Model save

README.md

@@ -0,0 +1,57 @@
+---
+base_model: liswei/OpenELM-1_1B-zh-base
+tags:
+- llama-factory
+- generated_from_trainer
+model-index:
+- name: OpenELM-1_1B-zh-cp
+  results: []
+---
+
+<!-- This model card has been generated automatically according to the information the Trainer had access to. You
+should probably proofread and complete it, then remove this comment. -->
+
+# OpenELM-1_1B-zh-cp
+
+This model is a fine-tuned version of [liswei/OpenELM-1_1B-zh-base](https://huggingface.co/liswei/OpenELM-1_1B-zh-base) on an unknown dataset.
+
+## Model description
+
+More information needed
+
+## Intended uses & limitations
+
+More information needed
+
+## Training and evaluation data
+
+More information needed
+
+## Training procedure
+
+### Training hyperparameters
+
+The following hyperparameters were used during training:
+- learning_rate: 0.0001
+- train_batch_size: 4
+- eval_batch_size: 8
+- seed: 42
+- distributed_type: multi-GPU
+- num_devices: 4
+- total_train_batch_size: 16
+- total_eval_batch_size: 32
+- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
+- lr_scheduler_type: cosine
+- lr_scheduler_warmup_ratio: 0.1
+- num_epochs: 1.0
+
+### Training results
+
+
+
+### Framework versions
+
+- Transformers 4.41.1
+- Pytorch 2.3.0+cu121
+- Datasets 2.19.1
+- Tokenizers 0.19.1

configuration_openelm.py

@@ -0,0 +1,318 @@
+#
+# For licensing see accompanying LICENSE file.
+# Copyright (C) 2024 Apple Inc. All Rights Reserved.
+#
+
+"""Implements HF OpenELMConfig based on PretrainedConfig"""
+from numbers import Number
+from typing import List, Optional, Union
+
+import numpy as np
+from transformers import PretrainedConfig
+
+
+def make_divisible(
+    v: Union[float, int],
+    divisor: Optional[int] = 8,
+    min_value: Optional[Union[float, int]] = None,
+) -> Union[float, int]:
+    """
+    This function is taken from the original tf repo.
+    It ensures that all layers have a channel number that is divisible by the divisor
+    It can be seen at:
+    https://github.com/tensorflow/models/blob/2cfc99eff5e5eb729c6793d2f3d03aa1c9be2b15/research/slim/nets/mobilenet/mobilenet.py#L62
+
+    Args:
+        v: input value
+        divisor: default to 8
+        min_value: minimum divisor value
+    Returns:
+        new_v: new divisible value
+    """
+    if min_value is None:
+        min_value = divisor
+    new_v = max(min_value, int(v + divisor / 2) // divisor * divisor)
+    # Make sure that round down does not go down by more than 10%.
+    if new_v < 0.9 * v:
+        new_v += divisor
+    return new_v
+
+
+def compute_heads(model_dim: int, head_dim: int) -> int:
+    """Compute the number of heads.
+
+    Args:
+        model_dim: Model dimension.
+        head_dim: Head dimension.
+
+    Returns:
+        An integer denoting number of heads in multi-head attention is returned.
+
+    Raises:
+        ValueError: if model dimension is not divisible by head dimension.
+    """
+    if model_dim % head_dim == 0:
+        return model_dim // head_dim
+    else:
+        raise ValueError(
+            f"Model dimension should be divisible by head dimension. Got: {model_dim} and {head_dim}."
+        )
+
+
+OpenELM_CONFIGS = {
+    "OpenELM-270M": dict(
+        num_transformer_layers=16,
+        model_dim=1280,
+        head_dim=64,
+        num_gqa_groups=4,
+        normalize_qk_projections=True,
+        share_input_output_layers=True,
+        # Vary the FFN and QKV multipliers to create variable FFN and attention layers respectively.
+        ffn_multipliers=(0.5, 4.0),
+        qkv_multipliers=(0.5, 1.0),
+    ),
+    "OpenELM-450M": dict(
+        num_transformer_layers=20,
+        model_dim=1536,
+        head_dim=64,
+        num_gqa_groups=4,
+        normalize_qk_projections=True,
+        share_input_output_layers=True,
+        # Vary the FFN and QKV multipliers to create variable FFN and attention layers respectively.
+        ffn_multipliers=(0.5, 4.0),
+        qkv_multipliers=(0.5, 1.0),
+    ),
+    "OpenELM-1_1B": dict(
+        num_transformer_layers=28,
+        model_dim=2048,
+        head_dim=64,
+        num_gqa_groups=4,
+        normalize_qk_projections=True,
+        share_input_output_layers=True,
+        # Vary the FFN and QKV multipliers to create variable FFN and attention layers respectively.
+        ffn_multipliers=(0.5, 4.0),
+        qkv_multipliers=(0.5, 1.0),
+    ),
+    "OpenELM-3B": dict(
+        num_transformer_layers=36,
+        model_dim=3072,
+        head_dim=128,
+        num_gqa_groups=4,
+        normalize_qk_projections=True,
+        share_input_output_layers=True,
+        # Vary the FFN and QKV multipliers to create variable FFN and attention layers respectively.
+        ffn_multipliers=(0.5, 4.0),
+        qkv_multipliers=(0.5, 1.0),
+    ),
+}
+
+
+class OpenELMConfig(PretrainedConfig):
+    r"""
+    This is the configuration class to store the configuration of a [`OpenELMModel`]. It is used to instantiate an OpenELM model according to the specified arguments, defining the model architecture.
+
+    Configuration objects inherit from [`PretrainedConfig`] and can be used to control the model outputs. Read the
+    documentation from [`PretrainedConfig`] for more information.
+
+    Args:
+        vocab_size (`int`, *optional*, defaults to 32000):
+            Vocabulary size of the OpenELM model.
+        max_context_length (`int`, *optional*, defaults to 2048):
+            Maximum number of input tokens.
+        num_transformer_layers (`int`, *optional*, defaults to 12):
+            Number of hidden layers in the Transformer decoder.
+        model_dim (`int`, *optional*, defaults to 2048):
+            Dimension of the hidden representations.
+        head_dim (`int`, *optional*, defaults to 128):
+            The attention head dimension.
+        qkv_multipliers (`Union[Number, List[Number]]`, *optional*, defaults to 1.0):
+            If the qkv_multipliers is a Number, then all attention layers have the same latent dimensions,
+            resulting in uniform allocation of parameters.
+            If the qkv_multipliers is a List of Number, then each attention layer have different latent dimensions
+            assuming qkv_multipliers[0] != qkv_multipliers[1]. This results in variable allocation of parameters in attention layer.
+            This scaling is known as layer-wise or block-wise scaling: https://arxiv.org/abs/2008.00623
+        num_query_heads (`Union[int, None]`, *optional*, defaults to None):
+            The number of query heads, computed from `compute_heads(model_dim=model_dim, head_dim=head_dim)`.
+        num_gqa_groups (`int`, *optional*, defaults to 1):
+            This variable allows to switch between multi-head attention, group query attention, and multi-query attention.
+            When num_gqa_groups == 1, then it is multi-head attention.
+            When 1 < num_gqa_groups < num_heads and num_heads is divisible by num_gqa_groups, then it is group query attention
+            When num_gqa_groups == num_heads, then it is multi-query attention
+        ffn_multipliers (`Union[Number, List[Number]]`, *optional*, defaults to 4.0):
+            Feed-forward network (FFN) multipliers.
+            If the ffn_multipliers is a Number, then all FFN layers have the same latent dimensions,
+            resulting in uniform allocation of parameters.
+            If the ffn_multipliers is a List of Number, then each FFN layer have different latent dimensions
+            assuming ffn_multipliers[0] != ffn_multipliers[1]. This results in variable allocation of parameters in FFN layer.
+            This scaling is known as layer-wise or block-wise scaling: https://arxiv.org/abs/2008.00623
+        ffn_with_glu (`bool`, *optional*, defaults to True):
+            Whether to use FFN with Gated Linear Unit (GLU)
+        ffn_dim_divisor (`int`, *optional*, defaults to 256):
+            The ffn layer dimension divisor.
+        activation_fn_name (`str` or `function`, *optional*, defaults to `"swish"`):
+            The non-linear activation function (function or string) in the decoder.
+        normalization_layer_name (`str` or `function`, *optional*, defaults to `"rms_norm"`):
+            Type of normalization layer.
+        normalize_qk_projections (`bool`, *optional*, defaults to False):
+            Whether to normalize queries and keys after projections
+        share_input_output_layers (`bool`, *optional*, defaults to False):
+            Whether to share the embedding between input and output linear layer
+        rope_freq_constant (`int`, *optional*, defaults to 10000):
+            The base period of the RoPE embeddings.
+        rope_max_length (`int`, *optional*, defaults to 4096):
+            That rope_max_length is set to twice of max_context_length.
+            This allows flexibility in token lengths during training or fine-tuning.
+        initializer_range (`float`, *optional*, defaults to 0.02):
+            The standard deviation of the truncated_normal_initializer for initializing all weight matrices.
+        use_cache (`bool`, *optional*, defaults to `True`):
+            Whether or not the model should return the last key/values attentions (not used by all models). Only
+            relevant if `config.is_decoder=True`.
+        bos_token_id (`int`, *optional*, defaults to 2):
+            Beginning of stream token id.
+        eos_token_id (`int`, *optional*, defaults to 1):
+            End of stream token id.
+    """
+
+    model_type = "openelm"
+
+    def __init__(
+        self,
+        vocab_size: int = 32000,
+        max_context_length: int = 2048,
+        num_transformer_layers: int = 12,
+        model_dim: int = 2048,
+        head_dim: int = 128,
+        qkv_multipliers: Union[Number, List[Number]] = 1.0,
+        num_query_heads: Union[int, None] = None,
+        num_gqa_groups: int = 1,
+        ffn_multipliers: Union[Number, List[Number]] = 4.0,
+        ffn_with_glu: bool = True,
+        ffn_dim_divisor: int = 256,
+        activation_fn_name: str = "swish",
+        normalization_layer_name: str = "rms_norm",
+        normalize_qk_projections: bool = False,
+        share_input_output_layers: bool = False,
+        rope_freq_constant: int = 10000,
+        rope_max_length: int = 4096,
+        initializer_range: float = 0.02,
+        use_cache: bool = True,
+        bos_token_id: int = 1,
+        eos_token_id: int = 2,
+        **kwargs,
+    ) -> None:
+        self.vocab_size = vocab_size
+        self.max_context_length = max_context_length
+        self.num_transformer_layers = num_transformer_layers
+        self.model_dim = model_dim
+        self.head_dim = head_dim
+        self.qkv_multipliers = qkv_multipliers
+        self.num_query_heads = num_query_heads
+        self.num_gqa_groups = num_gqa_groups
+        self.ffn_multipliers = ffn_multipliers
+        self.ffn_with_glu = ffn_with_glu
+        self.ffn_dim_divisor = ffn_dim_divisor
+        self.activation_fn_name = activation_fn_name
+        self.normalization_layer_name = normalization_layer_name
+        self.normalize_qk_projections = normalize_qk_projections
+        self.share_input_output_layers = share_input_output_layers
+        self.rope_freq_constant = rope_freq_constant
+        self.rope_max_length = rope_max_length
+        self.num_query_heads = (
+            compute_heads(model_dim=model_dim, head_dim=head_dim)
+            if num_query_heads is None
+            else num_query_heads
+        )
+        self.initializer_range = initializer_range
+
+        self.__post_init__()
+        super().__init__(
+            use_cache=use_cache,
+            bos_token_id=bos_token_id,
+            eos_token_id=eos_token_id,
+            **kwargs,
+        )
+
+    def __post_init__(self) -> None:
+        if self.num_gqa_groups is not None:
+            head_multiple_of = self.num_gqa_groups
+        else:
+            head_multiple_of = 2
+
+        if isinstance(self.qkv_multipliers, Number):
+            # All attention layers have the same latent dimensions, resulting in uniform allocation of parameters.
+            qkv_dim = make_divisible(
+                self.model_dim * self.qkv_multipliers,
+                divisor=self.head_dim * head_multiple_of,
+            )
+            query_dims = [int(qkv_dim)] * self.num_transformer_layers
+
+        elif (
+            isinstance(self.qkv_multipliers, (tuple, list))
+            and len(self.qkv_multipliers) == 2
+        ):
+            # Each attention layer have different latent dimensions assuming qkv_multipliers[0] != qkv_multipliers[1].
+            # This results in variable allocation of parameters in attention layer.
+            # This scaling is known as layer-wise or block-wise scaling: https://arxiv.org/abs/2008.00623
+            qkv_multipliers = [
+                round(v, 2)
+                for v in np.linspace(
+                    self.qkv_multipliers[0],
+                    self.qkv_multipliers[1],
+                    num=self.num_transformer_layers,
+                    dtype=float,
+                )
+            ]
+            # Make sure that scaled model dimension is divisible by scaled head dimension.
+            query_dims = [
+                int(
+                    make_divisible(
+                        self.model_dim * m, divisor=self.head_dim * head_multiple_of
+                    )
+                )
+                for m in qkv_multipliers
+            ]
+        else:
+            raise NotImplementedError(
+                f"QKV multipliers should be a single number or a list containing exactly two numbers. Got: {qkv_multipliers}."
+            )
+
+        # compute the number of query, key, and value heads
+        # For multi-head and multi-query attention, the number of heads for query, key, and value are the same.
+        # For group query attention, the number of key and value heads are the same.
+        self.num_query_heads = [
+            int(compute_heads(q_dim, self.head_dim)) for q_dim in query_dims
+        ]
+        self.num_kv_heads = [
+            q_heads // self.num_gqa_groups for q_heads in self.num_query_heads
+        ]
+
+        # Feed-forward network (FFN) multipliers
+        if isinstance(self.ffn_multipliers, Number):
+            # All FFN layers have the same latent dimensions, resulting in uniform allocation of parameters.
+            self.ffn_multipliers = [self.ffn_multipliers] * self.num_transformer_layers
+        elif isinstance(self.ffn_multipliers, (tuple, list)):
+            # Each FFN layer have different latent dimensions assuming ffn_multipliers[0] != ffn_multipliers[1].
+            # This results in variable allocation of parameters in FFN layer.
+            # This scaling is known as layer-wise or block-wise scaling: https://arxiv.org/abs/2008.00623
+            if len(self.ffn_multipliers) == 2:
+                self.ffn_multipliers = [
+                    round(v, 2)
+                    for v in np.linspace(
+                        self.ffn_multipliers[0],
+                        self.ffn_multipliers[1],
+                        num=self.num_transformer_layers,
+                        dtype=float,
+                    )
+                ]
+            else:
+                assert (
+                    len(self.ffn_multipliers) == self.num_transformer_layers
+                ), f"{len(self.ffn_multipliers)=}!={self.num_transformer_layers=}"
+        else:
+            raise NotImplementedError(
+                f"FFN multipliers should be a single number or a list containing exactly two numbers. Got: {qkv_multipliers}."
+            )
+
+        # check num_query_heads divisible by num_kv_heads for every layer
+        for layer_idx in range(len(query_dims)):
+            assert self.num_query_heads[layer_idx] % self.num_kv_heads[layer_idx] == 0

generation_config.json

@@ -0,0 +1,6 @@
+{
+  "_from_model_config": true,
+  "bos_token_id": 1,
+  "eos_token_id": 2,
+  "transformers_version": "4.41.1"
+}

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:83215022dc4410fc88da4af61b0e753ab0dbab067e0b84fca186aa8a68a4bb1c
+oid sha256:c5b0b679ba69dba67254c1c1e6797608e427a209cb8b3e7116925e851752f516
 size 4563369024

modeling_openelm.py

@@ -0,0 +1,1008 @@
+#
+# For licensing see accompanying LICENSE file.
+# Copyright (C) 2024 Apple Inc. All Rights Reserved.
+#
+
+from typing import List, Optional, Tuple, Union
+
+import torch
+import torch.utils.checkpoint
+from torch import Tensor, nn
+from torch.nn import CrossEntropyLoss
+from torch.nn import functional as F
+from transformers import PreTrainedModel
+from transformers.activations import ACT2FN
+from transformers.cache_utils import Cache, DynamicCache, StaticCache
+from transformers.modeling_outputs import (
+    BaseModelOutputWithPast,
+    CausalLMOutputWithPast,
+)
+from transformers.utils import logging
+
+logger = logging.get_logger(__name__)
+
+# this import has to be relative, otherwise, when setting trust_remote_code=True
+# huggingface transformers won't be able to load the module correctly
+from .configuration_openelm import OpenELMConfig, make_divisible
+
+
+class OpenELMRMSNorm(nn.Module):
+    def __init__(self, num_features: int, eps: float = 1e-6):
+        """
+        Initialize the OpenELMRMSNorm normalization layer.
+
+        Args:
+            dim (int): The dimension of the input tensor.
+            eps (float, optional): A small value added to the denominator for numerical stability. Default is 1e-6.
+
+        Attributes:
+            eps (float): A small value added to the denominator for numerical stability.
+            weight (nn.Parameter): Learnable scaling parameter.
+
+        """
+        super().__init__()
+        self.eps = eps
+        self.weight = nn.Parameter(torch.ones(num_features))
+        self.num_features = num_features
+
+    def _norm(self, x: Tensor) -> Tensor:
+        """
+        Apply the OpenELMRMSNorm normalization to the input tensor.
+
+        Args:
+            x (torch.Tensor): The input tensor.
+
+        Returns:
+            torch.Tensor: The normalized tensor.
+
+        """
+        return x * torch.rsqrt(x.pow(2).mean(-1, keepdim=True) + self.eps)
+
+    def forward(self, x: Tensor) -> Tensor:
+        """
+        Forward pass through the OpenELMRMSNorm layer.
+
+        Args:
+            x (torch.Tensor): The input tensor.
+
+        Returns:
+            torch.Tensor: The output tensor after applying OpenELMRMSNorm.
+
+        """
+        output = self._norm(x.float()).type_as(x)
+        return output * self.weight
+
+    def extra_repr(self) -> str:
+        return (
+            super().extra_repr() + f"num_features={self.num_features}, eps={self.eps}"
+        )
+
+
+class OpenELMPreTrainedModel(PreTrainedModel):
+    config_class = OpenELMConfig
+    base_model_prefix = "transformer"
+    supports_gradient_checkpointing = True
+    _no_split_modules = ["OpenELMDecoderLayer"]
+    _skip_keys_device_placement = "past_key_values"
+
+    def __init__(self, *inputs, **kwargs) -> None:
+        super().__init__(*inputs, **kwargs)
+
+    def _init_weights(self, module: nn.Module) -> None:
+        """Initialize the weights."""
+        if isinstance(module, nn.Linear):
+            # Slightly different from the TF version which uses truncated_normal for initialization
+            # cf https://github.com/pytorch/pytorch/pull/5617
+            module.weight.data.normal_(mean=0.0, std=self.config.initializer_range)
+            if module.bias is not None:
+                module.bias.data.zero_()
+        elif isinstance(module, nn.Embedding):
+            module.weight.data.normal_(mean=0.0, std=self.config.initializer_range)
+            if module.padding_idx is not None:
+                module.weight.data[module.padding_idx].zero_()
+        elif isinstance(module, OpenELMRMSNorm):
+            module.weight.data.fill_(1.0)
+
+
+def _rotate_half(x: Tensor) -> Tensor:
+    x1, x2 = x.chunk(2, dim=-1)
+    return torch.cat((-x2, x1), dim=-1)
+
+
+def _apply_rotary_pos_emb(x: Tensor, pos_sin: Tensor, pos_cos: Tensor) -> Tensor:
+    return (x * pos_cos) + (_rotate_half(x) * pos_sin)
+
+
+class OpenELMRotaryEmbedding(torch.nn.Module):
+    """
+    The rotary position embeddings (aka RoPE) from `RoFormer <https://arxiv.org/abs/2104.09864>`_.
+
+    RoPE encodes the position information of tokens using a rotation matrix, and is able to capture
+    explicit relative positional dependencies.
+
+    Args:
+        model_dim: The dimensionality of the model's hidden state.
+        max_seq_length: Maximum sequence length.
+        freq_constant: A constant used for computing frequencies.
+    """
+
+    def __init__(
+        self, model_dim: int, max_seq_length: int, freq_constant: int = 10000
+    ) -> None:
+        inv_freq = 1.0 / (
+            freq_constant
+            ** (torch.arange(0, model_dim, 2, dtype=torch.float32) / model_dim)
+        )
+        super().__init__()
+
+        self.model_dim = model_dim
+        self.freq_constant = freq_constant
+        self.max_seq_length = max_seq_length
+
+        self.register_buffer("inv_freq", inv_freq, persistent=False)
+        self._cached_cos = None
+        self._cached_sin = None
+        self._cached_seq_length = max_seq_length
+        self._compute_sin_cos_embeddings(max_seq_length)
+
+    def extra_repr(self) -> str:
+        return f"\tmodel_dim={self.model_dim}, max_seq_length={self.max_seq_length}, freq_constant={self.freq_constant}"
+
+    def _compute_sin_cos_embeddings(
+        self,
+        key_len: int,
+        key_device: torch.device = torch.device("cpu"),
+        key_dtype: torch.dtype = torch.float32,
+    ) -> None:
+        """
+        Compute sine and cos embeddings.
+
+        Args:
+            key_len: Number of tokens in the key embeddings in the transformer model.
+            device: Device where the key embeddings are stored.
+            key_dtype: Data type of the key embeddings.
+
+        Returns:
+            None
+
+        ...note:
+            We recalculate the sine and cosine embeddings if any of the following conditions are met:
+                1. The number of tokens in key embeddings are greater than the cached sequence length.
+                2. Sine and cosine caches are empty.
+                3. The device and data type of sine and cosine embeddings does not match with the key embeddings.
+        """
+        if (
+            key_len > self._cached_seq_length
+            or self._cached_cos is None
+            or (self._cached_cos is not None and self._cached_cos.device != key_device)
+            or (self._cached_cos is not None and self._cached_cos.dtype != key_dtype)
+            or self._cached_sin is None
+            or (self._cached_sin is not None and self._cached_sin.device != key_device)
+            or (self._cached_sin is not None and self._cached_sin.dtype != key_dtype)
+        ):
+            self._cached_seq_length = max(key_len, self._cached_seq_length)
+
+            # The shape of 'pos_index' is [number of key tokens]
+            pos_index = torch.arange(
+                self._cached_seq_length,
+                dtype=torch.float32,
+                device=self.inv_freq.device,
+            )
+            # The shape of 'pos_index_theta' is [number of key tokens, model dimension]
+            pos_index_theta = torch.einsum("i,j->ij", pos_index, self.inv_freq)
+            # The shape of 'emb' is [number of key tokens, model dimension]
+            emb = torch.cat((pos_index_theta, pos_index_theta), dim=-1)
+
+            # the shape of cos and sin embeddings is [number of key tokens, model_dim]
+            cos_emb = emb.cos().to(dtype=key_dtype, device=key_device)
+            sin_emb = emb.sin().to(dtype=key_dtype, device=key_device)
+
+            # the shape of cached cos and sin embeddings is [1, 1, number of key tokens, model_dim]
+            self._cached_cos = cos_emb[None, None, :, :]
+            self._cached_sin = sin_emb[None, None, :, :]
+
+    def forward(
+        self,
+        query: torch.Tensor,
+        key: torch.Tensor,
+    ) -> Tuple[torch.Tensor, torch.Tensor]:
+        """
+        The forward function of RoPE embeddings.
+
+        Args:
+            query: Query embeddings in the transformer model. The shape of query embeddings is
+                [Batch, number of query heads, number of query tokens, model dimension].
+            key: Key embeddings in the transformer model. The shape of key embeddings is
+                [Batch, number of key heads, number of key tokens, model dimension].
+
+        Returns:
+            A tuple containing the query and key embeddings with positional information. The shape of the returned query
+            and key embeddings is the same as the input query and key embeddings respectively.
+
+        ...note:
+            The RoPE embedding computation is done in full-precision. After the computation, input query and key tensors
+            are casted to original input datatype.
+        """
+        dim = key.shape[-1]
+        key_len = key.shape[2]
+        query_len = query.shape[2]
+
+        assert dim == self.model_dim
+        assert key.device == query.device
+        assert key.dtype == query.dtype
+
+        # In the context of self-attention, the lengths of keys and queries are equal.
+        # However, in generation tasks, such as predicting the next token in a sequence, the lengths of keys and queries
+        # can differ. For instance, when employing key-value (KV) caching for sequence prediction, the keys
+        # represent embeddings of previous tokens and the current token, while the query corresponds
+        # to the embedding of the current token only.
+        assert (
+            key_len >= query_len
+        ), "Number of keys has to be greater than or equal to number of queries."
+
+        query_float = query.float()
+        key_float = key.float()
+
+        self._compute_sin_cos_embeddings(
+            key_len, key_device=key_float.device, key_dtype=key_float.dtype
+        )
+        query_float = _apply_rotary_pos_emb(
+            x=query_float,
+            pos_sin=self._cached_sin[..., key_len - query_len : key_len, :],
+            pos_cos=self._cached_cos[..., key_len - query_len : key_len, :],
+        )
+        key_float = _apply_rotary_pos_emb(
+            x=key_float,
+            pos_sin=self._cached_sin[..., :key_len, :],
+            pos_cos=self._cached_cos[..., :key_len, :],
+        )
+
+        return query_float.type_as(query), key_float.type_as(key)
+
+
+class OpenELMMultiHeadCausalAttention(nn.Module):
+    def __init__(self, config: OpenELMConfig, layer_idx: int) -> None:
+        super().__init__()
+        self.layer_idx = layer_idx
+        head_dim = config.head_dim
+        q_heads = config.num_query_heads[layer_idx]
+        k_heads = config.num_kv_heads[layer_idx]
+        v_heads = config.num_kv_heads[layer_idx]
+
+        self.qkv_proj = nn.Linear(
+            in_features=config.model_dim,
+            out_features=(q_heads + k_heads + v_heads) * head_dim,
+            bias=False,
+        )
+
+        self.pos_embedding = OpenELMRotaryEmbedding(
+            model_dim=config.head_dim,
+            max_seq_length=config.rope_max_length,
+            freq_constant=config.rope_freq_constant,
+        )
+
+        if config.normalize_qk_projections:
+            self.q_norm = OpenELMRMSNorm(
+                num_features=config.head_dim,
+            )
+            self.k_norm = OpenELMRMSNorm(
+                num_features=config.head_dim,
+            )
+        else:
+            self.q_norm = None
+            self.k_norm = None
+
+        self.out_proj = nn.Linear(
+            in_features=q_heads * head_dim,
+            out_features=config.model_dim,
+            bias=False,
+        )
+
+        self.head_dim = config.head_dim
+        self.num_q_heads = q_heads
+        self.num_k_heads = k_heads
+        self.num_v_heads = v_heads
+        self.transformer_dim = config.model_dim
+        self.num_groups = self.num_q_heads // self.num_k_heads
+
+    def extra_repr(self) -> str:
+        return (
+            super().extra_repr()
+            + f"query_heads={self.num_q_heads}, key_heads={self.num_k_heads}, value_heads={self.num_v_heads}"
+        )
+
+    def forward(
+        self,
+        hidden_states: torch.Tensor,
+        attention_mask: Optional[torch.Tensor] = None,
+        past_key_value: Optional[Cache] = None,
+        output_attentions: bool = False,
+        use_cache: bool = False,
+        cache_position: Optional[torch.LongTensor] = None,
+    ) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor]]]:
+        """
+        Forward pass of multi-head self-attention.
+
+        Args:
+            hidden_states: Input tensor of the shape [batch size, sequence length, model dimension].
+            past_key_value: Tensor storing the cached keys and values.
+            output_attentions: output attention weights.
+            use_cache: Specifies whether to use kv-cache for generation.
+            cache_position: used for updating the kv-cache.
+
+        Returns:
+            The output of the same shape as the input, optionally with a tensor containing cached keys and values.
+        """
+
+        # scaled_dot_product_attention does not return attention weights, set output_attentions to False
+        output_attentions = False
+        batch_size, seq_length, d_model = hidden_states.size()
+
+        # [B, S, d] --> [B, S, (q_h + k_h + v_h) * h]
+        qkv = self.qkv_proj(hidden_states)
+        # [B, S, (q_h + k_h + v_h) * h] --> [B, S, (q_h + k_h + v_h), h]
+        qkv = qkv.reshape(
+            batch_size,
+            seq_length,
+            self.num_q_heads + self.num_k_heads + self.num_v_heads,
+            self.head_dim,
+        )
+        # [B, S, (q_h + k_h + v_h), h] --> [B, (q_h + k_h + v_h), S, h]
+        qkv = qkv.transpose(1, 2)
+        # [B, (q_h + k_h + v_h), S, h] --> [B, q_h, S h], [B, k_h, S, h], [B, v_h, S, h]
+        queries, keys, values = qkv.split(
+            [self.num_q_heads, self.num_k_heads, self.num_v_heads], dim=1
+        )
+
+        if self.q_norm is not None:
+            queries = self.q_norm(queries)
+
+        if self.k_norm is not None:
+            keys = self.k_norm(keys)
+
+        past_key_value = getattr(self, "past_key_value", past_key_value)
+
+        if past_key_value is not None:
+            # sin and cos are specific to RoPE models; position_ids needed for the static cache
+            # cache_kwargs = {"sin": sin, "cos": cos, "cache_position": cache_position}
+            cache_kwargs = {"cache_position": cache_position}
+            keys, values = past_key_value.update(
+                keys, values, self.layer_idx, cache_kwargs
+            )
+
+        # Add positional embedding
+        queries, keys = self.pos_embedding(queries, keys)
+
+        if self.num_groups != 1:
+            # GQA
+            # [B, k_h, S, h] --> [B, q_h, S, h]
+            keys = keys.repeat_interleave(self.num_groups, dim=1)
+            # [B, v_h, S, h] --> [B, q_h, S, h]
+            values = values.repeat_interleave(self.num_groups, dim=1)
+
+        causal_mask = attention_mask
+        if attention_mask is not None and cache_position is not None:
+            causal_mask = causal_mask[:, :, cache_position, : keys.shape[-2]]
+
+        attn_output = F.scaled_dot_product_attention(
+            queries,
+            keys,
+            values,
+            attn_mask=causal_mask,
+            dropout_p=0,
+        )
+
+        attn_output = attn_output.transpose(1, 2).contiguous()
+        attn_output = attn_output.reshape(
+            batch_size, seq_length, self.num_q_heads * self.head_dim
+        )
+        attn_output = self.out_proj(attn_output)
+        if not output_attentions:
+            attn_weights = None
+        return attn_output, attn_weights, past_key_value
+
+
+class OpenELMFeedForwardNetwork(nn.Module):
+    def __init__(self, config: OpenELMConfig, layer_idx: int) -> None:
+        super().__init__()
+        ffn_multiplier = config.ffn_multipliers[layer_idx]
+        intermediate_dim = int(
+            make_divisible(
+                ffn_multiplier * config.model_dim,
+                divisor=config.ffn_dim_divisor,
+            )
+        )
+        if config.ffn_with_glu:
+            # FFN with Gated linear unit, as described in https://arxiv.org/abs/2002.05202v1.
+            self.proj_1 = nn.Linear(
+                in_features=config.model_dim,
+                out_features=2 * intermediate_dim,
+                bias=False,
+            )
+            self.proj_2 = nn.Linear(
+                in_features=intermediate_dim,
+                out_features=config.model_dim,
+                bias=False,
+            )
+            self.ffn_with_glu = True
+        else:
+            # Standard FFN, as described in https://arxiv.org/abs/1706.03762
+            self.proj_1 = nn.Linear(
+                in_features=config.model_dim,
+                out_features=intermediate_dim,
+                bias=False,
+            )
+            self.proj_2 = nn.Linear(
+                in_features=intermediate_dim,
+                out_features=config.model_dim,
+                bias=False,
+            )
+            self.ffn_with_glu = False
+
+        self.act = ACT2FN[config.activation_fn_name]
+
+    def extra_repr(self) -> str:
+        return super().extra_repr() + f"(ffn_with_glu) : {self.ffn_with_glu}"
+
+    def forward(self, x: Tensor) -> Tensor:
+        """Forward function of FFN layer.
+
+        Args:
+            x: Input tensor of the shape [batch size, sequence length, model dimension].
+
+        Returns:
+            A tensor of the same shape as the input.
+        """
+        if self.ffn_with_glu:
+            y_12 = self.proj_1(x)
+            y_1, y_2 = y_12.chunk(2, dim=-1)
+            y = self.act(y_1) * y_2
+            return self.proj_2(y)
+        else:
+            return self.proj_2(self.act(self.proj_1(x)))
+
+
+class OpenELMDecoderLayer(nn.Module):
+    def __init__(self, config: OpenELMConfig, layer_idx: int) -> None:
+        super().__init__()
+        self.attn = OpenELMMultiHeadCausalAttention(config=config, layer_idx=layer_idx)
+        self.ffn = OpenELMFeedForwardNetwork(config=config, layer_idx=layer_idx)
+        self.ffn_norm = OpenELMRMSNorm(
+            num_features=config.model_dim,
+        )
+        self.attn_norm = OpenELMRMSNorm(
+            num_features=config.model_dim,
+        )
+
+    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: Optional[bool] = False,
+        use_cache: Optional[bool] = False,
+        cache_position: Optional[torch.LongTensor] = None,
+        **kwargs,
+    ) -> Tuple[
+        torch.FloatTensor, Optional[Tuple[torch.FloatTensor, torch.FloatTensor]]
+    ]:
+        """
+        Args:
+            hidden_states (`torch.FloatTensor`): input to the layer of shape `(batch, seq_len, embed_dim)`
+            attention_mask (`torch.FloatTensor`, *optional*):
+                attention mask of size `(batch_size, sequence_length)` if flash attention is used or `(batch_size, 1,
+                query_sequence_length, key_sequence_length)` if default attention is used.
+            output_attentions (`bool`, *optional*):
+                Whether or not to return the attentions tensors of all attention layers. See `attentions` under
+                returned tensors for more detail.
+            use_cache (`bool`, *optional*):
+                If set to `True`, `past_key_values` key value states are returned and can be used to speed up decoding
+                (see `past_key_values`).
+            past_key_value (`Tuple(torch.FloatTensor)`, *optional*): cached past key and value projection states
+        """
+        residual = hidden_states
+        hidden_states = self.attn_norm(hidden_states)
+
+        # Self Attention
+        hidden_states, self_attn_weights, present_key_value = self.attn(
+            hidden_states=hidden_states,
+            attention_mask=attention_mask,
+            past_key_value=past_key_value,
+            output_attentions=output_attentions,
+            use_cache=use_cache,
+            cache_position=cache_position,
+            **kwargs,
+        )
+        hidden_states = residual + hidden_states
+
+        # Fully Connected
+        residual = hidden_states
+        hidden_states = self.ffn_norm(hidden_states)
+        hidden_states = self.ffn(hidden_states)
+        hidden_states = residual + hidden_states
+
+        outputs = (hidden_states,)
+
+        if output_attentions:
+            outputs += (self_attn_weights,)
+
+        if use_cache:
+            outputs += (present_key_value,)
+
+        return outputs
+
+
+class OpenELMModel(OpenELMPreTrainedModel):
+    config_class = OpenELMConfig
+
+    def __init__(self, config: OpenELMConfig):
+        super().__init__(config)
+        self.config = config
+
+        self.token_embeddings = nn.Embedding(
+            embedding_dim=config.model_dim,
+            num_embeddings=config.vocab_size,
+        )
+
+        self.layers = nn.ModuleList(
+            OpenELMDecoderLayer(config=config, layer_idx=layer_idx)
+            for layer_idx in range(config.num_transformer_layers)
+        )
+        self.norm = OpenELMRMSNorm(num_features=config.model_dim)
+        if config.share_input_output_layers:
+            self.classifier = None
+        else:
+            self.classifier = nn.Linear(
+                in_features=config.model_dim,
+                out_features=config.vocab_size,
+                bias=False,
+            )
+        self.num_transformer_layers = config.num_transformer_layers
+        self.gradient_checkpointing = False
+
+        # Register a causal mask to separate causal and padding mask creation. Merging happens in the attention class.
+        # NOTE: This is not friendly with TorchScript, ONNX, ExportedProgram serialization for very large `max_context_length`.
+        causal_mask = torch.full(
+            (config.max_context_length, config.max_context_length),
+            fill_value=True,
+            dtype=torch.bool,
+        )
+        self.register_buffer(
+            "causal_mask", torch.triu(causal_mask, diagonal=1), persistent=False
+        )
+
+        # Initialize weights and apply final processing
+        self.post_init()
+        self.reset_parameters(config=config)
+
+    def get_input_embeddings(self):
+        return self.token_embeddings
+
+    def set_input_embeddings(self, new_embeddings: torch.Tensor):
+        self.token_embeddings = new_embeddings
+
+    def reset_parameters(self, config: OpenELMConfig) -> None:
+        """Initialize the layers in Language Model
+
+        The initialization scheme is followed, following `OPT <https://arxiv.org/pdf/2205.01068.pdf>`_.
+
+        Args:
+            use_megatron_std: Use standard deviation as described in Megatron-LM.
+
+        Returns:
+            None
+        """
+        for module in self.modules():
+            if isinstance(module, nn.Linear):
+                std = module.in_features**-0.5
+                torch.nn.init.normal_(module.weight, mean=0.0, std=std)
+                if module.bias is not None:
+                    torch.nn.init.zeros_(module.bias)
+            elif isinstance(module, nn.Embedding):
+                std = module.embedding_dim**-0.5
+                torch.nn.init.normal_(module.weight, mean=0.0, std=std)
+            elif isinstance(module, OpenELMRMSNorm):
+                if module.weight is not None:
+                    torch.nn.init.ones_(module.weight)
+                if hasattr(module, "bias") and module.bias is not None:
+                    torch.nn.init.zeros_(module.bias)
+
+        model_dim = config.model_dim
+        n_layers = config.num_transformer_layers
+        std = (model_dim**-0.5) * ((2 * n_layers) ** -0.5)
+        for param_name, param in self.named_parameters():
+            if param_name.endswith("out_proj.weight") or param_name.endswith(
+                "ffn.proj_2.weight"
+            ):
+                torch.nn.init.normal_(param, mean=0.0, std=std)
+
+    def forward(
+        self,
+        input_ids: 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,
+        cache_position: Optional[torch.LongTensor] = None,
+    ) -> Union[Tuple, BaseModelOutputWithPast]:
+        output_attentions = (
+            output_attentions
+            if output_attentions is not None
+            else self.config.output_attentions
+        )
+        output_hidden_states = (
+            output_hidden_states
+            if output_hidden_states is not None
+            else self.config.output_hidden_states
+        )
+        use_cache = use_cache if use_cache is not None else self.config.use_cache
+        return_dict = (
+            return_dict if return_dict is not None else self.config.use_return_dict
+        )
+
+        if (input_ids is None) ^ (inputs_embeds is not None):
+            raise ValueError(
+                "You cannot specify both input_ids and inputs_embeds at the same time, and must specify either one"
+            )
+
+        if self.gradient_checkpointing and self.training and use_cache:
+            logger.warning_once(
+                "`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`."
+            )
+            use_cache = False
+
+        if inputs_embeds is None:
+            inputs_embeds = self.token_embeddings(input_ids)
+
+        past_seen_tokens = 0
+        if use_cache:  # kept for BC (cache positions)
+            if not isinstance(past_key_values, StaticCache):
+                past_key_values = DynamicCache.from_legacy_cache(past_key_values)
+            past_seen_tokens = past_key_values.get_seq_length()
+
+        if cache_position is None:
+            cache_position = torch.arange(
+                past_seen_tokens,
+                past_seen_tokens + inputs_embeds.shape[1],
+                device=inputs_embeds.device,
+            )
+
+        if position_ids is None:
+            position_ids = cache_position.unsqueeze(0)
+
+        causal_mask = self._update_causal_mask(attention_mask, inputs_embeds)
+
+        # embed positions
+        hidden_states = inputs_embeds
+
+        # decoder layers
+        all_hidden_states = () if output_hidden_states else None
+        all_self_attns = () if output_attentions else None
+        next_decoder_cache = None
+
+        for decoder_layer in self.layers:
+            if output_hidden_states:
+                all_hidden_states += (hidden_states,)
+
+            if self.gradient_checkpointing and self.training:
+                layer_outputs = self._gradient_checkpointing_func(
+                    decoder_layer.__call__,
+                    hidden_states,
+                    causal_mask,
+                    position_ids,
+                    past_key_values,
+                    output_attentions,
+                    use_cache,
+                    cache_position,
+                )
+            else:
+                layer_outputs = decoder_layer(
+                    hidden_states,
+                    attention_mask=causal_mask,
+                    position_ids=position_ids,
+                    past_key_value=past_key_values,
+                    output_attentions=output_attentions,
+                    use_cache=use_cache,
+                    cache_position=cache_position,
+                )
+
+            hidden_states = layer_outputs[0]
+
+            if use_cache:
+                next_decoder_cache = layer_outputs[2 if output_attentions else 1]
+
+            if output_attentions:
+                all_self_attns += (layer_outputs[1],)
+
+        hidden_states = self.norm(hidden_states)
+
+        # add hidden states from the last decoder layer
+        if output_hidden_states:
+            all_hidden_states += (hidden_states,)
+
+        next_cache = None
+        if use_cache:
+            next_cache = (
+                next_decoder_cache.to_legacy_cache()
+                if isinstance(next_decoder_cache, Cache)
+                else next_decoder_cache
+            )
+        if not return_dict:
+            return tuple(
+                v
+                for v in [hidden_states, next_cache, all_hidden_states, all_self_attns]
+                if v is not None
+            )
+        return BaseModelOutputWithPast(
+            last_hidden_state=hidden_states,
+            past_key_values=next_cache,
+            hidden_states=all_hidden_states,
+            attentions=all_self_attns,
+        )
+
+    def _update_causal_mask(self, attention_mask, input_tensor):
+        if self.config._attn_implementation == "flash_attention_2":
+            if attention_mask is not None and 0.0 in attention_mask:
+                return attention_mask
+            return None
+
+        batch_size, seq_length = input_tensor.shape[:2]
+        dtype = input_tensor.dtype
+        device = input_tensor.device
+
+        # support going beyond cached `max_position_embedding`
+        if seq_length > self.causal_mask.shape[-1]:
+            causal_mask = torch.full(
+                (2 * self.causal_mask.shape[-1], 2 * self.causal_mask.shape[-1]),
+                fill_value=1,
+            )
+            self.register_buffer(
+                "causal_mask", torch.triu(causal_mask, diagonal=1), persistent=False
+            )
+
+        # We use the current dtype to avoid any overflows
+        min_dtype = torch.finfo(dtype).min
+        causal_mask = (
+            self.causal_mask[None, None, :, :].repeat(batch_size, 1, 1, 1).to(dtype)
+            * min_dtype
+        )
+
+        causal_mask = causal_mask.to(dtype=dtype, device=device)
+        if attention_mask is not None and attention_mask.dim() == 2:
+            mask_length = attention_mask.shape[-1]
+            padding_mask = causal_mask[..., :mask_length].eq(0.0) * attention_mask[
+                :, None, None, :
+            ].eq(0.0)
+            causal_mask[..., :mask_length] = causal_mask[..., :mask_length].masked_fill(
+                padding_mask, min_dtype
+            )
+
+        if self.config._attn_implementation == "sdpa" and attention_mask is not None:
+            # For dynamo, rather use a check on fullgraph=True once this is possible (https://github.com/pytorch/pytorch/pull/120400).
+            is_tracing = (
+                torch.jit.is_tracing()
+                or isinstance(input_tensor, torch.fx.Proxy)
+                or (hasattr(torch, "_dynamo") and torch._dynamo.is_compiling())
+            )
+            if not is_tracing and torch.any(attention_mask != 1):
+                # Attend to all tokens in masked rows from the causal_mask, for example the relevant first rows when
+                # using left padding. This is required by F.scaled_dot_product_attention memory-efficient attention path.
+                # Details: https://github.com/pytorch/pytorch/issues/110213
+                causal_mask = causal_mask.mul(
+                    ~torch.all(causal_mask == min_dtype, dim=-1, keepdim=True)
+                ).to(dtype)
+
+        return causal_mask
+
+
+class OpenELMForCausalLM(OpenELMPreTrainedModel):
+    _tied_weights_keys = ["lm_head.weight"]
+
+    def __init__(self, config: OpenELMConfig):
+        super().__init__(config)
+        self.transformer = OpenELMModel(config)
+        self.vocab_size = config.vocab_size
+        if config.share_input_output_layers:
+            self.lm_head = None
+        else:
+            self.lm_head = nn.Linear(config.model_dim, config.vocab_size, bias=False)
+
+        # Initialize weights and apply final processing
+        self.post_init()
+
+    def get_input_embeddings(self):
+        return self.transformer.token_embeddings
+
+    def set_input_embeddings(self, value):
+        self.transformer.token_embeddings = value
+
+    def get_output_embeddings(self):
+        return self.lm_head
+
+    def set_output_embeddings(self, new_embeddings):
+        self.lm_head = new_embeddings
+
+    def set_decoder(self, decoder):
+        self.transformer = decoder
+
+    def get_decoder(self):
+        return self.transformer
+
+    def forward(
+        self,
+        input_ids: 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,
+        labels: Optional[torch.LongTensor] = None,
+        use_cache: Optional[bool] = None,
+        output_attentions: Optional[bool] = None,
+        output_hidden_states: Optional[bool] = None,
+        return_dict: Optional[bool] = None,
+        cache_position: Optional[torch.LongTensor] = None,
+    ) -> Union[Tuple, CausalLMOutputWithPast]:
+        output_attentions = (
+            output_attentions
+            if output_attentions is not None
+            else self.config.output_attentions
+        )
+        output_hidden_states = (
+            output_hidden_states
+            if output_hidden_states is not None
+            else self.config.output_hidden_states
+        )
+        return_dict = (
+            return_dict if return_dict is not None else self.config.use_return_dict
+        )
+        # decoder outputs consists of (dec_features, layer_state, dec_hidden, dec_attn)
+        outputs = self.transformer(
+            input_ids=input_ids,
+            attention_mask=attention_mask,
+            position_ids=position_ids,
+            past_key_values=past_key_values,
+            inputs_embeds=inputs_embeds,
+            use_cache=use_cache,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+            return_dict=return_dict,
+            cache_position=cache_position,
+        )
+
+        hidden_states = outputs[0]
+        if self.lm_head is None:
+            # shared
+            logits = F.linear(
+                hidden_states, weight=self.transformer.token_embeddings.weight
+            )
+        else:
+            logits = self.lm_head(hidden_states)
+        logits = logits[:, : self.config.vocab_size]
+        loss = None
+        if labels is not None:
+            # Shift so that tokens < n predict n
+            shift_logits = logits[..., :-1, :].contiguous()
+            shift_labels = labels[..., 1:].contiguous()
+            # Flatten the tokens
+            loss_fct = CrossEntropyLoss()
+            shift_logits = shift_logits.view(-1, self.config.vocab_size)
+            shift_labels = shift_labels.view(-1)
+            # Enable model parallelism
+            shift_labels = shift_labels.to(shift_logits.device)
+            loss = loss_fct(shift_logits, shift_labels)
+
+        if not return_dict:
+            output = (logits,) + outputs[1:]
+            return (loss,) + output if loss is not None else output
+
+        return CausalLMOutputWithPast(
+            loss=loss,
+            logits=logits,
+            past_key_values=outputs.past_key_values,
+            hidden_states=outputs.hidden_states,
+            attentions=outputs.attentions,
+        )
+
+    def prepare_inputs_for_generation(
+        self,
+        input_ids,
+        past_key_values=None,
+        attention_mask=None,
+        inputs_embeds=None,
+        **kwargs,
+    ):
+        past_length = 0
+        if past_key_values is not None:
+            if isinstance(past_key_values, Cache):
+                cache_length = past_key_values.get_seq_length()
+                past_length = past_key_values.seen_tokens
+                max_cache_length = past_key_values.get_max_length()
+            else:
+                cache_length = past_length = past_key_values[0][0].shape[2]
+                max_cache_length = None
+
+            # Keep only the unprocessed tokens:
+            # 1 - If the length of the attention_mask exceeds the length of input_ids, then we are in a setting where
+            # some of the inputs are exclusively passed as part of the cache (e.g. when passing input_embeds as
+            # input)
+            if (
+                attention_mask is not None
+                and attention_mask.shape[1] > input_ids.shape[1]
+            ):
+                input_ids = input_ids[:, -(attention_mask.shape[1] - past_length) :]
+            # 2 - If the past_length is smaller than input_ids', then input_ids holds all input tokens. We can discard
+            # input_ids based on the past_length.
+            elif past_length < input_ids.shape[1]:
+                input_ids = input_ids[:, past_length:]
+            # 3 - Otherwise (past_length >= input_ids.shape[1]), let's assume input_ids only has unprocessed tokens.
+
+            # If we are about to go beyond the maximum cache length, we need to crop the input attention mask.
+            if (
+                max_cache_length is not None
+                and attention_mask is not None
+                and cache_length + input_ids.shape[1] > max_cache_length
+            ):
+                attention_mask = attention_mask[:, -max_cache_length:]
+
+        position_ids = kwargs.get("position_ids", None)
+        if attention_mask is not None and position_ids is None:
+            # create position_ids on the fly for batch generation
+            position_ids = attention_mask.long().cumsum(-1) - 1
+            position_ids.masked_fill_(attention_mask == 0, 1)
+            if past_key_values:
+                position_ids = position_ids[:, -input_ids.shape[1] :]
+
+        if self.generation_config.cache_implementation == "static":
+            # generation with static cache
+            cache_position = kwargs.get("cache_position", None)
+            if cache_position is None:
+                past_length = 0
+            else:
+                past_length = cache_position[-1] + 1
+            input_ids = input_ids[:, past_length:]
+            position_ids = position_ids[:, past_length:]
+
+        # we should only keep a `cache_position` in generate, and do +=1.
+        # same goes for position ids. Could also help with continued generation.
+        cache_position = torch.arange(
+            past_length,
+            past_length + position_ids.shape[-1],
+            device=position_ids.device,
+        )
+
+        # if `inputs_embeds` are passed, we only want to use them in the 1st generation step
+        if inputs_embeds is not None and past_key_values is None:
+            model_inputs = {"inputs_embeds": inputs_embeds}
+        else:
+            # The `contiguous()` here is necessary to have a static stride during decoding. torchdynamo otherwise
+            # recompiles graphs as the stride of the inputs is a guard. Ref: https://github.com/huggingface/transformers/pull/29114
+            # We could use `next_tokens` directly instead.
+            model_inputs = {"input_ids": input_ids.contiguous()}
+
+        model_inputs.update(
+            {
+                "position_ids": position_ids.contiguous(),
+                "cache_position": cache_position,
+                "past_key_values": past_key_values,
+                "use_cache": kwargs.get("use_cache"),
+                "attention_mask": attention_mask,
+            }
+        )
+        return model_inputs
+
+    @staticmethod
+    def _reorder_cache(past_key_values, beam_idx):
+        reordered_past = ()
+        for layer_past in past_key_values:
+            reordered_past += (
+                tuple(
+                    past_state.index_select(0, beam_idx.to(past_state.device))
+                    for past_state in layer_past
+                ),
+            )
+        return reordered_past

runs/May27_23-45-48_n0qbsictr1716813819608-jbj64/events.out.tfevents.1716824772.n0qbsictr1716813819608-jbj64.1320.0

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:186112c031ab2875f4ed132a505e75f2c27dbf37de7f820d3b7881f5ac93aaf6
-size 259252
+oid sha256:4832002fc33b179307877efe909c118aed5d2e1b750b1887d688125658e5126e
+size 292311

trainer_log.jsonl

@@ -1199,3 +1199,158 @@
 {"current_steps": 11990, "total_steps": 13550, "loss": 2.7833, "learning_rate": 3.983569534229864e-06, "epoch": 0.8848708487084871, "percentage": 88.49, "elapsed_time": "2 days, 23:57:28", "remaining_time": "9:21:44"}
 {"current_steps": 12000, "total_steps": 13550, "loss": 2.8684, "learning_rate": 3.933340092148202e-06, "epoch": 0.8856088560885609, "percentage": 88.56, "elapsed_time": "3 days, 0:01:05", "remaining_time": "9:18:08"}
 {"current_steps": 12010, "total_steps": 13550, "loss": 2.8399, "learning_rate": 3.883416369200399e-06, "epoch": 0.8863468634686347, "percentage": 88.63, "elapsed_time": "3 days, 0:05:05", "remaining_time": "9:14:35"}
+{"current_steps": 12020, "total_steps": 13550, "loss": 2.837, "learning_rate": 3.8337986967028e-06, "epoch": 0.8870848708487085, "percentage": 88.71, "elapsed_time": "3 days, 0:08:42", "remaining_time": "9:10:59"}
+{"current_steps": 12030, "total_steps": 13550, "loss": 2.8523, "learning_rate": 3.7844874039406674e-06, "epoch": 0.8878228782287823, "percentage": 88.78, "elapsed_time": "3 days, 0:12:18", "remaining_time": "9:07:23"}
+{"current_steps": 12040, "total_steps": 13550, "loss": 2.8815, "learning_rate": 3.7354828181659695e-06, "epoch": 0.888560885608856, "percentage": 88.86, "elapsed_time": "3 days, 0:15:55", "remaining_time": "9:03:47"}
+{"current_steps": 12050, "total_steps": 13550, "loss": 2.7918, "learning_rate": 3.6867852645952494e-06, "epoch": 0.8892988929889298, "percentage": 88.93, "elapsed_time": "3 days, 0:19:31", "remaining_time": "9:00:11"}
+{"current_steps": 12060, "total_steps": 13550, "loss": 2.8106, "learning_rate": 3.6383950664074405e-06, "epoch": 0.8900369003690037, "percentage": 89.0, "elapsed_time": "3 days, 0:23:08", "remaining_time": "8:56:35"}
+{"current_steps": 12070, "total_steps": 13550, "loss": 2.8244, "learning_rate": 3.5903125447417196e-06, "epoch": 0.8907749077490775, "percentage": 89.08, "elapsed_time": "3 days, 0:26:45", "remaining_time": "8:52:59"}
+{"current_steps": 12080, "total_steps": 13550, "loss": 2.8061, "learning_rate": 3.5425380186953904e-06, "epoch": 0.8915129151291513, "percentage": 89.15, "elapsed_time": "3 days, 0:30:22", "remaining_time": "8:49:23"}
+{"current_steps": 12090, "total_steps": 13550, "loss": 2.9384, "learning_rate": 3.495071805321759e-06, "epoch": 0.8922509225092251, "percentage": 89.23, "elapsed_time": "3 days, 0:33:58", "remaining_time": "8:45:47"}
+{"current_steps": 12100, "total_steps": 13550, "loss": 2.7863, "learning_rate": 3.447914219628029e-06, "epoch": 0.8929889298892989, "percentage": 89.3, "elapsed_time": "3 days, 0:37:35", "remaining_time": "8:42:11"}
+{"current_steps": 12110, "total_steps": 13550, "loss": 2.8553, "learning_rate": 3.4010655745731865e-06, "epoch": 0.8937269372693727, "percentage": 89.37, "elapsed_time": "3 days, 0:41:12", "remaining_time": "8:38:35"}
+{"current_steps": 12120, "total_steps": 13550, "loss": 2.7823, "learning_rate": 3.354526181066003e-06, "epoch": 0.8944649446494465, "percentage": 89.45, "elapsed_time": "3 days, 0:44:48", "remaining_time": "8:34:59"}
+{"current_steps": 12130, "total_steps": 13550, "loss": 2.7281, "learning_rate": 3.308296347962875e-06, "epoch": 0.8952029520295203, "percentage": 89.52, "elapsed_time": "3 days, 0:48:24", "remaining_time": "8:31:23"}
+{"current_steps": 12140, "total_steps": 13550, "loss": 2.8478, "learning_rate": 3.2623763820658237e-06, "epoch": 0.8959409594095941, "percentage": 89.59, "elapsed_time": "3 days, 0:52:01", "remaining_time": "8:27:47"}
+{"current_steps": 12150, "total_steps": 13550, "loss": 2.7823, "learning_rate": 3.2167665881204567e-06, "epoch": 0.8966789667896679, "percentage": 89.67, "elapsed_time": "3 days, 0:55:37", "remaining_time": "8:24:11"}
+{"current_steps": 12160, "total_steps": 13550, "loss": 2.8281, "learning_rate": 3.171467268813938e-06, "epoch": 0.8974169741697416, "percentage": 89.74, "elapsed_time": "3 days, 0:59:14", "remaining_time": "8:20:35"}
+{"current_steps": 12170, "total_steps": 13550, "loss": 2.7918, "learning_rate": 3.1264787247729908e-06, "epoch": 0.8981549815498155, "percentage": 89.82, "elapsed_time": "3 days, 1:02:51", "remaining_time": "8:16:59"}
+{"current_steps": 12180, "total_steps": 13550, "loss": 2.793, "learning_rate": 3.0818012545618835e-06, "epoch": 0.8988929889298893, "percentage": 89.89, "elapsed_time": "3 days, 1:06:27", "remaining_time": "8:13:23"}
+{"current_steps": 12190, "total_steps": 13550, "loss": 2.7829, "learning_rate": 3.0374351546804514e-06, "epoch": 0.8996309963099631, "percentage": 89.96, "elapsed_time": "3 days, 1:10:03", "remaining_time": "8:09:47"}
+{"current_steps": 12200, "total_steps": 13550, "loss": 2.8107, "learning_rate": 2.9933807195621445e-06, "epoch": 0.9003690036900369, "percentage": 90.04, "elapsed_time": "3 days, 1:13:40", "remaining_time": "8:06:11"}
+{"current_steps": 12210, "total_steps": 13550, "loss": 2.8524, "learning_rate": 2.9496382415720723e-06, "epoch": 0.9011070110701107, "percentage": 90.11, "elapsed_time": "3 days, 1:17:16", "remaining_time": "8:02:35"}
+{"current_steps": 12220, "total_steps": 13550, "loss": 2.8215, "learning_rate": 2.9062080110050515e-06, "epoch": 0.9018450184501845, "percentage": 90.18, "elapsed_time": "3 days, 1:20:53", "remaining_time": "7:58:59"}
+{"current_steps": 12230, "total_steps": 13550, "loss": 2.835, "learning_rate": 2.8630903160836773e-06, "epoch": 0.9025830258302583, "percentage": 90.26, "elapsed_time": "3 days, 1:24:29", "remaining_time": "7:55:22"}
+{"current_steps": 12240, "total_steps": 13550, "loss": 2.829, "learning_rate": 2.820285442956422e-06, "epoch": 0.9033210332103321, "percentage": 90.33, "elapsed_time": "3 days, 1:28:05", "remaining_time": "7:51:46"}
+{"current_steps": 12250, "total_steps": 13550, "loss": 2.7945, "learning_rate": 2.7777936756957333e-06, "epoch": 0.9040590405904059, "percentage": 90.41, "elapsed_time": "3 days, 1:31:42", "remaining_time": "7:48:10"}
+{"current_steps": 12260, "total_steps": 13550, "loss": 2.8904, "learning_rate": 2.7356152962961567e-06, "epoch": 0.9047970479704797, "percentage": 90.48, "elapsed_time": "3 days, 1:35:19", "remaining_time": "7:44:34"}
+{"current_steps": 12270, "total_steps": 13550, "loss": 2.8889, "learning_rate": 2.6937505846724165e-06, "epoch": 0.9055350553505535, "percentage": 90.55, "elapsed_time": "3 days, 1:38:55", "remaining_time": "7:40:58"}
+{"current_steps": 12280, "total_steps": 13550, "loss": 2.836, "learning_rate": 2.6521998186576357e-06, "epoch": 0.9062730627306274, "percentage": 90.63, "elapsed_time": "3 days, 1:42:32", "remaining_time": "7:37:22"}
+{"current_steps": 12290, "total_steps": 13550, "loss": 2.7639, "learning_rate": 2.610963274001438e-06, "epoch": 0.9070110701107011, "percentage": 90.7, "elapsed_time": "3 days, 1:46:08", "remaining_time": "7:33:46"}
+{"current_steps": 12300, "total_steps": 13550, "loss": 2.7735, "learning_rate": 2.5700412243681417e-06, "epoch": 0.9077490774907749, "percentage": 90.77, "elapsed_time": "3 days, 1:49:44", "remaining_time": "7:30:10"}
+{"current_steps": 12310, "total_steps": 13550, "loss": 2.8901, "learning_rate": 2.5294339413349076e-06, "epoch": 0.9084870848708487, "percentage": 90.85, "elapsed_time": "3 days, 1:53:21", "remaining_time": "7:26:34"}
+{"current_steps": 12320, "total_steps": 13550, "loss": 2.8662, "learning_rate": 2.4891416943900014e-06, "epoch": 0.9092250922509225, "percentage": 90.92, "elapsed_time": "3 days, 1:56:57", "remaining_time": "7:22:58"}
+{"current_steps": 12330, "total_steps": 13550, "loss": 2.8268, "learning_rate": 2.449164750930938e-06, "epoch": 0.9099630996309963, "percentage": 91.0, "elapsed_time": "3 days, 2:00:33", "remaining_time": "7:19:22"}
+{"current_steps": 12340, "total_steps": 13550, "loss": 2.8246, "learning_rate": 2.409503376262762e-06, "epoch": 0.9107011070110701, "percentage": 91.07, "elapsed_time": "3 days, 2:04:09", "remaining_time": "7:15:46"}
+{"current_steps": 12350, "total_steps": 13550, "loss": 2.7924, "learning_rate": 2.3701578335962206e-06, "epoch": 0.9114391143911439, "percentage": 91.14, "elapsed_time": "3 days, 2:07:46", "remaining_time": "7:12:10"}
+{"current_steps": 12360, "total_steps": 13550, "loss": 2.8639, "learning_rate": 2.3311283840460994e-06, "epoch": 0.9121771217712177, "percentage": 91.22, "elapsed_time": "3 days, 2:11:23", "remaining_time": "7:08:34"}
+{"current_steps": 12370, "total_steps": 13550, "loss": 2.8531, "learning_rate": 2.292415286629418e-06, "epoch": 0.9129151291512915, "percentage": 91.29, "elapsed_time": "3 days, 2:14:59", "remaining_time": "7:04:58"}
+{"current_steps": 12380, "total_steps": 13550, "loss": 2.8349, "learning_rate": 2.254018798263763e-06, "epoch": 0.9136531365313653, "percentage": 91.37, "elapsed_time": "3 days, 2:18:36", "remaining_time": "7:01:22"}
+{"current_steps": 12390, "total_steps": 13550, "loss": 2.8225, "learning_rate": 2.2159391737655466e-06, "epoch": 0.9143911439114392, "percentage": 91.44, "elapsed_time": "3 days, 2:22:12", "remaining_time": "6:57:46"}
+{"current_steps": 12400, "total_steps": 13550, "loss": 2.7716, "learning_rate": 2.1781766658483303e-06, "epoch": 0.915129151291513, "percentage": 91.51, "elapsed_time": "3 days, 2:25:49", "remaining_time": "6:54:10"}
+{"current_steps": 12410, "total_steps": 13550, "loss": 2.7796, "learning_rate": 2.1407315251211422e-06, "epoch": 0.9158671586715867, "percentage": 91.59, "elapsed_time": "3 days, 2:29:25", "remaining_time": "6:50:34"}
+{"current_steps": 12420, "total_steps": 13550, "loss": 2.8009, "learning_rate": 2.103604000086856e-06, "epoch": 0.9166051660516605, "percentage": 91.66, "elapsed_time": "3 days, 2:33:02", "remaining_time": "6:46:58"}
+{"current_steps": 12430, "total_steps": 13550, "loss": 2.8486, "learning_rate": 2.066794337140443e-06, "epoch": 0.9173431734317343, "percentage": 91.73, "elapsed_time": "3 days, 2:36:38", "remaining_time": "6:43:21"}
+{"current_steps": 12440, "total_steps": 13550, "loss": 2.7234, "learning_rate": 2.0303027805674445e-06, "epoch": 0.9180811808118081, "percentage": 91.81, "elapsed_time": "3 days, 2:40:15", "remaining_time": "6:39:45"}
+{"current_steps": 12450, "total_steps": 13550, "loss": 2.7963, "learning_rate": 1.994129572542286e-06, "epoch": 0.9188191881918819, "percentage": 91.88, "elapsed_time": "3 days, 2:43:51", "remaining_time": "6:36:09"}
+{"current_steps": 12460, "total_steps": 13550, "loss": 2.8314, "learning_rate": 1.958274953126693e-06, "epoch": 0.9195571955719557, "percentage": 91.96, "elapsed_time": "3 days, 2:47:28", "remaining_time": "6:32:33"}
+{"current_steps": 12470, "total_steps": 13550, "loss": 2.8796, "learning_rate": 1.922739160268089e-06, "epoch": 0.9202952029520295, "percentage": 92.03, "elapsed_time": "3 days, 2:51:04", "remaining_time": "6:28:57"}
+{"current_steps": 12480, "total_steps": 13550, "loss": 2.7904, "learning_rate": 1.8875224297980332e-06, "epoch": 0.9210332103321033, "percentage": 92.1, "elapsed_time": "3 days, 2:54:41", "remaining_time": "6:25:21"}
+{"current_steps": 12490, "total_steps": 13550, "loss": 2.7583, "learning_rate": 1.8526249954306241e-06, "epoch": 0.9217712177121771, "percentage": 92.18, "elapsed_time": "3 days, 2:58:18", "remaining_time": "6:21:45"}
+{"current_steps": 12500, "total_steps": 13550, "loss": 2.8608, "learning_rate": 1.8180470887609769e-06, "epoch": 0.922509225092251, "percentage": 92.25, "elapsed_time": "3 days, 3:01:54", "remaining_time": "6:18:09"}
+{"current_steps": 12510, "total_steps": 13550, "loss": 2.8282, "learning_rate": 1.7837889392636864e-06, "epoch": 0.9232472324723248, "percentage": 92.32, "elapsed_time": "3 days, 3:05:30", "remaining_time": "6:14:33"}
+{"current_steps": 12520, "total_steps": 13550, "loss": 2.8048, "learning_rate": 1.7498507742912784e-06, "epoch": 0.9239852398523986, "percentage": 92.4, "elapsed_time": "3 days, 3:09:07", "remaining_time": "6:10:57"}
+{"current_steps": 12530, "total_steps": 13550, "loss": 2.8095, "learning_rate": 1.7162328190727217e-06, "epoch": 0.9247232472324723, "percentage": 92.47, "elapsed_time": "3 days, 3:12:44", "remaining_time": "6:07:21"}
+{"current_steps": 12540, "total_steps": 13550, "loss": 2.7822, "learning_rate": 1.682935296711935e-06, "epoch": 0.9254612546125461, "percentage": 92.55, "elapsed_time": "3 days, 3:16:20", "remaining_time": "6:03:45"}
+{"current_steps": 12550, "total_steps": 13550, "loss": 2.8494, "learning_rate": 1.6499584281862935e-06, "epoch": 0.9261992619926199, "percentage": 92.62, "elapsed_time": "3 days, 3:19:56", "remaining_time": "6:00:09"}
+{"current_steps": 12560, "total_steps": 13550, "loss": 2.8629, "learning_rate": 1.6173024323451747e-06, "epoch": 0.9269372693726937, "percentage": 92.69, "elapsed_time": "3 days, 3:23:33", "remaining_time": "5:56:33"}
+{"current_steps": 12570, "total_steps": 13550, "loss": 2.8258, "learning_rate": 1.5849675259084872e-06, "epoch": 0.9276752767527675, "percentage": 92.77, "elapsed_time": "3 days, 3:27:09", "remaining_time": "5:52:57"}
+{"current_steps": 12580, "total_steps": 13550, "loss": 2.8093, "learning_rate": 1.5529539234652668e-06, "epoch": 0.9284132841328413, "percentage": 92.84, "elapsed_time": "3 days, 3:30:46", "remaining_time": "5:49:21"}
+{"current_steps": 12590, "total_steps": 13550, "loss": 2.828, "learning_rate": 1.5212618374722155e-06, "epoch": 0.9291512915129151, "percentage": 92.92, "elapsed_time": "3 days, 3:34:22", "remaining_time": "5:45:45"}
+{"current_steps": 12600, "total_steps": 13550, "loss": 2.8305, "learning_rate": 1.4898914782523143e-06, "epoch": 0.9298892988929889, "percentage": 92.99, "elapsed_time": "3 days, 3:37:59", "remaining_time": "5:42:09"}
+{"current_steps": 12610, "total_steps": 13550, "loss": 2.7875, "learning_rate": 1.458843053993403e-06, "epoch": 0.9306273062730628, "percentage": 93.06, "elapsed_time": "3 days, 3:41:35", "remaining_time": "5:38:32"}
+{"current_steps": 12620, "total_steps": 13550, "loss": 2.8113, "learning_rate": 1.4281167707468457e-06, "epoch": 0.9313653136531366, "percentage": 93.14, "elapsed_time": "3 days, 3:45:12", "remaining_time": "5:34:56"}
+{"current_steps": 12630, "total_steps": 13550, "loss": 2.8511, "learning_rate": 1.3977128324261068e-06, "epoch": 0.9321033210332104, "percentage": 93.21, "elapsed_time": "3 days, 3:48:48", "remaining_time": "5:31:20"}
+{"current_steps": 12640, "total_steps": 13550, "loss": 2.7979, "learning_rate": 1.3676314408054391e-06, "epoch": 0.9328413284132842, "percentage": 93.28, "elapsed_time": "3 days, 3:52:25", "remaining_time": "5:27:44"}
+{"current_steps": 12650, "total_steps": 13550, "loss": 2.8319, "learning_rate": 1.3378727955185244e-06, "epoch": 0.933579335793358, "percentage": 93.36, "elapsed_time": "3 days, 3:56:01", "remaining_time": "5:24:08"}
+{"current_steps": 12660, "total_steps": 13550, "loss": 2.8245, "learning_rate": 1.3084370940571577e-06, "epoch": 0.9343173431734317, "percentage": 93.43, "elapsed_time": "3 days, 3:59:37", "remaining_time": "5:20:32"}
+{"current_steps": 12670, "total_steps": 13550, "loss": 2.7542, "learning_rate": 1.2793245317699321e-06, "epoch": 0.9350553505535055, "percentage": 93.51, "elapsed_time": "3 days, 4:03:14", "remaining_time": "5:16:56"}
+{"current_steps": 12680, "total_steps": 13550, "loss": 2.7729, "learning_rate": 1.2505353018609444e-06, "epoch": 0.9357933579335793, "percentage": 93.58, "elapsed_time": "3 days, 4:06:50", "remaining_time": "5:13:20"}
+{"current_steps": 12690, "total_steps": 13550, "loss": 2.8164, "learning_rate": 1.2220695953885031e-06, "epoch": 0.9365313653136531, "percentage": 93.65, "elapsed_time": "3 days, 4:10:27", "remaining_time": "5:09:44"}
+{"current_steps": 12700, "total_steps": 13550, "loss": 2.8644, "learning_rate": 1.1939276012638723e-06, "epoch": 0.9372693726937269, "percentage": 93.73, "elapsed_time": "3 days, 4:14:03", "remaining_time": "5:06:08"}
+{"current_steps": 12710, "total_steps": 13550, "loss": 2.8716, "learning_rate": 1.1661095062500237e-06, "epoch": 0.9380073800738007, "percentage": 93.8, "elapsed_time": "3 days, 4:17:40", "remaining_time": "5:02:32"}
+{"current_steps": 12720, "total_steps": 13550, "loss": 2.8307, "learning_rate": 1.1386154949603934e-06, "epoch": 0.9387453874538746, "percentage": 93.87, "elapsed_time": "3 days, 4:21:16", "remaining_time": "4:58:56"}
+{"current_steps": 12730, "total_steps": 13550, "loss": 2.7868, "learning_rate": 1.1114457498576258e-06, "epoch": 0.9394833948339484, "percentage": 93.95, "elapsed_time": "3 days, 4:24:52", "remaining_time": "4:55:20"}
+{"current_steps": 12740, "total_steps": 13550, "loss": 2.8357, "learning_rate": 1.0846004512524211e-06, "epoch": 0.9402214022140222, "percentage": 94.02, "elapsed_time": "3 days, 4:28:28", "remaining_time": "4:51:43"}
+{"current_steps": 12750, "total_steps": 13550, "loss": 2.8843, "learning_rate": 1.0580797773022733e-06, "epoch": 0.940959409594096, "percentage": 94.1, "elapsed_time": "3 days, 4:32:05", "remaining_time": "4:48:07"}
+{"current_steps": 12760, "total_steps": 13550, "loss": 2.8038, "learning_rate": 1.03188390401035e-06, "epoch": 0.9416974169741698, "percentage": 94.17, "elapsed_time": "3 days, 4:35:41", "remaining_time": "4:44:31"}
+{"current_steps": 12770, "total_steps": 13550, "loss": 2.813, "learning_rate": 1.006013005224271e-06, "epoch": 0.9424354243542435, "percentage": 94.24, "elapsed_time": "3 days, 4:39:18", "remaining_time": "4:40:55"}
+{"current_steps": 12780, "total_steps": 13550, "loss": 2.8414, "learning_rate": 9.80467252634998e-07, "epoch": 0.9431734317343173, "percentage": 94.32, "elapsed_time": "3 days, 4:42:54", "remaining_time": "4:37:19"}
+{"current_steps": 12790, "total_steps": 13550, "loss": 2.7851, "learning_rate": 9.552468157756622e-07, "epoch": 0.9439114391143911, "percentage": 94.39, "elapsed_time": "3 days, 4:46:30", "remaining_time": "4:33:43"}
+{"current_steps": 12800, "total_steps": 13550, "loss": 2.8378, "learning_rate": 9.303518620204677e-07, "epoch": 0.9446494464944649, "percentage": 94.46, "elapsed_time": "3 days, 4:50:07", "remaining_time": "4:30:07"}
+{"current_steps": 12810, "total_steps": 13550, "loss": 2.7366, "learning_rate": 9.057825565835399e-07, "epoch": 0.9453874538745387, "percentage": 94.54, "elapsed_time": "3 days, 4:53:43", "remaining_time": "4:26:31"}
+{"current_steps": 12820, "total_steps": 13550, "loss": 2.7483, "learning_rate": 8.815390625178887e-07, "epoch": 0.9461254612546125, "percentage": 94.61, "elapsed_time": "3 days, 4:57:20", "remaining_time": "4:22:55"}
+{"current_steps": 12830, "total_steps": 13550, "loss": 2.7874, "learning_rate": 8.576215407142651e-07, "epoch": 0.9468634686346864, "percentage": 94.69, "elapsed_time": "3 days, 5:00:56", "remaining_time": "4:19:19"}
+{"current_steps": 12840, "total_steps": 13550, "loss": 2.8252, "learning_rate": 8.340301499001446e-07, "epoch": 0.9476014760147602, "percentage": 94.76, "elapsed_time": "3 days, 5:04:33", "remaining_time": "4:15:43"}
+{"current_steps": 12850, "total_steps": 13550, "loss": 2.8445, "learning_rate": 8.107650466386285e-07, "epoch": 0.948339483394834, "percentage": 94.83, "elapsed_time": "3 days, 5:08:09", "remaining_time": "4:12:07"}
+{"current_steps": 12860, "total_steps": 13550, "loss": 2.8411, "learning_rate": 7.878263853274281e-07, "epoch": 0.9490774907749078, "percentage": 94.91, "elapsed_time": "3 days, 5:11:46", "remaining_time": "4:08:30"}
+{"current_steps": 12870, "total_steps": 13550, "loss": 2.8118, "learning_rate": 7.652143181978655e-07, "epoch": 0.9498154981549816, "percentage": 94.98, "elapsed_time": "3 days, 5:15:22", "remaining_time": "4:04:54"}
+{"current_steps": 12880, "total_steps": 13550, "loss": 2.8086, "learning_rate": 7.429289953138019e-07, "epoch": 0.9505535055350554, "percentage": 95.06, "elapsed_time": "3 days, 5:18:58", "remaining_time": "4:01:18"}
+{"current_steps": 12890, "total_steps": 13550, "loss": 2.8468, "learning_rate": 7.209705645706944e-07, "epoch": 0.9512915129151291, "percentage": 95.13, "elapsed_time": "3 days, 5:22:35", "remaining_time": "3:57:42"}
+{"current_steps": 12900, "total_steps": 13550, "loss": 2.8114, "learning_rate": 6.993391716946019e-07, "epoch": 0.9520295202952029, "percentage": 95.2, "elapsed_time": "3 days, 5:26:11", "remaining_time": "3:54:06"}
+{"current_steps": 12910, "total_steps": 13550, "loss": 2.8352, "learning_rate": 6.780349602411918e-07, "epoch": 0.9527675276752767, "percentage": 95.28, "elapsed_time": "3 days, 5:29:48", "remaining_time": "3:50:30"}
+{"current_steps": 12920, "total_steps": 13550, "loss": 2.8013, "learning_rate": 6.570580715948404e-07, "epoch": 0.9535055350553505, "percentage": 95.35, "elapsed_time": "3 days, 5:33:24", "remaining_time": "3:46:54"}
+{"current_steps": 12930, "total_steps": 13550, "loss": 2.8368, "learning_rate": 6.364086449676232e-07, "epoch": 0.9542435424354243, "percentage": 95.42, "elapsed_time": "3 days, 5:37:01", "remaining_time": "3:43:18"}
+{"current_steps": 12940, "total_steps": 13550, "loss": 2.8559, "learning_rate": 6.160868173984591e-07, "epoch": 0.9549815498154982, "percentage": 95.5, "elapsed_time": "3 days, 5:40:37", "remaining_time": "3:39:42"}
+{"current_steps": 12950, "total_steps": 13550, "loss": 2.85, "learning_rate": 5.960927237521563e-07, "epoch": 0.955719557195572, "percentage": 95.57, "elapsed_time": "3 days, 5:44:14", "remaining_time": "3:36:06"}
+{"current_steps": 12960, "total_steps": 13550, "loss": 2.9074, "learning_rate": 5.764264967185462e-07, "epoch": 0.9564575645756458, "percentage": 95.65, "elapsed_time": "3 days, 5:47:50", "remaining_time": "3:32:30"}
+{"current_steps": 12970, "total_steps": 13550, "loss": 2.7595, "learning_rate": 5.570882668115784e-07, "epoch": 0.9571955719557196, "percentage": 95.72, "elapsed_time": "3 days, 5:51:27", "remaining_time": "3:28:54"}
+{"current_steps": 12980, "total_steps": 13550, "loss": 2.8024, "learning_rate": 5.380781623684661e-07, "epoch": 0.9579335793357934, "percentage": 95.79, "elapsed_time": "3 days, 5:55:04", "remaining_time": "3:25:17"}
+{"current_steps": 12990, "total_steps": 13550, "loss": 2.8231, "learning_rate": 5.193963095488419e-07, "epoch": 0.9586715867158672, "percentage": 95.87, "elapsed_time": "3 days, 5:58:40", "remaining_time": "3:21:41"}
+{"current_steps": 13000, "total_steps": 13550, "loss": 2.8898, "learning_rate": 5.010428323339033e-07, "epoch": 0.959409594095941, "percentage": 95.94, "elapsed_time": "3 days, 6:02:17", "remaining_time": "3:18:05"}
+{"current_steps": 13010, "total_steps": 13550, "loss": 2.8558, "learning_rate": 4.830178525256079e-07, "epoch": 0.9601476014760147, "percentage": 96.01, "elapsed_time": "3 days, 6:05:53", "remaining_time": "3:14:29"}
+{"current_steps": 13020, "total_steps": 13550, "loss": 2.8007, "learning_rate": 4.653214897458513e-07, "epoch": 0.9608856088560885, "percentage": 96.09, "elapsed_time": "3 days, 6:09:29", "remaining_time": "3:10:53"}
+{"current_steps": 13030, "total_steps": 13550, "loss": 2.8271, "learning_rate": 4.4795386143567374e-07, "epoch": 0.9616236162361623, "percentage": 96.16, "elapsed_time": "3 days, 6:13:06", "remaining_time": "3:07:17"}
+{"current_steps": 13040, "total_steps": 13550, "loss": 2.8371, "learning_rate": 4.309150828544939e-07, "epoch": 0.9623616236162361, "percentage": 96.24, "elapsed_time": "3 days, 6:16:42", "remaining_time": "3:03:41"}
+{"current_steps": 13050, "total_steps": 13550, "loss": 2.8808, "learning_rate": 4.1420526707933727e-07, "epoch": 0.9630996309963099, "percentage": 96.31, "elapsed_time": "3 days, 6:20:19", "remaining_time": "3:00:05"}
+{"current_steps": 13060, "total_steps": 13550, "loss": 2.8506, "learning_rate": 3.978245250040702e-07, "epoch": 0.9638376383763838, "percentage": 96.38, "elapsed_time": "3 days, 6:23:56", "remaining_time": "2:56:29"}
+{"current_steps": 13070, "total_steps": 13550, "loss": 2.8261, "learning_rate": 3.817729653386892e-07, "epoch": 0.9645756457564576, "percentage": 96.46, "elapsed_time": "3 days, 6:27:33", "remaining_time": "2:52:53"}
+{"current_steps": 13080, "total_steps": 13550, "loss": 2.8319, "learning_rate": 3.660506946085829e-07, "epoch": 0.9653136531365314, "percentage": 96.53, "elapsed_time": "3 days, 6:31:09", "remaining_time": "2:49:17"}
+{"current_steps": 13090, "total_steps": 13550, "loss": 2.8326, "learning_rate": 3.506578171538377e-07, "epoch": 0.9660516605166052, "percentage": 96.61, "elapsed_time": "3 days, 6:34:45", "remaining_time": "2:45:40"}
+{"current_steps": 13100, "total_steps": 13550, "loss": 2.7896, "learning_rate": 3.355944351285278e-07, "epoch": 0.966789667896679, "percentage": 96.68, "elapsed_time": "3 days, 6:38:21", "remaining_time": "2:42:04"}
+{"current_steps": 13110, "total_steps": 13550, "loss": 2.8499, "learning_rate": 3.2086064850004314e-07, "epoch": 0.9675276752767528, "percentage": 96.75, "elapsed_time": "3 days, 6:41:58", "remaining_time": "2:38:28"}
+{"current_steps": 13120, "total_steps": 13550, "loss": 2.8005, "learning_rate": 3.064565550484455e-07, "epoch": 0.9682656826568266, "percentage": 96.83, "elapsed_time": "3 days, 6:45:34", "remaining_time": "2:34:52"}
+{"current_steps": 13130, "total_steps": 13550, "loss": 2.8419, "learning_rate": 2.9238225036579693e-07, "epoch": 0.9690036900369003, "percentage": 96.9, "elapsed_time": "3 days, 6:49:11", "remaining_time": "2:31:16"}
+{"current_steps": 13140, "total_steps": 13550, "loss": 2.8581, "learning_rate": 2.7863782785552685e-07, "epoch": 0.9697416974169741, "percentage": 96.97, "elapsed_time": "3 days, 6:52:48", "remaining_time": "2:27:40"}
+{"current_steps": 13150, "total_steps": 13550, "loss": 2.8275, "learning_rate": 2.65223378731827e-07, "epoch": 0.9704797047970479, "percentage": 97.05, "elapsed_time": "3 days, 6:56:24", "remaining_time": "2:24:04"}
+{"current_steps": 13160, "total_steps": 13550, "loss": 2.8673, "learning_rate": 2.521389920190298e-07, "epoch": 0.9712177121771217, "percentage": 97.12, "elapsed_time": "3 days, 7:00:01", "remaining_time": "2:20:28"}
+{"current_steps": 13170, "total_steps": 13550, "loss": 2.9407, "learning_rate": 2.3938475455103083e-07, "epoch": 0.9719557195571956, "percentage": 97.2, "elapsed_time": "3 days, 7:03:37", "remaining_time": "2:16:52"}
+{"current_steps": 13180, "total_steps": 13550, "loss": 2.8481, "learning_rate": 2.269607509707006e-07, "epoch": 0.9726937269372694, "percentage": 97.27, "elapsed_time": "3 days, 7:07:13", "remaining_time": "2:13:16"}
+{"current_steps": 13190, "total_steps": 13550, "loss": 2.7954, "learning_rate": 2.1486706372932375e-07, "epoch": 0.9734317343173432, "percentage": 97.34, "elapsed_time": "3 days, 7:10:50", "remaining_time": "2:09:39"}
+{"current_steps": 13200, "total_steps": 13550, "loss": 2.8533, "learning_rate": 2.031037730860774e-07, "epoch": 0.974169741697417, "percentage": 97.42, "elapsed_time": "3 days, 7:14:26", "remaining_time": "2:06:03"}
+{"current_steps": 13210, "total_steps": 13550, "loss": 2.8151, "learning_rate": 1.916709571074482e-07, "epoch": 0.9749077490774908, "percentage": 97.49, "elapsed_time": "3 days, 7:18:03", "remaining_time": "2:02:27"}
+{"current_steps": 13220, "total_steps": 13550, "loss": 2.8355, "learning_rate": 1.8056869166677703e-07, "epoch": 0.9756457564575646, "percentage": 97.56, "elapsed_time": "3 days, 7:21:39", "remaining_time": "1:58:51"}
+{"current_steps": 13230, "total_steps": 13550, "loss": 2.8121, "learning_rate": 1.6979705044369297e-07, "epoch": 0.9763837638376384, "percentage": 97.64, "elapsed_time": "3 days, 7:25:15", "remaining_time": "1:55:15"}
+{"current_steps": 13240, "total_steps": 13550, "loss": 2.9067, "learning_rate": 1.5935610492366915e-07, "epoch": 0.9771217712177122, "percentage": 97.71, "elapsed_time": "3 days, 7:28:51", "remaining_time": "1:51:39"}
+{"current_steps": 13250, "total_steps": 13550, "loss": 2.7666, "learning_rate": 1.4924592439753416e-07, "epoch": 0.977859778597786, "percentage": 97.79, "elapsed_time": "3 days, 7:32:28", "remaining_time": "1:48:03"}
+{"current_steps": 13260, "total_steps": 13550, "loss": 2.7254, "learning_rate": 1.394665759610003e-07, "epoch": 0.9785977859778597, "percentage": 97.86, "elapsed_time": "3 days, 7:36:05", "remaining_time": "1:44:27"}
+{"current_steps": 13270, "total_steps": 13550, "loss": 2.778, "learning_rate": 1.3001812451423068e-07, "epoch": 0.9793357933579335, "percentage": 97.93, "elapsed_time": "3 days, 7:39:41", "remaining_time": "1:40:51"}
+{"current_steps": 13280, "total_steps": 13550, "loss": 2.809, "learning_rate": 1.209006327614226e-07, "epoch": 0.9800738007380074, "percentage": 98.01, "elapsed_time": "3 days, 7:43:17", "remaining_time": "1:37:15"}
+{"current_steps": 13290, "total_steps": 13550, "loss": 2.8325, "learning_rate": 1.1211416121035823e-07, "epoch": 0.9808118081180812, "percentage": 98.08, "elapsed_time": "3 days, 7:46:54", "remaining_time": "1:33:38"}
+{"current_steps": 13300, "total_steps": 13550, "loss": 2.7841, "learning_rate": 1.036587681720269e-07, "epoch": 0.981549815498155, "percentage": 98.15, "elapsed_time": "3 days, 7:50:30", "remaining_time": "1:30:02"}
+{"current_steps": 13310, "total_steps": 13550, "loss": 2.8358, "learning_rate": 9.55345097602256e-08, "epoch": 0.9822878228782288, "percentage": 98.23, "elapsed_time": "3 days, 7:54:06", "remaining_time": "1:26:26"}
+{"current_steps": 13320, "total_steps": 13550, "loss": 2.8313, "learning_rate": 8.774143989119798e-08, "epoch": 0.9830258302583026, "percentage": 98.3, "elapsed_time": "3 days, 7:57:43", "remaining_time": "1:22:50"}
+{"current_steps": 13330, "total_steps": 13550, "loss": 2.8781, "learning_rate": 8.027961028328479e-08, "epoch": 0.9837638376383764, "percentage": 98.38, "elapsed_time": "3 days, 8:01:19", "remaining_time": "1:19:14"}
+{"current_steps": 13340, "total_steps": 13550, "loss": 2.7926, "learning_rate": 7.314907045653519e-08, "epoch": 0.9845018450184502, "percentage": 98.45, "elapsed_time": "3 days, 8:04:56", "remaining_time": "1:15:38"}
+{"current_steps": 13350, "total_steps": 13550, "loss": 2.7885, "learning_rate": 6.634986773244034e-08, "epoch": 0.985239852398524, "percentage": 98.52, "elapsed_time": "3 days, 8:08:32", "remaining_time": "1:12:02"}
+{"current_steps": 13360, "total_steps": 13550, "loss": 2.7721, "learning_rate": 5.988204723356705e-08, "epoch": 0.9859778597785978, "percentage": 98.6, "elapsed_time": "3 days, 8:12:09", "remaining_time": "1:08:26"}
+{"current_steps": 13370, "total_steps": 13550, "loss": 2.8138, "learning_rate": 5.374565188329683e-08, "epoch": 0.9867158671586715, "percentage": 98.67, "elapsed_time": "3 days, 8:15:46", "remaining_time": "1:04:50"}
+{"current_steps": 13380, "total_steps": 13550, "loss": 2.7988, "learning_rate": 4.794072240550951e-08, "epoch": 0.9874538745387453, "percentage": 98.75, "elapsed_time": "3 days, 8:19:22", "remaining_time": "1:01:13"}
+{"current_steps": 13390, "total_steps": 13550, "loss": 2.7823, "learning_rate": 4.246729732434451e-08, "epoch": 0.9881918819188192, "percentage": 98.82, "elapsed_time": "3 days, 8:22:58", "remaining_time": "0:57:37"}
+{"current_steps": 13400, "total_steps": 13550, "loss": 2.872, "learning_rate": 3.7325412963912235e-08, "epoch": 0.988929889298893, "percentage": 98.89, "elapsed_time": "3 days, 8:26:35", "remaining_time": "0:54:01"}
+{"current_steps": 13410, "total_steps": 13550, "loss": 2.9374, "learning_rate": 3.251510344807751e-08, "epoch": 0.9896678966789668, "percentage": 98.97, "elapsed_time": "3 days, 8:30:11", "remaining_time": "0:50:25"}
+{"current_steps": 13420, "total_steps": 13550, "loss": 2.7839, "learning_rate": 2.8036400700232058e-08, "epoch": 0.9904059040590406, "percentage": 99.04, "elapsed_time": "3 days, 8:33:48", "remaining_time": "0:46:49"}
+{"current_steps": 13430, "total_steps": 13550, "loss": 2.8689, "learning_rate": 2.3889334443055744e-08, "epoch": 0.9911439114391144, "percentage": 99.11, "elapsed_time": "3 days, 8:37:24", "remaining_time": "0:43:13"}
+{"current_steps": 13440, "total_steps": 13550, "loss": 2.9239, "learning_rate": 2.007393219836118e-08, "epoch": 0.9918819188191882, "percentage": 99.19, "elapsed_time": "3 days, 8:41:01", "remaining_time": "0:39:37"}
+{"current_steps": 13450, "total_steps": 13550, "loss": 2.8412, "learning_rate": 1.6590219286871655e-08, "epoch": 0.992619926199262, "percentage": 99.26, "elapsed_time": "3 days, 8:44:38", "remaining_time": "0:36:01"}
+{"current_steps": 13460, "total_steps": 13550, "loss": 2.7462, "learning_rate": 1.3438218828076832e-08, "epoch": 0.9933579335793358, "percentage": 99.34, "elapsed_time": "3 days, 8:48:14", "remaining_time": "0:32:25"}
+{"current_steps": 13470, "total_steps": 13550, "loss": 2.8598, "learning_rate": 1.0617951740077292e-08, "epoch": 0.9940959409594096, "percentage": 99.41, "elapsed_time": "3 days, 8:51:50", "remaining_time": "0:28:48"}
+{"current_steps": 13480, "total_steps": 13550, "loss": 2.8083, "learning_rate": 8.12943673943467e-09, "epoch": 0.9948339483394834, "percentage": 99.48, "elapsed_time": "3 days, 8:55:27", "remaining_time": "0:25:12"}
+{"current_steps": 13490, "total_steps": 13550, "loss": 2.929, "learning_rate": 5.9726903410661786e-09, "epoch": 0.9955719557195571, "percentage": 99.56, "elapsed_time": "3 days, 8:59:03", "remaining_time": "0:21:36"}
+{"current_steps": 13500, "total_steps": 13550, "loss": 2.844, "learning_rate": 4.147726858100276e-09, "epoch": 0.996309963099631, "percentage": 99.63, "elapsed_time": "3 days, 9:02:40", "remaining_time": "0:18:00"}
+{"current_steps": 13510, "total_steps": 13550, "loss": 2.8096, "learning_rate": 2.6545584018211613e-09, "epoch": 0.9970479704797048, "percentage": 99.7, "elapsed_time": "3 days, 9:06:17", "remaining_time": "0:14:24"}
+{"current_steps": 13520, "total_steps": 13550, "loss": 2.8317, "learning_rate": 1.4931948815744e-09, "epoch": 0.9977859778597786, "percentage": 99.78, "elapsed_time": "3 days, 9:09:53", "remaining_time": "0:10:48"}
+{"current_steps": 13530, "total_steps": 13550, "loss": 2.8792, "learning_rate": 6.636440046892123e-10, "epoch": 0.9985239852398524, "percentage": 99.85, "elapsed_time": "3 days, 9:13:29", "remaining_time": "0:07:12"}
+{"current_steps": 13540, "total_steps": 13550, "loss": 2.8205, "learning_rate": 1.6591127643961202e-10, "epoch": 0.9992619926199262, "percentage": 99.93, "elapsed_time": "3 days, 9:17:06", "remaining_time": "0:03:36"}
+{"current_steps": 13550, "total_steps": 13550, "loss": 2.8161, "learning_rate": 0.0, "epoch": 1.0, "percentage": 100.0, "elapsed_time": "3 days, 9:20:43", "remaining_time": "0:00:00"}
+{"current_steps": 13550, "total_steps": 13550, "epoch": 1.0, "percentage": 100.0, "elapsed_time": "3 days, 9:20:43", "remaining_time": "0:00:00"}