k), v) for k, v in state_dict.items())

    # LayerNorm
    def key_mapping_ln(key):
        key = re.sub(r"^bert.embeddings.LayerNorm.", "bert.emb_ln.", key)
        key = re.sub(
            r"^bert.encoder.layers.(\d+).attention.output.LayerNorm.(weight|bias)",
            r"bert.encoder.layers.\1.norm1.\2",
            key,
        )
        key = re.sub(
            r"^bert.encoder.layers.(\d+).output.LayerNorm.(weight|bias)",
            r"bert.encoder.layers.\1.norm2.\2",
            key,
        )
        key = re.sub(
            r"^cls.predictions.transform.LayerNorm.(weight|bias)",
            r"cls.predictions.transform.layer_norm.\1",
            key,
        )
        return key

    state_dict = OrderedDict((key_mapping_ln(k), v) for k, v in state_dict.items())

    # MLP
    def key_mapping_mlp(key):
        key = re.sub(
            r"^bert.encoder.layers.(\d+).intermediate.dense.(weight|bias)",
            r"bert.encoder.layers.\1.mlp.fc1.\2",
            key,
        )
        key = re.sub(
            r"^bert.encoder.layers.(\d+).output.dense.(weight|bias)",
            r"bert.encoder.layers.\1.mlp.fc2.\2",
            key,
        )
        return key

    state_dict = OrderedDict((key_mapping_mlp(k), v) for k, v in state_dict.items())

    # Attention
    last_layer_subset = getattr(config, "last_layer_subset", False)
    for d in range(config.num_hidden_layers):
        if f"bert.encoder.layers.{d}.attention.self.query.weight" not in state_dict:
            continue
        Wq = state_dict.pop(f"bert.encoder.layers.{d}.attention.self.query.weight")
        Wk = state_dict.pop(f"bert.encoder.layers.{d}.attention.self.key.weight")
        Wv = state_dict.pop(f"bert.encoder.layers.{d}.attention.self.value.weight")
        bq = state_dict.pop(f"bert.encoder.layers.{d}.attention.self.query.bias")
        bk = state_dict.pop(f"bert.encoder.layers.{d}.attention.self.key.bias")
        bv = state_dict.pop(f"bert.encoder.layers.{d}.attention.self.value.bias")
        if not (last_layer_subset and d == config.num_hidden_layers - 1):
            state_dict[f"bert.encoder.layers.{d}.attn.Wqkv.weight"] = torch.cat([Wq, Wk, Wv], dim=0)
            state_dict[f"bert.encoder.layers.{d}.attn.Wqkv.bias"] = torch.cat([bq, bk, bv], dim=0)
        else:
            state_dict[f"bert.encoder.layers.{d}.attn.Wq.weight"] = Wq
            state_dict[f"bert.encoder.layers.{d}.attn.Wkv.weight"] = torch.cat([Wk, Wv], dim=0)
            state_dict[f"bert.encoder.layers.{d}.attn.Wq.bias"] = bq
            state_dict[f"bert.encoder.layers.{d}.attn.Wkv.bias"] = torch.cat([bk, bv], dim=0)

    def key_mapping_attn(key):
        return re.sub(
            r"^bert.encoder.layers.(\d+).attention.output.dense.(weight|bias)",
            r"bert.encoder.layers.\1.attn.out_proj.\2",
            key,
        )

    state_dict = OrderedDict((key_mapping_attn(k), v) for k, v in state_dict.items())

    def key_mapping_decoder_bias(key):
        return re.sub(r"^cls.predictions.bias", "cls.predictions.decoder.bias", key)

    # remove nsp weights, we don't use
    state_dict.pop("cls.seq_relationship.weight", None)
    state_dict.pop("cls.seq_relationship.bias", None)
    state_dict.pop("bert.embeddings.position_ids", None)

    state_dict = OrderedDict((key_mapping_decoder_bias(k), v) for k, v in state_dict.items())

    if remove_cls_weights:
        cls_weights = [
            "cls.predictions.decoder.bias",
            "cls.predictions.transform.dense.weight",
            "cls.predictions.transform.dense.bias",
            "cls.predictions.transform.layer_norm.weight",
            "cls.predictions.transform.layer_norm.bias",
            "cls.predictions.decoder.weight",
        ]
        for weight in cls_weights:
            state_dict.pop(weight, None)

    # Word embedding
    pad_vocab_size_multiple = getattr(config, "pad_vocab_size_multiple", 1)
    if pad_vocab_size_multiple > 1:
        word_embeddings = state_dict["bert.embeddings.word_embeddings.weight"]
        state_dict["bert.embeddings.word_embeddings.weight"] = F.pad(
            word_embeddings, (0, 0, 0, config.vocab_size - word_embeddings.shape[0])
        )
        if not remove_cls_weights:
            decoder_weight = state_dict["cls.predictions.decoder.weight"]
            state_dict["cls.predictions.decoder.weight"] = F.pad(
                decoder_weight, (0, 0, 0, config.vocab_size - decoder_weight.shape[0])
            )
            # If the vocab was padded, we want to set the decoder bias for those padded indices to be
            # strongly negative (i.e. the decoder shouldn't predict those indices).
            # TD [2022-05-09]: I don't think it affects the MLPerf training.
            if "cls.predictions.decoder.bias" in state_dict:
                decoder_bias = state_dict["cls.predictions.decoder.bias"]
                state_dict["cls.predictions.decoder.bias"] = F.pad(
                    decoder_bias, (0, config.vocab_size - decoder_bias.shape[0]), value=-100.0
                )

    if add_pooling_layer is False:
        pooler_weights = [
            "bert.pooler.dense.weight",
            "bert.pooler.dense.bias",
        ]
        for key in pooler_weights:
            state_dict.pop(key, None)

    if remove_bert:

        def remove_bert_prefix(key):
            key = re.sub(r"^bert.", "", key)
            return key

        state_dict = OrderedDict((remove_bert_prefix(k), v) for k, v in state_dict.items())

    return state_dict


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):
        print(
            "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
    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
    tensor.clamp_(min=a, max=b)
    return tensor


def trunc_normal_tf_(tensor, mean=0.0, std=1.0, a=-2.0, b=2.0):
    r"""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
    Examples:
        >>> w = torch.empty(3, 5)
        >>> nn.init.trunc_normal_(w)
    """
    with torch.no_grad():
        _trunc_normal_(tensor, 0, 1.0, a, b)
        tensor.mul_(std).add_(mean)
    return tensor


class ContextualNomicBertPreTrainedModel(PreTrainedModel):
    """An abstract class to handle weights initialization and
    a simple interface for dowloading and loading pretrained models.
    """

    config_class = ContextualNomicBertConfig
    base_model_prefix = "model"
    supports_gradient_checkpointing = True
    _no_split_modules = ["Block"]
    _skip_keys_device_placement = "past_key_values"

    def __init__(self, config, *inputs, **kwargs):
        super().__init__(config)
        if not isinstance(config, GPT2Config):
            raise ValueError(
                "Parameter config in `{}(config)` should be an instance of class `GPT2Config`. "
                "To create a model from a Google pretrained model use "
                "`model = {}.from_pretrained(PRETRAINED_MODEL_NAME)`".format(
                    self.__class__.__name__, self.__class__.__name__
                )
            )
        self.config = config

    @classmethod
    def from_pretrained(cls, model_name, config=None, *inputs, **kwargs):
        """
        Instantiate a ContextualNomicBertPreTrainedModel from a pre-trained model file or a pytorch state dict.
        Download and cache the pre-trained model file if needed.
        Params:
            pretrained_model_name_or_path: either:
                - a path or url to a pretrained model archive containing:
                    . `bert_config.json` a configuration file for the model
                    . `pytorch_model.bin` a PyTorch dump of a ContextualNomicBertForPretraining instance
                - a path or url to a pretrained model archive containing:
                    . `bert_config.json` a configuration file for the model
                    . `model.chkpt` a TensorFlow checkpoint
            *inputs, **kwargs: additional input for the specific ContextualNomicBert class
                (ex: num_labels for ContextualNomicBertForSequenceClassification)
        """
        # Instantiate model.
        if config is None:
            config = cls.config_class.from_pretrained(model_name)
        remove_cls = cls != ContextualNomicBertForPreTraining
        remove_bert_prefix = cls not in [ContextualNomicBertForPreTraining, ContextualNomicBertForSequenceClassification, ContextualNomicBertForTokenClassification, ContextualNomicBertForMultipleChoice, ContextualNomicBertForQuestionAnswering]
        ignore_mismatched_shapes = kwargs.pop("ignore_mismatched_sizes", False)
        num_labels = kwargs.pop("num_labels", None)
        rotary_scaling_factor = kwargs.pop("rotary_scaling_factor", None)
        strict = kwargs.pop("strict", True)
        dtype = kwargs.pop("torch_dtype", None)
        if rotary_scaling_factor:
            config.rotary_scaling_factor = rotary_scaling_factor

        if config.n_positions <= 0 and config.rotary_emb_fraction > 0:
            config.n_positions = 2048
        if num_labels:
            config.num_labels = num_labels

        if "add_pooling_layer" in kwargs:
            model = cls(config, *inputs, add_pooling_layer=kwargs.pop("add_pooling_layer"))
        else:
            if cls == ContextualNomicBertModel:
                model = cls(config, *inputs, add_pooling_layer=False)
            else:
                model = cls(config, *inputs)

        if dtype is not None:
            model = model.to(dtype=dtype)
        # TODO: fix this
        # Assuming we know what we're doing when loading from disk
        # Prob a bad assumption but i'm tired and want to train this asap
        if os.path.exists(model_name):
            model_path = f"{model_name}/pytorch_model.bin"
            if os.path.exists(model_path):
                state_dict = torch.load(f"{model_name}/pytorch_model.bin")
            else:
                model_path = f"{model_name}/model.safetensors"
                if not os.path.exists(model_path):
                    raise ValueError(f"Model path {model_path} not found")
                state_dict = safe_load_file(model_path)

            if ignore_mismatched_shapes:
                state_dict = filter_shapes(state_dict, model)
            load_return = model.load_state_dict(state_dict, strict=False)
        else:
            # TODO: can probably check config class and see if we need to remap from a bert model
            state_dict = state_dict_from_pretrained(model_name, dtype=dtype)
            state_dict = remap_bert_state_dict(
                state_dict,
                config,
                remove_bert=remove_bert_prefix,
                remove_cls_weights=remove_cls,
                add_pooling_layer=getattr(config, "add_pooling_layer", False),
            )
            if ignore_mismatched_shapes:
                state_dict = filter_shapes(state_dict, model)

            load_return = model.load_state_dict(state_dict, strict=strict)
        logger.warning(load_return)
        return model

    def _set_gradient_checkpointing(self, module, value=False):
        if isinstance(module, ContextualNomicBertEncoder):
            module.gradient_checkpointing = value


# https://github.com/huggingface/transformers/blob/7032e0203262ebb2ebf55da8d2e01f873973e835/src/transformers/models/bert/modeling_bert.py#L748
def _init_weights(module, initializer_range=0.02):
    if isinstance(module, nn.Linear):
        nn.init.normal_(module.weight, std=initializer_range)
        if module.bias is not None:
            nn.init.zeros_(module.bias)
    elif isinstance(module, nn.Embedding):
        nn.init.normal_(module.weight, std=initializer_range)
        if module.padding_idx is not None:
            nn.init.zeros_(module.weight[module.padding_idx])


def _ntuple(n):
    def parse(x):
        if isinstance(x, collections.abc.Iterable) and not isinstance(x, str):
            return tuple(x)
        return tuple(repeat(x, n))

    return parse


to_1tuple = _ntuple(1)
to_2tuple = _ntuple(2)
to_3tuple = _ntuple(3)
to_4tuple = _ntuple(4)
to_ntuple = _ntuple


def get_2d_sincos_pos_embed(embed_dim, grid_size, add_cls_token=False):
    """
    Create 2D sin/cos positional embeddings.
    Args:
        embed_dim (`int`):
            Embedding dimension.
        grid_size (`int`):
            The grid height and width.
        add_cls_token (`bool`, *optional*, defaults to `False`):
            Whether or not to add a classification (CLS) token.
    Returns:
        (`torch.FloatTensor` of shape (grid_size*grid_size, embed_dim) or (1+grid_size*grid_size, embed_dim): the
        position embeddings (with or without classification token)
    """
    grid_h = np.arange(grid_size, dtype=np.float32)

    grid_w = np.arange(grid_size, dtype=np.float32)
    grid = np.meshgrid(grid_w, grid_h)  # here w goes first
    grid = np.stack(grid, axis=0)

    grid = grid.reshape([2, 1, grid_size, grid_size])
    pos_embed = get_2d_sincos_pos_embed_from_grid(embed_dim, grid)
    if add_cls_token:
        pos_embed = np.concatenate([np.zeros([1, embed_dim]), pos_embed], axis=0)
    return pos_embed


def get_2d_sincos_pos_embed_from_grid(embed_dim, grid):
    if embed_dim % 2 != 0:
        raise ValueError("embed_dim must be even")

    # use half of dimensions to encode grid_h
    emb_h = get_1d_sincos_pos_embed_from_grid(embed_dim // 2, grid[0])  # (H*W, D/2)
    emb_w = get_1d_sincos_pos_embed_from_grid(embed_dim // 2, grid[1])  # (H*W, D/2)

    emb = np.concatenate([emb_h, emb_w], axis=1)  # (H*W, D)
    return emb


def get_1d_sincos_pos_embed_from_grid(embed_dim, pos):
    """
    embed_dim: output dimension for each position pos: a list of positions to be encoded: size (M,) out: (M, D)
    """
    if embed_dim % 2 != 0:
        raise ValueError("embed_dim must be even")

    omega = np.arange(embed_dim // 2, dtype=float)
    omega /= embed_dim / 2.0
    omega = 1.0 / 10000**omega  # (D/2,)

    pos = pos.reshape(-1)  # (M,)
    out = np.einsum("m,d->md", pos, omega)  # (M, D/2), outer product

    emb_sin = np.sin(out)  # (M, D/2)
    emb_cos = np.cos(out)  # (M, D/2)

    emb = np.concatenate([emb_sin, emb_cos], axis=1)  # (M, D)
    return emb


def ndgrid(*tensors) -> Tuple[torch.Tensor, ...]:
    """generate N-D grid in dimension order.
    The ndgrid function is like meshgrid except that the order of the first two input arguments are switched.
    That is, the statement
    [X1,X2,X3] = ndgrid(x1,x2,x3)
    produces the same result as
    [X2,X1,X3] = meshgrid(x2,x1,x3)
    This naming is based on MATLAB, the purpose is to avoid confusion due to torch's change to make
    torch.meshgrid behaviour move from matching ndgrid ('ij') indexing to numpy meshgrid defaults of ('xy').
    """
    try:
        return torch.meshgrid(*tensors, indexing='ij')
    except TypeError:
        # old PyTorch < 1.10 will follow this path as it does not have indexing arg,
        # the old behaviour of meshgrid was 'ij'
        return torch.meshgrid(*tensors)


def build_fourier_pos_embed(
    feat_shape: List[int],
    bands: Optional[torch.Tensor] = None,
    num_bands: int = 64,
    max_res: int = 224,
    temperature: float = 10000.0,
    linear_bands: bool = False,
    include_grid: bool = False,
    in_pixels: bool = True,
    ref_feat_shape: Optional[List[int]] = None,
    dtype: torch.dtype = torch.float32,
    device: Optional[torch.device] = None,
) -> List[torch.Tensor]:
    """
    Args:
        feat_shape: Feature shape for embedding.
        bands: Pre-calculated frequency bands.
        num_bands: Number of frequency bands (determines output dim).
        max_res: Maximum resolution for pixel based freq.
        temperature: Temperature for non-pixel freq.
        linear_bands: Linear band spacing for pixel based freq.
        include_grid: Include the spatial grid in output.
        in_pixels: Output in pixel freq.
        ref_feat_shape: Reference feature shape for resize / fine-tune.
        dtype: Output dtype.
        device: Output device.
    Returns:
    """
    if bands is None:
        if in_pixels:
            bands = pixel_freq_bands(
                num_bands,
                float(max_res),
                linear_bands=linear_bands,
                device=device,
            )
        else:
            bands = freq_bands(
                num_bands,
                temperature=temperature,
                step=1,
                device=device,
            )
    else:
        if device is None:
            device = bands.device
        if dtype is None:
            dtype = bands.dtype

    if in_pixels:
        t = [torch.linspace(-1.0, 1.0, steps=s, device=device, dtype=torch.float32) for s in feat_shape]
    else:
        t = [torch.arange(s, device=device, dtype=torch.int64).to(torch.float32) for s in feat_shape]

    if ref_feat_shape is not None:
        # eva's scheme for resizing rope embeddings (ref shape = pretrain)
        t = [x / f * r for x, f, r in zip(t, feat_shape, ref_feat_shape)]

    grid = torch.stack(ndgrid(t), dim=-1)
    grid = grid.unsqueeze(-1)
    pos = grid * bands

    pos_sin, pos_cos = pos.sin().to(dtype=dtype), pos.cos().to(dtype)
    out = [grid, pos_sin, pos_cos] if include_grid else [pos_sin, pos_cos]
    return out


def build_rotary_pos_embed(
    feat_shape: List[int],
    bands: Optional[torch.Tensor] = None,
    dim: int = 64,
    max_res: int = 224,
    temperature: float = 10000.0,
    linear_bands: bool = False,
    in_pixels: bool = True,
    ref_feat_shape: Optional[List[int]] = None,
    dtype: torch.dtype = torch.float32,
    device: Optional[torch.device] = None,
):
    """
    Args:
        feat_shape: Spatial shape of the target tensor for embedding.
        bands: Optional pre-generated frequency bands
        dim: Output dimension of embedding tensor.
        max_res: Maximum resolution for pixel mode.
        temperature: Temperature (inv freq) for non-pixel mode
        linear_bands: Linearly (instead of log) spaced bands for pixel mode
        in_pixels: Pixel vs language (inv freq) mode.
        dtype: Output dtype.
        device: Output device.
    Returns:
    """
    sin_emb, cos_emb = build_fourier_pos_embed(
        feat_shape,
        bands=bands,
        num_bands=dim // 4,
        max_res=max_res,
        temperature=temperature,
        linear_bands=linear_bands,
        in_pixels=in_pixels,
        ref_feat_shape=ref_feat_shape,
        device=device,
        dtype=dtype,
    )
    num_spatial_dim = 1
    # this would be much nicer as a .numel() call to torch.Size(), but torchscript sucks
    for x in feat_shape:
        num_spatial_dim *= x
    sin_emb = sin_emb.reshape(num_spatial_dim, -1).repeat_interleave(2, -1)
    cos_emb = cos_emb.reshape(num_spatial_dim, -1).repeat_interleave(2, -1)
    return sin_emb, cos_emb


def freq_bands(
    num_bands: int,
    temperature: float = 10000.0,
    step: int = 2,
    device: Optional[torch.device] = None,
) -> torch.Tensor:
    exp = torch.arange(0, num_bands, step, dtype=torch.int64, device=device).to(torch.float32) / num_bands
    bands = 1.0 / (temperature**exp)
    return bands


def pixel_freq_bands(
    num_bands: int,
    max_freq: float = 224.0,
    linear_bands: bool = True,
    device: Optional[torch.device] = None,
):
    if linear_bands:
        bands = torch.linspace(1.0, max_freq / 2, num_bands, dtype=torch.float32, device=device)
    else:
        bands = 2 ** torch.linspace(0, math.log(max_freq, 2) - 1, num_bands, dtype=torch.float32, device=device)
    return bands * torch.pi


def rot(x):
    return torch.stack([-x[..., 1::2], x[..., ::2]], -1).reshape(x.shape)


def apply_rot_embed_cat(x: torch.Tensor, emb):
    sin_emb, cos_emb = emb.tensor_split(2, -1)
    if sin_emb.ndim == 3:
        return x * cos_emb.unsqueeze(1).expand_as(x) + rot(x) * sin_emb.unsqueeze(1).expand_as(x)
    return x * cos_emb + rot(x) * sin_emb


class ContextualNomicBertEmbeddings(nn.Module):
    def __init__(self, config):
        """
        If max_position_embeddings <= 0, there's no position embeddings
        If type_vocab_size <= 0, there's no token type embeddings
        """
        super().__init__()
        self.word_embeddings = nn.Embedding(config.vocab_size, config.hidden_size, padding_idx=config.pad_token_id)
        self.max_position_embeddings = config.max_position_embeddings if config.rotary_emb_fraction <= 0 else 0
        self.type_vocab_size = config.type_vocab_size
        if self.max_position_embeddings > 0 and config.rotary_emb_fraction <= 0:
            self.position_embeddings = nn.Embedding(
                config.max_position_embeddings,
                config.hidden_size,
            )
        if self.type_vocab_size > 0:
            self.token_type_embeddings = nn.Embedding(config.type_vocab_size, config.hidden_size)

    def forward(self, input_ids=None, position_ids=None, token_type_ids=None, inputs_embeds=None):
        """
        input_ids: (batch, seqlen)
        position_ids: (batch, seqlen)
        token_type_ids: (batch, seqlen)
        """
        if inputs_embeds is None:
            embeddings = self.word_embeddings(input_ids)
        else:
            embeddings = inputs_embeds
        batch_size, seqlen, _ = embeddings.shape
        
        if self.type_vocab_size > 0:
            if token_type_ids is None:
                token_type_ids = torch.zeros(seqlen, dtype=torch.long, device=embeddings.device)
            token_type_embeddings = self.token_type_embeddings(token_type_ids)
            embeddings = embeddings + token_type_embeddings

        if self.max_position_embeddings > 0:
            if position_ids is None:
                position_ids = torch.arange(seqlen, dtype=torch.long, device=embeddings.device)
            position_embeddings = self.position_embeddings(position_ids)
            embeddings = embeddings + position_embeddings
        return embeddings


class ContextualNomicBertMLP(nn.Module):
    def __init__(
        self,
        in_features,
        hidden_features=None,
        out_features=None,
        activation=F.gelu,
        bias1=True,
        bias2=True,
        return_residual=False,
        fused_bias_fc=False,
    ):
        super().__init__()
        out_features = out_features if out_features is not None else in_features
        hidden_features = hidden_features if hidden_features is not None else in_features * 4
        self.return_residual = return_residual
        self.fc1 = nn.Linear(in_features, hidden_features, bias=bias1)
        approximate = "tanh" if activation in ["gelu_new", "gelu_fast", "gelu_pytorch_tanh"] else "none"
        self.activation = nn.GELU(approximate=approximate) if activation == "gelu" else activation
        self.fc2 = nn.Linear(hidden_features, out_features, bias=bias2)

    def forward(self, x):
        y = self.fc1(x)
        y = self.activation(y)
        y = self.fc2(y)
        return y if not self.return_residual else (y, x)


class NomciBertGatedMLP(nn.Module):
    def __init__(
        self,
        in_features,
        hidden_features=None,
        out_features=None,
        activation=F.sigmoid,
        bias1=True,
        bias2=True,
        multiple_of=256,
        return_residual=False,
        fused_bias_fc=True,
        device=None,
        dtype=None,
        norm_layer=False,
    ):
        super().__init__()
        out_features = out_features if out_features is not None else in_features
        hidden_features = hidden_features if hidden_features is not None else int(8 * in_features / 3)
        hidden_features = int((hidden_features + multiple_of - 1) // multiple_of * multiple_of)
        self.return_residual = return_residual

        self.fc11 = nn.Linear(in_features, hidden_features, bias=bias1)
        self.fc12 = nn.Linear(in_features, hidden_features, bias=bias1)
        self.activation = activation
        self.fc2 = nn.Linear(hidden_features, out_features, bias=bias2)
        self.norm = nn.LayerNorm(hidden_features) if norm_layer else nn.Identity()

    def forward(self, x):
        y = self.fc11(x)
        gate = self.fc12(x)
        if self.activation == F.sigmoid:  # Special case for GLU
            y = F.glu(torch.cat([y, gate], dim=-1), dim=-1)
        else:
            y = y * self.activation(gate)

        # eva uses layer norm after the activation
        y = self.norm(y)

        y = self.fc2(y)
        return y if not self.return_residual else (y, x)


def rotate_half(x, interleaved=False):
    if not interleaved:
        x1, x2 = x.chunk(2, dim=-1)
        return torch.cat((-x2, x1), dim=-1)
    else:
        x1, x2 = x[..., ::2], x[..., 1::2]
        return rearrange(torch.stack((-x2, x1), dim=-1), "... d two -> ... (d two)", two=2)


def apply_rotary_emb(x, cos, sin, offset=0, interleaved=False):
    """
    x: (batch_size, seqlen, nheads, headdim)
    cos, sin: (seqlen, rotary_dim / 2) or (batch_size, seqlen, rotary_dim / 2)
    """
    ro_dim = cos.shape[-1] * 2
    assert ro_dim <= x.shape[-1]
    cos, sin = (
        cos[offset : offset + x.shape[1]],
        sin[offset : offset + x.shape[1]],
    )
    cos = repeat(cos, "... d -> ... 1 (2 d)" if not interleaved else "... d -> ... 1 (d 2)")
    sin = repeat(sin, "... d -> ... 1 (2 d)" if not interleaved else "... d -> ... 1 (d 2)")
    return torch.cat(
        [x[..., :ro_dim] * cos + rotate_half(x[..., :ro_dim], interleaved) * sin, x[..., ro_dim:]],
        dim=-1,
    )


class ContextualNomicBertRotaryEmbedding(nn.Module):
    def __init__(
        self,
        dim: int,
        base=10000.0,
        interleaved=False,
        scale_base=None,
        pos_idx_in_fp32=True,
        device=None,
    ):
        """
        interleaved: if True, rotate pairs of even and odd dimensions (GPT-J style) instead
            of 1st half and 2nd half (GPT-NeoX style).
        pos_idx_in_fp32: if True, the position indices [0.0, ..., seqlen - 1] are in fp32,
            otherwise they might be in lower precision.
            This option was added because previously (before 2023-07-02), when we construct
            the position indices, we use the dtype of self.inv_freq. In most cases this would
            be fp32, but if the model is trained in pure bf16 (not mixed precision), then
            self.inv_freq would be bf16, and the position indices are also in bf16.
            Because of the limited precision of bf16 (e.g. 1995.0 is rounded to 2000.0), the
            embeddings for some positions will coincide.
            To maintain compatibility with models previously trained in pure bf16,
            we add this option.
        """
        super().__init__()
        self.dim = dim
        self.base = float(base)
        self.pos_idx_in_fp32 = pos_idx_in_fp32
        # Generate and save the inverse frequency buffer (non trainable)
        inv_freq = self._compute_inv_freq(device)
        self.register_buffer("inv_freq", inv_freq, persistent=False)
        self.interleaved = interleaved
        self.scale_base = scale_base
        scale = (
            (torch.arange(0, dim, 2, device=device, dtype=torch.float32) + 0.4 * dim) / (1.4 * dim)
            if scale_base is not None
            else None
        )
        self.register_buffer("scale", scale, persistent=False)

        self._seq_len_cached = 0
        self._cos_cached = None
        self._sin_cached = None
        self._cos_k_cached = None
        self._sin_k_cached = None

    def _compute_inv_freq(self, device=None):
        return 1.0 / (self.base ** (torch.arange(0, self.dim, 2, device=device, dtype=torch.float32) / self.dim))

    def _update_cos_sin_cache(self, seqlen, device=None, dtype=None):
        # Reset the tables if the sequence length has changed,
        # if we're on a new device (possibly due to tracing for instance),
        # or if we're switching from inference mode to training
        if (
            seqlen > self._seq_len_cached
            or self._cos_cached is None
            or self._cos_cached.device != device
            or self._cos_cached.dtype != dtype
            or (self.training and self._cos_cached.is_inference())
        ):
            self._seq_len_cached = seqlen
            # We want fp32 here, not self.inv_freq.dtype, since the model could be loaded in bf16
            # And the output of arange can be quite large, so bf16 would lose a lot of precision.
            # However, for compatibility reason, we add an option to use the dtype of self.inv_freq.
            if self.pos_idx_in_fp32:
                t = torch.arange(seqlen, device=device, dtype=torch.float32)
                # We want fp32 here as well since inv_freq will be multiplied with t, and the output
                # will be large. Having it in bf16 will lose a lot of precision and cause the
                # cos & sin output to change significantly.
                # We want to recompute self.inv_freq if it was not loaded in fp32
                if self.inv_freq.dtype != torch.float32:
                    inv_freq = self._compute_inv_freq(device=device)
                else:
                    inv_freq = self.inv_freq
            else:
                t = torch.arange(seqlen, device=device, dtype=self.inv_freq.dtype)
                inv_freq = self.inv_freq
            # Don't do einsum, it converts fp32 to fp16 under AMP
            # freqs = torch.einsum("i,j->ij", t, self.inv_freq)
            freqs = torch.outer(t, inv_freq)
            self._cos_cached = torch.cos(freqs).to(dtype)
            self._sin_cached = torch.sin(freqs).to(dtype)

    def forward(
        self,
        qkv: torch.Tensor,
        kv: Optional[torch.Tensor] = None,
        seqlen_offset: Union[int, torch.Tensor] = 0,
        max_seqlen: Optional[int] = None,
    ) -> Tuple[torch.Tensor, torch.Tensor]:
        """
        qkv: (batch, seqlen, 3, nheads, headdim) if kv is none,
             else it's just q of shape (batch, seqlen, nheads, headdim)
        kv: (batch, seqlen, 2, nheads, headdim)
        seqlen_offset: (batch_size,) or int. Each sequence in x is shifted by this amount.
            Most commonly used in inference when we have KV cache.
            If it's a tensor of shape (batch_size,), then to update the cos / sin cache, one
            should pass in max_seqlen, which will update the cos / sin cache up to that length.
        Apply rotary embedding *inplace* to qkv and / or kv.
        """
        seqlen = qkv.shape[1]
        if seqlen > self._seq_len_cached:
            self._update_cos_sin_cache(seqlen, device=qkv.device, dtype=qkv.dtype)
        elif max_seqlen is not None:
            self._update_cos_sin_cache(max_seqlen, device=qkv.device, dtype=qkv.dtype)
        elif isinstance(seqlen_offset, int):
            self._update_cos_sin_cache(seqlen + seqlen_offset, device=qkv.device, dtype=qkv.dtype)

        q_rot = apply_rotary_emb(qkv[:, :, 0], self._cos_cached, self._sin_cached, seqlen_offset, self.interleaved)
        k_rot = apply_rotary_emb(qkv[:, :, 1], self._cos_cached, self._sin_cached, seqlen_offset, self.interleaved)
        return torch.stack((q_rot, k_rot, qkv[:, :, 2]), dim=2)


class ContextualNomicBertDynamicNTKRotaryEmbedding(ContextualNomicBertRotaryEmbedding):
    def __init__(self, rotary_scaling_factor, max_position_embeddings, **kwargs):
        super().__init__(**kwargs)
        self.rotary_scaling_factor = rotary_scaling_factor
        self.max_position_embeddings = max_position_embeddings

    def _compute_inv_freq(self, base=None, device=None):
        if base is None:
            base = self.base
        return 1.0 / (base ** (torch.arange(0, self.dim, 2, device=device, dtype=torch.float32) / self.dim))

    def _update_cos_sin_cache(self, seqlen, device=None, dtype=None):
        # Reset the tables if the sequence length has changed,
        # if we're on a new device (possibly due to tracing for instance),
        # or if we're switching from inference mode to training
        if seqlen > self.max_position_embeddings:
            base = self.base * (
                (self.rotary_scaling_factor * seqlen / self.max_position_embeddings) - (self.rotary_scaling_factor - 1)
            ) ** (self.dim / (self.dim - 2))
            inv_freq = self._compute_inv_freq(base=base, device=device)
            self.register_buffer("inv_freq", inv_freq, persistent=False)

        if (
            seqlen > self._seq_len_cached
            or self._cos_cached is None
            or self._cos_cached.device != device
            or self._cos_cached.dtype != dtype
            or (self.training and self._cos_cached.is_inference())
        ):
            self._seq_len_cached = seqlen
            # We want fp32 here, not self.inv_freq.dtype, since the model could be loaded in bf16
            # And the output of arange can be quite large, so bf16 would lose a lot of precision.
            # However, for compatibility reason, we add an option to use the dtype of self.inv_freq.
            if self.pos_idx_in_fp32:
                t = torch.arange(seqlen, device=device, dtype=torch.float32)
                # We want fp32 here as well since inv_freq will be multiplied with t, and the output
                # will be large. Having it in bf16 will lose a lot of precision and cause the
                # cos & sin output to change significantly.
                # We want to recompute self.inv_freq if it was not loaded in fp32
                if self.inv_freq.dtype != torch.float32:
                    if seqlen > self.max_position_embeddings:
                        base = self.base * (
                            (self.scaling_factor * seqlen / self.max_position_embeddings) - (self.scaling_factor - 1)
                        ) ** (self.dim / (self.dim - 2))
                    else:
                        base = self.base
                    inv_freq = self._compute_inv_freq(device=device, base=base)
                else:
                    inv_freq = self.inv_freq
            else:
                t = torch.arange(seqlen, device=device, dtype=self.inv_freq.dtype)
                inv_freq = self.inv_freq
            # Don't do einsum, it converts fp32 to fp16 under AMP
            # freqs = torch.einsum("i,j->ij", t, self.inv_freq)
            freqs = torch.outer(t, inv_freq)
            if self.scale is None:
                self._cos_cached = torch.cos(freqs).to(dtype)
                self._sin_cached = torch.sin(freqs).to(dtype)
            else:
                power = (
                    torch.arange(seqlen, dtype=self.scale.dtype, device=self.scale.device) - seqlen // 2
                ) / self.scale_base
                scale = self.scale.to(device=power.device) ** rearrange(power, "s -> s 1")
                # We want the multiplication by scale to happen in fp32
                self._cos_cached = (torch.cos(freqs) * scale).to(dtype)
                self._sin_cached = (torch.sin(freqs) * scale).to(dtype)
                self._cos_k_cached = (torch.cos(freqs) / scale).to(dtype)
                self._sin_k_cached = (torch.sin(freqs) / scale).to(dtype)


class ContextualNomicBertAttention(nn.Module):
    """Multi-head self-attention and cross-attention"""

    def __init__(
        self,
        config,
    ) -> None:
        """
        num_heads_kv: can be used to toggle MQA / GQA. If None, use num_heads.
        return_residual: whether to return the input x along with the output. This is for
            performance reason: for post-norm architecture, returning the input allows us
            to fuse the backward of nn.Linear with the residual connection.
        """
        super().__init__()
        self.embed_dim = config.n_embd
        self.use_flash_attn = config.use_flash_attn
        self.fused_bias_fc = config.fused_bias_fc

        self.num_heads = config.n_head
        self.num_heads_kv = config.num_heads_kv if getattr(config, "num_heads_kv", None) is not None else self.num_heads
        assert self.embed_dim % self.num_heads == 0, "embed_dim must be divisible by num_heads"
        self.head_dim = self.embed_dim // self.num_heads
        # we don't really support mqa / gqa for now
        qkv_dim = self.head_dim * (self.num_heads + 2 * self.num_heads_kv)

        self.register_buffer(
            "norm_factor",
            torch.sqrt(torch.tensor(self.head_dim, dtype=torch.float32)).to(torch.get_default_dtype()),
            persistent=False,
        )

        self.rotary_emb_dim = self.head_dim * config.rotary_emb_fraction
        if self.rotary_emb_dim > 0:
            if getattr(config, "rotary_scaling_factor", None):
                self.rotary_emb = ContextualNomicBertDynamicNTKRotaryEmbedding(
                    dim=self.rotary_emb_dim,
                    base=config.rotary_emb_base,
                    scale_base=config.rotary_emb_scale_base,
                    interleaved=config.rotary_emb_interleaved,
                    rotary_scaling_factor=config.rotary_scaling_factor,
                    max_position_embeddings=config.max_trained_positions,
                )
            else:
                self.rotary_emb = ContextualNomicBertRotaryEmbedding(
                    dim=self.rotary_emb_dim,
                    base=config.rotary_emb_base,
                    scale_base=config.rotary_emb_scale_base,
                    interleaved=config.rotary_emb_interleaved,
                )
            # bug in xformers: https://github.com/facebookresearch/xformers/issues/841
            # uses the head dimension instead of the sequence dimension
            self.rotary_head_dim = getattr(config, "rotary_head_dim", False)

        self.Wqkv = nn.Linear(self.embed_dim, qkv_dim, bias=config.qkv_proj_bias)

        self.out_proj = nn.Linear(self.embed_dim, self.embed_dim, bias=config.qkv_proj_bias)
        self.causal = config.causal
        self.drop = nn.Dropout(config.attn_pdrop)
        self.num_prefix_tokens = max(getattr(config, "register_tokens", 1), 1)
        self.rotary_start_pos = 0

    def forward(
        self,
        hidden_states: torch.Tensor,
        attention_mask: Optional[torch.Tensor] = None,
        position_ids: Optional[torch.LongTensor] = None,
        past_key_value: Optional[Tuple[torch.Tensor]] = None,
        output_attentions: bool = False,
        use_cache: bool = False,
        is_padded_inputs: Optional[bool] = True,
        cu_seqlens: Optional[torch.Tensor] = None,
        max_seq_len: Optional[int] = None,
        rope: Optional[torch.Tensor] = None,
    ) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor]]]:

        has_layer_past = past_key_value is not None

        if has_layer_past:
            past_key_value = past_key_value[0]
            past_len = past_key_value[1]
        else:
            past_len = 0

        qkv = self.Wqkv(hidden_states)

        ######################### {1/2}  Remove embeddings that don't get rotary ##########################
        if self.rotary_start_pos > 0:
            ############## FIRST NEW PART ##############
            assert len(qkv.shape) == 3 # (b, s, dim)
            # full_seq_len = qkv.shape[0]
            original_qkv = qkv.clone()
            # no_rotary_qkv = original_qkv[no_rotary_token_mask]
            # qkv = original_qkv[~no_rotary_token_mask]
            qkv_zeros = torch.zeros_like(qkv, device=qkv.device)

            is_contextual_token_mask = torch.arange(qkv.shape[1], device=qkv.device) < self.rotary_start_pos
            qkv = qkv_zeros.where(
                is_contextual_token_mask[None, :, None].expand_as(qkv),
                qkv
            )
        qkv = rearrange(qkv, "... (three h d) -> ... three h d", three=3, d=self.head_dim)
        
        past_key_value = (past_key_value, past_len + qkv.size(1)) if use_cache else None

        assert self.rotary_emb_dim > 0
        qkv = rearrange(qkv, "b s three h d -> b h three s d")
        qkv = self.rotary_emb(qkv, seqlen_offset=past_len)

        qkv = rearrange(qkv, "b h three s d -> b s three h d")

        ########################## {2/2}  Restore embeddings that don't get rotary ##########################
        if self.rotary_start_pos > 0:
            ############## SECOND NEW PART ##############
            # take the original (pre-rotary) QKV for contextual tokens
            original_qkv = original_qkv.reshape(qkv.shape)
            qkv = original_qkv.where(
                is_contextual_token_mask[None, :, None, None, None].expand_as(qkv), 
                qkv
            )

        query, key, value = qkv[:, :, 0], qkv[:, :, 1], qkv[:, :, 2]

        query = query.permute(0, 2, 1, 3)
        key = key.permute(0, 2, 1, 3)
        value = value.permute(0, 2, 1, 3)

        if scaled_dot_product_attention is not None:
            attn_output = F.scaled_dot_product_attention(
                query, key, value, attn_mask=attention_mask, dropout_p=self.drop.p, is_causal=False
            )
        else:
            attention_scores = torch.matmul(query, key.transpose(-1, -2)) / self.norm_factor
            if attention_mask is not None:
                attention_scores = attention_scores + attention_mask

            attentions_probs = F.softmax(attention_scores, dim=-1)
            attentions_probs = self.drop(attentions_probs)

            attn_output = torch.matmul(attentions_probs, value)

        attn_output = rearrange(attn_output.permute(0, 2, 1, 3), "... h d -> ... (h d)")

        attn_output = self.out_proj(attn_output)

        return attn_output


class ContextualNomicBertBlock(ContextualNomicBertPreTrainedModel):
    def __init__(
        self,
        config,
    ):
        super().__init__(config=config)
        self.prenorm = config.prenorm
        self.fused_dropout_add_ln = config.fused_dropout_add_ln

        self.attn = ContextualNomicBertAttention(config)
        activation = (
            F.sigmoid
            if config.activation_function == "glu"
            else (F.silu if config.activation_function == "swiglu" else F.gelu)
        )
        if config.activation_function in ["glu", "swiglu", "geglu"]:
            self.mlp = NomciBertGatedMLP(
                config.n_embd,
                hidden_features=config.n_inner,
                bias1=config.mlp_fc1_bias,
                bias2=config.mlp_fc2_bias,
                activation=activation,
                fused_bias_fc=config.fused_bias_fc,
                norm_layer=getattr(config, "norm_mlp", False),
            )
        else:
            self.mlp = ContextualNomicBertMLP(
                config.n_embd,
                hidden_features=config.n_inner,
                bias1=config.mlp_fc1_bias,
                bias2=config.mlp_fc2_bias,
                activation=activation,
                fused_bias_fc=config.fused_bias_fc,
            )

        self.dropout1 = nn.Dropout(config.resid_pdrop)
        self.norm1 = nn.LayerNorm(config.n_embd, eps=config.layer_norm_epsilon)
        self.norm2 = nn.LayerNorm(config.n_embd, eps=config.layer_norm_epsilon)
        self.dropout2 = nn.Dropout(config.resid_pdrop)

    def forward(
        self,
        hidden_states: torch.Tensor,
        hidden_states2: torch.Tensor,
        residual: Optional[torch.Tensor] = None,
        attention_mask: Optional[torch.Tensor] = None,
        position_ids: Optional[torch.LongTensor] = None,
        past_key_value: Optional[Tuple[torch.Tensor]] = None,
        is_padded_inputs: Optional[bool] = True,
        output_attentions: Optional[bool] = False,
        use_cache: Optional[bool] = False,
        cu_seqlens: Optional[torch.Tensor] = None,
        max_seq_len: Optional[int] = None,
        rope: Optional[torch.Tensor] = None,
    ):
        r"""Pass the input through the encoder layer.
        Args:
            hidden_states: the sequence to the encoder layer (required).
            residual: if postnorm, residual=None, If prenorm, hidden_states = Attn/MLP(LN(residual))
            mixer_subset: for cross-attention only. If not None, will take a subset of x
                before applying the query projection. Useful for e.g., ViT where we only care
                about the CLS token in the last layer.
        """
        if self.prenorm:
            dropped = self.dropout1(hidden_states)
            residual = (dropped + residual) if residual is not None else dropped
            hidden_states = self.norm1(residual.to(dtype=self.norm1.weight.dtype))
            hidden_states = self.attn(
                hidden_states,
                attention_mask=attention_mask,
                is_padded_inputs=is_padded_inputs,
                cu_seqlens=cu_seqlens,
                max_seq_len=max_seq_len,
                rope=rope,
            )

            dropped = self.dropout2(hidden_states)
            residual = (dropped + residual) if residual is not None else dropped
            hidden_states = self.norm2(residual.to(dtype=self.norm2.weight.dtype))
            hidden_states = self.mlp(hidden_states)

            return hidden_states, None, residual
        else:
            assert residual is None
            attn_outputs = self.attn(
                hidden_states,
                attention_mask=attention_mask,
                is_padded_inputs=is_padded_inputs,
                cu_seqlens=cu_seqlens,
                max_seq_len=max_seq_len,
                rope=rope,
            )
            hidden_states = self.norm1((self.dropout1(attn_outputs) + hidden_states).to(dtype=self.norm1.weight.dtype))
            mlp_out = self.mlp(hidden_states)

            hidden_states = self.norm2((self.dropout2(mlp_out) + hidden_states).to(dtype=self.norm2.weight.dtype))
            return hidden_states, None, None


class ContextualNomicBertEncoder(nn.Module):
    def __init__(self, config: GPT2Config):
        super().__init__()
        self.layers = nn.ModuleList([ContextualNomicBertBlock(config) for _ in range(config.n_layer)])
        self.gradient_checkpointing = False
        self.config = config

    def forward(
        self,
        hidden_states: torch.LongTensor = None,
        attention_mask: Optional[torch.Tensor] = None,
        position_ids: Optional[torch.LongTensor] = None,
        past_key_values: Optional[List[torch.FloatTensor]] = None,
        inputs_embeds: Optional[torch.FloatTensor] = None,
        use_cache: Optional[bool] = None,
        output_attentions: Optional[bool] = None,
        output_hidden_states: Optional[bool] = None,
        return_dict: Optional[bool] = None,
        is_padded_inputs: Optional[bool] = True,
        rope: Optional[torch.Tensor] = None,
    ):
        """If subset_mask is not None, we only want output for the subset of the sequence.
        This means that we only compute the last layer output for these tokens.
        subset_mask: (batch, seqlen), dtype=torch.bool
        """
        hidden_states2 = None
        residual = None

        for _, layer in enumerate(self.layers):
            if self.gradient_checkpointing and self.training:

                def create_custom_forward(module):
                    def custom_forward(*inputs):
                        # None for past_key_value
                        return module(*inputs)

                    return custom_forward

                hidden_states, hidden_states2, residual = torch.utils.checkpoint.checkpoint(
                    create_custom_forward(layer),
                    hidden_states,
                    hidden_states2,
                    residual,
                    attention_mask,
                    position_ids,
                    past_key_values,
                    is_padded_inputs,
                    output_attentions,
                    use_cache,
                    None,
                    None,
                    rope,
                    # if you freeze ANY layers, you need `use_reentrant=False`
                    # https://github.com/huggingface/transformers/issues/21381
                    # https://discuss.pytorch.org/t/checkpoint-with-no-grad-requiring-inputs-problem/19117/7
                    use_reentrant=False,
                )

            else:
                hidden_states, hidden_states2, residual = layer(
                    hidden_states,
                    hidden_states2,
                    residual,
                    attention_mask,
                    position_ids,
                    None,
                    is_padded_inputs,
                    output_attentions,
                    use_cache,
                    rope=rope,
                )
        return hidden_states


class ContextualNomicBertPooler(nn.Module):
    def __init__(self, config):
        super().__init__()
        self.dense = nn.Linear(config.n_embd, config.n_embd)
        self.activation = nn.Tanh()

    def forward(self, hidden_states, pool=True):
        # We "pool" the model by simply taking the hidden state corresponding
        # to the first token.
        first_token_tensor = hidden_states[:, 0] if pool else hidden_states
        pooled_output = self.dense(first_token_tensor)
        pooled_output = self.activation(pooled_output)
        return pooled_output


class ContextualNomicBertPredictionHeadTransform(nn.Module):
    def __init__(self, config):
        super().__init__()
        self.dense = nn.Linear(config.n_embd, config.n_embd, bias=config.mlp_fc1_bias)
        approximate = "tanh" if config.activation_function in ["gelu_new", "gelu_fast", "gelu_pytorch_tanh"] else "none"
        if config.activation_function == "swiglu":
            self.transform_act_fn = F.silu
        else:
            self.transform_act_fn = nn.GELU(approximate=approximate)

        self.layer_norm = nn.LayerNorm(config.n_embd, eps=config.layer_norm_epsilon)

    def forward(self, hidden_states: torch.Tensor) -> torch.Tensor:
        hidden_states = self.dense(hidden_states)
        hidden_states = self.transform_act_fn(hidden_states)
        hidden_states = self.layer_norm(hidden_states)

        return hidden_states


class ContextualNomicBertLMPredictionHead(nn.Module):
    def __init__(self, config):
        super().__init__()

        self.transform = ContextualNomicBertPredictionHeadTransform(config)

        self.decoder = nn.Linear(config.n_embd, config.vocab_size, bias=config.mlp_fc1_bias)

    def forward(self, hidden_states):
        hidden_states = self.transform(hidden_states)
        hidden_states = self.decoder(hidden_states)
        return hidden_states


class ContextualNomicBertPreTrainingHeads(nn.Module):
    def __init__(self, config):
        super().__init__()
        self.predictions = ContextualNomicBertLMPredictionHead(config)

    def forward(self, sequence_output):
        prediction_scores = self.predictions(sequence_output)
        return prediction_scores


class ContextualNomicBertModel(ContextualNomicBertPreTrainedModel):
    def __init__(self, config: GPT2Config, add_pooling_layer=True):
        super().__init__(config)
        self.pad_vocab_size_multiple = getattr(config, "pad_vocab_size_multiple", 1)
        if config.vocab_size % self.pad_vocab_size_multiple != 0:
            config.vocab_size += self.pad_vocab_size_multiple - (config.vocab_size % self.pad_vocab_size_multiple)

        assert config.activation_function in [
            "gelu",
            "gelu_new",
            "gelu_fast",
            "gelu_pytorch_tanh",
            "swiglu",
            "geglu",
            "glu",
        ]

        self.embeddings = ContextualNomicBertEmbeddings(config)
        self.emb_drop = nn.Dropout(config.resid_pdrop)
        self.emb_ln = nn.LayerNorm(config.n_embd, eps=config.layer_norm_epsilon)
        self.encoder = ContextualNomicBertEncoder(config)
        self.pooler = ContextualNomicBertPooler(config) if add_pooling_layer else None

        self.apply(partial(_init_weights, initializer_range=config.initializer_range))

    def forward(
        self,
        input_ids=None,
        attention_mask=None,
        position_ids=None,
        token_type_ids=None,
        return_dict=None,
        matryoshka_dim=None,
        inputs_embeds=None,
    ):
        if input_ids is not None and inputs_embeds is not None:
            raise ValueError("You cannot specify both input_ids and inputs_embeds at the same time")
        hidden_states = self.embeddings(
            input_ids=input_ids, 
            position_ids=position_ids, 
            token_type_ids=token_type_ids,
            inputs_embeds=inputs_embeds,
        )
        hidden_states = self.emb_ln(hidden_states)
        hidden_states = self.emb_drop(hidden_states)

        attention_mask = self.get_extended_attention_mask(attention_mask, hidden_states.shape[:-1])
        sequence_output = self.encoder(hidden_states, attention_mask=attention_mask, return_dict=return_dict)

        pooled_output = self.pooler(sequence_output) if self.pooler is not None else None

        if matryoshka_dim:
            sequence_output = sequence_output[:, :matryoshka_dim]

        return BaseModelOutputWithPoolingAndCrossAttentions(
            last_hidden_state=sequence_output,
            pooler_output=pooled_output,
        )


class ContextualNomicBertForPreTraining(ContextualNomicBertPreTrainedModel):
    _tied_weights_keys = ["predictions.decoder.bias", "cls.predictions.decoder.weight"]

    def __init__(self, config: GPT2Config):
        super().__init__(config)

        self.bert = ContextualNomicBertModel(config, add_pooling_layer=getattr(config, "add_pooling_layer", False))
        self.cls = ContextualNomicBertPreTrainingHeads(config)
        self.mlm_loss = nn.CrossEntropyLoss()

        # Initialize weights and apply final processing
        self.apply(partial(_init_weights, initializer_range=config.initializer_range))
        self.tie_weights()

    def tie_weights(self):
        self.cls.predictions.decoder.weight = self.bert.embeddings.word_embeddings.weight

    def forward(
        self,
        input_ids,
        position_ids=None,
        token_type_ids=None,
        attention_mask=None,
        labels=None,
    ):
        """
        If labels are provided, they must be -100 for masked out tokens (as specified in the attention
        mask).
        Outputs:
            if `labels` and `next_sentence_label` are not `None`:
                Outputs the total_loss which is the sum of the masked language modeling loss and the next
                sentence classification loss.
            if `labels` or `next_sentence_label` is `None`:
                Outputs a tuple comprising
                - the masked language modeling logits of shape [batch_size, sequence_length, vocab_size], and
                - the next sentence classification logits of shape [batch_size, 2].
        """
        outputs = self.bert(
            input_ids,
            position_ids=position_ids,
            token_type_ids=token_type_ids,
            attention_mask=attention_mask.bool() if attention_mask is not None else None,
        )
        sequence_output, _ = outputs.last_hidden_state, outputs.pooler_output

        prediction_scores = self.cls(sequence_output)

        total_loss = None
        if labels is not None:
            masked_lm_loss = self.mlm_loss(
                rearrange(prediction_scores, "... v -> (...) v"),
                rearrange(labels, "... -> (...)"),
            )
            total_loss = masked_lm_loss.float()

        return MaskedLMOutput(
            loss=total_loss,
            logits=prediction_scores,
            hidden_states=outputs.hidden_states,
            attentions=None,
        )


class ContextualNomicBertForSequenceClassification(ContextualNomicBertPreTrainedModel):
    def __init__(self, config):
        super().__init__(config)
        self.num_labels = config.num_labels
        self.config = config

        self.bert = ContextualNomicBertModel(config)
        classifier_dropout = getattr(config, "classifier_dropout", config.embd_pdrop)
        self.dropout = nn.Dropout(classifier_dropout)
        self.classifier = nn.Linear(config.n_embd, config.num_labels)

        # Initialize weights and apply final processing
        self.post_init()

    def forward(
        self,
        input_ids: Optional[torch.Tensor] = None,
        attention_mask: Optional[torch.Tensor] = None,
        token_type_ids: Optional[torch.Tensor] = None,
        position_ids: Optional[torch.Tensor] = None,
        head_mask: Optional[torch.Tensor] = None,
        inputs_embeds: Optional[torch.Tensor] = None,
        labels: Optional[torch.Tensor] = None,
        output_attentions: Optional[bool] = None,
        output_hidden_states: Optional[bool] = None,
        return_dict: Optional[bool] = None,
    ):
        r"""
        labels (`torch.LongTensor` of shape `(batch_size,)`, *optional*):
            Labels for computing the sequence classification/regression loss. Indices should be in `[0, ...,
            config.num_labels - 1]`. If `config.num_labels == 1` a regression loss is computed (Mean-Square loss), If
            `config.num_labels > 1` a classification loss is computed (Cross-Entropy).
        """
        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
        outputs = self.bert(
            input_ids,
            position_ids=position_ids,
            token_type_ids=token_type_ids,
            attention_mask=attention_mask.bool() if attention_mask is not None else None,
        )

        pooled_output = outputs[1]

        pooled_output = self.dropout(pooled_output)
        logits = self.classifier(pooled_output)

        loss = None
        if labels is not None:
            if self.config.problem_type is None:
                if self.num_labels == 1:
                    self.config.problem_type = "regression"
                elif self.num_labels > 1 and (labels.dtype == torch.long or labels.dtype == torch.int):
                    self.config.problem_type = "single_label_classification"
                else:
                    self.config.problem_type = "multi_label_classification"

            if self.config.problem_type == "regression":
                loss_fct = nn.MSELoss()
                if self.num_labels == 1:
                    loss = loss_fct(logits.squeeze(), labels.squeeze())
                else:
                    loss = loss_fct(logits, labels)
            elif self.config.problem_type == "single_label_classification":
                loss_fct = nn.CrossEntropyLoss()
                loss = loss_fct(logits.view(-1, self.num_labels), labels.view(-1))
            elif self.config.problem_type == "multi_label_classification":
                loss_fct = nn.BCEWithLogitsLoss()
                loss = loss_fct(logits, labels)
        if not return_dict:
            output = (logits,) + outputs[2:]
            return ((loss,) + output) if loss is not None else output

        return SequenceClassifierOutput(
            loss=loss,
            logits=logits,
            hidden_states=outputs.hidden_states,
            attentions=outputs.attentions,
        )

class ContextualNomicBertForMultipleChoice(ContextualNomicBertPreTrainedModel):
    def __init__(self, config):
        super().__init__(config)

        self.bert = ContextualNomicBertModel(config, add_pooling_layer=True)
        classifier_dropout = (
            getattr(config, "classifier_dropout", config.resid_pdrop)
        )
        self.dropout = nn.Dropout(classifier_dropout)
        self.classifier = nn.Linear(config.hidden_size, 1)

        # Initialize weights and apply final processing
        self.post_init()

    def forward(
        self,
        input_ids: Optional[torch.Tensor] = None,
        attention_mask: Optional[torch.Tensor] = None,
        token_type_ids: Optional[torch.Tensor] = None,
        position_ids: Optional[torch.Tensor] = None,
        head_mask: Optional[torch.Tensor] = None,
        inputs_embeds: Optional[torch.Tensor] = None,
        labels: Optional[torch.Tensor] = None,
        output_attentions: Optional[bool] = None,
        output_hidden_states: Optional[bool] = None,
        return_dict: Optional[bool] = None,
        unpad_inputs: Optional[bool] = None,
    ) -> Union[Tuple[torch.Tensor], MultipleChoiceModelOutput]:
        r"""
        labels (`torch.LongTensor` of shape `(batch_size,)`, *optional*):
            Labels for computing the multiple choice classification loss. Indices should be in `[0, ...,
            num_choices-1]` where `num_choices` is the size of the second dimension of the input tensors. (See
            `input_ids` above)
        """
        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
        num_choices = input_ids.shape[1] if input_ids is not None else inputs_embeds.shape[1]

        input_ids = input_ids.view(-1, input_ids.size(-1)) if input_ids is not None else None
        attention_mask = attention_mask.view(-1, attention_mask.size(-1)) if attention_mask is not None else None
        token_type_ids = token_type_ids.view(-1, token_type_ids.size(-1)) if token_type_ids is not None else None
        position_ids = position_ids.view(-1, position_ids.size(-1)) if position_ids is not None else None
        inputs_embeds = (
            inputs_embeds.view(-1, inputs_embeds.size(-2), inputs_embeds.size(-1))
            if inputs_embeds is not None
            else None
        )

        outputs = self.bert(
            input_ids,
            attention_mask=attention_mask,
            token_type_ids=token_type_ids,
            position_ids=position_ids,
            inputs_embeds=inputs_embeds,
        )

        pooled_output = outputs[1]

        pooled_output = self.dropout(pooled_output)
        logits = self.classifier(pooled_output)
        reshaped_logits = logits.view(-1, num_choices)

        loss = None
        if labels is not None:
            loss_fct = nn.CrossEntropyLoss()
            loss = loss_fct(reshaped_logits, labels)

        if not return_dict:
            output = (reshaped_logits,) + outputs[2:]
            return ((loss,) + output) if loss is not None else output

        return MultipleChoiceModelOutput(
            loss=loss,
            logits=reshaped_logits,
            hidden_states=outputs.hidden_states,
            attentions=outputs.attentions,
        )

class ContextualNomicBertForTokenClassification(ContextualNomicBertPreTrainedModel):
    def __init__(self, config):
        super().__init__(config)
        self.num_labels = config.num_labels

        self.bert = ContextualNomicBertModel(config, add_pooling_layer=False)
        classifier_dropout = (
            getattr(config, "classifier_dropout", config.resid_pdrop)
        )
        self.dropout = nn.Dropout(classifier_dropout)
        self.classifier = nn.Linear(config.hidden_size, config.num_labels)

        # Initialize weights and apply final processing
        self.post_init()

    def forward(
        self,
        input_ids: Optional[torch.Tensor] = None,
        attention_mask: Optional[torch.Tensor] = None,
        token_type_ids: Optional[torch.Tensor] = None,
        position_ids: Optional[torch.Tensor] = None,
        head_mask: Optional[torch.Tensor] = None,
        inputs_embeds: Optional[torch.Tensor] = None,
        labels: Optional[torch.Tensor] = None,
        output_attentions: Optional[bool] = None,
        output_hidden_states: Optional[bool] = None,
        return_dict: Optional[bool] = None,
    ) -> Union[Tuple[torch.Tensor], TokenClassifierOutput]:
        r"""
        labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
            Labels for computing the token classification loss. Indices should be in `[0, ..., config.num_labels - 1]`.
        """
        return_dict = return_dict if return_dict is not None else self.config.use_return_dict

        outputs = self.bert(
            input_ids,
            attention_mask=attention_mask,
            token_type_ids=token_type_ids,
            position_ids=position_ids,
            inputs_embeds=inputs_embeds,
        )

        sequence_output = outputs[0]

        sequence_output = self.dropout(sequence_output)
        logits = self.classifier(sequence_output)

        loss = None
        if labels is not None:
            loss_fct = nn.CrossEntropyLoss()
            loss = loss_fct(logits.view(-1, self.num_labels), labels.view(-1))

        if not return_dict:
            output = (logits,) + outputs[2:]
            return ((loss,) + output) if loss is not None else output

        return TokenClassifierOutput(
            loss=loss,
            logits=logits,
            hidden_states=outputs.hidden_states,
            attentions=outputs.attentions,
        )

class ContextualNomicBertForQuestionAnswering(ContextualNomicBertPreTrainedModel):
    def __init__(self, config):
        super().__init__(config)
        self.num_labels = config.num_labels

        self.bert = ContextualNomicBertModel(config, add_pooling_layer=False)
        self.qa_outputs = nn.Linear(config.hidden_size, config.num_labels)

        # Initialize weights and apply final processing
        self.post_init()

    def forward(
        self,
        input_ids: Optional[torch.Tensor] = None,
        attention_mask: Optional[torch.Tensor] = None,
        token_type_ids: Optional[torch.Tensor] = None,
        position_ids: Optional[torch.Tensor] = None,
        head_mask: Optional[torch.Tensor] = None,
        inputs_embeds: Optional[torch.Tensor] = None,
        start_positions: Optional[torch.Tensor] = None,
        end_positions: Optional[torch.Tensor] = None,
        output_attentions: Optional[bool] = None,
        output_hidden_states: Optional[bool] = None,
        return_dict: Optional[bool] = None,
    ) -> Union[Tuple[torch.Tensor], QuestionAnsweringModelOutput]:
        r"""
        start_positions (`torch.LongTensor` of shape `(batch_size,)`, *optional*):
            Labels for position (index) of the start of the labelled span for computing the token classification loss.
            Positions are clamped to the length of the sequence (`sequence_length`). Position outside of the sequence
            are not taken into account for computing the loss.
        end_positions (`torch.LongTensor` of shape `(batch_size,)`, *optional*):
            Labels for position (index) of the end of the labelled span for computing the token classification loss.
            Positions are clamped to the length of the sequence (`sequence_length`). Position outside of the sequence
            are not taken into account for computing the loss.
        """
        return_dict = return_dict if return_dict is not None else self.config.use_return_dict

        outputs = self.bert(
            input_ids,
            attention_mask=attention_mask,
            token_type_ids=token_type_ids,
            position_ids=position_ids,
            inputs_embeds=inputs_embeds,
        )

        sequence_output = outputs[0]

        logits = self.qa_outputs(sequence_output)
        start_logits, end_logits = logits.split(1, dim=-1)
        start_logits = start_logits.squeeze(-1).contiguous()
        end_logits = end_logits.squeeze(-1).contiguous()

        total_loss = None
        if start_positions is not None and end_positions is not None:
            # If we are on multi-GPU, split add a dimension
            if len(start_positions.size()) > 1:
                start_positions = start_positions.squeeze(-1)
            if len(end_positions.size()) > 1:
                end_positions = end_positions.squeeze(-1)
            # sometimes the start/end positions are outside our model inputs, we ignore these terms
            ignored_index = start_logits.size(1)
            start_positions = start_positions.clamp(0, ignored_index)
            end_positions = end_positions.clamp(0, ignored_index)

            loss_fct = nn.CrossEntropyLoss(ignore_index=ignored_index)
            start_loss = loss_fct(start_logits, start_positions)
            end_loss = loss_fct(end_logits, end_positions)
            total_loss = (start_loss + end_loss) / 2

        if not return_dict:
            output = (start_logits, end_logits) + outputs[2:]
            return ((total_loss,) + output) if total_loss is not None else output

        return QuestionAnsweringModelOutput(
            loss=total_loss,
            start_logits=start_logits,
            end_logits=end_logits,
            hidden_states=outputs.hidden_states,
            attentions=outputs.attentions,
        )

def hf_vit_config_to_vit_config(vit_config: ViTConfig) -> GPT2Config:
    return GPT2Config(
        n_embd=vit_config.hidden_size,
        n_layer=vit_config.num_hidden_layers,
        n_head=vit_config.num_attention_heads,
        n_inner=vit_config.intermediate_size,
        activation_function=vit_config.hidden_act,
        vocab_size=0,  # no vocab since using patches
        n_positions=0,  # No absolute position embedding
        resid_pdrop=0.0,  # No dropout
        embd_pdrop=getattr(vit_config, "dropout", 0.0),
        attn_pdrop=vit_config.attention_probs_dropout_prob,
        layer_norm_epsilon=vit_config.layer_norm_eps,
        initializer_range=vit_config.initializer_range,
        bos_token_id=None,
        eos_token_id=None,
        # These are new arguments not in the original GPT2Config
        drop_path_rate=0.0,
        # Why is there double layer norm??
        prepre_layernom=False,
        layer_scale=False,
        layer_scale_init=None,
        img_size=vit_config.image_size,
        patch_size=vit_config.patch_size,
        num_channels=vit_config.num_channels,
        prenorm=True,
        parallel_block=False,
        parallel_block_tied_norm=False,
        rotary_emb_fraction=0,
        tie_word_embeddings=False,
        fused_dropout_add_ln=True,
        fused_bias_fc=True,
        patch_embed_bias=True,
        use_flash_attn=True,
        qkv_proj_bias=True,
        mlp_fc1_bias=getattr(vit_config, "mlp_fc1_bias", True),
        mlp_fc2_bias=getattr(vit_config, "mlp_fc2_bias", True),
        use_rms_norm=False,
        causal=False,
        hidden_features_scaling_factor=1.0,
        mask_token=False,
        learned_pos_embedding=False,
        patch_dropout=0,
        sinusoidal_pos_embedding=vit_config.model_type == "vit_mae",
    )


class ContextualNomicAttentionPooling(nn.Module):
    def __init__(self, config):
        super().__init__()
        self.embed_dim = config.n_embd
        self.use_flash_attn = config.use_flash_attn
        self.fused_bias_fc = config.fused_bias_fc

        self.num_heads = config.n_head
        self.num_heads_kv = config.num_heads_kv if getattr(config, "num_heads_kv", None) is not None else self.num_heads
        assert self.embed_dim % self.num_heads == 0, "embed_dim must be divisible by num_heads"
        self.head_dim = self.embed_dim // self.num_heads
        # we don't really support mqa / gqa for now
        kv_dim = 2 * self.head_dim * self.num_heads_kv

        self.register_buffer(
            "norm_factor",
            torch.sqrt(torch.tensor(self.head_dim, dtype=torch.float32)).to(torch.get_default_dtype()),
            persistent=False,
        )

        self.Wq = nn.Linear(self.embed_dim, self.embed_dim, bias=config.qkv_proj_bias)
        self.Wkv = nn.Linear(self.embed_dim, kv_dim, bias=config.qkv_proj_bias)

        self.latent = nn.Parameter(torch.zeros(1, 1, self.embed_dim))

        self.out_proj = nn.Linear(self.embed_dim, self.embed_dim, bias=config.qkv_proj_bias)
        self.causal = config.causal
        self.drop = nn.Dropout(config.attn_pdrop)

    def init_weights(self):
        trunc_normal_tf_(self.latent, std=self.embed_dim**-0.5)

    def forward(
        self,
        kv,
        attention_mask=None,
        cu_seqlens_k=None,
        max_seqlen_k=None,
        is_padded_inputs: Optional[bool] = True,
        output_attentions: bool = False,
    ):
        """Implements the multihead softmax attention.
        Arguments
        ---------
            q: The tensor containing the query. (B, Sq, H, D)
            kv: The tensor containing the key and value. (B, Sk, 2, H_k, D)
            causal: if passed, will override self.causal
            cu_seqlens: (batch_size + 1,), dtype torch.int32. The cumulative sequence lengths
                of the sequences in the batch, used to index into q.
            max_seqlen: int. Maximum sequence length in the batch of q.
            cu_seqlens_k: (batch_size + 1,), dtype torch.int32. The cumulative sequence lengths
                of the sequences in the batch, used to index into kv.
            max_seqlen_k: int. Maximum sequence length in the batch of k and v.
        """
        q_latent = self.latent.expand(kv.size(0), -1, -1)
        q = self.Wq(q_latent)
        bsz, q_len, h_size = q.shape
        kv = self.Wkv(kv)
        query = rearrange(q, "... (h d) -> ... h d", d=self.head_dim)
        kv = rearrange(kv, "... (two hkv d) -> ... two hkv d", two=2, d=self.head_dim)

        key, value = kv[:, :, 0], kv[:, :, 1]

        query = query.permute(0, 2, 1, 3)
        key = key.permute(0, 2, 1, 3)
        value = value.permute(0, 2, 1, 3)

        attention_scores = torch.matmul(query, key.transpose(-1, -2)) / self.norm_factor
        if attention_mask is not None:
            attention_scores = attention_scores + attention_mask

        attentions_probs = F.softmax(attention_scores, dim=-1)
        attentions_probs = self.drop(attentions_probs)

        attn_output = torch.matmul(attentions_probs, value)
        attn_output = rearrange(attn_output.permute(0, 2, 1, 3), "... h d -> ... (h d)")

        attn_output = self.out_proj(attn_output)

        return attn_output


class ContextualNomicMultiHeadAttentionPooling(nn.Module):
    def __init__(
        self,
        config,
    ):
        super().__init__()
        self.prenorm = config.prenorm
        self.fused_dropout_add_ln = config.fused_dropout_add_ln

        self.attn = ContextualNomicAttentionPooling(config)
        activation = (
            F.sigmoid
            if config.activation_function == "glu"
            else (F.silu if config.activation_function == "swiglu" else F.gelu)
        )
        if config.activation_function in ["glu", "swiglu", "geglu"]:
            self.mlp = NomciBertGatedMLP(
                config.n_embd,
                hidden_features=config.n_inner,
                bias1=config.mlp_fc1_bias,
                bias2=config.mlp_fc2_bias,
                activation=activation,
                fused_bias_fc=config.fused_bias_fc,
            )
        else:
            self.mlp = ContextualNomicBertMLP(
                config.n_embd,
                hidden_features=config.n_inner,
                bias1=config.mlp_fc1_bias,
                bias2=config.mlp_fc2_bias,
                activation=activation,
                fused_bias_fc=config.fused_bias_fc,
            )

        self.dropout1 = nn.Dropout(config.resid_pdrop)
        self.norm1 = nn.LayerNorm(config.n_embd, eps=config.layer_norm_epsilon)
        self.dropout2 = nn.Dropout(config.resid_pdrop)

    def forward(
        self,
        hidden_states: torch.Tensor,
        attention_mask: Optional[torch.Tensor] = None,
    ):
        r"""Pass the input through the encoder layer.
        Args:
            hidden_states: the sequence to the encoder layer (required).
            residual: if postnorm, residual=None, If prenorm, hidden_states = Attn/MLP(LN(residual))
            mixer_subset: for cross-attention only. If not None, will take a subset of x
                before applying the query projection. Useful for e.g., ViT where we only care
                about the CLS token in the last layer.
        """

        attn_outputs = self.attn(
            hidden_states,
            attention_mask=attention_mask,
        )

        normed = self.norm1(attn_outputs)
        hidden_states = hidden_states + self.mlp(normed)

        return hidden_states




########################################################
########################################################
########################################################
########################################################


from typing import Callable, Dict, Optional, Union, Tuple
import copy
import math
import multiprocessing
import os

import torch
import torch.nn as nn
import transformers


class ContextualModelConfig(transformers.configuration_utils.PretrainedConfig):
    """We create a dummy configuration class that will just set properties
    based on whatever kwargs we pass in.

    When this class is initialized (see experiments.py) we pass in the
    union of all data, model, and training args, all of which should
    get saved to the config json.
    """

    def __init__(self, **kwargs):
        for key, value in kwargs.items():
            try:
                json.dumps(value)
                setattr(self, key, value)
            except TypeError:
                # value was not JSON-serializable, skip
                continue
        super().__init__()


def load_embedder_and_tokenizer(name: str) -> Tuple[
        transformers.PreTrainedModel, 
        transformers.PreTrainedTokenizer
]:
    print("Loading model:", name)
    if name.startswith("nomic") or (name == "bert-base-uncased"):
        model = ContextualNomicBertForPreTraining.from_pretrained(name, trust_remote_code=True).bert
        tokenizer = transformers.AutoTokenizer.from_pretrained(name)
    elif name in ["gtr-base", "gtr_base"]:
        model = transformers.AutoModel.from_pretrained(
            "sentence-transformers/gtr-t5-base"
        ).encoder
        tokenizer = transformers.AutoTokenizer.from_pretrained(
            "sentence-transformers/gtr-t5-base"
        )
    elif name == "pile-t5-base-encoder":
        model = transformers.AutoModel.from_pretrained(
            "EleutherAI/pile-t5-base"
        ).encoder
        tokenizer = transformers.AutoTokenizer.from_pretrained(
            "EleutherAI/pile-t5-base"
        )
        tokenizer.pad_token = tokenizer.eos_token
    elif name == "pile-t5-base-decoder":
        model = transformers.AutoModel.from_pretrained(
            "EleutherAI/pile-t5-base"
        ).decoder
        tokenizer = transformers.AutoTokenizer.from_pretrained(
            "EleutherAI/pile-t5-base"
        )
        tokenizer.pad_token = tokenizer.eos_token
    elif name.startswith("gpt2") or name.startswith("meta-llama") or ("Llama" in name):
        model = transformers.AutoModelForCausalLM.from_pretrained(
            name, 
            # torch_dtype=torch.bfloat16,
            attn_implementation="flash_attention_2",
            low_cpu_mem_usage=True,
            # device_map="auto",
        )
        model.padding_side = "right"
        tokenizer = transformers.AutoTokenizer.from_pretrained(name)
        tokenizer.pad_token = tokenizer.eos_token
        tokenizer.add_eos_token = True
    else:
        model = transformers.AutoModel.from_pretrained(name, trust_remote_code=True)
        tokenizer = transformers.AutoTokenizer.from_pretrained(name)

        # if use_bettertransformer:
        #     from optimum.bettertransformer import BetterTransformer
        #     model = BetterTransformer.transform(model)
    return model, tokenizer


def get_world_size() -> int:
    try:
        return torch.distributed.get_world_size()
    except (RuntimeError, ValueError):
        return 1


def get_rank() -> int:
    try:
        return torch.distributed.get_rank()
    except (RuntimeError, ValueError):
        return 0
    
def gather(t: torch.Tensor) -> torch.Tensor:
    # torch.distributed.nn.all_gather scales by world size since the reduce op is SUM
    # https://github.com/pytorch/pytorch/issues/58005
    # only should use torch.distributed.nn.all_gather if we implement a `local_loss`
    # like: https://github.com/mlfoundations/open_clip/issues/616
    world_size = get_world_size()
    if world_size == 1:
        return t

    if t.ndim == 0:
        t = t.unsqueeze(0)

    gathered = [torch.empty_like(t) for _ in range(world_size)]
    torch.distributed.all_gather(gathered, t)
    gathered[get_rank()] = t
    return torch.cat(gathered, dim=0)


def gather_sum(t: torch.Tensor) -> torch.Tensor:
    # torch.distributed.nn.all_gather scales by world size since the reduce op is SUM
    # https://github.com/pytorch/pytorch/issues/58005
    # only should use torch.distributed.nn.all_gather if we implement a `local_loss`
    # like: https://github.com/mlfoundations/open_clip/issues/616
    world_size = get_world_size()
    if world_size == 1:
        return t

    if t.ndim == 0:
        t = t.unsqueeze(0)

    gathered = [torch.empty_like(t) for _ in range(world_size)]
    torch.distributed.all_gather(gathered, t)
    gathered = torch.stack(gathered, dim=0)
    return gathered.sum(dim=0) # Sum across workers


def get_num_proc() -> int:
    world_size: int = get_world_size()
    try:
        # os.sched_getaffinity respects schedulers, unlike cpu_count(), but it's only available
        # on some Unix platforms, so we support both!
        return len(os.sched_getaffinity(0)) // world_size  # type: ignore[attr-defined]
    except AttributeError:
        return multiprocessing.cpu_count() // world_size


def torch_main_worker_finish_first(func: Callable):
    def wrapper(*args, **kwargs):
        # Get local rank (need to support non-DDP).
        try:
            local_rank = torch.distributed.get_rank()
            ddp_enabled = True
        except (RuntimeError, ValueError):
            local_rank = -1
            ddp_enabled = False
        is_main_worker = local_rank <= 0
        # Run on main worker first.
        if is_main_worker:
            result = func(*args, **kwargs)
        # Then everyone waits.
        if ddp_enabled:
            torch.distributed.barrier()
        # Run on other workers now.
        if not is_main_worker:
            result = func(*args, **kwargs)
        # Now everyone waits again.
        if ddp_enabled:
            torch.distributed.barrier()
        return result

    return wrapper


def print0(*args, **kwargs) -> None:
    if get_rank() == 0:
        print(*args, **kwargs)


def verify_ddp_weights_equal(model: torch.nn.Module, atol: float = 1e-5) -> None:
    if hasattr(model, "module"):
        model = model.module
    
    world_size = get_world_size()

    if world_size > 8:
        print0(f"[verify_ddp_weights_equal] Skipping with world_size={world_size} ⚠️")
        return

    for name, param in model.named_parameters():
        if param is None: continue
        if param.grad is None: 
            print0(f"[verify_ddp_weights_equal] Skipping param [{name}] with no grad")
            continue
        gathered_param = gather(param).reshape((world_size, -1))
        absolute_diffs = (gathered_param[None, 0, :] - gathered_param).abs()
        rank_params_eq = (absolute_diffs < atol).all()
        assert rank_params_eq, f"❌ param [{name}] not equal - got max_absolute_diff={absolute_diffs.max()}"
        ###################################################################################################################
        gathered_param_grad = gather(param.grad).reshape((world_size, -1))
        absolute_grad_diffs = (gathered_param_grad[None, 0, :] - gathered_param_grad).abs()
        rank_grad_params_eq = (absolute_grad_diffs < atol).all()
        assert rank_grad_params_eq, f"❌ param [{name}] grad not equal - got max_absolute_diff={absolute_grad_diffs.max()}"
        ###################################################################################################################
        
    
    print0("[verify_ddp_weights_equal] Verified DDP parameter correctness ✅")
    


def mean_pool_3d(
    hidden_states: torch.Tensor, attention_mask: torch.Tensor
) -> torch.Tensor:
    B, T, S, D = hidden_states.shape
    unmasked_outputs = hidden_states * attention_mask[..., None]
    pooled_outputs = unmasked_outputs.sum(dim=2) / (attention_mask.sum(dim=2)[..., None] + 1e-9)

    # fix for gradient flow: fill empty rows with the mean of the rest of the sequence
    sequence_means = (
        hidden_states.reshape((B, S * T, D))
            .mean(dim=1, keepdim=True)
            .expand(-1, T, -1)
    )
    pooled_outputs = pooled_outputs.where(
        (attention_mask.sum(dim=2)[..., None] > 0), 
        sequence_means
    )
    assert pooled_outputs.shape == (B, T, D)

    return pooled_outputs

def mean_pool(
    hidden_states: torch.Tensor, attention_mask: torch.Tensor
) -> torch.Tensor:
    B, _S, D = hidden_states.shape
    unmasked_outputs = hidden_states * attention_mask[..., None]
    pooled_outputs = unmasked_outputs.sum(dim=1) / (attention_mask.sum(dim=1)[:, None] + 1e-20)
    
    assert pooled_outputs.shape == (B, D)
    return pooled_outputs


def mean_pool_weighted(
    hidden_states: torch.Tensor, attention_mask: torch.Tensor
) -> torch.Tensor:
    B, _S, D = hidden_states.shape
    attention_mask *= attention_mask.cumsum(dim=1) # [0,1,1,1,0,0] -> [0,1,2,3,0,0]
    s = torch.sum(hidden_states * attention_mask.unsqueeze(-1).float(), dim=1)
    d = attention_mask.sum(dim=1, keepdim=True).float()
    return s / d


def slice_sparse_tensor_rows(t: torch.sparse.Tensor, min_row: int, max_row: int) -> torch.sparse.Tensor:
    assert min_row < max_row, f"can't slice from row {min_row} to {max_row}"
    t = t.coalesce()
    row_idxs = t.indices()[0]
    index_mask = (min_row <= row_idxs) & (row_idxs < max_row)

    num_rows = (max_row - min_row)
    num_cols = t.shape[1]

    idxs = t.indices()[:, index_mask]
    vals = t.values()[index_mask]
    return torch.sparse_coo_tensor(idxs, vals, size=(num_rows, num_cols)).coalesce()


def slice_tensor_rows(t: torch.Tensor, min_row: int, max_row: int) -> torch.Tensor:
    if t.is_sparse:
        return slice_sparse_tensor_rows(t=t, min_row=min_row, max_row=max_row)
    else:
        return t[min_row:max_row]


@torch.no_grad
def maxsim(
    X: torch.Tensor, y: torch.Tensor, 
    maximize: bool, chunk_size: int = 8_000,
    debug_mem_usage: bool = False) -> torch.Tensor:
    device = X.device
    n_samples = X.shape[0]

    max_sim_v = torch.zeros(n_samples, device=device, dtype=X.dtype)
    max_sim_i = torch.zeros(n_samples, device=device, dtype=torch.int64)

    # TODO: Implement faster max (without going to dense tensors).
    # TODO: Use multiple GPUs.
    rank = get_rank()
    world_size = get_world_size()

    worker_worklist_size = int(math.ceil(n_samples / world_size))
    splits_start_idx = worker_worklist_size * rank
    splits_end_idx = worker_worklist_size * (rank + 1)

    for i in range(splits_start_idx, splits_end_idx, chunk_size):
        start, end = i, min(i + chunk_size, n_samples)
        sub_x = slice_tensor_rows(X, start, end)
        if debug_mem_usage: print(f"[maxsim] step {i} cuda mem free/total = {torch.cuda.mem_get_info()}")
        if debug_mem_usage: print("[maxsim] sub_x.shape:", sub_x.shape, "//", "y.shape:", y.shape)
        sub_sim = sub_x @ y # TODO – Implement sparse max here to save mem!
        sub_sim = sub_sim
        if maximize:
            sub_max_sim_v, sub_max_sim_i = sub_sim.to_dense().max(dim=-1)
        else:
            sub_max_sim_v, sub_max_sim_i = sub_sim.to_dense().min(dim=-1)
        del sub_sim
        del sub_x
        torch.cuda.empty_cache() # needs to happen after maxsim for some reason.
        max_sim_v[start: end] = sub_max_sim_v
        max_sim_i[start: end] = sub_max_sim_i
    
    # gather
    max_sim_v = gather_sum(max_sim_v)
    max_sim_i = gather_sum(max_sim_i)
    k = y.shape[1]

    assert max_sim_v.shape == (n_samples,)
    assert max_sim_i.shape == (n_samples,)
    assert max_sim_i.min() >= 0
    assert max_sim_i.max() <= k

    return max_sim_v, max_sim_i


def forward_batched(
        model: torch.nn.Module,
        input_ids: torch.Tensor,
        attention_mask: torch.Tensor,
        batch_size: int,
        dataset_input_ids: Optional[torch.Tensor] = None,
        dataset_attention_mask: Optional[torch.Tensor] = None,
        **second_stage_model_kwargs,
) -> torch.Tensor:
    if hasattr(model, "module"):
        model = model.module
    
    if hasattr(model, "first_stage_model"):
        # Support pooling over 3D dataset_input_ids inputs.
        if len(dataset_input_ids.shape) == 2:
            dataset_input_ids = dataset_input_ids[None]
            dataset_attention_mask = dataset_attention_mask[None]

        dataset_embeddings = []
        for j in range(len(dataset_input_ids)):
            i = 0
            dataset_embeddings_batch = []
            while i < dataset_input_ids.shape[1]:
                dataset_embeddings_batch.append(
                    model.first_stage_model(
                        input_ids=dataset_input_ids[j][i:i+batch_size],
                        attention_mask=dataset_attention_mask[j][i:i+batch_size],
                    )
                )
                i += batch_size
            dataset_embeddings.append(
                torch.cat(dataset_embeddings_batch, dim=0)
            )
       
        # Automatically pool over 3D dataset_input_ids.
        dataset_embeddings = torch.stack(dataset_embeddings, dim=0).mean(dim=0)

        j = 0
        outputs = []
        while j < len(input_ids):
            outputs.append(
                model.second_stage_model(
                    input_ids=input_ids[j:j+batch_size],
                    attention_mask=attention_mask[j:j+batch_size],
                    dataset_embeddings=dataset_embeddings,
                    **second_stage_model_kwargs,
                )
            )
            j += batch_size
        return torch.cat(outputs, dim=0)

    else:
        i = 0
        outputs = []
        while i < len(input_ids):
            outputs.append(
                model(
                    input_ids=input_ids[i:i+batch_size],
                    attention_mask=attention_mask[i:i+batch_size],
                    **second_stage_model_kwargs,
                )
            )
            i += batch_size
        return torch.cat(outputs, dim=0)


def last_token_pool(hidden_state: torch.Tensor, attention_mask: torch.Tensor) -> torch.Tensor:
    # https://github.com/ContextualAI/gritlm/blob/main/gritlm/gritlm.py#L190
    b, n, d = hidden_state.size()
    # Get the last `1` in the attention mask of each item
    # Often it is just `gather_indices = torch.argmin(attention_mask, 1, keepdim=False) - 1`
    # except when 1) There's all 1's 2) There's 0's before the 1's
    reversed_mask = torch.flip(attention_mask, dims=(1,))
    argmax_reverse = torch.argmax(reversed_mask, dim=1, keepdim=False)
    gather_indices = attention_mask.size(1) - argmax_reverse - 1
    # If there are empty sequences, where the index would become -1 it will crash so set them to 0
    gather_indices = torch.clamp(gather_indices, min=0)
    # Turn indices from shape [b] -> [b, 1, d]
    gather_indices = gather_indices.unsqueeze(-1).repeat(1, d)
    gather_indices = gather_indices.unsqueeze(1)
    assert gather_indices.shape == (b, 1, d)
    # Gather along the seq len: [b, n, d] -> [b, d]
    # Actually no need for the attention mask as we gather the last token where attn_mask=1 but
    # as some indices (which shouldn't be attended to) may be 0 due to clamp, use mask to ignore them again
    input_mask_expanded = attention_mask.unsqueeze(-1).expand((b, n, d)).float()
    return torch.gather(hidden_state * input_mask_expanded, 1, gather_indices).squeeze(dim=1)

def print0(*args, **kwargs) -> None:
    if get_rank() == 0:
        print(*args, **kwargs)


def limit_layers(model: transformers.PreTrainedModel, n_layers: int) -> None:
    if hasattr(model, 'transformer'):
        if hasattr(model.transformer, 'h'):
            # gpt2
            model.transformer.h = model.transformer.h[:n_layers]
        else:
            model.transformer.layer = model.transformer.layer[:n_layers]
    elif hasattr(model, 'encoder'):
        if hasattr(model.encoder, 'layers'):
            model.encoder.layers = model.encoder.layers[:n_layers]
        else:
            model.encoder.layer = model.encoder.layer[:n_layers]
    else:
        raise RuntimeError(f"unknown how to limit layers of model {type(model)}")
    


def disable_dropout(model: torch.nn.Module):
    dropout_modules = [m for m in model.modules() if isinstance(m, torch.nn.Dropout)]
    for m in dropout_modules:
        m.p = 0.0
    print0(
        f"Disabled {len(dropout_modules)} dropout modules from model type {type(model)}"
    )


def disable_causality(model: torch.nn.Module):
    disabled_modules = 0
    for m in model.modules():
        if hasattr(m, "is_causal"):
            m.is_causal = False
            disabled_modules += 1
    print0(
        f"Set is_causal=False in {disabled_modules} modules from model type {type(model)}"
    )


class ContextualModelMixin(nn.Module):
    @property
    def num_corpus_tokens(self) -> int:
        return self.transductive_corpus_size * self.transductive_tokens_per_document

    def contextual_init(self):
        self.n_soft_prompt = 8
        self.prompt_projection = torch.nn.Sequential(
            torch.nn.Linear(self.hidden_size, self.hidden_size),
            torch.nn.ReLU(),
            torch.nn.Linear(self.hidden_size, self.hidden_size * self.n_soft_prompt)
        )
        self.transductive_corpus_size = vars(self.config).get("transductive_corpus_size", 1)
        self.transductive_tokens_per_document = vars(self.config).get("transductive_tokens_per_document", 1)
        self.randomize_dataset_sequence_order = True
        self.sequence_dropout_prob = vars(self.config).get("transductive_sequence_dropout_prob", 0.0)
        if self.sequence_dropout_prob > 0.0:
            self.sequence_dropout_null_embedding = torch.nn.Parameter(
                torch.randn(self.hidden_size) * 0.01,
                requires_grad = True
            )       
        self.output_projection = torch.nn.Sequential(
            torch.nn.Linear(self.hidden_size, self.hidden_size),
            torch.nn.ReLU(),
            torch.nn.Linear(self.hidden_size, self.hidden_size)
        )

    def _prepare_dataset_embeddings(
            self, 
            input_ids: torch.Tensor, dataset_embeddings: torch.Tensor,
            null_dataset_embedding: bool = False,
        ) -> torch.Tensor:
        if not isinstance(dataset_embeddings, torch.Tensor):
            dataset_embeddings = torch.tensor(dataset_embeddings)

        if len(dataset_embeddings.shape) == 2:
            # Auto-expand for a batch.
            dataset_embeddings = dataset_embeddings[None, :, :] # (b, d) -> (1, b, d)
        dataset_embeddings = dataset_embeddings.to(input_ids.device)

        if len(dataset_embeddings.shape) < 3:
            raise ValueError(f"dataset_embeddings must have at least 3 dimensions, got {dataset_embeddings.shape}")
    
        batch_size = input_ids.shape[0]
        if (self.transductive_tokens_per_document > 1):
            if self.training:
                # Choose N random documents to fill our context window with.
                # This logic is a little confusing but allows us to sample a
                # different batch *per-document*
                assert dataset_embeddings.shape[1] == self.transductive_tokens_per_document
                R = torch.randint(
                    low=0, 
                    high=len(dataset_embeddings), 
                    size=(batch_size, self.config.transductive_corpus_size), 
                    device=dataset_embeddings.device
                )
                # TODO make this deterministic somehow for evaluation?
                dataset_embeddings = dataset_embeddings[R].reshape((batch_size, self.num_corpus_tokens, self.hidden_size))
            else:
                dataset_embeddings = dataset_embeddings.reshape((1, self.num_corpus_tokens, self.hidden_size))


        if dataset_embeddings.shape[1] < self.num_corpus_tokens:
            raise ValueError(f"dataset_embeddings must have at least {self.num_corpus_tokens} tokens, got {dataset_embeddings.shape[1]}")
        elif dataset_embeddings.shape[1] > self.num_corpus_tokens:
            # If too many dataset embeddings are passed in, just take the first N until
            # we have the proper number.
            dataset_embeddings = dataset_embeddings[:, :self.num_corpus_tokens, :]
        
        _, corpus_size, _hidden_size = dataset_embeddings.shape
        if _ == 1:
            # Auto-expand for a batch.
            dataset_embeddings = dataset_embeddings.expand((batch_size, -1, -1))

        if self.training and self.sequence_dropout_prob > 0.0:
            sequence_dropout_mask = (
                torch.rand((batch_size, corpus_size), device=dataset_embeddings.device) < self.sequence_dropout_prob
            )
            null_embeddings = self.sequence_dropout_null_embedding[None, None].expand(batch_size, corpus_size, -1)
            dataset_embeddings = torch.where(
                sequence_dropout_mask[..., None], null_embeddings, dataset_embeddings
            )
        elif null_dataset_embedding:
            null_embeddings = self.sequence_dropout_null_embedding[None, None].expand(batch_size, corpus_size, -1)
            dataset_embeddings = null_embeddings
        
        # backbone_max_seq_length = self.backbone.config.max_trained_positions
        # assert batch_size + (2 * self.n_soft_prompt + corpus_size) <= backbone_max_seq_length, "too many hard negatives for backbone model"
        soft_prompt = torch.ones((1, self.hidden_size), device=dataset_embeddings.device, dtype=dataset_embeddings.dtype)
        soft_prompt = self.prompt_projection(soft_prompt).reshape((1, self.n_soft_prompt, self.hidden_size))
        soft_prompt = soft_prompt.expand((len(dataset_embeddings), -1, -1)) # -> (b, 4+b, d) # soft_prompt.repeat((len(input_ids), 1, 1))  
        soft_prompt = torch.cat((dataset_embeddings, soft_prompt), dim=1)

        # print(f"[ContextualModelMixin] soft_prompt.shape = {soft_prompt.shape}")

        if self.training and self.randomize_dataset_sequence_order:
            randomized_order = torch.stack(
                [
                    torch.cat(
                        (
                            torch.randperm(corpus_size, device=soft_prompt.device), 
                            torch.arange(self.n_soft_prompt, device=soft_prompt.device) + corpus_size
                        ), dim=0) 
                        for _ in range(batch_size)])
            randomized_order = randomized_order.to(soft_prompt.device)
            soft_prompt = soft_prompt.gather(1, randomized_order[..., None].expand_as(soft_prompt))
        
        return soft_prompt

class BiEncoder(transformers.PreTrainedModel):
    embedder: transformers.PreTrainedModel
    def __init__(
            self, 
            config, #: transformers.PreTrainedConfig, 
        ):
        super().__init__(config=config)
        embedder, _ = load_embedder_and_tokenizer(
            config.embedder,
        )

        if config.limit_layers:
            print0(f"Limiting layers to {config.limit_layers}")
            limit_layers(embedder, config.limit_layers)
    
        self.embedder = embedder
        # if ("t5" in embedder.config.model_type):
        #     print0(f"using torch.compile() on embedder of type `{embedder.config.model_type}`")
        #     self.embedder = torch.compile(self.embedder) 
        self.hidden_size = self.embedder.config.hidden_size
        # Allow pooling to multiple tokens per document
        self.transductive_tokens_per_document = vars(self.config).get("transductive_tokens_per_document", 1)
        self.mlp = torch.nn.Sequential(
            torch.nn.Linear(self.hidden_size, self.hidden_size),
            torch.nn.GELU(),
            torch.nn.Linear(self.hidden_size, self.config.embedding_output_dim or self.hidden_size),
        )
        self.temp = config.logit_scale

        if config.disable_dropout:
            disable_dropout(self)
        self.pooling_strategy = vars(config).get("pooling_strategy", "mean")

    def forward(
            self, 
            input_ids: torch.Tensor,
            attention_mask: torch.Tensor,
            dataset_input_ids: Optional[torch.Tensor] = None,
            dataset_attention_mask: Optional[torch.Tensor] = None,
            token_type_ids = None,
            output_hidden_states: bool = False,
        ) -> torch.Tensor:
        """
        query_embedding (float torch.Tensor) - shape (batch_size, embedding_dim)
        document_embeddings (float torch.Tensor) - shape (corpus_size, embedding_dim)
            where the corpus_size >= batch_size and is structured like this:
                [d1, d2, d3, hn1_1, hn1_2, hn2_1, hn2_2, hn3_1, hn3_2]
                for a corpus with three documents and two hard negatives per document
        """
        del token_type_ids

        outputs = (
            self.embedder(
                input_ids=input_ids,
                attention_mask=attention_mask,
            ).last_hidden_state
        )

        if self.transductive_tokens_per_document > 1:
            document_embeddings = None
            batch_size, seq_length, output_dim = outputs.shape

            if seq_length % self.transductive_tokens_per_document != 0:
                # Pad to nearest multiple
                n_extra_embeds = self.transductive_tokens_per_document - (seq_length % self.transductive_tokens_per_document)
                outputs = torch.cat(
                    (outputs, torch.zeros((batch_size, n_extra_embeds, output_dim), device=outputs.device)),
                    dim=1
                )
                attention_mask = torch.cat(
                    (attention_mask, torch.zeros((batch_size, n_extra_embeds), device=attention_mask.device)),
                    dim=1
                )
                seq_length += n_extra_embeds
                print(f"Added {n_extra_embeds} padding tokens to input_ids and attention_mask")
            
            # print("ftransductive_tokens_per_document {self.transductive_tokens_per_document} outputs.shape =", outputs.shape)

            outputs = outputs.reshape(
                (batch_size,  self.transductive_tokens_per_document, seq_length // self.transductive_tokens_per_document, output_dim)
            )

            attention_mask = attention_mask.reshape((batch_size, self.transductive_tokens_per_document, -1))
            document_embeddings = mean_pool_3d(outputs, attention_mask)
            
            document_embeddings = document_embeddings.reshape((batch_size, self.transductive_tokens_per_document, output_dim))
        else:
            if self.pooling_strategy == "mean":
                document_embeddings = mean_pool(outputs, attention_mask)
            else:
                document_embeddings = document_embeddings.max(dim=1)
        output = self.mlp(document_embeddings)

        if output_hidden_states:
            return {
                "hidden_states": outputs,
                "pooled": output,
            }
        else:
            return output


class DatasetConditionedAutoregressive(transformers.PreTrainedModel, ContextualModelMixin):
    def __init__(
            self, 
            config,
            dataset_backbone: transformers.PreTrainedModel,
            first_stage_hidden_size: int,
        ):
        super().__init__(config=config)
        self.backbone = dataset_backbone
        self.backbone_hidden_size = self.backbone.config.hidden_size
        self.hidden_size = first_stage_hidden_size # Input token size
        self.contextual_init()
        disable_causality(self.backbone)
        
        self.input_ln = torch.nn.LayerNorm(
            self.backbone_hidden_size, 
            eps=1e-5
        )
        
        # Override contextual init
        self.output_projection = torch.nn.Sequential(
            torch.nn.Linear(self.backbone_hidden_size, self.backbone_hidden_size),
            torch.nn.ReLU(),
            torch.nn.Linear(self.backbone_hidden_size, self.backbone_hidden_size)
        )
        self._shift_rotary_embedding()
                
    @property
    def num_corpus_tokens(self) -> int:
        return self.config.transductive_corpus_size * self.transductive_tokens_per_document

    @property
    def corpus_token_ratio(self) -> float:
        # How many tokens from the first stage make one token in the second
        # stage?
        return self.backbone_hidden_size / self.hidden_size
    
    def corpus_token_pad_size(self, n_tokens: int) -> int:
        return self.hidden_size % self.backbone_hidden_size
    
    def _shift_rotary_embedding(self) -> None:
        disable_transductive_rotary_embedding = vars(self.config).get("disable_transductive_rotary_embedding", True)
        # TODO: Can we do this for LLAMA?
        print("Warning: Positional embedding disabling not implemented for LLAMA.")
    
    def forward(
            self, 
            input_ids: torch.Tensor,
            attention_mask: torch.Tensor,
            dataset_embeddings: torch.Tensor,
            output_hidden_states: bool = False,
            null_dataset_embedding: bool = False,
        ) -> torch.Tensor:
        soft_prompt = self._prepare_dataset_embeddings(
            input_ids=input_ids,
            dataset_embeddings=dataset_embeddings,
            null_dataset_embedding=null_dataset_embedding,
        )
        
        # Reshape for this model.
        # print("[DatasetConditionedAutoregressive] 1 -> soft_prompt.shape =", soft_prompt.shape)
        num_soft_elements = torch.prod(torch.tensor(soft_prompt.shape[1:])).item()
        soft_prompt = soft_prompt.reshape((soft_prompt.shape[0], num_soft_elements))
        num_padding_elements = self.backbone_hidden_size - (num_soft_elements % self.backbone_hidden_size)
        padding = torch.ones((soft_prompt.shape[0], num_padding_elements), device=soft_prompt.device)
        soft_prompt = torch.cat((soft_prompt, padding), dim=1)
        soft_prompt = soft_prompt.reshape(
            (soft_prompt.shape[0], -1, self.backbone_hidden_size)
        )
        soft_prompt = self.input_ln(soft_prompt)
        # print("[DatasetConditionedAutoregressive] 2 -> soft_prompt.shape =", soft_prompt.shape)

        backbone_attention_mask = torch.ones(
            soft_prompt.shape[0:2],
            dtype=torch.long,
            device=soft_prompt.device,
        )
        token_embeddings = self.backbone.get_input_embeddings()
        inputs_embeds = token_embeddings(input_ids) # (b, s) -> (b, s, d)
        # print("[2] inputs_embeds.shape =", inputs_embeds.shape)
        inputs_embeds = torch.cat((soft_prompt, inputs_embeds), dim=1) # (v, 4+b+s, d)
        # print("[3.a] inputs_embeds.shape =", inputs_embeds.shape)
        input_attention_mask = torch.cat((backbone_attention_mask, attention_mask), dim=1)
        # print("[3.b] attention_mask.shape =", attention_mask.shape)

        output = self.backbone(
            inputs_embeds=inputs_embeds,
            attention_mask=input_attention_mask,
            output_hidden_states=True,
        ) # (1, 4 + b + s, d)
        # trim soft prompt
        last_hidden_state = output.hidden_states[-1]
        n_soft_prompt_tokens = soft_prompt.shape[1]

        output_vectors = last_hidden_state[:, n_soft_prompt_tokens:, :]
        output_attention_mask = input_attention_mask[:, n_soft_prompt_tokens:]

        # Take last token position
        if vars(self.config).get("pooling_strategy") == "last_token":
            output_pooled = last_token_pool(output_vectors, output_attention_mask)
        elif vars(self.config).get("pooling_strategy") == "mean":
            output_pooled = mean_pool(output_vectors, output_attention_mask)
        else:
            output_pooled = mean_pool_weighted(output_vectors, output_attention_mask)

        # average with original vectors
        # TODO: Argparse for pooling strategy.
        output = self.output_projection(output_pooled) # (b, 2d) -> (b, d)

        if output_hidden_states:
            return {
                "hidden_states": output_vectors,
                "pooled": output,
            }
        else:
            return output


class DatasetConditionedBiencoder(transformers.PreTrainedModel, ContextualModelMixin):
    def __init__(
            self, 
            config,
            dataset_backbone: transformers.PreTrainedModel,
        ):
        super().__init__(config=config)
        self.backbone = dataset_backbone
        self.hidden_size = self.backbone.config.hidden_size
        self.hidden_size = dataset_backbone.config.hidden_size
        # self.input_ln = torch.nn.LayerNorm(
        #     self.hidden_size, 
        #     eps=self.backbone.config.layer_norm_epsilon
        # )
        self.contextual_init()
        self._shift_rotary_embedding()
                
    @property
    def num_corpus_tokens(self) -> int:
        return self.config.transductive_corpus_size * self.transductive_tokens_per_document
    
    def _shift_rotary_embedding(self) -> None:
        disable_transductive_rotary_embedding = vars(self.config).get("disable_transductive_rotary_embedding", True)
        if self.backbone.config.model_type.startswith("nomic") and disable_transductive_rotary_embedding:
            # We only want to apply positional embeddings to the
            # *text* portion of the backbone network.
            self.backbone.config.rotary_start_pos = 0.0
            rotary_disabled = 0

            rotary_start_pos = self.num_corpus_tokens
            for module in self.backbone.modules():
                if hasattr(module, "rotary_emb_dim"):
                    print(f"editing module", type(module))
                    module.rotary_start_pos = rotary_start_pos
                    rotary_disabled += 1
            print0(f"modified {rotary_disabled} rotary modules – set rotary_start_pos to {rotary_start_pos}")
    
    def forward(
            self, 
            input_ids: torch.Tensor,
            attention_mask: torch.Tensor,
            dataset_embeddings: torch.Tensor,
            output_hidden_states: bool = False,
            null_dataset_embedding: bool = False,
        ) -> torch.Tensor:
        # print(f"[DatasetConditionedBiencoder - 0] input_ids.shape => {input_ids.shape} // dataset_embeddings.shape =", dataset_embeddings.shape)
        soft_prompt = self._prepare_dataset_embeddings(
            input_ids=input_ids,
            dataset_embeddings=dataset_embeddings,
            null_dataset_embedding=null_dataset_embedding,
        )
        # print(f"[DatasetConditionedBiencoder - 1] soft_prompt.shape => {soft_prompt.shape}")
        backbone_attention_mask = torch.ones(
            soft_prompt.shape[0:2],
            dtype=torch.long,
            device=soft_prompt.device,
        )
        inputs_embeds = self.backbone.embeddings(input_ids) # (b, s) -> (b, s, d)
        # print("[2] inputs_embeds.shape =", inputs_embeds.shape)
        inputs_embeds = torch.cat((soft_prompt, inputs_embeds), dim=1) # (v, 4+b+s, d)
        # print("[3.a] inputs_embeds.shape =", inputs_embeds.shape)
        attention_mask = torch.cat((backbone_attention_mask, attention_mask), dim=1)
        # print("[3.b] attention_mask.shape =", attention_mask.shape)
        output = self.backbone(
            inputs_embeds=inputs_embeds,
            attention_mask=attention_mask,
        ) # (1, 4 + b + s, d)
        # trim soft prompt
        output_vectors = output.last_hidden_state

        # use only these tokens
        n_soft_prompt_tokens = soft_prompt.shape[1]
        # print("n_soft_prompt_tokens =", n_soft_prompt_tokens)

        output_vectors = output.last_hidden_state[:, n_soft_prompt_tokens:, :]
        output_attention_mask = attention_mask[:, n_soft_prompt_tokens:]

        # print("pooling output_vectors.shape =", output_vectors.shape, "and output_attention_mask.shape =", output_attention_mask.shape)
        output_pooled = mean_pool(output_vectors, output_attention_mask)

        # average with original vectors
        # TODO: Argparse for pooling strategy.
        # output_vectors = torch.cat((soft_prompt_pooled, output_pooled), dim=1) # (b, d) + (b, d) -> (b, 2d)
        # print("output_pooled.shape =", output_pooled.shape)
        output = self.output_projection(output_pooled) # (b, 2d) -> (b, d)

        # print("returning output.shape =", output.shape)

        if output_hidden_states:
            return {
                "hidden_states": output_vectors,
                "pooled": output,
            }
        else:
            return output


class DatasetPrefixBiencoder(transformers.PreTrainedModel, ContextualModelMixin):
    def __init__(
            self, 
            config, #: transformers.PreTrainedConfig, 
            embedder: transformers.PreTrainedModel, 
        ):
        super().__init__(config=config)
        self.embedder = embedder
        self.hidden_size = self.embedder.config.hidden_size
        self.contextual_init()
    
    def forward(
            self, 
            input_ids: torch.Tensor,
            attention_mask: torch.Tensor,
            dataset_input_ids: torch.Tensor,
            dataset_attention_mask: torch.Tensor,
            output_hidden_states: bool = False,
        ) -> torch.Tensor:
        R = torch.randint(low=0, high=len(dataset_input_ids), size=(len(input_ids),), device=dataset_input_ids.device)
        
        dataset_input_ids = dataset_input_ids[R]
        input_ids = torch.cat((dataset_input_ids, input_ids), dim=1)

        dataset_attention_mask = torch.ones_like(dataset_attention_mask, device=dataset_attention_mask.device)
        input_attention_mask = torch.cat((dataset_attention_mask, attention_mask), dim=1)
        output_attention_mask = torch.cat(
            (torch.zeros_like(dataset_input_ids), attention_mask), dim=1
        )

        output = self.embedder(
            input_ids=input_ids,
            attention_mask=input_attention_mask,
        ) 
        
        output_vectors = output.last_hidden_state
        output_pooled = mean_pool(output_vectors, output_attention_mask)
        output = self.output_projection(output_pooled) # (b, 2d) -> (b, d)

        if output_hidden_states:
            S_d = dataset_attention_mask.shape[1]
            output_vectors = output_vectors[:, S_d:, :]
            return {
                "hidden_states": output_vectors,
                "pooled": output,
            }
        else:
            return output


class ContextualDocumentEmbeddingTransformer(transformers.PreTrainedModel):
    config_class = ContextualModelConfig
    embedder: transformers.PreTrainedModel
    dataset_backbone: transformers.PreTrainedModel
    def __init__(
            self, 
            config,
        ):
        super().__init__(config=config)
        dataset_backbone, _ = load_embedder_and_tokenizer(
            vars(config).get("dataset_backbone", config.embedder)
        )

        if config.limit_layers:
            print0(f"Limiting layers to {config.limit_layers}")
            limit_layers(dataset_backbone, config.limit_layers)
        
        biencoder_config = copy.deepcopy(config)
        biencoder_config.embedding_output_dim = None
        biencoder_config.limit_layers = vars(self.config).get("limit_layers_first_stage", None)
        self.first_stage_model = BiEncoder(
            config=biencoder_config,
        )

        if vars(config).get("autoregressive_backbone", False):
            self.second_stage_model = DatasetConditionedAutoregressive(
                config=config,
                dataset_backbone=dataset_backbone,
                first_stage_hidden_size=self.first_stage_model.hidden_size,
            )
        else:
            self.second_stage_model = DatasetConditionedBiencoder(
                config=config,
                dataset_backbone=dataset_backbone
            )
        
        self.temp = config.logit_scale
        if config.disable_dropout:
            disable_dropout(self)
        
        transductive_tie_token_embeddings = vars(self.config).get("transductive_tie_token_embeddings", False)
        if transductive_tie_token_embeddings:
            self.second_stage_model.backbone.embeddings.word_embeddings.weight = (
                self.first_stage_model.embedder.embeddings.word_embeddings.weight
            )

    def forward(
            self, 
            input_ids: torch.Tensor,
            attention_mask: torch.Tensor,
            dataset_input_ids: Optional[torch.Tensor],
            dataset_attention_mask: Optional[torch.Tensor],
            output_hidden_states: bool = False,
        ) -> torch.Tensor:
        """
        input_ids (long torch.Tensor) – ids of input tokens
        attention_mask (bool torch.Tensor)
        """
        dataset_embeddings = self.first_stage_model(
            input_ids=dataset_input_ids, 
            attention_mask=dataset_attention_mask
        )
        return self.second_stage_model(
            input_ids=input_ids,
            attention_mask=attention_mask,
            dataset_embeddings=dataset_embeddings,
            output_hidden_states=output_hidden_states,
        )



def get_model_class(name: str):
    if name in 'transductive': 
        return ContextualDocumentEmbeddingTransformer
    elif name == 'biencoder':
        return BiEncoder
    elif name == "dataset_prefix_biencoder":
        return DatasetPrefixBiencoder
    else:
        raise ValueError(f'unknown model cls {name}')