First Upload.

.gitattributes

@@ -33,3 +33,25 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+pytorch_model-00001-of-00022.bin filter=lfs diff=lfs merge=lfs -text
+pytorch_model-00016-of-00022.bin filter=lfs diff=lfs merge=lfs -text
+pytorch_model-00020-of-00022.bin filter=lfs diff=lfs merge=lfs -text
+pytorch_model-00002-of-00022.bin filter=lfs diff=lfs merge=lfs -text
+pytorch_model-00004-of-00022.bin filter=lfs diff=lfs merge=lfs -text
+pytorch_model-00008-of-00022.bin filter=lfs diff=lfs merge=lfs -text
+pytorch_model-00013-of-00022.bin filter=lfs diff=lfs merge=lfs -text
+pytorch_model-00014-of-00022.bin filter=lfs diff=lfs merge=lfs -text
+pytorch_model-00015-of-00022.bin filter=lfs diff=lfs merge=lfs -text
+pytorch_model-00021-of-00022.bin filter=lfs diff=lfs merge=lfs -text
+pytorch_model-00022-of-00022.bin filter=lfs diff=lfs merge=lfs -text
+pytorch_model-00005-of-00022.bin filter=lfs diff=lfs merge=lfs -text
+pytorch_model-00010-of-00022.bin filter=lfs diff=lfs merge=lfs -text
+pytorch_model-00012-of-00022.bin filter=lfs diff=lfs merge=lfs -text
+pytorch_model-00019-of-00022.bin filter=lfs diff=lfs merge=lfs -text
+pytorch_model-00007-of-00022.bin filter=lfs diff=lfs merge=lfs -text
+pytorch_model-00009-of-00022.bin filter=lfs diff=lfs merge=lfs -text
+pytorch_model-00017-of-00022.bin filter=lfs diff=lfs merge=lfs -text
+pytorch_model-00018-of-00022.bin filter=lfs diff=lfs merge=lfs -text
+pytorch_model-00003-of-00022.bin filter=lfs diff=lfs merge=lfs -text
+pytorch_model-00006-of-00022.bin filter=lfs diff=lfs merge=lfs -text
+pytorch_model-00011-of-00022.bin filter=lfs diff=lfs merge=lfs -text

added_tokens.json

@@ -0,0 +1,6 @@
+{
+  "[cls]": 100259,
+  "[pad]": 100258,
+  "[sep]": 100260,
+  "[tsk]": 100257
+}

config.json

@@ -0,0 +1,43 @@
+{
+  "activation_function": "gelu_fast",
+  "attn_pdrop": 0.1,
+  "auto_map": {
+    "AutoConfig": "configuration_flm.FLMConfig",
+    "AutoModel": "modeling_flm.FLM",
+    "AutoModelForCausalLM": "modeling_flm.FLM"
+  },
+  "bos_token_id": 100351,
+  "cls_token_id": 100351,
+  "embd_pdrop": 0.1,
+  "eos_token_id": 100351,
+  "initializer_range": 0.02,
+  "input_mult": 1.0,
+  "layer_norm_epsilon": 1e-05,
+  "model_type": "flm",
+  "mup_base_width": 256,
+  "mup_scale_factor": 40.0,
+  "n_embd": 10240,
+  "n_head": 80,
+  "n_inner": 40960,
+  "n_layer": 80,
+  "n_positions": 2048,
+  "output_mult": 2.0,
+  "relative_encoding": "rotary",
+  "reorder_and_upcast_attn": true,
+  "resid_pdrop": 0.1,
+  "rotary_theta": 10000,
+  "rotary_use_xpos": true,
+  "rotary_xpos_scale_base": 512,
+  "scale_attn_by_inverse_layer_idx": true,
+  "scale_attn_weights": true,
+  "summary_activation": null,
+  "summary_first_dropout": 0.1,
+  "summary_proj_to_labels": true,
+  "summary_type": "cls_index",
+  "summary_use_proj": true,
+  "tokenizer_class": "FLMTokenizer",
+  "transformers_version": "4.30.2",
+  "use_cache": true,
+  "use_mup": true,
+  "vocab_size": 100352
+}

configuration_flm.py

@@ -0,0 +1,208 @@
+# coding=utf-8
+# Copyright
+
+""" Cofe-AI FLM configuration"""
+
+from transformers.configuration_utils import PretrainedConfig
+from transformers.utils import logging
+
+
+
+logger = logging.get_logger(__name__)
+
+FLM_PRETRAINED_CONFIG_ARCHIVE_MAP = {
+    # "freelm": "xxxx/config.json",
+    # "freelm-medium": "xxxx/config.json",
+    # "freelm-large": "xxxx/config.json",
+    # "freelm-xl": "xxxx/config.json",
+}
+
+
+class FLMConfig(PretrainedConfig):
+    """
+    xxxxxx
+
+    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 50257):
+            Vocabulary size of the GPT-2 model. Defines the number of different tokens that can be represented by the
+            `inputs_ids` passed when calling [`GPT2Model`] or [`TFGPT2Model`].
+        n_positions (`int`, *optional*, defaults to 1024):
+            The maximum sequence length that this model might ever be used with. Typically set this to something large
+            just in case (e.g., 512 or 1024 or 2048).
+        n_embd (`int`, *optional*, defaults to 768):
+            Dimensionality of the embeddings and hidden states.
+        n_layer (`int`, *optional*, defaults to 12):
+            Number of hidden layers in the Transformer encoder.
+        n_head (`int`, *optional*, defaults to 12):
+            Number of attention heads for each attention layer in the Transformer encoder.
+        n_inner (`int`, *optional*, defaults to None):
+            Dimensionality of the inner feed-forward layers. `None` will set it to 4 times n_embd
+        activation_function (`str`, *optional*, defaults to `"gelu"`):
+            Activation function, to be selected in the list `["relu", "silu", "gelu", "tanh", "gelu_new"]`.
+        resid_pdrop (`float`, *optional*, defaults to 0.1):
+            The dropout probability for all fully connected layers in the embeddings, encoder, and pooler.
+        embd_pdrop (`int`, *optional*, defaults to 0.1):
+            The dropout ratio for the embeddings.
+        attn_pdrop (`float`, *optional*, defaults to 0.1):
+            The dropout ratio for the attention.
+        layer_norm_epsilon (`float`, *optional*, defaults to 1e-5):
+            The epsilon to use in the layer normalization layers.
+        initializer_range (`float`, *optional*, defaults to 0.02):
+            The standard deviation of the truncated_normal_initializer for initializing all weight matrices.
+        summary_type (`string`, *optional*, defaults to `"cls_index"`):
+            Argument used when doing sequence summary, used in the models [`GPT2DoubleHeadsModel`] and
+            [`TFGPT2DoubleHeadsModel`].
+
+            Has to be one of the following options:
+
+                - `"last"`: Take the last token hidden state (like XLNet).
+                - `"first"`: Take the first token hidden state (like BERT).
+                - `"mean"`: Take the mean of all tokens hidden states.
+                - `"cls_index"`: Supply a Tensor of classification token position (like GPT/GPT-2).
+                - `"attn"`: Not implemented now, use multi-head attention.
+        summary_use_proj (`bool`, *optional*, defaults to `True`):
+            Argument used when doing sequence summary, used in the models [`GPT2DoubleHeadsModel`] and
+            [`TFGPT2DoubleHeadsModel`].
+
+            Whether or not to add a projection after the vector extraction.
+        summary_activation (`str`, *optional*):
+            Argument used when doing sequence summary. Used in for the multiple choice head in
+            [`GPT2DoubleHeadsModel`].
+
+            Pass `"tanh"` for a tanh activation to the output, any other value will result in no activation.
+        summary_proj_to_labels (`bool`, *optional*, defaults to `True`):
+            Argument used when doing sequence summary, used in the models [`GPT2DoubleHeadsModel`] and
+            [`TFGPT2DoubleHeadsModel`].
+
+            Whether the projection outputs should have `config.num_labels` or `config.hidden_size` classes.
+        summary_first_dropout (`float`, *optional*, defaults to 0.1):
+            Argument used when doing sequence summary, used in the models [`GPT2DoubleHeadsModel`] and
+            [`TFGPT2DoubleHeadsModel`].
+
+            The dropout ratio to be used after the projection and activation.
+        scale_attn_weights (`bool`, *optional*, defaults to `True`):
+            Scale attention weights by dividing by sqrt(hidden_size)..
+        use_cache (`bool`, *optional*, defaults to `True`):
+            Whether or not the model should return the last key/values attentions (not used by all models).
+        scale_attn_by_inverse_layer_idx (`bool`, *optional*, defaults to `False`):
+            Whether to additionally scale attention weights by `1 / layer_idx + 1`.
+        reorder_and_upcast_attn (`bool`, *optional*, defaults to `False`):
+            Whether to scale keys (K) prior to computing attention (dot-product) and upcast attention
+            dot-product/softmax to float() when training with mixed precision.
+
+    Example:
+
+    ```python
+    >>> from transformers import GPT2Config, GPT2Model
+
+    >>> # Initializing a GPT2 configuration
+    >>> configuration = GPT2Config()
+
+    >>> # Initializing a model (with random weights) from the configuration
+    >>> model = GPT2Model(configuration)
+
+    >>> # Accessing the model configuration
+    >>> configuration = model.config
+    ```"""
+
+    model_type = "flm"
+    keys_to_ignore_at_inference = ["past_key_values"]
+    attribute_map = {
+        "hidden_size": "n_embd",
+        "max_position_embeddings": "n_positions",
+        "num_attention_heads": "n_head",
+        "num_hidden_layers": "n_layer",
+    }
+
+    def __init__(
+            self,
+            vocab_size=50257,
+            n_positions=1024,
+            n_embd=768,
+            n_layer=12,
+            n_head=12,
+            n_inner=None,
+            activation_function="gelu_new",
+            resid_pdrop=0.1,
+            embd_pdrop=0.1,
+            attn_pdrop=0.1,
+            layer_norm_epsilon=1e-5,
+            initializer_range=0.02,
+            summary_type="cls_index",
+            summary_use_proj=True,
+            summary_activation=None,
+            summary_proj_to_labels=True,
+            summary_first_dropout=0.1,
+            scale_attn_weights=True,
+            use_cache=True,
+            bos_token_id=50256,
+            eos_token_id=50256,
+            cls_token_id=50256,
+            sep_token_id=50256,
+            pad_token_id=50256,
+            scale_attn_by_inverse_layer_idx=False,
+            reorder_and_upcast_attn=False,
+            relative_encoding=None,
+            rotary_theta=10000,
+            rotary_use_xpos=True,
+            rotary_xpos_scale_base=512,
+            use_mup=False,
+            mup_scale_factor=1.0,
+            output_mult=1.0,
+            input_mult=1.0,
+            mup_base_width=256,
+            **kwargs,
+    ):
+        self.vocab_size = vocab_size
+        self.n_positions = n_positions
+        self.n_embd = n_embd
+        self.n_layer = n_layer
+        self.n_head = n_head
+        self.n_inner = n_inner
+        self.activation_function = activation_function
+        self.resid_pdrop = resid_pdrop
+        self.embd_pdrop = embd_pdrop
+        self.attn_pdrop = attn_pdrop
+        self.layer_norm_epsilon = layer_norm_epsilon
+        self.initializer_range = initializer_range
+        self.summary_type = summary_type
+        self.summary_use_proj = summary_use_proj
+        self.summary_activation = summary_activation
+        self.summary_first_dropout = summary_first_dropout
+        self.summary_proj_to_labels = summary_proj_to_labels
+        self.scale_attn_weights = scale_attn_weights
+        self.use_cache = use_cache
+        self.scale_attn_by_inverse_layer_idx = scale_attn_by_inverse_layer_idx
+        self.reorder_and_upcast_attn = reorder_and_upcast_attn
+        self.relative_encoding = relative_encoding
+
+        # for rotary
+        self.rotary_theta = rotary_theta
+        self.rotary_use_xpos = rotary_use_xpos
+        self.rotary_xpos_scale_base = rotary_xpos_scale_base
+
+        # for mup
+        self.use_mup = use_mup
+        self.mup_scale_factor = mup_scale_factor
+        self.output_mult = output_mult
+        self.input_mult = input_mult
+        self.mup_base_width = mup_base_width
+
+        self.bos_token_id = bos_token_id
+        self.eos_token_id = eos_token_id
+        self.cls_token_id = cls_token_id
+        self.sep_token_id = sep_token_id
+        self.pad_token_id = pad_token_id
+
+
+        self.architectures = ["FLM"]
+        self.auto_map = {
+            "AutoConfig": "configuration_flm.FLMConfig",
+            "AutoModel": "modeling_flm.FLM",
+            "AutoModelForCausalLM": "modeling_flm.FLM"
+        }
+
+        super().__init__(bos_token_id=bos_token_id, eos_token_id=eos_token_id, **kwargs)

modeling_flm.py

@@ -0,0 +1,954 @@
+# coding=utf-8
+# Copyright 2023 EleutherAI and the HuggingFace Inc. team. All rights reserved.
+#
+# This code is based on OpenAI's GPT-2 library. It has been modified from its
+# original forms to accommodate minor architectural differences compared to GPT-2.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+from typing import Optional, Tuple, Union
+
+import math
+import torch
+import torch.nn.functional as f
+from einops import rearrange, repeat
+from torch import einsum, nn
+from torch.cuda.amp import autocast
+from transformers.activations import ACT2FN
+from transformers.modeling_outputs import (
+    BaseModelOutputWithPastAndCrossAttentions,
+    CausalLMOutputWithCrossAttentions,
+    SequenceClassifierOutputWithPast,
+)
+from transformers.modeling_utils import PreTrainedModel
+from transformers.pytorch_utils import find_pruneable_heads_and_indices, prune_conv1d_layer
+from transformers.utils import add_start_docstrings, add_start_docstrings_to_model_forward, logging, replace_return_docstrings
+from transformers.utils.model_parallel_utils import assert_device_map, get_device_map
+from .configuration_flm import FLMConfig
+
+
+class Conv1D(nn.Module):
+
+    def __init__(self, nf, nx):
+        super().__init__()
+        self.nf = nf
+        self.weight = nn.Parameter(torch.empty(nx, nf))
+        self.bias = nn.Parameter(torch.zeros(nf))
+        nn.init.normal_(self.weight, std=0.02)
+
+    def forward(self, x):
+        x = torch.matmul(x, self.weight) + self.bias
+        return x
+
+
+logger = logging.get_logger(__name__)
+
+
+def exists(v):
+    return v is not None
+
+
+class RotaryEmbedding(nn.Module):
+    def __init__(self, dim, use_xpos=False, xpos_scale_base=512, theta=10000):
+        super().__init__()
+        inv_freq = 1.0 / (theta ** (torch.arange(0, dim, 2).float() / dim))
+        self.register_buffer('inv_freq', inv_freq)
+        self.cache = dict()
+        self.cache_scale = dict()
+        self.use_xpos = use_xpos
+        if not use_xpos:
+            self.register_buffer('scale', None)
+            return
+        scale = (torch.arange(0, dim, 2) + 0.4 * dim) / (1.4 * dim)
+        self.register_buffer('scale', scale)
+        self.scale_base = xpos_scale_base
+
+    def forward(self, seq, cache_key=None):
+
+        if cache_key is not None and cache_key in self.cache:
+            return self.cache[cache_key]
+
+        inv_freq = self.inv_freq.to(device=seq.device)
+        freqs = einsum('i , j -> i j', seq, inv_freq)
+        # first part even vector components, second part odd vector components,
+        #  2 * dim in dimension size
+        scale = torch.cat((freqs, freqs), dim=-1)
+        if exists(cache_key):
+            self.cache[cache_key] = scale
+        return scale
+
+    def rotate_queries_and_keys(self, q, k, seq_dim=-2):
+        """
+        use this only when xpos is activated.
+        """
+        assert self.use_xpos and q.device == k.device
+        device, seq_len_k, seq_len_q = k.device, k.shape[seq_dim], q.shape[seq_dim]
+        pos_seq_k = torch.arange(seq_len_k, device=device, dtype=torch.float32)
+        pos_seq_q = torch.arange(seq_len_k - seq_len_q, seq_len_k, device=device, dtype=torch.float32)
+        freqs_k = self.forward(pos_seq_k, cache_key=f"{0}:{seq_len_k}")
+        freqs_q = self.forward(pos_seq_q, cache_key=f"{seq_len_k - seq_len_q}:{seq_len_k}")
+        scale_k = self.get_scale(pos_seq_k)
+        scale_q = self.get_scale(pos_seq_q, offset=seq_len_k - seq_len_q)  # 这里的offset是Q相对于K的offset
+        rotated_q = apply_rotary_emb(freqs_q, q, scale=scale_q)
+        rotated_k = apply_rotary_emb(freqs_k, k, scale=scale_k ** -1)
+        return rotated_q, rotated_k
+
+    def get_scale(self, t, cache_key=None, offset=0, ):
+        assert self.use_xpos, 'This function is only useful for xpos.'
+        if exists(cache_key) and cache_key in self.cache_scale:
+            return self.cache_scale[cache_key]
+        if callable(t):
+            t = t()
+        length = len(t)
+        min_pos = -(length + offset) // 2
+        max_pos = length + offset + min_pos
+        power = torch.arange(min_pos, max_pos, 1).to(device=self.scale.device) / self.scale_base
+        scale = self.scale ** rearrange(power, 'n -> n 1')
+        scale = scale[-length:, :]
+        scale = torch.cat((scale, scale), dim=-1)
+        if exists(cache_key):
+            self.cache_scale[cache_key] = scale
+        return scale
+
+
+def rotate_half(x):
+    """
+    change sign so the last dimension becomes [-odd, +even]
+    """
+    x1, x2 = torch.chunk(x, 2, dim=-1)
+    return torch.cat((-x2, x1), dim=-1)
+
+
+def apply_rotary_emb(freqs, t, start_index=0, scale=1.):
+    """
+    freq: seqlen  x  dim
+       t: [batchsize  *  headnum  ,  seqlen  , dim (dim_of_head actually)]
+    """
+    dtype_t = t.dtype
+    freqs = freqs.to(device=t.device)
+    if isinstance(scale, torch.Tensor):
+        scale = scale.to(device=t.device)
+    rot_dim = freqs.shape[-1]
+    end_index = start_index + rot_dim
+    t_left, t, t_right = t[..., :start_index], t[..., start_index:end_index], t[..., end_index:]
+    t = (t * freqs.cos() + rotate_half(t) * freqs.sin()) * scale
+    rotated = torch.cat((t_left, t, t_right), dim=-1)
+    rotated = rotated.to(dtype=dtype_t)
+    return rotated
+
+
+class FLMAttention(nn.Module):
+    def __init__(self, config, is_cross_attention=False, layer_idx=None):
+        super().__init__()
+
+        max_positions = config.max_position_embeddings
+        self.register_buffer(
+            "bias",
+            torch.tril(torch.ones((max_positions, max_positions), dtype=torch.bool)).view(
+                1, 1, max_positions, max_positions
+            ),
+        )
+        self.register_buffer("masked_bias", torch.tensor(-1e4))
+
+        self.embed_dim = config.hidden_size
+        self.num_heads = config.num_attention_heads
+        self.head_dim = self.embed_dim // self.num_heads
+        self.split_size = self.embed_dim
+        if self.head_dim * self.num_heads != self.embed_dim:
+            raise ValueError(
+                f"`embed_dim` must be divisible by num_heads (got `embed_dim`: {self.embed_dim} and `num_heads`:"
+                f" {self.num_heads})."
+            )
+
+        self.scale_attn_weights = config.scale_attn_weights
+        self.is_cross_attention = is_cross_attention
+
+        # Layer-wise attention scaling, reordering, and upcasting
+        self.scale_attn_by_inverse_layer_idx = config.scale_attn_by_inverse_layer_idx
+        # for alignment with megatron-lm in softmax scale
+        self.layer_idx = max(1, layer_idx)
+        self.reorder_and_upcast_attn = config.reorder_and_upcast_attn
+
+        self.relative_encoding = config.relative_encoding
+        self.rotary_use_xpos = config.rotary_use_xpos
+
+        self.use_mup = config.use_mup
+
+        if self.is_cross_attention:
+            self.c_attn = Conv1D(2 * self.embed_dim, self.embed_dim)
+            self.q_attn = Conv1D(self.embed_dim, self.embed_dim)
+        else:
+            self.c_attn = Conv1D(3 * self.embed_dim, self.embed_dim)
+        self.c_proj = Conv1D(self.embed_dim, self.embed_dim)
+
+        self.attn_dropout = nn.Dropout(config.attn_pdrop)
+        self.resid_dropout = nn.Dropout(config.resid_pdrop)
+
+        self.pruned_heads = set()
+
+    def set_max_positions(self, max_positions, device='cuda'):
+        self.max_positions = max_positions
+        self.register_buffer(
+            "bias",
+            torch.tril(torch.ones((self.max_positions, self.max_positions), dtype=torch.bool)).view(
+                1, 1, self.max_positions, self.max_positions
+            ).to(device=device)
+        )
+
+    def prune_heads(self, heads):
+        if len(heads) == 0:
+            return
+        heads, index = find_pruneable_heads_and_indices(heads, self.num_heads, self.head_dim, self.pruned_heads)
+        index_attn = torch.cat([index, index + self.split_size, index + (2 * self.split_size)])
+
+        # Prune conv1d layers
+        self.c_attn = prune_conv1d_layer(self.c_attn, index_attn, dim=1)
+        self.c_proj = prune_conv1d_layer(self.c_proj, index, dim=0)
+
+        # Update hyper params
+        self.split_size = (self.split_size // self.num_heads) * (self.num_heads - len(heads))
+        self.num_heads = self.num_heads - len(heads)
+        self.pruned_heads = self.pruned_heads.union(heads)
+
+    def _attn(self, query, key, value, attention_mask=None, head_mask=None):
+        # (batch, head, seq_length, head_features)
+        # batch_size, head_num, k_seq_len(q_seq_len), head_features
+        batch_size, head_num, k_seq_len, head_features = key.shape
+        _, _, q_seq_len, _ = query.shape
+        attn_weights = torch.matmul(query, key.transpose(-1, -2))
+
+        if self.scale_attn_weights:
+            if self.use_mup:
+                attn_weights = attn_weights / torch.full(
+                    [], value.size(-1) / (value.size(-1) ** 0.5), dtype=attn_weights.dtype,
+                    device=attn_weights.device
+                )
+            else:
+                attn_weights = attn_weights / torch.full(
+                    [], value.size(-1) ** 0.5, dtype=attn_weights.dtype, device=attn_weights.device
+                )
+
+        if not self.is_cross_attention:
+            # if only "normal" attention layer implements causal mask
+            query_length, key_length = query.size(-2), key.size(-2)
+            causal_mask = self.bias[:, :, key_length - query_length: key_length, :key_length]
+            mask_value = torch.finfo(attn_weights.dtype).min
+            # Need to be a tensor, otherwise we get error: `RuntimeError: expected scalar type float but found double`.
+            # Need to be on the same device, otherwise `RuntimeError: ..., x and y to be on the same device`
+            mask_value = torch.full([], mask_value, dtype=attn_weights.dtype).to(attn_weights.device)
+            attn_weights = torch.where(causal_mask, attn_weights.to(attn_weights.dtype), mask_value)
+
+        if attention_mask is not None:
+            # Apply the attention mask
+            attn_weights = attn_weights + attention_mask
+
+        attn_weights = nn.functional.softmax(attn_weights, dim=-1)
+
+        # Downcast (if necessary) back to V's dtype (if in mixed-precision) -- No-Op otherwise
+        attn_weights = attn_weights.type(value.dtype)
+        attn_weights = self.attn_dropout(attn_weights)
+
+        # Mask heads if we want to
+        if head_mask is not None:
+            attn_weights = attn_weights * head_mask
+
+        attn_output = torch.matmul(attn_weights, value)
+
+        return attn_output, attn_weights
+
+    def _upcast_and_reordered_attn(self, query, key, value, attention_mask=None, head_mask=None):
+        # Use `torch.baddbmm` (a bit more efficient w/ alpha param for scaling -- from Megatron-LM)
+        bsz, num_heads, q_seq_len, dk = query.size()
+        _, _, k_seq_len, _ = key.size()
+
+        # Preallocate attn_weights for `baddbmm`
+        attn_weights = torch.empty(bsz * num_heads, q_seq_len, k_seq_len, dtype=query.dtype, device=query.device)
+
+        # Compute Scale Factor
+        scale_factor = 1.0
+        if self.scale_attn_weights:
+            scale_factor /= float(value.size(-1)) ** 0.5
+
+        if self.scale_attn_by_inverse_layer_idx:
+            scale_factor /= float(self.layer_idx)
+        # Upcast (turn off autocast) and reorder (Scale K by 1 / root(dk))
+        with autocast(enabled=False):
+            q, k = query.reshape(-1, q_seq_len, dk), key.transpose(-1, -2).reshape(-1, dk, k_seq_len)
+            attn_weights = torch.baddbmm(attn_weights, q, k, beta=0, alpha=scale_factor)
+            attn_weights = attn_weights.reshape(bsz, num_heads, q_seq_len, k_seq_len)
+
+        if not self.is_cross_attention:
+            attn_weights = attn_weights.float()
+            if self.scale_attn_by_inverse_layer_idx:
+                attn_weights *= self.layer_idx
+            # if only "normal" attention layer implements causal mask
+            query_length, key_length = query.size(-2), key.size(-2)
+            causal_mask = self.bias[:, :, key_length - query_length: key_length, :key_length]
+            mask_value = -10000.0  # align with megatron-lm
+            # Need to be a tensor, otherwise we get error: `RuntimeError: expected scalar type float but found double`.
+            # Need to be on the same device, otherwise `RuntimeError: ..., x and y to be on the same device`
+            mask_value = torch.tensor(mask_value, dtype=attn_weights.dtype).to(attn_weights.device)
+            attn_weights = torch.where(causal_mask, attn_weights, mask_value)
+
+        if attention_mask is not None:
+            # Apply the attention mask
+            attn_weights = attn_weights + attention_mask
+        attn_weights = nn.functional.softmax(attn_weights, dim=-1)
+        # Downcast (if necessary) back to V's dtype (if in mixed-precision) -- No-Op if otherwise
+        if attn_weights.dtype != torch.float32:
+            raise RuntimeError("Error with upcasting, attn_weights does not have dtype torch.float32")
+        attn_weights = attn_weights.type(value.dtype)
+        attn_weights = self.attn_dropout(attn_weights)
+
+        # Mask heads if we want to
+        if head_mask is not None:
+            attn_weights = attn_weights * head_mask
+        attn_output = torch.matmul(attn_weights, value)
+        return attn_output, attn_weights
+
+    def _split_heads(self, tensor, num_heads, attn_head_size):
+        """
+        Splits hidden_size dim into attn_head_size and num_heads
+        """
+        new_shape = tensor.size()[:-1] + (num_heads, attn_head_size)
+        tensor = tensor.view(new_shape)
+        return tensor.permute(0, 2, 1, 3)  # (batch, head, seq_length, head_features)
+
+    def _merge_heads(self, tensor, num_heads, attn_head_size):
+        """
+        Merges attn_head_size dim and num_attn_heads dim into hidden_size
+        """
+        tensor = tensor.permute(0, 2, 1, 3).contiguous()
+        new_shape = tensor.size()[:-2] + (num_heads * attn_head_size,)
+        return tensor.view(new_shape)
+
+    def forward(
+            self,
+            hidden_states: Optional[Tuple[torch.FloatTensor]],
+            layer_past: Optional[Tuple[torch.Tensor]] = None,
+            attention_mask: Optional[torch.FloatTensor] = None,
+            head_mask: Optional[torch.FloatTensor] = None,
+            encoder_hidden_states: Optional[torch.Tensor] = None,
+            encoder_attention_mask: Optional[torch.FloatTensor] = None,
+            rotary_embedding: Optional[RotaryEmbedding] = None,
+            use_cache: Optional[bool] = False,
+            output_attentions: Optional[bool] = False,
+    ) -> Tuple[Union[torch.Tensor, Tuple[torch.Tensor]], ...]:
+        if encoder_hidden_states is not None:
+            if not hasattr(self, "q_attn"):
+                raise ValueError(
+                    "If class is used as cross attention, the weights `q_attn` have to be defined. "
+                    "Please make sure to instantiate class with `GPT2Attention(..., is_cross_attention=True)`."
+                )
+
+            query = self.q_attn(hidden_states)
+            key, value = self.c_attn(encoder_hidden_states).split(self.split_size, dim=2)
+            attention_mask = encoder_attention_mask
+        else:
+            query, key, value = self.c_attn(hidden_states).split(self.split_size, dim=2)
+
+        query = self._split_heads(query, self.num_heads, self.head_dim)
+        key = self._split_heads(key, self.num_heads, self.head_dim)
+        value = self._split_heads(value, self.num_heads, self.head_dim)
+
+        if layer_past is not None:
+            past_key, past_value = layer_past
+            key = torch.cat((past_key, key), dim=-2)
+            value = torch.cat((past_value, value), dim=-2)
+
+        if use_cache is True:
+            present = (key, value)
+        else:
+            present = None
+
+        batch_size, head_num, k_seq_len, head_features = key.shape
+        _, _, q_seq_len, _ = query.shape
+        if rotary_embedding is not None:
+            query = query.contiguous().view(batch_size * head_num, q_seq_len, head_features)
+            key = key.contiguous().view(batch_size * head_num, k_seq_len, head_features)
+
+            # batch_size * head_num,  k_seq_len(q_seq_len), head_features
+            if self.rotary_use_xpos:
+                # query: [batch_size * head_num, seqlen, hn]
+                query, key = rotary_embedding.rotate_queries_and_keys(query, key)
+            else:
+                query = rotary_embedding.rotate_queries_or_keys(query)
+                key = rotary_embedding.rotate_queries_or_keys(key)
+            # batch_size * head_num, k_seq_len(q_seq_len), head_features
+            query = query.view(batch_size, head_num, q_seq_len, head_features)
+            key = key.view(batch_size, head_num, k_seq_len, head_features)
+
+        if self.reorder_and_upcast_attn:
+            attn_output, attn_weights = self._upcast_and_reordered_attn(query, key, value, attention_mask, head_mask)
+        else:
+            attn_output, attn_weights = self._attn(query, key, value, attention_mask, head_mask)
+        attn_output = self._merge_heads(attn_output, self.num_heads, self.head_dim)
+        attn_output = self.c_proj(attn_output)
+        attn_output = self.resid_dropout(attn_output)
+        outputs = (attn_output, present)
+        if output_attentions:
+            outputs += (attn_weights,)
+
+        return outputs
+
+
+class FLMMLP(nn.Module):
+    def __init__(self, intermediate_size, config):
+        super().__init__()
+        embed_dim = config.hidden_size
+        self.c_fc = Conv1D(intermediate_size, embed_dim)
+        self.c_proj = Conv1D(embed_dim, intermediate_size)
+        self.act = ACT2FN[config.activation_function]
+        self.dropout = nn.Dropout(config.resid_pdrop)
+
+    def forward(self, hidden_states: Optional[Tuple[torch.FloatTensor]]) -> torch.FloatTensor:
+        hidden_states = self.c_fc(hidden_states)
+        hidden_states = self.act(hidden_states)
+        hidden_states = self.c_proj(hidden_states)
+        hidden_states = self.dropout(hidden_states)
+        return hidden_states
+
+
+class FLMBlock(nn.Module):
+    def __init__(self, config, layer_idx=None):
+        super().__init__()
+        hidden_size = config.hidden_size
+        inner_dim = config.n_inner if config.n_inner is not None else 4 * hidden_size
+        self.layer_idx = layer_idx
+        self.ln_1 = nn.LayerNorm(hidden_size, eps=config.layer_norm_epsilon)
+        self.attn = FLMAttention(config, layer_idx=layer_idx)
+        self.ln_2 = nn.LayerNorm(hidden_size, eps=config.layer_norm_epsilon)
+
+        if config.add_cross_attention:
+            self.crossattention = FLMAttention(config, is_cross_attention=True, layer_idx=layer_idx)
+            self.ln_cross_attn = nn.LayerNorm(hidden_size, eps=config.layer_norm_epsilon)
+
+        self.mlp = FLMMLP(inner_dim, config)
+
+    def forward(
+            self,
+            hidden_states: Optional[Tuple[torch.FloatTensor]],
+            layer_past: Optional[Tuple[torch.Tensor]] = None,
+            attention_mask: Optional[torch.FloatTensor] = None,
+            head_mask: Optional[torch.FloatTensor] = None,
+            encoder_hidden_states: Optional[torch.Tensor] = None,
+            encoder_attention_mask: Optional[torch.FloatTensor] = None,
+            rotary_embedding: Optional[RotaryEmbedding] = None,
+            use_cache: Optional[bool] = False,
+            output_attentions: Optional[bool] = False,
+    ) -> Union[Tuple[torch.Tensor], Optional[Tuple[torch.Tensor, Tuple[torch.FloatTensor, ...]]]]:
+        residual = hidden_states
+        hidden_states = self.ln_1(hidden_states)
+        attn_outputs = self.attn(
+            hidden_states,
+            layer_past=layer_past,
+            attention_mask=attention_mask,
+            head_mask=head_mask,
+            rotary_embedding=rotary_embedding,
+            use_cache=use_cache,
+            output_attentions=output_attentions
+        )
+        attn_output = attn_outputs[0]  # output_attn: a, present, (attentions)
+        outputs = attn_outputs[1:]
+        # residual connection
+        hidden_states = attn_output + residual
+
+        residual = hidden_states
+        hidden_states = self.ln_2(hidden_states)
+        feed_forward_hidden_states = self.mlp(hidden_states)
+        # residual connection
+        hidden_states = residual + feed_forward_hidden_states
+        if use_cache:
+            outputs = (hidden_states,) + outputs
+        else:
+            outputs = (hidden_states,) + outputs[1:]
+
+        return outputs
+
+
+class FLMPretrainedModel(PreTrainedModel):
+    """
+    An abstract class to handle weights initialization and a simple interface for downloading and loading pretrained
+    models.
+    """
+
+    config_class = FLMConfig
+    load_tf_weights = None
+    base_model_prefix = "transformer"
+    is_parallelizable = True
+    supports_gradient_checkpointing = True
+    _no_split_modules = ["FLMBlock"]
+
+    def __init__(self, *inputs, **kwargs):
+        super().__init__(*inputs, **kwargs)
+
+    def _init_weights(self, module):
+        """Initialize the weights."""
+        if isinstance(module, (nn.Linear, Conv1D)):
+            # 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, nn.LayerNorm):
+            module.bias.data.zero_()
+            module.weight.data.fill_(1.0)
+
+        # Reinitialize selected weights subject to the OpenAI GPT-2 Paper Scheme:
+        #   > A modified initialization which accounts for the accumulation on the residual path with model depth. Scale
+        #   > the weights of residual layers at initialization by a factor of 1/√N where N is the # of residual layers.
+        #   >   -- GPT-2 :: https://openai.com/blog/better-language-models/
+        #
+        # Reference (Megatron-LM): https://github.com/NVIDIA/Megatron-LM/blob/main/megatron/model/gpt_model.py
+        for name, p in module.named_parameters():
+            if name == "c_proj.weight":
+                # Special Scaled Initialization --> There are 2 Layer Norms per Transformer Block
+                p.data.normal_(mean=0.0, std=(self.config.initializer_range / math.sqrt(2 * self.config.n_layer)))
+
+    def _set_gradient_checkpointing(self, module, value=False):
+        if isinstance(module, FLMTransformer):
+            module.gradient_checkpointing = value
+
+
+class FLMTransformer(FLMPretrainedModel):
+    _keys_to_ignore_on_load_missing = ["attn.masked_bias"]
+
+    def __init__(self, config):
+        super().__init__(config)
+
+        self.embed_dim = config.hidden_size
+
+        self.relative_encoding = config.relative_encoding
+        self.wte = nn.Embedding(config.vocab_size, self.embed_dim)
+
+        self.use_mup = config.use_mup
+        if self.use_mup:
+            self.input_mult = config.input_mult
+
+        if self.relative_encoding is None:
+            self.wpe = nn.Embedding(config.max_position_embeddings, self.embed_dim)
+        elif self.relative_encoding == 'rotary':
+            pe_dim = config.n_embd // config.n_head
+            self.wpe = RotaryEmbedding(pe_dim,
+                                       use_xpos=config.rotary_use_xpos,
+                                       xpos_scale_base=config.rotary_xpos_scale_base,
+                                       theta=config.rotary_theta
+                                       )
+
+        else:
+            raise RuntimeError(
+                f'Unknown relative positional encoding type: `relative_encoding`={self.relative_encoding}')
+        self.drop = nn.Dropout(config.embd_pdrop)
+        self.h = nn.ModuleList([FLMBlock(config, layer_idx=i + 1) for i in range(config.num_hidden_layers)])
+        self.ln_f = nn.LayerNorm(self.embed_dim, eps=config.layer_norm_epsilon)
+
+        # Model parallel
+        self.model_parallel = False
+        self.device_map = None
+        self.gradient_checkpointing = False
+
+        # Initialize weights and apply final processing
+        self.post_init()
+
+    # @add_start_docstrings(PARALLELIZE_DOCSTRING)
+    def parallelize(self, device_map=None):
+        # Check validity of device_map
+        self.device_map = (
+            get_device_map(len(self.h), range(torch.cuda.device_count())) if device_map is None else device_map
+        )
+        assert_device_map(self.device_map, len(self.h))
+        self.model_parallel = True
+        self.first_device = "cpu" if "cpu" in self.device_map.keys() else "cuda:" + str(min(self.device_map.keys()))
+        self.last_device = "cuda:" + str(max(self.device_map.keys()))
+        self.wte = self.wte.to(self.first_device)
+        self.wpe = self.wpe.to(self.first_device)
+        # Load onto devices
+        for k, v in self.device_map.items():
+            for block in v:
+                cuda_device = "cuda:" + str(k)
+                self.h[block] = self.h[block].to(cuda_device)
+        # ln_f to last
+        self.ln_f = self.ln_f.to(self.last_device)
+
+    def deparallelize(self):
+        self.model_parallel = False
+        self.device_map = None
+        self.first_device = "cpu"
+        self.last_device = "cpu"
+        self.wte = self.wte.to("cpu")
+        self.wpe = self.wpe.to("cpu")
+        for index in range(len(self.h)):
+            self.h[index] = self.h[index].to("cpu")
+        self.ln_f = self.ln_f.to("cpu")
+        torch.cuda.empty_cache()
+
+    def get_input_embeddings(self):
+        return self.wte
+
+    def set_input_embeddings(self, new_embeddings):
+        self.wte = new_embeddings
+
+    def _prune_heads(self, heads_to_prune):
+        """
+        Prunes heads of the model. heads_to_prune: dict of {layer_num: list of heads to prune in this layer}
+        """
+        for layer, heads in heads_to_prune.items():
+            self.h[layer].attn.prune_heads(heads)
+
+    def forward(
+            self,
+            input_ids: Optional[torch.LongTensor] = None,
+            past_key_values: Optional[Tuple[Tuple[torch.Tensor]]] = None,
+            attention_mask: Optional[torch.FloatTensor] = None,
+            token_type_ids: Optional[torch.LongTensor] = None,
+            position_ids: Optional[torch.LongTensor] = None,
+            head_mask: Optional[torch.FloatTensor] = None,
+            inputs_embeds: Optional[torch.FloatTensor] = None,
+            encoder_hidden_states: Optional[torch.Tensor] = None,
+            encoder_attention_mask: 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,
+    ) -> Union[Tuple, BaseModelOutputWithPastAndCrossAttentions]:
+        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 not None and inputs_embeds is not None:
+            raise ValueError("You cannot specify both input_ids and inputs_embeds at the same time")
+        elif input_ids is not None:
+            input_shape = input_ids.size()
+            input_ids = input_ids.view(-1, input_shape[-1])
+            batch_size = input_ids.shape[0]
+        elif inputs_embeds is not None:
+            input_shape = inputs_embeds.size()[:-1]
+            batch_size = inputs_embeds.shape[0]
+        else:
+            raise ValueError("You have to specify either input_ids or inputs_embeds")
+
+        device = input_ids.device if input_ids is not None else inputs_embeds.device
+
+        if token_type_ids is not None:
+            token_type_ids = token_type_ids.view(-1, input_shape[-1])
+        if position_ids is not None:
+            position_ids = position_ids.view(-1, input_shape[-1])
+
+        if past_key_values is None:
+            past_length = 0
+            past_key_values = tuple([None] * len(self.h))
+        else:
+            past_length = past_key_values[0][0].size(-2)
+        if position_ids is None:
+            position_ids = torch.arange(past_length, input_shape[-1] + past_length, dtype=torch.long, device=device)
+            position_ids = position_ids.unsqueeze(0).view(-1, input_shape[-1])
+
+        # GPT2Attention mask.
+        if attention_mask is not None:
+            if batch_size <= 0:
+                raise ValueError("batch_size has to be defined and > 0")
+            attention_mask = attention_mask.view(batch_size, -1)
+            # We create a 3D attention mask from a 2D tensor mask.
+            # Sizes are [batch_size, 1, 1, to_seq_length]
+            # So we can broadcast to [batch_size, num_heads, from_seq_length, to_seq_length]
+            # this attention mask is more simple than the triangular masking of causal attention
+            # used in OpenAI GPT, we just need to prepare the broadcast dimension here.
+            attention_mask = attention_mask[:, None, None, :]
+
+            # Since attention_mask is 1.0 for positions we want to attend and 0.0 for
+            # masked positions, this operation will create a tensor which is 0.0 for
+            # positions we want to attend and the dtype's smallest value for masked positions.
+            # Since we are adding it to the raw scores before the softmax, this is
+            # effectively the same as removing these entirely.
+            attention_mask = attention_mask.to(dtype=self.dtype)  # fp16 compatibility
+            attention_mask = (1.0 - attention_mask) * torch.finfo(self.dtype).min
+
+        # If a 2D or 3D attention mask is provided for the cross-attention
+        # we need to make broadcastable to [batch_size, num_heads, seq_length, seq_length]
+        if self.config.add_cross_attention and encoder_hidden_states is not None:
+            encoder_batch_size, encoder_sequence_length, _ = encoder_hidden_states.size()
+            encoder_hidden_shape = (encoder_batch_size, encoder_sequence_length)
+            if encoder_attention_mask is None:
+                encoder_attention_mask = torch.ones(encoder_hidden_shape, device=device)
+            encoder_attention_mask = self.invert_attention_mask(encoder_attention_mask)
+        else:
+            encoder_attention_mask = None
+
+        # Prepare head mask if needed
+        # 1.0 in head_mask indicate we keep the head
+        # attention_probs has shape bsz x n_heads x N x N
+        # head_mask has shape n_layer x batch x n_heads x N x N
+        head_mask = self.get_head_mask(head_mask, self.config.n_layer)
+
+        if inputs_embeds is None:
+            inputs_embeds = self.wte(input_ids)
+
+        # Mup
+        if self.use_mup:
+            inputs_embeds = inputs_embeds * self.input_mult
+        if self.relative_encoding is None:
+            position_embeds = self.wpe(position_ids)
+            hidden_states = inputs_embeds + position_embeds
+        elif self.relative_encoding == 'rotary':
+            hidden_states = inputs_embeds
+        if token_type_ids is not None:
+            token_type_embeds = self.wte(token_type_ids)
+            hidden_states = hidden_states + token_type_embeds
+        hidden_states = self.drop(hidden_states)
+
+        output_shape = input_shape + (hidden_states.size(-1),)
+
+        presents = () if use_cache else None
+        all_self_attentions = () if output_attentions else None
+        all_cross_attentions = () if output_attentions and self.config.add_cross_attention else None
+        all_hidden_states = () if output_hidden_states else None
+        for i, (block, layer_past) in enumerate(zip(self.h, past_key_values)):
+
+            # Model parallel
+            if self.model_parallel:
+                torch.cuda.set_device(hidden_states.device)
+                # Ensure layer_past is on same device as hidden_states (might not be correct)
+                if layer_past is not None:
+                    layer_past = tuple(past_state.to(hidden_states.device) for past_state in layer_past)
+                # Ensure that attention_mask is always on the same device as hidden_states
+                if attention_mask is not None:
+                    attention_mask = attention_mask.to(hidden_states.device)
+                if isinstance(head_mask, torch.Tensor):
+                    head_mask = head_mask.to(hidden_states.device)
+            if output_hidden_states:
+                all_hidden_states = all_hidden_states + (hidden_states,)
+
+            if self.gradient_checkpointing and self.training:
+
+                if use_cache:
+                    logger.warning(
+                        "`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`..."
+                    )
+                    use_cache = False
+
+                def create_custom_forward(module):
+                    def custom_forward(*inputs):
+                        # None for past_key_value
+                        return module(*inputs, use_cache, output_attentions)
+
+                    return custom_forward
+
+                outputs = torch.utils.checkpoint.checkpoint(
+                    create_custom_forward(block),
+                    hidden_states,
+                    None,
+                    attention_mask,
+                    head_mask[i],
+                    encoder_hidden_states,
+                    encoder_attention_mask,
+                )
+            else:
+                outputs = block(
+                    hidden_states,
+                    layer_past=layer_past,
+                    attention_mask=attention_mask,
+                    head_mask=head_mask[i],
+                    encoder_hidden_states=encoder_hidden_states,
+                    encoder_attention_mask=encoder_attention_mask,
+                    rotary_embedding=self.wpe if self.relative_encoding == 'rotary' else None,
+                    use_cache=use_cache,
+                    output_attentions=output_attentions
+                )
+
+            hidden_states = outputs[0]
+            if use_cache is True:
+                presents = presents + (outputs[1],)
+
+            if output_attentions:
+                all_self_attentions = all_self_attentions + (outputs[2 if use_cache else 1],)
+                if self.config.add_cross_attention:
+                    all_cross_attentions = all_cross_attentions + (outputs[3 if use_cache else 2],)
+
+            # Model Parallel: If it's the last layer for that device, put things on the next device
+            if self.model_parallel:
+                for k, v in self.device_map.items():
+                    if i == v[-1] and "cuda:" + str(k) != self.last_device:
+                        hidden_states = hidden_states.to("cuda:" + str(k + 1))
+
+        hidden_states = self.ln_f(hidden_states)
+
+        hidden_states = hidden_states.view(output_shape)
+        # Add last hidden state
+        if output_hidden_states:
+            all_hidden_states = all_hidden_states + (hidden_states,)
+
+        if not return_dict:
+            return tuple(
+                v
+                for v in [hidden_states, presents, all_hidden_states, all_self_attentions, all_cross_attentions]
+                if v is not None
+            )
+
+        return BaseModelOutputWithPastAndCrossAttentions(
+            last_hidden_state=hidden_states,
+            past_key_values=presents,
+            hidden_states=all_hidden_states,
+            attentions=all_self_attentions,
+            cross_attentions=all_cross_attentions,
+        )
+
+
+class FLM(FLMPretrainedModel):
+    _keys_to_ignore_on_load_missing = [r"attn.masked_bias", r"attn.bias", r"lm_head.weight"]
+
+    def __init__(self, config):
+        super().__init__(config)
+        self.transformer = FLMTransformer(config)
+        self.lm_head = nn.Linear(config.n_embd, config.vocab_size, bias=False)
+        self.use_mup = config.use_mup
+        if self.use_mup:
+            self.mup_scale_factor = config.mup_scale_factor
+            self.output_mult = config.output_mult / self.mup_scale_factor
+
+        # Model parallel
+        self.model_parallel = False
+        self.device_map = None
+
+        # Initialize weights and apply final processing
+        self.post_init()
+
+    def set_max_positions(self, max_positions):
+        for layer in self.transformer.h:
+            device = layer.ln_1.weight.device
+            layer.attn.set_max_positions(max_positions, device=device)
+
+    def parallelize(self, device_map=None):
+        self.device_map = (
+            get_device_map(len(self.transformer.h), range(torch.cuda.device_count()))
+            if device_map is None
+            else device_map
+        )
+        assert_device_map(self.device_map, len(self.transformer.h))
+        self.transformer.parallelize(self.device_map)
+        self.lm_head = self.lm_head.to(self.transformer.first_device)
+        self.model_parallel = True
+
+    def deparallelize(self):
+        self.transformer.deparallelize()
+        self.transformer = self.transformer.to("cpu")
+        self.lm_head = self.lm_head.to("cpu")
+        self.model_parallel = False
+        torch.cuda.empty_cache()
+
+    def get_output_embeddings(self):
+        return self.lm_head
+
+    def set_output_embeddings(self, new_embeddings):
+        self.lm_head = new_embeddings
+
+    def prepare_inputs_for_generation(self, input_ids, past=None, **kwargs):
+        token_type_ids = kwargs.get("token_type_ids", None)
+        # only last token for inputs_ids if past is defined in kwargs
+        if past:
+            input_ids = input_ids[:, -1].unsqueeze(-1)
+            if token_type_ids is not None:
+                token_type_ids = token_type_ids[:, -1].unsqueeze(-1)
+
+        attention_mask = kwargs.get("attention_mask", None)
+        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:
+                position_ids = position_ids[:, -1].unsqueeze(-1)
+        else:
+            position_ids = None
+        return {
+            "input_ids": input_ids,
+            "past_key_values": past,
+            "use_cache": kwargs.get("use_cache"),
+            "position_ids": position_ids,
+            "attention_mask": attention_mask,
+            "token_type_ids": token_type_ids,
+        }
+
+    def forward(
+            self,
+            input_ids: Optional[torch.LongTensor] = None,
+            past_key_values: Optional[Tuple[Tuple[torch.Tensor]]] = None,
+            attention_mask: Optional[torch.FloatTensor] = None,
+            token_type_ids: Optional[torch.LongTensor] = None,
+            position_ids: Optional[torch.LongTensor] = None,
+            head_mask: Optional[torch.FloatTensor] = None,
+            inputs_embeds: Optional[torch.FloatTensor] = None,
+            encoder_hidden_states: Optional[torch.Tensor] = None,
+            encoder_attention_mask: 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,
+    ) -> Union[Tuple, CausalLMOutputWithCrossAttentions, SequenceClassifierOutputWithPast]:
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+        transformer_outputs = self.transformer(
+            input_ids,
+            past_key_values=past_key_values,
+            attention_mask=attention_mask,
+            token_type_ids=token_type_ids,
+            position_ids=position_ids,
+            head_mask=head_mask,
+            inputs_embeds=inputs_embeds,
+            encoder_hidden_states=encoder_hidden_states,
+            encoder_attention_mask=encoder_attention_mask,
+            use_cache=use_cache,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+            return_dict=return_dict
+        )
+        hidden_states = transformer_outputs[0]
+
+        # Set device for model parallelism
+        if self.model_parallel:
+            torch.cuda.set_device(self.transformer.first_device)
+            hidden_states = hidden_states.to(self.lm_head.weight.device)
+
+        lm_logits = self.lm_head(hidden_states)
+        # Mup
+        if self.use_mup:
+            lm_logits = lm_logits * self.output_mult
+
+        loss = None
+        if labels is not None:
+            # Shift so that tokens < n predict n
+            shift_logits = lm_logits[..., :-1, :].contiguous()
+            shift_labels = labels[..., 1:].contiguous()
+            # Flatten the tokens
+            loss_fct = nn.CrossEntropyLoss()
+            loss = loss_fct(shift_logits.view(-1, shift_logits.size(-1)), shift_labels.view(-1))
+
+        if not return_dict:
+            output = (lm_logits,) + transformer_outputs[1:]
+            return ((loss,) + output) if loss is not None else output
+
+        return CausalLMOutputWithCrossAttentions(
+            loss=loss,
+            logits=lm_logits,
+            past_key_values=transformer_outputs.past_key_values,
+            hidden_states=transformer_outputs.hidden_states,
+            attentions=transformer_outputs.attentions,
+            cross_attentions=transformer_outputs.cross_attentions,
+        )

pytorch_model-00001-of-00022.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9f9e2c9695e2a3fe04b3067055d69faf5acc844fad447595b7f3b0203cdc31f0
+size 9639365909

pytorch_model-00002-of-00022.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d3fb35d432efe84988fba6b3c42c99fd0e5d29075c8c0f53ae982ee5ef535e55
+size 9253654563

pytorch_model-00003-of-00022.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:354fe82894eb784714d8656abc405be0e11ce8083bb39de3a8e5329caffccab4
+size 9262024111

pytorch_model-00004-of-00022.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7dc627ef3223487a59562a0ec2cc20321169546846787ab2b744e972d239345f
+size 9891189603

pytorch_model-00005-of-00022.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:650b9c4491b6e36c537430e2938c372a0ebb8bd40c1fe43d2c36f80e0b773023
+size 9471759795

pytorch_model-00006-of-00022.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:18f884ca0727c74dacf5d3848f2db7655565285c4dd8ee0c2225b5cc1021db9c
+size 9253654627

pytorch_model-00007-of-00022.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0b9b869f1c22ff677d05e47e3f7fc7931940fede69f935abd021c8c32e81d897
+size 9262024175

pytorch_model-00008-of-00022.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1698c72d96d2a6e1180535ad12cf7fd9ac6a7bda028899bfbbcb9c82280afbe2
+size 9891189603

pytorch_model-00009-of-00022.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:18e065d2a6d2c514e228dcb7584bf6a5d096194d31cd2c6a70cbd72dabd68625
+size 9471759795

pytorch_model-00010-of-00022.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:480f83a3cba0ed3acff65c023ec50cf8a5f8f1fc64993b4812568fcefa32a4cd
+size 9253654627

pytorch_model-00011-of-00022.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:004fa22a47e6e56c02370cda5daad0c593fa531885d582d9a6dc8ea45707780e
+size 9262024175

pytorch_model-00012-of-00022.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:701720da23937731d07a636b133a0a80b269753a08f0158b506df29f6664fb4d
+size 9891189603

pytorch_model-00013-of-00022.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6ab873f367cd52ef369dee9c2cbbe855562c5c0d2bbb93923fc6b92fa63b4442
+size 9471759795

pytorch_model-00014-of-00022.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:eb1ffd9796a55082db31ba8f3ed301cea7db12ffbcc412111ae141334194d772
+size 9253654627

pytorch_model-00015-of-00022.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fd5f5820df358c81b6d103dee27e875228a32494e40790449d60aa719d0d9d0a
+size 9262024175

pytorch_model-00016-of-00022.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1c23ec06c1e4079a288f04d9182a31c53bf09cccf5bef8bad249ce1bbb40c671
+size 9891189603

pytorch_model-00017-of-00022.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2be582fe182b80f9943d9c271377654c3c401158c302a99cb6da67afd66497b5
+size 9471759795

pytorch_model-00018-of-00022.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:121011d9c151b31a303dec52d929f9f7072c318ef3ed405caeaaee8b95b4f34d
+size 9253654627

pytorch_model-00019-of-00022.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cf41a8c67f850478de690fda435ed8ad49cda96b95854dc241bb876234bc0cb6
+size 9262024175

pytorch_model-00020-of-00022.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:39d64a3505241ca634d82330a9208d95dcb65e499894c3dd857b09394c3b4995
+size 9891189603

pytorch_model-00021-of-00022.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b93ce871a511a23158e20decee9164981db9bb2c8aa82bcc8a17e21e775880e1
+size 9471759795

pytorch_model-00022-of-00022.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6060cef05f3a10c6395ed8668f0dd3004eb4964df1791c4f6273187f28fe0cae
+size 5042094403

pytorch_model.bin.index.json

@@ -0,0 +1,1131 @@
+{
+  "metadata": {
+    "total_size": 204074229920
+  },
+  "weight_map": {
+    "lm_head.weight": "pytorch_model-00001-of-00022.bin",
+    "transformer.h.0.attn.bias": "pytorch_model-00001-of-00022.bin",
+    "transformer.h.0.attn.c_attn.bias": "pytorch_model-00001-of-00022.bin",
+    "transformer.h.0.attn.c_attn.weight": "pytorch_model-00001-of-00022.bin",
+    "transformer.h.0.attn.c_proj.bias": "pytorch_model-00001-of-00022.bin",
+    "transformer.h.0.attn.c_proj.weight": "pytorch_model-00001-of-00022.bin",
+    "transformer.h.0.attn.masked_bias": "pytorch_model-00001-of-00022.bin",
+    "transformer.h.0.ln_1.bias": "pytorch_model-00001-of-00022.bin",
+    "transformer.h.0.ln_1.weight": "pytorch_model-00001-of-00022.bin",
+    "transformer.h.0.ln_2.bias": "pytorch_model-00001-of-00022.bin",
+    "transformer.h.0.ln_2.weight": "pytorch_model-00001-of-00022.bin",
+    "transformer.h.0.mlp.c_fc.bias": "pytorch_model-00001-of-00022.bin",
+    "transformer.h.0.mlp.c_fc.weight": "pytorch_model-00001-of-00022.bin",
+    "transformer.h.0.mlp.c_proj.bias": "pytorch_model-00001-of-00022.bin",
+    "transformer.h.0.mlp.c_proj.weight": "pytorch_model-00001-of-00022.bin",
+    "transformer.h.1.attn.bias": "pytorch_model-00001-of-00022.bin",
+    "transformer.h.1.attn.c_attn.bias": "pytorch_model-00001-of-00022.bin",
+    "transformer.h.1.attn.c_attn.weight": "pytorch_model-00001-of-00022.bin",
+    "transformer.h.1.attn.c_proj.bias": "pytorch_model-00001-of-00022.bin",
+    "transformer.h.1.attn.c_proj.weight": "pytorch_model-00001-of-00022.bin",
+    "transformer.h.1.attn.masked_bias": "pytorch_model-00001-of-00022.bin",
+    "transformer.h.1.ln_1.bias": "pytorch_model-00001-of-00022.bin",
+    "transformer.h.1.ln_1.weight": "pytorch_model-00001-of-00022.bin",
+    "transformer.h.1.ln_2.bias": "pytorch_model-00001-of-00022.bin",
+    "transformer.h.1.ln_2.weight": "pytorch_model-00001-of-00022.bin",
+    "transformer.h.1.mlp.c_fc.bias": "pytorch_model-00001-of-00022.bin",
+    "transformer.h.1.mlp.c_fc.weight": "pytorch_model-00001-of-00022.bin",
+    "transformer.h.1.mlp.c_proj.bias": "pytorch_model-00001-of-00022.bin",
+    "transformer.h.1.mlp.c_proj.weight": "pytorch_model-00001-of-00022.bin",
+    "transformer.h.10.attn.bias": "pytorch_model-00003-of-00022.bin",
+    "transformer.h.10.attn.c_attn.bias": "pytorch_model-00003-of-00022.bin",
+    "transformer.h.10.attn.c_attn.weight": "pytorch_model-00003-of-00022.bin",
+    "transformer.h.10.attn.c_proj.bias": "pytorch_model-00003-of-00022.bin",
+    "transformer.h.10.attn.c_proj.weight": "pytorch_model-00003-of-00022.bin",
+    "transformer.h.10.attn.masked_bias": "pytorch_model-00003-of-00022.bin",
+    "transformer.h.10.ln_1.bias": "pytorch_model-00003-of-00022.bin",
+    "transformer.h.10.ln_1.weight": "pytorch_model-00003-of-00022.bin",
+    "transformer.h.10.ln_2.bias": "pytorch_model-00003-of-00022.bin",
+    "transformer.h.10.ln_2.weight": "pytorch_model-00003-of-00022.bin",
+    "transformer.h.10.mlp.c_fc.bias": "pytorch_model-00004-of-00022.bin",
+    "transformer.h.10.mlp.c_fc.weight": "pytorch_model-00004-of-00022.bin",
+    "transformer.h.10.mlp.c_proj.bias": "pytorch_model-00004-of-00022.bin",
+    "transformer.h.10.mlp.c_proj.weight": "pytorch_model-00004-of-00022.bin",
+    "transformer.h.11.attn.bias": "pytorch_model-00004-of-00022.bin",
+    "transformer.h.11.attn.c_attn.bias": "pytorch_model-00004-of-00022.bin",
+    "transformer.h.11.attn.c_attn.weight": "pytorch_model-00004-of-00022.bin",
+    "transformer.h.11.attn.c_proj.bias": "pytorch_model-00004-of-00022.bin",
+    "transformer.h.11.attn.c_proj.weight": "pytorch_model-00004-of-00022.bin",
+    "transformer.h.11.attn.masked_bias": "pytorch_model-00004-of-00022.bin",
+    "transformer.h.11.ln_1.bias": "pytorch_model-00004-of-00022.bin",
+    "transformer.h.11.ln_1.weight": "pytorch_model-00004-of-00022.bin",
+    "transformer.h.11.ln_2.bias": "pytorch_model-00004-of-00022.bin",
+    "transformer.h.11.ln_2.weight": "pytorch_model-00004-of-00022.bin",
+    "transformer.h.11.mlp.c_fc.bias": "pytorch_model-00004-of-00022.bin",
+    "transformer.h.11.mlp.c_fc.weight": "pytorch_model-00004-of-00022.bin",
+    "transformer.h.11.mlp.c_proj.bias": "pytorch_model-00004-of-00022.bin",
+    "transformer.h.11.mlp.c_proj.weight": "pytorch_model-00004-of-00022.bin",
+    "transformer.h.12.attn.bias": "pytorch_model-00004-of-00022.bin",
+    "transformer.h.12.attn.c_attn.bias": "pytorch_model-00004-of-00022.bin",
+    "transformer.h.12.attn.c_attn.weight": "pytorch_model-00004-of-00022.bin",
+    "transformer.h.12.attn.c_proj.bias": "pytorch_model-00004-of-00022.bin",
+    "transformer.h.12.attn.c_proj.weight": "pytorch_model-00004-of-00022.bin",
+    "transformer.h.12.attn.masked_bias": "pytorch_model-00004-of-00022.bin",
+    "transformer.h.12.ln_1.bias": "pytorch_model-00004-of-00022.bin",
+    "transformer.h.12.ln_1.weight": "pytorch_model-00004-of-00022.bin",
+    "transformer.h.12.ln_2.bias": "pytorch_model-00004-of-00022.bin",
+    "transformer.h.12.ln_2.weight": "pytorch_model-00004-of-00022.bin",
+    "transformer.h.12.mlp.c_fc.bias": "pytorch_model-00004-of-00022.bin",
+    "transformer.h.12.mlp.c_fc.weight": "pytorch_model-00004-of-00022.bin",
+    "transformer.h.12.mlp.c_proj.bias": "pytorch_model-00004-of-00022.bin",
+    "transformer.h.12.mlp.c_proj.weight": "pytorch_model-00004-of-00022.bin",
+    "transformer.h.13.attn.bias": "pytorch_model-00004-of-00022.bin",
+    "transformer.h.13.attn.c_attn.bias": "pytorch_model-00004-of-00022.bin",
+    "transformer.h.13.attn.c_attn.weight": "pytorch_model-00004-of-00022.bin",
+    "transformer.h.13.attn.c_proj.bias": "pytorch_model-00004-of-00022.bin",
+    "transformer.h.13.attn.c_proj.weight": "pytorch_model-00004-of-00022.bin",
+    "transformer.h.13.attn.masked_bias": "pytorch_model-00004-of-00022.bin",
+    "transformer.h.13.ln_1.bias": "pytorch_model-00004-of-00022.bin",
+    "transformer.h.13.ln_1.weight": "pytorch_model-00004-of-00022.bin",
+    "transformer.h.13.ln_2.bias": "pytorch_model-00004-of-00022.bin",
+    "transformer.h.13.ln_2.weight": "pytorch_model-00004-of-00022.bin",
+    "transformer.h.13.mlp.c_fc.bias": "pytorch_model-00004-of-00022.bin",
+    "transformer.h.13.mlp.c_fc.weight": "pytorch_model-00004-of-00022.bin",
+    "transformer.h.13.mlp.c_proj.bias": "pytorch_model-00004-of-00022.bin",
+    "transformer.h.13.mlp.c_proj.weight": "pytorch_model-00004-of-00022.bin",
+    "transformer.h.14.attn.bias": "pytorch_model-00004-of-00022.bin",
+    "transformer.h.14.attn.c_attn.bias": "pytorch_model-00004-of-00022.bin",
+    "transformer.h.14.attn.c_attn.weight": "pytorch_model-00004-of-00022.bin",
+    "transformer.h.14.attn.c_proj.bias": "pytorch_model-00005-of-00022.bin",
+    "transformer.h.14.attn.c_proj.weight": "pytorch_model-00005-of-00022.bin",
+    "transformer.h.14.attn.masked_bias": "pytorch_model-00004-of-00022.bin",
+    "transformer.h.14.ln_1.bias": "pytorch_model-00004-of-00022.bin",
+    "transformer.h.14.ln_1.weight": "pytorch_model-00004-of-00022.bin",
+    "transformer.h.14.ln_2.bias": "pytorch_model-00005-of-00022.bin",
+    "transformer.h.14.ln_2.weight": "pytorch_model-00005-of-00022.bin",
+    "transformer.h.14.mlp.c_fc.bias": "pytorch_model-00005-of-00022.bin",
+    "transformer.h.14.mlp.c_fc.weight": "pytorch_model-00005-of-00022.bin",
+    "transformer.h.14.mlp.c_proj.bias": "pytorch_model-00005-of-00022.bin",
+    "transformer.h.14.mlp.c_proj.weight": "pytorch_model-00005-of-00022.bin",
+    "transformer.h.15.attn.bias": "pytorch_model-00005-of-00022.bin",
+    "transformer.h.15.attn.c_attn.bias": "pytorch_model-00005-of-00022.bin",
+    "transformer.h.15.attn.c_attn.weight": "pytorch_model-00005-of-00022.bin",
+    "transformer.h.15.attn.c_proj.bias": "pytorch_model-00005-of-00022.bin",
+    "transformer.h.15.attn.c_proj.weight": "pytorch_model-00005-of-00022.bin",
+    "transformer.h.15.attn.masked_bias": "pytorch_model-00005-of-00022.bin",
+    "transformer.h.15.ln_1.bias": "pytorch_model-00005-of-00022.bin",
+    "transformer.h.15.ln_1.weight": "pytorch_model-00005-of-00022.bin",
+    "transformer.h.15.ln_2.bias": "pytorch_model-00005-of-00022.bin",
+    "transformer.h.15.ln_2.weight": "pytorch_model-00005-of-00022.bin",
+    "transformer.h.15.mlp.c_fc.bias": "pytorch_model-00005-of-00022.bin",
+    "transformer.h.15.mlp.c_fc.weight": "pytorch_model-00005-of-00022.bin",
+    "transformer.h.15.mlp.c_proj.bias": "pytorch_model-00005-of-00022.bin",
+    "transformer.h.15.mlp.c_proj.weight": "pytorch_model-00005-of-00022.bin",
+    "transformer.h.16.attn.bias": "pytorch_model-00005-of-00022.bin",
+    "transformer.h.16.attn.c_attn.bias": "pytorch_model-00005-of-00022.bin",
+    "transformer.h.16.attn.c_attn.weight": "pytorch_model-00005-of-00022.bin",
+    "transformer.h.16.attn.c_proj.bias": "pytorch_model-00005-of-00022.bin",
+    "transformer.h.16.attn.c_proj.weight": "pytorch_model-00005-of-00022.bin",
+    "transformer.h.16.attn.masked_bias": "pytorch_model-00005-of-00022.bin",
+    "transformer.h.16.ln_1.bias": "pytorch_model-00005-of-00022.bin",
+    "transformer.h.16.ln_1.weight": "pytorch_model-00005-of-00022.bin",
+    "transformer.h.16.ln_2.bias": "pytorch_model-00005-of-00022.bin",
+    "transformer.h.16.ln_2.weight": "pytorch_model-00005-of-00022.bin",
+    "transformer.h.16.mlp.c_fc.bias": "pytorch_model-00005-of-00022.bin",
+    "transformer.h.16.mlp.c_fc.weight": "pytorch_model-00005-of-00022.bin",
+    "transformer.h.16.mlp.c_proj.bias": "pytorch_model-00005-of-00022.bin",
+    "transformer.h.16.mlp.c_proj.weight": "pytorch_model-00005-of-00022.bin",
+    "transformer.h.17.attn.bias": "pytorch_model-00005-of-00022.bin",
+    "transformer.h.17.attn.c_attn.bias": "pytorch_model-00005-of-00022.bin",
+    "transformer.h.17.attn.c_attn.weight": "pytorch_model-00005-of-00022.bin",
+    "transformer.h.17.attn.c_proj.bias": "pytorch_model-00005-of-00022.bin",
+    "transformer.h.17.attn.c_proj.weight": "pytorch_model-00005-of-00022.bin",
+    "transformer.h.17.attn.masked_bias": "pytorch_model-00005-of-00022.bin",
+    "transformer.h.17.ln_1.bias": "pytorch_model-00005-of-00022.bin",
+    "transformer.h.17.ln_1.weight": "pytorch_model-00005-of-00022.bin",
+    "transformer.h.17.ln_2.bias": "pytorch_model-00005-of-00022.bin",
+    "transformer.h.17.ln_2.weight": "pytorch_model-00005-of-00022.bin",
+    "transformer.h.17.mlp.c_fc.bias": "pytorch_model-00005-of-00022.bin",
+    "transformer.h.17.mlp.c_fc.weight": "pytorch_model-00005-of-00022.bin",
+    "transformer.h.17.mlp.c_proj.bias": "pytorch_model-00005-of-00022.bin",
+    "transformer.h.17.mlp.c_proj.weight": "pytorch_model-00005-of-00022.bin",
+    "transformer.h.18.attn.bias": "pytorch_model-00005-of-00022.bin",
+    "transformer.h.18.attn.c_attn.bias": "pytorch_model-00006-of-00022.bin",
+    "transformer.h.18.attn.c_attn.weight": "pytorch_model-00006-of-00022.bin",
+    "transformer.h.18.attn.c_proj.bias": "pytorch_model-00006-of-00022.bin",
+    "transformer.h.18.attn.c_proj.weight": "pytorch_model-00006-of-00022.bin",
+    "transformer.h.18.attn.masked_bias": "pytorch_model-00005-of-00022.bin",
+    "transformer.h.18.ln_1.bias": "pytorch_model-00005-of-00022.bin",
+    "transformer.h.18.ln_1.weight": "pytorch_model-00005-of-00022.bin",
+    "transformer.h.18.ln_2.bias": "pytorch_model-00006-of-00022.bin",
+    "transformer.h.18.ln_2.weight": "pytorch_model-00006-of-00022.bin",
+    "transformer.h.18.mlp.c_fc.bias": "pytorch_model-00006-of-00022.bin",
+    "transformer.h.18.mlp.c_fc.weight": "pytorch_model-00006-of-00022.bin",
+    "transformer.h.18.mlp.c_proj.bias": "pytorch_model-00006-of-00022.bin",
+    "transformer.h.18.mlp.c_proj.weight": "pytorch_model-00006-of-00022.bin",
+    "transformer.h.19.attn.bias": "pytorch_model-00006-of-00022.bin",
+    "transformer.h.19.attn.c_attn.bias": "pytorch_model-00006-of-00022.bin",
+    "transformer.h.19.attn.c_attn.weight": "pytorch_model-00006-of-00022.bin",
+    "transformer.h.19.attn.c_proj.bias": "pytorch_model-00006-of-00022.bin",
+    "transformer.h.19.attn.c_proj.weight": "pytorch_model-00006-of-00022.bin",
+    "transformer.h.19.attn.masked_bias": "pytorch_model-00006-of-00022.bin",
+    "transformer.h.19.ln_1.bias": "pytorch_model-00006-of-00022.bin",
+    "transformer.h.19.ln_1.weight": "pytorch_model-00006-of-00022.bin",
+    "transformer.h.19.ln_2.bias": "pytorch_model-00006-of-00022.bin",
+    "transformer.h.19.ln_2.weight": "pytorch_model-00006-of-00022.bin",
+    "transformer.h.19.mlp.c_fc.bias": "pytorch_model-00006-of-00022.bin",
+    "transformer.h.19.mlp.c_fc.weight": "pytorch_model-00006-of-00022.bin",
+    "transformer.h.19.mlp.c_proj.bias": "pytorch_model-00006-of-00022.bin",
+    "transformer.h.19.mlp.c_proj.weight": "pytorch_model-00006-of-00022.bin",
+    "transformer.h.2.attn.bias": "pytorch_model-00001-of-00022.bin",
+    "transformer.h.2.attn.c_attn.bias": "pytorch_model-00001-of-00022.bin",
+    "transformer.h.2.attn.c_attn.weight": "pytorch_model-00001-of-00022.bin",
+    "transformer.h.2.attn.c_proj.bias": "pytorch_model-00001-of-00022.bin",
+    "transformer.h.2.attn.c_proj.weight": "pytorch_model-00001-of-00022.bin",
+    "transformer.h.2.attn.masked_bias": "pytorch_model-00001-of-00022.bin",
+    "transformer.h.2.ln_1.bias": "pytorch_model-00001-of-00022.bin",
+    "transformer.h.2.ln_1.weight": "pytorch_model-00001-of-00022.bin",
+    "transformer.h.2.ln_2.bias": "pytorch_model-00001-of-00022.bin",
+    "transformer.h.2.ln_2.weight": "pytorch_model-00001-of-00022.bin",
+    "transformer.h.2.mlp.c_fc.bias": "pytorch_model-00001-of-00022.bin",
+    "transformer.h.2.mlp.c_fc.weight": "pytorch_model-00001-of-00022.bin",
+    "transformer.h.2.mlp.c_proj.bias": "pytorch_model-00001-of-00022.bin",
+    "transformer.h.2.mlp.c_proj.weight": "pytorch_model-00001-of-00022.bin",
+    "transformer.h.20.attn.bias": "pytorch_model-00006-of-00022.bin",
+    "transformer.h.20.attn.c_attn.bias": "pytorch_model-00006-of-00022.bin",
+    "transformer.h.20.attn.c_attn.weight": "pytorch_model-00006-of-00022.bin",
+    "transformer.h.20.attn.c_proj.bias": "pytorch_model-00006-of-00022.bin",
+    "transformer.h.20.attn.c_proj.weight": "pytorch_model-00006-of-00022.bin",
+    "transformer.h.20.attn.masked_bias": "pytorch_model-00006-of-00022.bin",
+    "transformer.h.20.ln_1.bias": "pytorch_model-00006-of-00022.bin",
+    "transformer.h.20.ln_1.weight": "pytorch_model-00006-of-00022.bin",
+    "transformer.h.20.ln_2.bias": "pytorch_model-00006-of-00022.bin",
+    "transformer.h.20.ln_2.weight": "pytorch_model-00006-of-00022.bin",
+    "transformer.h.20.mlp.c_fc.bias": "pytorch_model-00006-of-00022.bin",
+    "transformer.h.20.mlp.c_fc.weight": "pytorch_model-00006-of-00022.bin",
+    "transformer.h.20.mlp.c_proj.bias": "pytorch_model-00006-of-00022.bin",
+    "transformer.h.20.mlp.c_proj.weight": "pytorch_model-00006-of-00022.bin",
+    "transformer.h.21.attn.bias": "pytorch_model-00006-of-00022.bin",
+    "transformer.h.21.attn.c_attn.bias": "pytorch_model-00006-of-00022.bin",
+    "transformer.h.21.attn.c_attn.weight": "pytorch_model-00006-of-00022.bin",
+    "transformer.h.21.attn.c_proj.bias": "pytorch_model-00006-of-00022.bin",
+    "transformer.h.21.attn.c_proj.weight": "pytorch_model-00006-of-00022.bin",
+    "transformer.h.21.attn.masked_bias": "pytorch_model-00006-of-00022.bin",
+    "transformer.h.21.ln_1.bias": "pytorch_model-00006-of-00022.bin",
+    "transformer.h.21.ln_1.weight": "pytorch_model-00006-of-00022.bin",
+    "transformer.h.21.ln_2.bias": "pytorch_model-00006-of-00022.bin",
+    "transformer.h.21.ln_2.weight": "pytorch_model-00006-of-00022.bin",
+    "transformer.h.21.mlp.c_fc.bias": "pytorch_model-00006-of-00022.bin",
+    "transformer.h.21.mlp.c_fc.weight": "pytorch_model-00006-of-00022.bin",
+    "transformer.h.21.mlp.c_proj.bias": "pytorch_model-00007-of-00022.bin",
+    "transformer.h.21.mlp.c_proj.weight": "pytorch_model-00007-of-00022.bin",
+    "transformer.h.22.attn.bias": "pytorch_model-00007-of-00022.bin",
+    "transformer.h.22.attn.c_attn.bias": "pytorch_model-00007-of-00022.bin",
+    "transformer.h.22.attn.c_attn.weight": "pytorch_model-00007-of-00022.bin",
+    "transformer.h.22.attn.c_proj.bias": "pytorch_model-00007-of-00022.bin",
+    "transformer.h.22.attn.c_proj.weight": "pytorch_model-00007-of-00022.bin",
+    "transformer.h.22.attn.masked_bias": "pytorch_model-00007-of-00022.bin",
+    "transformer.h.22.ln_1.bias": "pytorch_model-00007-of-00022.bin",
+    "transformer.h.22.ln_1.weight": "pytorch_model-00007-of-00022.bin",
+    "transformer.h.22.ln_2.bias": "pytorch_model-00007-of-00022.bin",
+    "transformer.h.22.ln_2.weight": "pytorch_model-00007-of-00022.bin",
+    "transformer.h.22.mlp.c_fc.bias": "pytorch_model-00007-of-00022.bin",
+    "transformer.h.22.mlp.c_fc.weight": "pytorch_model-00007-of-00022.bin",
+    "transformer.h.22.mlp.c_proj.bias": "pytorch_model-00007-of-00022.bin",
+    "transformer.h.22.mlp.c_proj.weight": "pytorch_model-00007-of-00022.bin",
+    "transformer.h.23.attn.bias": "pytorch_model-00007-of-00022.bin",
+    "transformer.h.23.attn.c_attn.bias": "pytorch_model-00007-of-00022.bin",
+    "transformer.h.23.attn.c_attn.weight": "pytorch_model-00007-of-00022.bin",
+    "transformer.h.23.attn.c_proj.bias": "pytorch_model-00007-of-00022.bin",
+    "transformer.h.23.attn.c_proj.weight": "pytorch_model-00007-of-00022.bin",
+    "transformer.h.23.attn.masked_bias": "pytorch_model-00007-of-00022.bin",
+    "transformer.h.23.ln_1.bias": "pytorch_model-00007-of-00022.bin",
+    "transformer.h.23.ln_1.weight": "pytorch_model-00007-of-00022.bin",
+    "transformer.h.23.ln_2.bias": "pytorch_model-00007-of-00022.bin",
+    "transformer.h.23.ln_2.weight": "pytorch_model-00007-of-00022.bin",
+    "transformer.h.23.mlp.c_fc.bias": "pytorch_model-00007-of-00022.bin",
+    "transformer.h.23.mlp.c_fc.weight": "pytorch_model-00007-of-00022.bin",
+    "transformer.h.23.mlp.c_proj.bias": "pytorch_model-00007-of-00022.bin",
+    "transformer.h.23.mlp.c_proj.weight": "pytorch_model-00007-of-00022.bin",
+    "transformer.h.24.attn.bias": "pytorch_model-00007-of-00022.bin",
+    "transformer.h.24.attn.c_attn.bias": "pytorch_model-00007-of-00022.bin",
+    "transformer.h.24.attn.c_attn.weight": "pytorch_model-00007-of-00022.bin",
+    "transformer.h.24.attn.c_proj.bias": "pytorch_model-00007-of-00022.bin",
+    "transformer.h.24.attn.c_proj.weight": "pytorch_model-00007-of-00022.bin",
+    "transformer.h.24.attn.masked_bias": "pytorch_model-00007-of-00022.bin",
+    "transformer.h.24.ln_1.bias": "pytorch_model-00007-of-00022.bin",
+    "transformer.h.24.ln_1.weight": "pytorch_model-00007-of-00022.bin",
+    "transformer.h.24.ln_2.bias": "pytorch_model-00007-of-00022.bin",
+    "transformer.h.24.ln_2.weight": "pytorch_model-00007-of-00022.bin",
+    "transformer.h.24.mlp.c_fc.bias": "pytorch_model-00007-of-00022.bin",
+    "transformer.h.24.mlp.c_fc.weight": "pytorch_model-00007-of-00022.bin",
+    "transformer.h.24.mlp.c_proj.bias": "pytorch_model-00007-of-00022.bin",
+    "transformer.h.24.mlp.c_proj.weight": "pytorch_model-00007-of-00022.bin",
+    "transformer.h.25.attn.bias": "pytorch_model-00007-of-00022.bin",
+    "transformer.h.25.attn.c_attn.bias": "pytorch_model-00007-of-00022.bin",
+    "transformer.h.25.attn.c_attn.weight": "pytorch_model-00007-of-00022.bin",
+    "transformer.h.25.attn.c_proj.bias": "pytorch_model-00007-of-00022.bin",
+    "transformer.h.25.attn.c_proj.weight": "pytorch_model-00007-of-00022.bin",
+    "transformer.h.25.attn.masked_bias": "pytorch_model-00007-of-00022.bin",
+    "transformer.h.25.ln_1.bias": "pytorch_model-00007-of-00022.bin",
+    "transformer.h.25.ln_1.weight": "pytorch_model-00007-of-00022.bin",
+    "transformer.h.25.ln_2.bias": "pytorch_model-00007-of-00022.bin",
+    "transformer.h.25.ln_2.weight": "pytorch_model-00007-of-00022.bin",
+    "transformer.h.25.mlp.c_fc.bias": "pytorch_model-00008-of-00022.bin",
+    "transformer.h.25.mlp.c_fc.weight": "pytorch_model-00008-of-00022.bin",
+    "transformer.h.25.mlp.c_proj.bias": "pytorch_model-00008-of-00022.bin",
+    "transformer.h.25.mlp.c_proj.weight": "pytorch_model-00008-of-00022.bin",
+    "transformer.h.26.attn.bias": "pytorch_model-00008-of-00022.bin",
+    "transformer.h.26.attn.c_attn.bias": "pytorch_model-00008-of-00022.bin",
+    "transformer.h.26.attn.c_attn.weight": "pytorch_model-00008-of-00022.bin",
+    "transformer.h.26.attn.c_proj.bias": "pytorch_model-00008-of-00022.bin",
+    "transformer.h.26.attn.c_proj.weight": "pytorch_model-00008-of-00022.bin",
+    "transformer.h.26.attn.masked_bias": "pytorch_model-00008-of-00022.bin",
+    "transformer.h.26.ln_1.bias": "pytorch_model-00008-of-00022.bin",
+    "transformer.h.26.ln_1.weight": "pytorch_model-00008-of-00022.bin",
+    "transformer.h.26.ln_2.bias": "pytorch_model-00008-of-00022.bin",
+    "transformer.h.26.ln_2.weight": "pytorch_model-00008-of-00022.bin",
+    "transformer.h.26.mlp.c_fc.bias": "pytorch_model-00008-of-00022.bin",
+    "transformer.h.26.mlp.c_fc.weight": "pytorch_model-00008-of-00022.bin",
+    "transformer.h.26.mlp.c_proj.bias": "pytorch_model-00008-of-00022.bin",
+    "transformer.h.26.mlp.c_proj.weight": "pytorch_model-00008-of-00022.bin",
+    "transformer.h.27.attn.bias": "pytorch_model-00008-of-00022.bin",
+    "transformer.h.27.attn.c_attn.bias": "pytorch_model-00008-of-00022.bin",
+    "transformer.h.27.attn.c_attn.weight": "pytorch_model-00008-of-00022.bin",
+    "transformer.h.27.attn.c_proj.bias": "pytorch_model-00008-of-00022.bin",
+    "transformer.h.27.attn.c_proj.weight": "pytorch_model-00008-of-00022.bin",
+    "transformer.h.27.attn.masked_bias": "pytorch_model-00008-of-00022.bin",
+    "transformer.h.27.ln_1.bias": "pytorch_model-00008-of-00022.bin",
+    "transformer.h.27.ln_1.weight": "pytorch_model-00008-of-00022.bin",
+    "transformer.h.27.ln_2.bias": "pytorch_model-00008-of-00022.bin",
+    "transformer.h.27.ln_2.weight": "pytorch_model-00008-of-00022.bin",
+    "transformer.h.27.mlp.c_fc.bias": "pytorch_model-00008-of-00022.bin",
+    "transformer.h.27.mlp.c_fc.weight": "pytorch_model-00008-of-00022.bin",
+    "transformer.h.27.mlp.c_proj.bias": "pytorch_model-00008-of-00022.bin",
+    "transformer.h.27.mlp.c_proj.weight": "pytorch_model-00008-of-00022.bin",
+    "transformer.h.28.attn.bias": "pytorch_model-00008-of-00022.bin",
+    "transformer.h.28.attn.c_attn.bias": "pytorch_model-00008-of-00022.bin",
+    "transformer.h.28.attn.c_attn.weight": "pytorch_model-00008-of-00022.bin",
+    "transformer.h.28.attn.c_proj.bias": "pytorch_model-00008-of-00022.bin",
+    "transformer.h.28.attn.c_proj.weight": "pytorch_model-00008-of-00022.bin",
+    "transformer.h.28.attn.masked_bias": "pytorch_model-00008-of-00022.bin",
+    "transformer.h.28.ln_1.bias": "pytorch_model-00008-of-00022.bin",
+    "transformer.h.28.ln_1.weight": "pytorch_model-00008-of-00022.bin",
+    "transformer.h.28.ln_2.bias": "pytorch_model-00008-of-00022.bin",
+    "transformer.h.28.ln_2.weight": "pytorch_model-00008-of-00022.bin",
+    "transformer.h.28.mlp.c_fc.bias": "pytorch_model-00008-of-00022.bin",
+    "transformer.h.28.mlp.c_fc.weight": "pytorch_model-00008-of-00022.bin",
+    "transformer.h.28.mlp.c_proj.bias": "pytorch_model-00008-of-00022.bin",
+    "transformer.h.28.mlp.c_proj.weight": "pytorch_model-00008-of-00022.bin",
+    "transformer.h.29.attn.bias": "pytorch_model-00008-of-00022.bin",
+    "transformer.h.29.attn.c_attn.bias": "pytorch_model-00008-of-00022.bin",
+    "transformer.h.29.attn.c_attn.weight": "pytorch_model-00008-of-00022.bin",
+    "transformer.h.29.attn.c_proj.bias": "pytorch_model-00009-of-00022.bin",
+    "transformer.h.29.attn.c_proj.weight": "pytorch_model-00009-of-00022.bin",
+    "transformer.h.29.attn.masked_bias": "pytorch_model-00008-of-00022.bin",
+    "transformer.h.29.ln_1.bias": "pytorch_model-00008-of-00022.bin",
+    "transformer.h.29.ln_1.weight": "pytorch_model-00008-of-00022.bin",
+    "transformer.h.29.ln_2.bias": "pytorch_model-00009-of-00022.bin",
+    "transformer.h.29.ln_2.weight": "pytorch_model-00009-of-00022.bin",
+    "transformer.h.29.mlp.c_fc.bias": "pytorch_model-00009-of-00022.bin",
+    "transformer.h.29.mlp.c_fc.weight": "pytorch_model-00009-of-00022.bin",
+    "transformer.h.29.mlp.c_proj.bias": "pytorch_model-00009-of-00022.bin",
+    "transformer.h.29.mlp.c_proj.weight": "pytorch_model-00009-of-00022.bin",
+    "transformer.h.3.attn.bias": "pytorch_model-00001-of-00022.bin",
+    "transformer.h.3.attn.c_attn.bias": "pytorch_model-00002-of-00022.bin",
+    "transformer.h.3.attn.c_attn.weight": "pytorch_model-00002-of-00022.bin",
+    "transformer.h.3.attn.c_proj.bias": "pytorch_model-00002-of-00022.bin",
+    "transformer.h.3.attn.c_proj.weight": "pytorch_model-00002-of-00022.bin",
+    "transformer.h.3.attn.masked_bias": "pytorch_model-00001-of-00022.bin",
+    "transformer.h.3.ln_1.bias": "pytorch_model-00001-of-00022.bin",
+    "transformer.h.3.ln_1.weight": "pytorch_model-00001-of-00022.bin",
+    "transformer.h.3.ln_2.bias": "pytorch_model-00002-of-00022.bin",
+    "transformer.h.3.ln_2.weight": "pytorch_model-00002-of-00022.bin",
+    "transformer.h.3.mlp.c_fc.bias": "pytorch_model-00002-of-00022.bin",
+    "transformer.h.3.mlp.c_fc.weight": "pytorch_model-00002-of-00022.bin",
+    "transformer.h.3.mlp.c_proj.bias": "pytorch_model-00002-of-00022.bin",
+    "transformer.h.3.mlp.c_proj.weight": "pytorch_model-00002-of-00022.bin",
+    "transformer.h.30.attn.bias": "pytorch_model-00009-of-00022.bin",
+    "transformer.h.30.attn.c_attn.bias": "pytorch_model-00009-of-00022.bin",
+    "transformer.h.30.attn.c_attn.weight": "pytorch_model-00009-of-00022.bin",
+    "transformer.h.30.attn.c_proj.bias": "pytorch_model-00009-of-00022.bin",
+    "transformer.h.30.attn.c_proj.weight": "pytorch_model-00009-of-00022.bin",
+    "transformer.h.30.attn.masked_bias": "pytorch_model-00009-of-00022.bin",
+    "transformer.h.30.ln_1.bias": "pytorch_model-00009-of-00022.bin",
+    "transformer.h.30.ln_1.weight": "pytorch_model-00009-of-00022.bin",
+    "transformer.h.30.ln_2.bias": "pytorch_model-00009-of-00022.bin",
+    "transformer.h.30.ln_2.weight": "pytorch_model-00009-of-00022.bin",
+    "transformer.h.30.mlp.c_fc.bias": "pytorch_model-00009-of-00022.bin",
+    "transformer.h.30.mlp.c_fc.weight": "pytorch_model-00009-of-00022.bin",
+    "transformer.h.30.mlp.c_proj.bias": "pytorch_model-00009-of-00022.bin",
+    "transformer.h.30.mlp.c_proj.weight": "pytorch_model-00009-of-00022.bin",
+    "transformer.h.31.attn.bias": "pytorch_model-00009-of-00022.bin",
+    "transformer.h.31.attn.c_attn.bias": "pytorch_model-00009-of-00022.bin",
+    "transformer.h.31.attn.c_attn.weight": "pytorch_model-00009-of-00022.bin",
+    "transformer.h.31.attn.c_proj.bias": "pytorch_model-00009-of-00022.bin",
+    "transformer.h.31.attn.c_proj.weight": "pytorch_model-00009-of-00022.bin",
+    "transformer.h.31.attn.masked_bias": "pytorch_model-00009-of-00022.bin",
+    "transformer.h.31.ln_1.bias": "pytorch_model-00009-of-00022.bin",
+    "transformer.h.31.ln_1.weight": "pytorch_model-00009-of-00022.bin",
+    "transformer.h.31.ln_2.bias": "pytorch_model-00009-of-00022.bin",
+    "transformer.h.31.ln_2.weight": "pytorch_model-00009-of-00022.bin",
+    "transformer.h.31.mlp.c_fc.bias": "pytorch_model-00009-of-00022.bin",
+    "transformer.h.31.mlp.c_fc.weight": "pytorch_model-00009-of-00022.bin",
+    "transformer.h.31.mlp.c_proj.bias": "pytorch_model-00009-of-00022.bin",
+    "transformer.h.31.mlp.c_proj.weight": "pytorch_model-00009-of-00022.bin",
+    "transformer.h.32.attn.bias": "pytorch_model-00009-of-00022.bin",
+    "transformer.h.32.attn.c_attn.bias": "pytorch_model-00009-of-00022.bin",
+    "transformer.h.32.attn.c_attn.weight": "pytorch_model-00009-of-00022.bin",
+    "transformer.h.32.attn.c_proj.bias": "pytorch_model-00009-of-00022.bin",
+    "transformer.h.32.attn.c_proj.weight": "pytorch_model-00009-of-00022.bin",
+    "transformer.h.32.attn.masked_bias": "pytorch_model-00009-of-00022.bin",
+    "transformer.h.32.ln_1.bias": "pytorch_model-00009-of-00022.bin",
+    "transformer.h.32.ln_1.weight": "pytorch_model-00009-of-00022.bin",
+    "transformer.h.32.ln_2.bias": "pytorch_model-00009-of-00022.bin",
+    "transformer.h.32.ln_2.weight": "pytorch_model-00009-of-00022.bin",
+    "transformer.h.32.mlp.c_fc.bias": "pytorch_model-00009-of-00022.bin",
+    "transformer.h.32.mlp.c_fc.weight": "pytorch_model-00009-of-00022.bin",
+    "transformer.h.32.mlp.c_proj.bias": "pytorch_model-00009-of-00022.bin",
+    "transformer.h.32.mlp.c_proj.weight": "pytorch_model-00009-of-00022.bin",
+    "transformer.h.33.attn.bias": "pytorch_model-00009-of-00022.bin",
+    "transformer.h.33.attn.c_attn.bias": "pytorch_model-00010-of-00022.bin",
+    "transformer.h.33.attn.c_attn.weight": "pytorch_model-00010-of-00022.bin",
+    "transformer.h.33.attn.c_proj.bias": "pytorch_model-00010-of-00022.bin",
+    "transformer.h.33.attn.c_proj.weight": "pytorch_model-00010-of-00022.bin",
+    "transformer.h.33.attn.masked_bias": "pytorch_model-00009-of-00022.bin",
+    "transformer.h.33.ln_1.bias": "pytorch_model-00009-of-00022.bin",
+    "transformer.h.33.ln_1.weight": "pytorch_model-00009-of-00022.bin",
+    "transformer.h.33.ln_2.bias": "pytorch_model-00010-of-00022.bin",
+    "transformer.h.33.ln_2.weight": "pytorch_model-00010-of-00022.bin",
+    "transformer.h.33.mlp.c_fc.bias": "pytorch_model-00010-of-00022.bin",
+    "transformer.h.33.mlp.c_fc.weight": "pytorch_model-00010-of-00022.bin",
+    "transformer.h.33.mlp.c_proj.bias": "pytorch_model-00010-of-00022.bin",
+    "transformer.h.33.mlp.c_proj.weight": "pytorch_model-00010-of-00022.bin",
+    "transformer.h.34.attn.bias": "pytorch_model-00010-of-00022.bin",
+    "transformer.h.34.attn.c_attn.bias": "pytorch_model-00010-of-00022.bin",
+    "transformer.h.34.attn.c_attn.weight": "pytorch_model-00010-of-00022.bin",
+    "transformer.h.34.attn.c_proj.bias": "pytorch_model-00010-of-00022.bin",
+    "transformer.h.34.attn.c_proj.weight": "pytorch_model-00010-of-00022.bin",
+    "transformer.h.34.attn.masked_bias": "pytorch_model-00010-of-00022.bin",
+    "transformer.h.34.ln_1.bias": "pytorch_model-00010-of-00022.bin",
+    "transformer.h.34.ln_1.weight": "pytorch_model-00010-of-00022.bin",
+    "transformer.h.34.ln_2.bias": "pytorch_model-00010-of-00022.bin",
+    "transformer.h.34.ln_2.weight": "pytorch_model-00010-of-00022.bin",
+    "transformer.h.34.mlp.c_fc.bias": "pytorch_model-00010-of-00022.bin",
+    "transformer.h.34.mlp.c_fc.weight": "pytorch_model-00010-of-00022.bin",
+    "transformer.h.34.mlp.c_proj.bias": "pytorch_model-00010-of-00022.bin",
+    "transformer.h.34.mlp.c_proj.weight": "pytorch_model-00010-of-00022.bin",
+    "transformer.h.35.attn.bias": "pytorch_model-00010-of-00022.bin",
+    "transformer.h.35.attn.c_attn.bias": "pytorch_model-00010-of-00022.bin",
+    "transformer.h.35.attn.c_attn.weight": "pytorch_model-00010-of-00022.bin",
+    "transformer.h.35.attn.c_proj.bias": "pytorch_model-00010-of-00022.bin",
+    "transformer.h.35.attn.c_proj.weight": "pytorch_model-00010-of-00022.bin",
+    "transformer.h.35.attn.masked_bias": "pytorch_model-00010-of-00022.bin",
+    "transformer.h.35.ln_1.bias": "pytorch_model-00010-of-00022.bin",
+    "transformer.h.35.ln_1.weight": "pytorch_model-00010-of-00022.bin",
+    "transformer.h.35.ln_2.bias": "pytorch_model-00010-of-00022.bin",
+    "transformer.h.35.ln_2.weight": "pytorch_model-00010-of-00022.bin",
+    "transformer.h.35.mlp.c_fc.bias": "pytorch_model-00010-of-00022.bin",
+    "transformer.h.35.mlp.c_fc.weight": "pytorch_model-00010-of-00022.bin",
+    "transformer.h.35.mlp.c_proj.bias": "pytorch_model-00010-of-00022.bin",
+    "transformer.h.35.mlp.c_proj.weight": "pytorch_model-00010-of-00022.bin",
+    "transformer.h.36.attn.bias": "pytorch_model-00010-of-00022.bin",
+    "transformer.h.36.attn.c_attn.bias": "pytorch_model-00010-of-00022.bin",
+    "transformer.h.36.attn.c_attn.weight": "pytorch_model-00010-of-00022.bin",
+    "transformer.h.36.attn.c_proj.bias": "pytorch_model-00010-of-00022.bin",
+    "transformer.h.36.attn.c_proj.weight": "pytorch_model-00010-of-00022.bin",
+    "transformer.h.36.attn.masked_bias": "pytorch_model-00010-of-00022.bin",
+    "transformer.h.36.ln_1.bias": "pytorch_model-00010-of-00022.bin",
+    "transformer.h.36.ln_1.weight": "pytorch_model-00010-of-00022.bin",
+    "transformer.h.36.ln_2.bias": "pytorch_model-00010-of-00022.bin",
+    "transformer.h.36.ln_2.weight": "pytorch_model-00010-of-00022.bin",
+    "transformer.h.36.mlp.c_fc.bias": "pytorch_model-00010-of-00022.bin",
+    "transformer.h.36.mlp.c_fc.weight": "pytorch_model-00010-of-00022.bin",
+    "transformer.h.36.mlp.c_proj.bias": "pytorch_model-00011-of-00022.bin",
+    "transformer.h.36.mlp.c_proj.weight": "pytorch_model-00011-of-00022.bin",
+    "transformer.h.37.attn.bias": "pytorch_model-00011-of-00022.bin",
+    "transformer.h.37.attn.c_attn.bias": "pytorch_model-00011-of-00022.bin",
+    "transformer.h.37.attn.c_attn.weight": "pytorch_model-00011-of-00022.bin",
+    "transformer.h.37.attn.c_proj.bias": "pytorch_model-00011-of-00022.bin",
+    "transformer.h.37.attn.c_proj.weight": "pytorch_model-00011-of-00022.bin",
+    "transformer.h.37.attn.masked_bias": "pytorch_model-00011-of-00022.bin",
+    "transformer.h.37.ln_1.bias": "pytorch_model-00011-of-00022.bin",
+    "transformer.h.37.ln_1.weight": "pytorch_model-00011-of-00022.bin",
+    "transformer.h.37.ln_2.bias": "pytorch_model-00011-of-00022.bin",
+    "transformer.h.37.ln_2.weight": "pytorch_model-00011-of-00022.bin",
+    "transformer.h.37.mlp.c_fc.bias": "pytorch_model-00011-of-00022.bin",
+    "transformer.h.37.mlp.c_fc.weight": "pytorch_model-00011-of-00022.bin",
+    "transformer.h.37.mlp.c_proj.bias": "pytorch_model-00011-of-00022.bin",
+    "transformer.h.37.mlp.c_proj.weight": "pytorch_model-00011-of-00022.bin",
+    "transformer.h.38.attn.bias": "pytorch_model-00011-of-00022.bin",
+    "transformer.h.38.attn.c_attn.bias": "pytorch_model-00011-of-00022.bin",
+    "transformer.h.38.attn.c_attn.weight": "pytorch_model-00011-of-00022.bin",
+    "transformer.h.38.attn.c_proj.bias": "pytorch_model-00011-of-00022.bin",
+    "transformer.h.38.attn.c_proj.weight": "pytorch_model-00011-of-00022.bin",
+    "transformer.h.38.attn.masked_bias": "pytorch_model-00011-of-00022.bin",
+    "transformer.h.38.ln_1.bias": "pytorch_model-00011-of-00022.bin",
+    "transformer.h.38.ln_1.weight": "pytorch_model-00011-of-00022.bin",
+    "transformer.h.38.ln_2.bias": "pytorch_model-00011-of-00022.bin",
+    "transformer.h.38.ln_2.weight": "pytorch_model-00011-of-00022.bin",
+    "transformer.h.38.mlp.c_fc.bias": "pytorch_model-00011-of-00022.bin",
+    "transformer.h.38.mlp.c_fc.weight": "pytorch_model-00011-of-00022.bin",
+    "transformer.h.38.mlp.c_proj.bias": "pytorch_model-00011-of-00022.bin",
+    "transformer.h.38.mlp.c_proj.weight": "pytorch_model-00011-of-00022.bin",
+    "transformer.h.39.attn.bias": "pytorch_model-00011-of-00022.bin",
+    "transformer.h.39.attn.c_attn.bias": "pytorch_model-00011-of-00022.bin",
+    "transformer.h.39.attn.c_attn.weight": "pytorch_model-00011-of-00022.bin",
+    "transformer.h.39.attn.c_proj.bias": "pytorch_model-00011-of-00022.bin",
+    "transformer.h.39.attn.c_proj.weight": "pytorch_model-00011-of-00022.bin",
+    "transformer.h.39.attn.masked_bias": "pytorch_model-00011-of-00022.bin",
+    "transformer.h.39.ln_1.bias": "pytorch_model-00011-of-00022.bin",
+    "transformer.h.39.ln_1.weight": "pytorch_model-00011-of-00022.bin",
+    "transformer.h.39.ln_2.bias": "pytorch_model-00011-of-00022.bin",
+    "transformer.h.39.ln_2.weight": "pytorch_model-00011-of-00022.bin",
+    "transformer.h.39.mlp.c_fc.bias": "pytorch_model-00011-of-00022.bin",
+    "transformer.h.39.mlp.c_fc.weight": "pytorch_model-00011-of-00022.bin",
+    "transformer.h.39.mlp.c_proj.bias": "pytorch_model-00011-of-00022.bin",
+    "transformer.h.39.mlp.c_proj.weight": "pytorch_model-00011-of-00022.bin",
+    "transformer.h.4.attn.bias": "pytorch_model-00002-of-00022.bin",
+    "transformer.h.4.attn.c_attn.bias": "pytorch_model-00002-of-00022.bin",
+    "transformer.h.4.attn.c_attn.weight": "pytorch_model-00002-of-00022.bin",
+    "transformer.h.4.attn.c_proj.bias": "pytorch_model-00002-of-00022.bin",
+    "transformer.h.4.attn.c_proj.weight": "pytorch_model-00002-of-00022.bin",
+    "transformer.h.4.attn.masked_bias": "pytorch_model-00002-of-00022.bin",
+    "transformer.h.4.ln_1.bias": "pytorch_model-00002-of-00022.bin",
+    "transformer.h.4.ln_1.weight": "pytorch_model-00002-of-00022.bin",
+    "transformer.h.4.ln_2.bias": "pytorch_model-00002-of-00022.bin",
+    "transformer.h.4.ln_2.weight": "pytorch_model-00002-of-00022.bin",
+    "transformer.h.4.mlp.c_fc.bias": "pytorch_model-00002-of-00022.bin",
+    "transformer.h.4.mlp.c_fc.weight": "pytorch_model-00002-of-00022.bin",
+    "transformer.h.4.mlp.c_proj.bias": "pytorch_model-00002-of-00022.bin",
+    "transformer.h.4.mlp.c_proj.weight": "pytorch_model-00002-of-00022.bin",
+    "transformer.h.40.attn.bias": "pytorch_model-00011-of-00022.bin",
+    "transformer.h.40.attn.c_attn.bias": "pytorch_model-00011-of-00022.bin",
+    "transformer.h.40.attn.c_attn.weight": "pytorch_model-00011-of-00022.bin",
+    "transformer.h.40.attn.c_proj.bias": "pytorch_model-00011-of-00022.bin",
+    "transformer.h.40.attn.c_proj.weight": "pytorch_model-00011-of-00022.bin",
+    "transformer.h.40.attn.masked_bias": "pytorch_model-00011-of-00022.bin",
+    "transformer.h.40.ln_1.bias": "pytorch_model-00011-of-00022.bin",
+    "transformer.h.40.ln_1.weight": "pytorch_model-00011-of-00022.bin",
+    "transformer.h.40.ln_2.bias": "pytorch_model-00011-of-00022.bin",
+    "transformer.h.40.ln_2.weight": "pytorch_model-00011-of-00022.bin",
+    "transformer.h.40.mlp.c_fc.bias": "pytorch_model-00012-of-00022.bin",
+    "transformer.h.40.mlp.c_fc.weight": "pytorch_model-00012-of-00022.bin",
+    "transformer.h.40.mlp.c_proj.bias": "pytorch_model-00012-of-00022.bin",
+    "transformer.h.40.mlp.c_proj.weight": "pytorch_model-00012-of-00022.bin",
+    "transformer.h.41.attn.bias": "pytorch_model-00012-of-00022.bin",
+    "transformer.h.41.attn.c_attn.bias": "pytorch_model-00012-of-00022.bin",
+    "transformer.h.41.attn.c_attn.weight": "pytorch_model-00012-of-00022.bin",
+    "transformer.h.41.attn.c_proj.bias": "pytorch_model-00012-of-00022.bin",
+    "transformer.h.41.attn.c_proj.weight": "pytorch_model-00012-of-00022.bin",
+    "transformer.h.41.attn.masked_bias": "pytorch_model-00012-of-00022.bin",
+    "transformer.h.41.ln_1.bias": "pytorch_model-00012-of-00022.bin",
+    "transformer.h.41.ln_1.weight": "pytorch_model-00012-of-00022.bin",
+    "transformer.h.41.ln_2.bias": "pytorch_model-00012-of-00022.bin",
+    "transformer.h.41.ln_2.weight": "pytorch_model-00012-of-00022.bin",
+    "transformer.h.41.mlp.c_fc.bias": "pytorch_model-00012-of-00022.bin",
+    "transformer.h.41.mlp.c_fc.weight": "pytorch_model-00012-of-00022.bin",
+    "transformer.h.41.mlp.c_proj.bias": "pytorch_model-00012-of-00022.bin",
+    "transformer.h.41.mlp.c_proj.weight": "pytorch_model-00012-of-00022.bin",
+    "transformer.h.42.attn.bias": "pytorch_model-00012-of-00022.bin",
+    "transformer.h.42.attn.c_attn.bias": "pytorch_model-00012-of-00022.bin",
+    "transformer.h.42.attn.c_attn.weight": "pytorch_model-00012-of-00022.bin",
+    "transformer.h.42.attn.c_proj.bias": "pytorch_model-00012-of-00022.bin",
+    "transformer.h.42.attn.c_proj.weight": "pytorch_model-00012-of-00022.bin",
+    "transformer.h.42.attn.masked_bias": "pytorch_model-00012-of-00022.bin",
+    "transformer.h.42.ln_1.bias": "pytorch_model-00012-of-00022.bin",
+    "transformer.h.42.ln_1.weight": "pytorch_model-00012-of-00022.bin",
+    "transformer.h.42.ln_2.bias": "pytorch_model-00012-of-00022.bin",
+    "transformer.h.42.ln_2.weight": "pytorch_model-00012-of-00022.bin",
+    "transformer.h.42.mlp.c_fc.bias": "pytorch_model-00012-of-00022.bin",
+    "transformer.h.42.mlp.c_fc.weight": "pytorch_model-00012-of-00022.bin",
+    "transformer.h.42.mlp.c_proj.bias": "pytorch_model-00012-of-00022.bin",
+    "transformer.h.42.mlp.c_proj.weight": "pytorch_model-00012-of-00022.bin",
+    "transformer.h.43.attn.bias": "pytorch_model-00012-of-00022.bin",
+    "transformer.h.43.attn.c_attn.bias": "pytorch_model-00012-of-00022.bin",
+    "transformer.h.43.attn.c_attn.weight": "pytorch_model-00012-of-00022.bin",
+    "transformer.h.43.attn.c_proj.bias": "pytorch_model-00012-of-00022.bin",
+    "transformer.h.43.attn.c_proj.weight": "pytorch_model-00012-of-00022.bin",
+    "transformer.h.43.attn.masked_bias": "pytorch_model-00012-of-00022.bin",
+    "transformer.h.43.ln_1.bias": "pytorch_model-00012-of-00022.bin",
+    "transformer.h.43.ln_1.weight": "pytorch_model-00012-of-00022.bin",
+    "transformer.h.43.ln_2.bias": "pytorch_model-00012-of-00022.bin",
+    "transformer.h.43.ln_2.weight": "pytorch_model-00012-of-00022.bin",
+    "transformer.h.43.mlp.c_fc.bias": "pytorch_model-00012-of-00022.bin",
+    "transformer.h.43.mlp.c_fc.weight": "pytorch_model-00012-of-00022.bin",
+    "transformer.h.43.mlp.c_proj.bias": "pytorch_model-00012-of-00022.bin",
+    "transformer.h.43.mlp.c_proj.weight": "pytorch_model-00012-of-00022.bin",
+    "transformer.h.44.attn.bias": "pytorch_model-00012-of-00022.bin",
+    "transformer.h.44.attn.c_attn.bias": "pytorch_model-00012-of-00022.bin",
+    "transformer.h.44.attn.c_attn.weight": "pytorch_model-00012-of-00022.bin",
+    "transformer.h.44.attn.c_proj.bias": "pytorch_model-00013-of-00022.bin",
+    "transformer.h.44.attn.c_proj.weight": "pytorch_model-00013-of-00022.bin",
+    "transformer.h.44.attn.masked_bias": "pytorch_model-00012-of-00022.bin",
+    "transformer.h.44.ln_1.bias": "pytorch_model-00012-of-00022.bin",
+    "transformer.h.44.ln_1.weight": "pytorch_model-00012-of-00022.bin",
+    "transformer.h.44.ln_2.bias": "pytorch_model-00013-of-00022.bin",
+    "transformer.h.44.ln_2.weight": "pytorch_model-00013-of-00022.bin",
+    "transformer.h.44.mlp.c_fc.bias": "pytorch_model-00013-of-00022.bin",
+    "transformer.h.44.mlp.c_fc.weight": "pytorch_model-00013-of-00022.bin",
+    "transformer.h.44.mlp.c_proj.bias": "pytorch_model-00013-of-00022.bin",
+    "transformer.h.44.mlp.c_proj.weight": "pytorch_model-00013-of-00022.bin",
+    "transformer.h.45.attn.bias": "pytorch_model-00013-of-00022.bin",
+    "transformer.h.45.attn.c_attn.bias": "pytorch_model-00013-of-00022.bin",
+    "transformer.h.45.attn.c_attn.weight": "pytorch_model-00013-of-00022.bin",
+    "transformer.h.45.attn.c_proj.bias": "pytorch_model-00013-of-00022.bin",
+    "transformer.h.45.attn.c_proj.weight": "pytorch_model-00013-of-00022.bin",
+    "transformer.h.45.attn.masked_bias": "pytorch_model-00013-of-00022.bin",
+    "transformer.h.45.ln_1.bias": "pytorch_model-00013-of-00022.bin",
+    "transformer.h.45.ln_1.weight": "pytorch_model-00013-of-00022.bin",
+    "transformer.h.45.ln_2.bias": "pytorch_model-00013-of-00022.bin",
+    "transformer.h.45.ln_2.weight": "pytorch_model-00013-of-00022.bin",
+    "transformer.h.45.mlp.c_fc.bias": "pytorch_model-00013-of-00022.bin",
+    "transformer.h.45.mlp.c_fc.weight": "pytorch_model-00013-of-00022.bin",
+    "transformer.h.45.mlp.c_proj.bias": "pytorch_model-00013-of-00022.bin",
+    "transformer.h.45.mlp.c_proj.weight": "pytorch_model-00013-of-00022.bin",
+    "transformer.h.46.attn.bias": "pytorch_model-00013-of-00022.bin",
+    "transformer.h.46.attn.c_attn.bias": "pytorch_model-00013-of-00022.bin",
+    "transformer.h.46.attn.c_attn.weight": "pytorch_model-00013-of-00022.bin",
+    "transformer.h.46.attn.c_proj.bias": "pytorch_model-00013-of-00022.bin",
+    "transformer.h.46.attn.c_proj.weight": "pytorch_model-00013-of-00022.bin",
+    "transformer.h.46.attn.masked_bias": "pytorch_model-00013-of-00022.bin",
+    "transformer.h.46.ln_1.bias": "pytorch_model-00013-of-00022.bin",
+    "transformer.h.46.ln_1.weight": "pytorch_model-00013-of-00022.bin",
+    "transformer.h.46.ln_2.bias": "pytorch_model-00013-of-00022.bin",
+    "transformer.h.46.ln_2.weight": "pytorch_model-00013-of-00022.bin",
+    "transformer.h.46.mlp.c_fc.bias": "pytorch_model-00013-of-00022.bin",
+    "transformer.h.46.mlp.c_fc.weight": "pytorch_model-00013-of-00022.bin",
+    "transformer.h.46.mlp.c_proj.bias": "pytorch_model-00013-of-00022.bin",
+    "transformer.h.46.mlp.c_proj.weight": "pytorch_model-00013-of-00022.bin",
+    "transformer.h.47.attn.bias": "pytorch_model-00013-of-00022.bin",
+    "transformer.h.47.attn.c_attn.bias": "pytorch_model-00013-of-00022.bin",
+    "transformer.h.47.attn.c_attn.weight": "pytorch_model-00013-of-00022.bin",
+    "transformer.h.47.attn.c_proj.bias": "pytorch_model-00013-of-00022.bin",
+    "transformer.h.47.attn.c_proj.weight": "pytorch_model-00013-of-00022.bin",
+    "transformer.h.47.attn.masked_bias": "pytorch_model-00013-of-00022.bin",
+    "transformer.h.47.ln_1.bias": "pytorch_model-00013-of-00022.bin",
+    "transformer.h.47.ln_1.weight": "pytorch_model-00013-of-00022.bin",
+    "transformer.h.47.ln_2.bias": "pytorch_model-00013-of-00022.bin",
+    "transformer.h.47.ln_2.weight": "pytorch_model-00013-of-00022.bin",
+    "transformer.h.47.mlp.c_fc.bias": "pytorch_model-00013-of-00022.bin",
+    "transformer.h.47.mlp.c_fc.weight": "pytorch_model-00013-of-00022.bin",
+    "transformer.h.47.mlp.c_proj.bias": "pytorch_model-00013-of-00022.bin",
+    "transformer.h.47.mlp.c_proj.weight": "pytorch_model-00013-of-00022.bin",
+    "transformer.h.48.attn.bias": "pytorch_model-00013-of-00022.bin",
+    "transformer.h.48.attn.c_attn.bias": "pytorch_model-00014-of-00022.bin",
+    "transformer.h.48.attn.c_attn.weight": "pytorch_model-00014-of-00022.bin",
+    "transformer.h.48.attn.c_proj.bias": "pytorch_model-00014-of-00022.bin",
+    "transformer.h.48.attn.c_proj.weight": "pytorch_model-00014-of-00022.bin",
+    "transformer.h.48.attn.masked_bias": "pytorch_model-00013-of-00022.bin",
+    "transformer.h.48.ln_1.bias": "pytorch_model-00013-of-00022.bin",
+    "transformer.h.48.ln_1.weight": "pytorch_model-00013-of-00022.bin",
+    "transformer.h.48.ln_2.bias": "pytorch_model-00014-of-00022.bin",
+    "transformer.h.48.ln_2.weight": "pytorch_model-00014-of-00022.bin",
+    "transformer.h.48.mlp.c_fc.bias": "pytorch_model-00014-of-00022.bin",
+    "transformer.h.48.mlp.c_fc.weight": "pytorch_model-00014-of-00022.bin",
+    "transformer.h.48.mlp.c_proj.bias": "pytorch_model-00014-of-00022.bin",
+    "transformer.h.48.mlp.c_proj.weight": "pytorch_model-00014-of-00022.bin",
+    "transformer.h.49.attn.bias": "pytorch_model-00014-of-00022.bin",
+    "transformer.h.49.attn.c_attn.bias": "pytorch_model-00014-of-00022.bin",
+    "transformer.h.49.attn.c_attn.weight": "pytorch_model-00014-of-00022.bin",
+    "transformer.h.49.attn.c_proj.bias": "pytorch_model-00014-of-00022.bin",
+    "transformer.h.49.attn.c_proj.weight": "pytorch_model-00014-of-00022.bin",
+    "transformer.h.49.attn.masked_bias": "pytorch_model-00014-of-00022.bin",
+    "transformer.h.49.ln_1.bias": "pytorch_model-00014-of-00022.bin",
+    "transformer.h.49.ln_1.weight": "pytorch_model-00014-of-00022.bin",
+    "transformer.h.49.ln_2.bias": "pytorch_model-00014-of-00022.bin",
+    "transformer.h.49.ln_2.weight": "pytorch_model-00014-of-00022.bin",
+    "transformer.h.49.mlp.c_fc.bias": "pytorch_model-00014-of-00022.bin",
+    "transformer.h.49.mlp.c_fc.weight": "pytorch_model-00014-of-00022.bin",
+    "transformer.h.49.mlp.c_proj.bias": "pytorch_model-00014-of-00022.bin",
+    "transformer.h.49.mlp.c_proj.weight": "pytorch_model-00014-of-00022.bin",
+    "transformer.h.5.attn.bias": "pytorch_model-00002-of-00022.bin",
+    "transformer.h.5.attn.c_attn.bias": "pytorch_model-00002-of-00022.bin",
+    "transformer.h.5.attn.c_attn.weight": "pytorch_model-00002-of-00022.bin",
+    "transformer.h.5.attn.c_proj.bias": "pytorch_model-00002-of-00022.bin",
+    "transformer.h.5.attn.c_proj.weight": "pytorch_model-00002-of-00022.bin",
+    "transformer.h.5.attn.masked_bias": "pytorch_model-00002-of-00022.bin",
+    "transformer.h.5.ln_1.bias": "pytorch_model-00002-of-00022.bin",
+    "transformer.h.5.ln_1.weight": "pytorch_model-00002-of-00022.bin",
+    "transformer.h.5.ln_2.bias": "pytorch_model-00002-of-00022.bin",
+    "transformer.h.5.ln_2.weight": "pytorch_model-00002-of-00022.bin",
+    "transformer.h.5.mlp.c_fc.bias": "pytorch_model-00002-of-00022.bin",
+    "transformer.h.5.mlp.c_fc.weight": "pytorch_model-00002-of-00022.bin",
+    "transformer.h.5.mlp.c_proj.bias": "pytorch_model-00002-of-00022.bin",
+    "transformer.h.5.mlp.c_proj.weight": "pytorch_model-00002-of-00022.bin",
+    "transformer.h.50.attn.bias": "pytorch_model-00014-of-00022.bin",
+    "transformer.h.50.attn.c_attn.bias": "pytorch_model-00014-of-00022.bin",
+    "transformer.h.50.attn.c_attn.weight": "pytorch_model-00014-of-00022.bin",
+    "transformer.h.50.attn.c_proj.bias": "pytorch_model-00014-of-00022.bin",
+    "transformer.h.50.attn.c_proj.weight": "pytorch_model-00014-of-00022.bin",
+    "transformer.h.50.attn.masked_bias": "pytorch_model-00014-of-00022.bin",
+    "transformer.h.50.ln_1.bias": "pytorch_model-00014-of-00022.bin",
+    "transformer.h.50.ln_1.weight": "pytorch_model-00014-of-00022.bin",
+    "transformer.h.50.ln_2.bias": "pytorch_model-00014-of-00022.bin",
+    "transformer.h.50.ln_2.weight": "pytorch_model-00014-of-00022.bin",
+    "transformer.h.50.mlp.c_fc.bias": "pytorch_model-00014-of-00022.bin",
+    "transformer.h.50.mlp.c_fc.weight": "pytorch_model-00014-of-00022.bin",
+    "transformer.h.50.mlp.c_proj.bias": "pytorch_model-00014-of-00022.bin",
+    "transformer.h.50.mlp.c_proj.weight": "pytorch_model-00014-of-00022.bin",
+    "transformer.h.51.attn.bias": "pytorch_model-00014-of-00022.bin",
+    "transformer.h.51.attn.c_attn.bias": "pytorch_model-00014-of-00022.bin",
+    "transformer.h.51.attn.c_attn.weight": "pytorch_model-00014-of-00022.bin",
+    "transformer.h.51.attn.c_proj.bias": "pytorch_model-00014-of-00022.bin",
+    "transformer.h.51.attn.c_proj.weight": "pytorch_model-00014-of-00022.bin",
+    "transformer.h.51.attn.masked_bias": "pytorch_model-00014-of-00022.bin",
+    "transformer.h.51.ln_1.bias": "pytorch_model-00014-of-00022.bin",
+    "transformer.h.51.ln_1.weight": "pytorch_model-00014-of-00022.bin",
+    "transformer.h.51.ln_2.bias": "pytorch_model-00014-of-00022.bin",
+    "transformer.h.51.ln_2.weight": "pytorch_model-00014-of-00022.bin",
+    "transformer.h.51.mlp.c_fc.bias": "pytorch_model-00014-of-00022.bin",
+    "transformer.h.51.mlp.c_fc.weight": "pytorch_model-00014-of-00022.bin",
+    "transformer.h.51.mlp.c_proj.bias": "pytorch_model-00015-of-00022.bin",
+    "transformer.h.51.mlp.c_proj.weight": "pytorch_model-00015-of-00022.bin",
+    "transformer.h.52.attn.bias": "pytorch_model-00015-of-00022.bin",
+    "transformer.h.52.attn.c_attn.bias": "pytorch_model-00015-of-00022.bin",
+    "transformer.h.52.attn.c_attn.weight": "pytorch_model-00015-of-00022.bin",
+    "transformer.h.52.attn.c_proj.bias": "pytorch_model-00015-of-00022.bin",
+    "transformer.h.52.attn.c_proj.weight": "pytorch_model-00015-of-00022.bin",
+    "transformer.h.52.attn.masked_bias": "pytorch_model-00015-of-00022.bin",
+    "transformer.h.52.ln_1.bias": "pytorch_model-00015-of-00022.bin",
+    "transformer.h.52.ln_1.weight": "pytorch_model-00015-of-00022.bin",
+    "transformer.h.52.ln_2.bias": "pytorch_model-00015-of-00022.bin",
+    "transformer.h.52.ln_2.weight": "pytorch_model-00015-of-00022.bin",
+    "transformer.h.52.mlp.c_fc.bias": "pytorch_model-00015-of-00022.bin",
+    "transformer.h.52.mlp.c_fc.weight": "pytorch_model-00015-of-00022.bin",
+    "transformer.h.52.mlp.c_proj.bias": "pytorch_model-00015-of-00022.bin",
+    "transformer.h.52.mlp.c_proj.weight": "pytorch_model-00015-of-00022.bin",
+    "transformer.h.53.attn.bias": "pytorch_model-00015-of-00022.bin",
+    "transformer.h.53.attn.c_attn.bias": "pytorch_model-00015-of-00022.bin",
+    "transformer.h.53.attn.c_attn.weight": "pytorch_model-00015-of-00022.bin",
+    "transformer.h.53.attn.c_proj.bias": "pytorch_model-00015-of-00022.bin",
+    "transformer.h.53.attn.c_proj.weight": "pytorch_model-00015-of-00022.bin",
+    "transformer.h.53.attn.masked_bias": "pytorch_model-00015-of-00022.bin",
+    "transformer.h.53.ln_1.bias": "pytorch_model-00015-of-00022.bin",
+    "transformer.h.53.ln_1.weight": "pytorch_model-00015-of-00022.bin",
+    "transformer.h.53.ln_2.bias": "pytorch_model-00015-of-00022.bin",
+    "transformer.h.53.ln_2.weight": "pytorch_model-00015-of-00022.bin",
+    "transformer.h.53.mlp.c_fc.bias": "pytorch_model-00015-of-00022.bin",
+    "transformer.h.53.mlp.c_fc.weight": "pytorch_model-00015-of-00022.bin",
+    "transformer.h.53.mlp.c_proj.bias": "pytorch_model-00015-of-00022.bin",
+    "transformer.h.53.mlp.c_proj.weight": "pytorch_model-00015-of-00022.bin",
+    "transformer.h.54.attn.bias": "pytorch_model-00015-of-00022.bin",
+    "transformer.h.54.attn.c_attn.bias": "pytorch_model-00015-of-00022.bin",
+    "transformer.h.54.attn.c_attn.weight": "pytorch_model-00015-of-00022.bin",
+    "transformer.h.54.attn.c_proj.bias": "pytorch_model-00015-of-00022.bin",
+    "transformer.h.54.attn.c_proj.weight": "pytorch_model-00015-of-00022.bin",
+    "transformer.h.54.attn.masked_bias": "pytorch_model-00015-of-00022.bin",
+    "transformer.h.54.ln_1.bias": "pytorch_model-00015-of-00022.bin",
+    "transformer.h.54.ln_1.weight": "pytorch_model-00015-of-00022.bin",
+    "transformer.h.54.ln_2.bias": "pytorch_model-00015-of-00022.bin",
+    "transformer.h.54.ln_2.weight": "pytorch_model-00015-of-00022.bin",
+    "transformer.h.54.mlp.c_fc.bias": "pytorch_model-00015-of-00022.bin",
+    "transformer.h.54.mlp.c_fc.weight": "pytorch_model-00015-of-00022.bin",
+    "transformer.h.54.mlp.c_proj.bias": "pytorch_model-00015-of-00022.bin",
+    "transformer.h.54.mlp.c_proj.weight": "pytorch_model-00015-of-00022.bin",
+    "transformer.h.55.attn.bias": "pytorch_model-00015-of-00022.bin",
+    "transformer.h.55.attn.c_attn.bias": "pytorch_model-00015-of-00022.bin",
+    "transformer.h.55.attn.c_attn.weight": "pytorch_model-00015-of-00022.bin",
+    "transformer.h.55.attn.c_proj.bias": "pytorch_model-00015-of-00022.bin",
+    "transformer.h.55.attn.c_proj.weight": "pytorch_model-00015-of-00022.bin",
+    "transformer.h.55.attn.masked_bias": "pytorch_model-00015-of-00022.bin",
+    "transformer.h.55.ln_1.bias": "pytorch_model-00015-of-00022.bin",
+    "transformer.h.55.ln_1.weight": "pytorch_model-00015-of-00022.bin",
+    "transformer.h.55.ln_2.bias": "pytorch_model-00015-of-00022.bin",
+    "transformer.h.55.ln_2.weight": "pytorch_model-00015-of-00022.bin",
+    "transformer.h.55.mlp.c_fc.bias": "pytorch_model-00016-of-00022.bin",
+    "transformer.h.55.mlp.c_fc.weight": "pytorch_model-00016-of-00022.bin",
+    "transformer.h.55.mlp.c_proj.bias": "pytorch_model-00016-of-00022.bin",
+    "transformer.h.55.mlp.c_proj.weight": "pytorch_model-00016-of-00022.bin",
+    "transformer.h.56.attn.bias": "pytorch_model-00016-of-00022.bin",
+    "transformer.h.56.attn.c_attn.bias": "pytorch_model-00016-of-00022.bin",
+    "transformer.h.56.attn.c_attn.weight": "pytorch_model-00016-of-00022.bin",
+    "transformer.h.56.attn.c_proj.bias": "pytorch_model-00016-of-00022.bin",
+    "transformer.h.56.attn.c_proj.weight": "pytorch_model-00016-of-00022.bin",
+    "transformer.h.56.attn.masked_bias": "pytorch_model-00016-of-00022.bin",
+    "transformer.h.56.ln_1.bias": "pytorch_model-00016-of-00022.bin",
+    "transformer.h.56.ln_1.weight": "pytorch_model-00016-of-00022.bin",
+    "transformer.h.56.ln_2.bias": "pytorch_model-00016-of-00022.bin",
+    "transformer.h.56.ln_2.weight": "pytorch_model-00016-of-00022.bin",
+    "transformer.h.56.mlp.c_fc.bias": "pytorch_model-00016-of-00022.bin",
+    "transformer.h.56.mlp.c_fc.weight": "pytorch_model-00016-of-00022.bin",
+    "transformer.h.56.mlp.c_proj.bias": "pytorch_model-00016-of-00022.bin",
+    "transformer.h.56.mlp.c_proj.weight": "pytorch_model-00016-of-00022.bin",
+    "transformer.h.57.attn.bias": "pytorch_model-00016-of-00022.bin",
+    "transformer.h.57.attn.c_attn.bias": "pytorch_model-00016-of-00022.bin",
+    "transformer.h.57.attn.c_attn.weight": "pytorch_model-00016-of-00022.bin",
+    "transformer.h.57.attn.c_proj.bias": "pytorch_model-00016-of-00022.bin",
+    "transformer.h.57.attn.c_proj.weight": "pytorch_model-00016-of-00022.bin",
+    "transformer.h.57.attn.masked_bias": "pytorch_model-00016-of-00022.bin",
+    "transformer.h.57.ln_1.bias": "pytorch_model-00016-of-00022.bin",
+    "transformer.h.57.ln_1.weight": "pytorch_model-00016-of-00022.bin",
+    "transformer.h.57.ln_2.bias": "pytorch_model-00016-of-00022.bin",
+    "transformer.h.57.ln_2.weight": "pytorch_model-00016-of-00022.bin",
+    "transformer.h.57.mlp.c_fc.bias": "pytorch_model-00016-of-00022.bin",
+    "transformer.h.57.mlp.c_fc.weight": "pytorch_model-00016-of-00022.bin",
+    "transformer.h.57.mlp.c_proj.bias": "pytorch_model-00016-of-00022.bin",
+    "transformer.h.57.mlp.c_proj.weight": "pytorch_model-00016-of-00022.bin",
+    "transformer.h.58.attn.bias": "pytorch_model-00016-of-00022.bin",
+    "transformer.h.58.attn.c_attn.bias": "pytorch_model-00016-of-00022.bin",
+    "transformer.h.58.attn.c_attn.weight": "pytorch_model-00016-of-00022.bin",
+    "transformer.h.58.attn.c_proj.bias": "pytorch_model-00016-of-00022.bin",
+    "transformer.h.58.attn.c_proj.weight": "pytorch_model-00016-of-00022.bin",
+    "transformer.h.58.attn.masked_bias": "pytorch_model-00016-of-00022.bin",
+    "transformer.h.58.ln_1.bias": "pytorch_model-00016-of-00022.bin",
+    "transformer.h.58.ln_1.weight": "pytorch_model-00016-of-00022.bin",
+    "transformer.h.58.ln_2.bias": "pytorch_model-00016-of-00022.bin",
+    "transformer.h.58.ln_2.weight": "pytorch_model-00016-of-00022.bin",
+    "transformer.h.58.mlp.c_fc.bias": "pytorch_model-00016-of-00022.bin",
+    "transformer.h.58.mlp.c_fc.weight": "pytorch_model-00016-of-00022.bin",
+    "transformer.h.58.mlp.c_proj.bias": "pytorch_model-00016-of-00022.bin",
+    "transformer.h.58.mlp.c_proj.weight": "pytorch_model-00016-of-00022.bin",
+    "transformer.h.59.attn.bias": "pytorch_model-00016-of-00022.bin",
+    "transformer.h.59.attn.c_attn.bias": "pytorch_model-00016-of-00022.bin",
+    "transformer.h.59.attn.c_attn.weight": "pytorch_model-00016-of-00022.bin",
+    "transformer.h.59.attn.c_proj.bias": "pytorch_model-00017-of-00022.bin",
+    "transformer.h.59.attn.c_proj.weight": "pytorch_model-00017-of-00022.bin",
+    "transformer.h.59.attn.masked_bias": "pytorch_model-00016-of-00022.bin",
+    "transformer.h.59.ln_1.bias": "pytorch_model-00016-of-00022.bin",
+    "transformer.h.59.ln_1.weight": "pytorch_model-00016-of-00022.bin",
+    "transformer.h.59.ln_2.bias": "pytorch_model-00017-of-00022.bin",
+    "transformer.h.59.ln_2.weight": "pytorch_model-00017-of-00022.bin",
+    "transformer.h.59.mlp.c_fc.bias": "pytorch_model-00017-of-00022.bin",
+    "transformer.h.59.mlp.c_fc.weight": "pytorch_model-00017-of-00022.bin",
+    "transformer.h.59.mlp.c_proj.bias": "pytorch_model-00017-of-00022.bin",
+    "transformer.h.59.mlp.c_proj.weight": "pytorch_model-00017-of-00022.bin",
+    "transformer.h.6.attn.bias": "pytorch_model-00002-of-00022.bin",
+    "transformer.h.6.attn.c_attn.bias": "pytorch_model-00002-of-00022.bin",
+    "transformer.h.6.attn.c_attn.weight": "pytorch_model-00002-of-00022.bin",
+    "transformer.h.6.attn.c_proj.bias": "pytorch_model-00002-of-00022.bin",
+    "transformer.h.6.attn.c_proj.weight": "pytorch_model-00002-of-00022.bin",
+    "transformer.h.6.attn.masked_bias": "pytorch_model-00002-of-00022.bin",
+    "transformer.h.6.ln_1.bias": "pytorch_model-00002-of-00022.bin",
+    "transformer.h.6.ln_1.weight": "pytorch_model-00002-of-00022.bin",
+    "transformer.h.6.ln_2.bias": "pytorch_model-00002-of-00022.bin",
+    "transformer.h.6.ln_2.weight": "pytorch_model-00002-of-00022.bin",
+    "transformer.h.6.mlp.c_fc.bias": "pytorch_model-00002-of-00022.bin",
+    "transformer.h.6.mlp.c_fc.weight": "pytorch_model-00002-of-00022.bin",
+    "transformer.h.6.mlp.c_proj.bias": "pytorch_model-00003-of-00022.bin",
+    "transformer.h.6.mlp.c_proj.weight": "pytorch_model-00003-of-00022.bin",
+    "transformer.h.60.attn.bias": "pytorch_model-00017-of-00022.bin",
+    "transformer.h.60.attn.c_attn.bias": "pytorch_model-00017-of-00022.bin",
+    "transformer.h.60.attn.c_attn.weight": "pytorch_model-00017-of-00022.bin",
+    "transformer.h.60.attn.c_proj.bias": "pytorch_model-00017-of-00022.bin",
+    "transformer.h.60.attn.c_proj.weight": "pytorch_model-00017-of-00022.bin",
+    "transformer.h.60.attn.masked_bias": "pytorch_model-00017-of-00022.bin",
+    "transformer.h.60.ln_1.bias": "pytorch_model-00017-of-00022.bin",
+    "transformer.h.60.ln_1.weight": "pytorch_model-00017-of-00022.bin",
+    "transformer.h.60.ln_2.bias": "pytorch_model-00017-of-00022.bin",
+    "transformer.h.60.ln_2.weight": "pytorch_model-00017-of-00022.bin",
+    "transformer.h.60.mlp.c_fc.bias": "pytorch_model-00017-of-00022.bin",
+    "transformer.h.60.mlp.c_fc.weight": "pytorch_model-00017-of-00022.bin",
+    "transformer.h.60.mlp.c_proj.bias": "pytorch_model-00017-of-00022.bin",
+    "transformer.h.60.mlp.c_proj.weight": "pytorch_model-00017-of-00022.bin",
+    "transformer.h.61.attn.bias": "pytorch_model-00017-of-00022.bin",
+    "transformer.h.61.attn.c_attn.bias": "pytorch_model-00017-of-00022.bin",
+    "transformer.h.61.attn.c_attn.weight": "pytorch_model-00017-of-00022.bin",
+    "transformer.h.61.attn.c_proj.bias": "pytorch_model-00017-of-00022.bin",
+    "transformer.h.61.attn.c_proj.weight": "pytorch_model-00017-of-00022.bin",
+    "transformer.h.61.attn.masked_bias": "pytorch_model-00017-of-00022.bin",
+    "transformer.h.61.ln_1.bias": "pytorch_model-00017-of-00022.bin",
+    "transformer.h.61.ln_1.weight": "pytorch_model-00017-of-00022.bin",
+    "transformer.h.61.ln_2.bias": "pytorch_model-00017-of-00022.bin",
+    "transformer.h.61.ln_2.weight": "pytorch_model-00017-of-00022.bin",
+    "transformer.h.61.mlp.c_fc.bias": "pytorch_model-00017-of-00022.bin",
+    "transformer.h.61.mlp.c_fc.weight": "pytorch_model-00017-of-00022.bin",
+    "transformer.h.61.mlp.c_proj.bias": "pytorch_model-00017-of-00022.bin",
+    "transformer.h.61.mlp.c_proj.weight": "pytorch_model-00017-of-00022.bin",
+    "transformer.h.62.attn.bias": "pytorch_model-00017-of-00022.bin",
+    "transformer.h.62.attn.c_attn.bias": "pytorch_model-00017-of-00022.bin",
+    "transformer.h.62.attn.c_attn.weight": "pytorch_model-00017-of-00022.bin",
+    "transformer.h.62.attn.c_proj.bias": "pytorch_model-00017-of-00022.bin",
+    "transformer.h.62.attn.c_proj.weight": "pytorch_model-00017-of-00022.bin",
+    "transformer.h.62.attn.masked_bias": "pytorch_model-00017-of-00022.bin",
+    "transformer.h.62.ln_1.bias": "pytorch_model-00017-of-00022.bin",
+    "transformer.h.62.ln_1.weight": "pytorch_model-00017-of-00022.bin",
+    "transformer.h.62.ln_2.bias": "pytorch_model-00017-of-00022.bin",
+    "transformer.h.62.ln_2.weight": "pytorch_model-00017-of-00022.bin",
+    "transformer.h.62.mlp.c_fc.bias": "pytorch_model-00017-of-00022.bin",
+    "transformer.h.62.mlp.c_fc.weight": "pytorch_model-00017-of-00022.bin",
+    "transformer.h.62.mlp.c_proj.bias": "pytorch_model-00017-of-00022.bin",
+    "transformer.h.62.mlp.c_proj.weight": "pytorch_model-00017-of-00022.bin",
+    "transformer.h.63.attn.bias": "pytorch_model-00017-of-00022.bin",
+    "transformer.h.63.attn.c_attn.bias": "pytorch_model-00018-of-00022.bin",
+    "transformer.h.63.attn.c_attn.weight": "pytorch_model-00018-of-00022.bin",
+    "transformer.h.63.attn.c_proj.bias": "pytorch_model-00018-of-00022.bin",
+    "transformer.h.63.attn.c_proj.weight": "pytorch_model-00018-of-00022.bin",
+    "transformer.h.63.attn.masked_bias": "pytorch_model-00017-of-00022.bin",
+    "transformer.h.63.ln_1.bias": "pytorch_model-00017-of-00022.bin",
+    "transformer.h.63.ln_1.weight": "pytorch_model-00017-of-00022.bin",
+    "transformer.h.63.ln_2.bias": "pytorch_model-00018-of-00022.bin",
+    "transformer.h.63.ln_2.weight": "pytorch_model-00018-of-00022.bin",
+    "transformer.h.63.mlp.c_fc.bias": "pytorch_model-00018-of-00022.bin",
+    "transformer.h.63.mlp.c_fc.weight": "pytorch_model-00018-of-00022.bin",
+    "transformer.h.63.mlp.c_proj.bias": "pytorch_model-00018-of-00022.bin",
+    "transformer.h.63.mlp.c_proj.weight": "pytorch_model-00018-of-00022.bin",
+    "transformer.h.64.attn.bias": "pytorch_model-00018-of-00022.bin",
+    "transformer.h.64.attn.c_attn.bias": "pytorch_model-00018-of-00022.bin",
+    "transformer.h.64.attn.c_attn.weight": "pytorch_model-00018-of-00022.bin",
+    "transformer.h.64.attn.c_proj.bias": "pytorch_model-00018-of-00022.bin",
+    "transformer.h.64.attn.c_proj.weight": "pytorch_model-00018-of-00022.bin",
+    "transformer.h.64.attn.masked_bias": "pytorch_model-00018-of-00022.bin",
+    "transformer.h.64.ln_1.bias": "pytorch_model-00018-of-00022.bin",
+    "transformer.h.64.ln_1.weight": "pytorch_model-00018-of-00022.bin",
+    "transformer.h.64.ln_2.bias": "pytorch_model-00018-of-00022.bin",
+    "transformer.h.64.ln_2.weight": "pytorch_model-00018-of-00022.bin",
+    "transformer.h.64.mlp.c_fc.bias": "pytorch_model-00018-of-00022.bin",
+    "transformer.h.64.mlp.c_fc.weight": "pytorch_model-00018-of-00022.bin",
+    "transformer.h.64.mlp.c_proj.bias": "pytorch_model-00018-of-00022.bin",
+    "transformer.h.64.mlp.c_proj.weight": "pytorch_model-00018-of-00022.bin",
+    "transformer.h.65.attn.bias": "pytorch_model-00018-of-00022.bin",
+    "transformer.h.65.attn.c_attn.bias": "pytorch_model-00018-of-00022.bin",
+    "transformer.h.65.attn.c_attn.weight": "pytorch_model-00018-of-00022.bin",
+    "transformer.h.65.attn.c_proj.bias": "pytorch_model-00018-of-00022.bin",
+    "transformer.h.65.attn.c_proj.weight": "pytorch_model-00018-of-00022.bin",
+    "transformer.h.65.attn.masked_bias": "pytorch_model-00018-of-00022.bin",
+    "transformer.h.65.ln_1.bias": "pytorch_model-00018-of-00022.bin",
+    "transformer.h.65.ln_1.weight": "pytorch_model-00018-of-00022.bin",
+    "transformer.h.65.ln_2.bias": "pytorch_model-00018-of-00022.bin",
+    "transformer.h.65.ln_2.weight": "pytorch_model-00018-of-00022.bin",
+    "transformer.h.65.mlp.c_fc.bias": "pytorch_model-00018-of-00022.bin",
+    "transformer.h.65.mlp.c_fc.weight": "pytorch_model-00018-of-00022.bin",
+    "transformer.h.65.mlp.c_proj.bias": "pytorch_model-00018-of-00022.bin",
+    "transformer.h.65.mlp.c_proj.weight": "pytorch_model-00018-of-00022.bin",
+    "transformer.h.66.attn.bias": "pytorch_model-00018-of-00022.bin",
+    "transformer.h.66.attn.c_attn.bias": "pytorch_model-00018-of-00022.bin",
+    "transformer.h.66.attn.c_attn.weight": "pytorch_model-00018-of-00022.bin",
+    "transformer.h.66.attn.c_proj.bias": "pytorch_model-00018-of-00022.bin",
+    "transformer.h.66.attn.c_proj.weight": "pytorch_model-00018-of-00022.bin",
+    "transformer.h.66.attn.masked_bias": "pytorch_model-00018-of-00022.bin",
+    "transformer.h.66.ln_1.bias": "pytorch_model-00018-of-00022.bin",
+    "transformer.h.66.ln_1.weight": "pytorch_model-00018-of-00022.bin",
+    "transformer.h.66.ln_2.bias": "pytorch_model-00018-of-00022.bin",
+    "transformer.h.66.ln_2.weight": "pytorch_model-00018-of-00022.bin",
+    "transformer.h.66.mlp.c_fc.bias": "pytorch_model-00018-of-00022.bin",
+    "transformer.h.66.mlp.c_fc.weight": "pytorch_model-00018-of-00022.bin",
+    "transformer.h.66.mlp.c_proj.bias": "pytorch_model-00019-of-00022.bin",
+    "transformer.h.66.mlp.c_proj.weight": "pytorch_model-00019-of-00022.bin",
+    "transformer.h.67.attn.bias": "pytorch_model-00019-of-00022.bin",
+    "transformer.h.67.attn.c_attn.bias": "pytorch_model-00019-of-00022.bin",
+    "transformer.h.67.attn.c_attn.weight": "pytorch_model-00019-of-00022.bin",
+    "transformer.h.67.attn.c_proj.bias": "pytorch_model-00019-of-00022.bin",
+    "transformer.h.67.attn.c_proj.weight": "pytorch_model-00019-of-00022.bin",
+    "transformer.h.67.attn.masked_bias": "pytorch_model-00019-of-00022.bin",
+    "transformer.h.67.ln_1.bias": "pytorch_model-00019-of-00022.bin",
+    "transformer.h.67.ln_1.weight": "pytorch_model-00019-of-00022.bin",
+    "transformer.h.67.ln_2.bias": "pytorch_model-00019-of-00022.bin",
+    "transformer.h.67.ln_2.weight": "pytorch_model-00019-of-00022.bin",
+    "transformer.h.67.mlp.c_fc.bias": "pytorch_model-00019-of-00022.bin",
+    "transformer.h.67.mlp.c_fc.weight": "pytorch_model-00019-of-00022.bin",
+    "transformer.h.67.mlp.c_proj.bias": "pytorch_model-00019-of-00022.bin",
+    "transformer.h.67.mlp.c_proj.weight": "pytorch_model-00019-of-00022.bin",
+    "transformer.h.68.attn.bias": "pytorch_model-00019-of-00022.bin",
+    "transformer.h.68.attn.c_attn.bias": "pytorch_model-00019-of-00022.bin",
+    "transformer.h.68.attn.c_attn.weight": "pytorch_model-00019-of-00022.bin",
+    "transformer.h.68.attn.c_proj.bias": "pytorch_model-00019-of-00022.bin",
+    "transformer.h.68.attn.c_proj.weight": "pytorch_model-00019-of-00022.bin",
+    "transformer.h.68.attn.masked_bias": "pytorch_model-00019-of-00022.bin",
+    "transformer.h.68.ln_1.bias": "pytorch_model-00019-of-00022.bin",
+    "transformer.h.68.ln_1.weight": "pytorch_model-00019-of-00022.bin",
+    "transformer.h.68.ln_2.bias": "pytorch_model-00019-of-00022.bin",
+    "transformer.h.68.ln_2.weight": "pytorch_model-00019-of-00022.bin",
+    "transformer.h.68.mlp.c_fc.bias": "pytorch_model-00019-of-00022.bin",
+    "transformer.h.68.mlp.c_fc.weight": "pytorch_model-00019-of-00022.bin",
+    "transformer.h.68.mlp.c_proj.bias": "pytorch_model-00019-of-00022.bin",
+    "transformer.h.68.mlp.c_proj.weight": "pytorch_model-00019-of-00022.bin",
+    "transformer.h.69.attn.bias": "pytorch_model-00019-of-00022.bin",
+    "transformer.h.69.attn.c_attn.bias": "pytorch_model-00019-of-00022.bin",
+    "transformer.h.69.attn.c_attn.weight": "pytorch_model-00019-of-00022.bin",
+    "transformer.h.69.attn.c_proj.bias": "pytorch_model-00019-of-00022.bin",
+    "transformer.h.69.attn.c_proj.weight": "pytorch_model-00019-of-00022.bin",
+    "transformer.h.69.attn.masked_bias": "pytorch_model-00019-of-00022.bin",
+    "transformer.h.69.ln_1.bias": "pytorch_model-00019-of-00022.bin",
+    "transformer.h.69.ln_1.weight": "pytorch_model-00019-of-00022.bin",
+    "transformer.h.69.ln_2.bias": "pytorch_model-00019-of-00022.bin",
+    "transformer.h.69.ln_2.weight": "pytorch_model-00019-of-00022.bin",
+    "transformer.h.69.mlp.c_fc.bias": "pytorch_model-00019-of-00022.bin",
+    "transformer.h.69.mlp.c_fc.weight": "pytorch_model-00019-of-00022.bin",
+    "transformer.h.69.mlp.c_proj.bias": "pytorch_model-00019-of-00022.bin",
+    "transformer.h.69.mlp.c_proj.weight": "pytorch_model-00019-of-00022.bin",
+    "transformer.h.7.attn.bias": "pytorch_model-00003-of-00022.bin",
+    "transformer.h.7.attn.c_attn.bias": "pytorch_model-00003-of-00022.bin",
+    "transformer.h.7.attn.c_attn.weight": "pytorch_model-00003-of-00022.bin",
+    "transformer.h.7.attn.c_proj.bias": "pytorch_model-00003-of-00022.bin",
+    "transformer.h.7.attn.c_proj.weight": "pytorch_model-00003-of-00022.bin",
+    "transformer.h.7.attn.masked_bias": "pytorch_model-00003-of-00022.bin",
+    "transformer.h.7.ln_1.bias": "pytorch_model-00003-of-00022.bin",
+    "transformer.h.7.ln_1.weight": "pytorch_model-00003-of-00022.bin",
+    "transformer.h.7.ln_2.bias": "pytorch_model-00003-of-00022.bin",
+    "transformer.h.7.ln_2.weight": "pytorch_model-00003-of-00022.bin",
+    "transformer.h.7.mlp.c_fc.bias": "pytorch_model-00003-of-00022.bin",
+    "transformer.h.7.mlp.c_fc.weight": "pytorch_model-00003-of-00022.bin",
+    "transformer.h.7.mlp.c_proj.bias": "pytorch_model-00003-of-00022.bin",
+    "transformer.h.7.mlp.c_proj.weight": "pytorch_model-00003-of-00022.bin",
+    "transformer.h.70.attn.bias": "pytorch_model-00019-of-00022.bin",
+    "transformer.h.70.attn.c_attn.bias": "pytorch_model-00019-of-00022.bin",
+    "transformer.h.70.attn.c_attn.weight": "pytorch_model-00019-of-00022.bin",
+    "transformer.h.70.attn.c_proj.bias": "pytorch_model-00019-of-00022.bin",
+    "transformer.h.70.attn.c_proj.weight": "pytorch_model-00019-of-00022.bin",
+    "transformer.h.70.attn.masked_bias": "pytorch_model-00019-of-00022.bin",
+    "transformer.h.70.ln_1.bias": "pytorch_model-00019-of-00022.bin",
+    "transformer.h.70.ln_1.weight": "pytorch_model-00019-of-00022.bin",
+    "transformer.h.70.ln_2.bias": "pytorch_model-00019-of-00022.bin",
+    "transformer.h.70.ln_2.weight": "pytorch_model-00019-of-00022.bin",
+    "transformer.h.70.mlp.c_fc.bias": "pytorch_model-00020-of-00022.bin",
+    "transformer.h.70.mlp.c_fc.weight": "pytorch_model-00020-of-00022.bin",
+    "transformer.h.70.mlp.c_proj.bias": "pytorch_model-00020-of-00022.bin",
+    "transformer.h.70.mlp.c_proj.weight": "pytorch_model-00020-of-00022.bin",
+    "transformer.h.71.attn.bias": "pytorch_model-00020-of-00022.bin",
+    "transformer.h.71.attn.c_attn.bias": "pytorch_model-00020-of-00022.bin",
+    "transformer.h.71.attn.c_attn.weight": "pytorch_model-00020-of-00022.bin",
+    "transformer.h.71.attn.c_proj.bias": "pytorch_model-00020-of-00022.bin",
+    "transformer.h.71.attn.c_proj.weight": "pytorch_model-00020-of-00022.bin",
+    "transformer.h.71.attn.masked_bias": "pytorch_model-00020-of-00022.bin",
+    "transformer.h.71.ln_1.bias": "pytorch_model-00020-of-00022.bin",
+    "transformer.h.71.ln_1.weight": "pytorch_model-00020-of-00022.bin",
+    "transformer.h.71.ln_2.bias": "pytorch_model-00020-of-00022.bin",
+    "transformer.h.71.ln_2.weight": "pytorch_model-00020-of-00022.bin",
+    "transformer.h.71.mlp.c_fc.bias": "pytorch_model-00020-of-00022.bin",
+    "transformer.h.71.mlp.c_fc.weight": "pytorch_model-00020-of-00022.bin",
+    "transformer.h.71.mlp.c_proj.bias": "pytorch_model-00020-of-00022.bin",
+    "transformer.h.71.mlp.c_proj.weight": "pytorch_model-00020-of-00022.bin",
+    "transformer.h.72.attn.bias": "pytorch_model-00020-of-00022.bin",
+    "transformer.h.72.attn.c_attn.bias": "pytorch_model-00020-of-00022.bin",
+    "transformer.h.72.attn.c_attn.weight": "pytorch_model-00020-of-00022.bin",
+    "transformer.h.72.attn.c_proj.bias": "pytorch_model-00020-of-00022.bin",
+    "transformer.h.72.attn.c_proj.weight": "pytorch_model-00020-of-00022.bin",
+    "transformer.h.72.attn.masked_bias": "pytorch_model-00020-of-00022.bin",
+    "transformer.h.72.ln_1.bias": "pytorch_model-00020-of-00022.bin",
+    "transformer.h.72.ln_1.weight": "pytorch_model-00020-of-00022.bin",
+    "transformer.h.72.ln_2.bias": "pytorch_model-00020-of-00022.bin",
+    "transformer.h.72.ln_2.weight": "pytorch_model-00020-of-00022.bin",
+    "transformer.h.72.mlp.c_fc.bias": "pytorch_model-00020-of-00022.bin",
+    "transformer.h.72.mlp.c_fc.weight": "pytorch_model-00020-of-00022.bin",
+    "transformer.h.72.mlp.c_proj.bias": "pytorch_model-00020-of-00022.bin",
+    "transformer.h.72.mlp.c_proj.weight": "pytorch_model-00020-of-00022.bin",
+    "transformer.h.73.attn.bias": "pytorch_model-00020-of-00022.bin",
+    "transformer.h.73.attn.c_attn.bias": "pytorch_model-00020-of-00022.bin",
+    "transformer.h.73.attn.c_attn.weight": "pytorch_model-00020-of-00022.bin",
+    "transformer.h.73.attn.c_proj.bias": "pytorch_model-00020-of-00022.bin",
+    "transformer.h.73.attn.c_proj.weight": "pytorch_model-00020-of-00022.bin",
+    "transformer.h.73.attn.masked_bias": "pytorch_model-00020-of-00022.bin",
+    "transformer.h.73.ln_1.bias": "pytorch_model-00020-of-00022.bin",
+    "transformer.h.73.ln_1.weight": "pytorch_model-00020-of-00022.bin",
+    "transformer.h.73.ln_2.bias": "pytorch_model-00020-of-00022.bin",
+    "transformer.h.73.ln_2.weight": "pytorch_model-00020-of-00022.bin",
+    "transformer.h.73.mlp.c_fc.bias": "pytorch_model-00020-of-00022.bin",
+    "transformer.h.73.mlp.c_fc.weight": "pytorch_model-00020-of-00022.bin",
+    "transformer.h.73.mlp.c_proj.bias": "pytorch_model-00020-of-00022.bin",
+    "transformer.h.73.mlp.c_proj.weight": "pytorch_model-00020-of-00022.bin",
+    "transformer.h.74.attn.bias": "pytorch_model-00020-of-00022.bin",
+    "transformer.h.74.attn.c_attn.bias": "pytorch_model-00020-of-00022.bin",
+    "transformer.h.74.attn.c_attn.weight": "pytorch_model-00020-of-00022.bin",
+    "transformer.h.74.attn.c_proj.bias": "pytorch_model-00021-of-00022.bin",
+    "transformer.h.74.attn.c_proj.weight": "pytorch_model-00021-of-00022.bin",
+    "transformer.h.74.attn.masked_bias": "pytorch_model-00020-of-00022.bin",
+    "transformer.h.74.ln_1.bias": "pytorch_model-00020-of-00022.bin",
+    "transformer.h.74.ln_1.weight": "pytorch_model-00020-of-00022.bin",
+    "transformer.h.74.ln_2.bias": "pytorch_model-00021-of-00022.bin",
+    "transformer.h.74.ln_2.weight": "pytorch_model-00021-of-00022.bin",
+    "transformer.h.74.mlp.c_fc.bias": "pytorch_model-00021-of-00022.bin",
+    "transformer.h.74.mlp.c_fc.weight": "pytorch_model-00021-of-00022.bin",
+    "transformer.h.74.mlp.c_proj.bias": "pytorch_model-00021-of-00022.bin",
+    "transformer.h.74.mlp.c_proj.weight": "pytorch_model-00021-of-00022.bin",
+    "transformer.h.75.attn.bias": "pytorch_model-00021-of-00022.bin",
+    "transformer.h.75.attn.c_attn.bias": "pytorch_model-00021-of-00022.bin",
+    "transformer.h.75.attn.c_attn.weight": "pytorch_model-00021-of-00022.bin",
+    "transformer.h.75.attn.c_proj.bias": "pytorch_model-00021-of-00022.bin",
+    "transformer.h.75.attn.c_proj.weight": "pytorch_model-00021-of-00022.bin",
+    "transformer.h.75.attn.masked_bias": "pytorch_model-00021-of-00022.bin",
+    "transformer.h.75.ln_1.bias": "pytorch_model-00021-of-00022.bin",
+    "transformer.h.75.ln_1.weight": "pytorch_model-00021-of-00022.bin",
+    "transformer.h.75.ln_2.bias": "pytorch_model-00021-of-00022.bin",
+    "transformer.h.75.ln_2.weight": "pytorch_model-00021-of-00022.bin",
+    "transformer.h.75.mlp.c_fc.bias": "pytorch_model-00021-of-00022.bin",
+    "transformer.h.75.mlp.c_fc.weight": "pytorch_model-00021-of-00022.bin",
+    "transformer.h.75.mlp.c_proj.bias": "pytorch_model-00021-of-00022.bin",
+    "transformer.h.75.mlp.c_proj.weight": "pytorch_model-00021-of-00022.bin",
+    "transformer.h.76.attn.bias": "pytorch_model-00021-of-00022.bin",
+    "transformer.h.76.attn.c_attn.bias": "pytorch_model-00021-of-00022.bin",
+    "transformer.h.76.attn.c_attn.weight": "pytorch_model-00021-of-00022.bin",
+    "transformer.h.76.attn.c_proj.bias": "pytorch_model-00021-of-00022.bin",
+    "transformer.h.76.attn.c_proj.weight": "pytorch_model-00021-of-00022.bin",
+    "transformer.h.76.attn.masked_bias": "pytorch_model-00021-of-00022.bin",
+    "transformer.h.76.ln_1.bias": "pytorch_model-00021-of-00022.bin",
+    "transformer.h.76.ln_1.weight": "pytorch_model-00021-of-00022.bin",
+    "transformer.h.76.ln_2.bias": "pytorch_model-00021-of-00022.bin",
+    "transformer.h.76.ln_2.weight": "pytorch_model-00021-of-00022.bin",
+    "transformer.h.76.mlp.c_fc.bias": "pytorch_model-00021-of-00022.bin",
+    "transformer.h.76.mlp.c_fc.weight": "pytorch_model-00021-of-00022.bin",
+    "transformer.h.76.mlp.c_proj.bias": "pytorch_model-00021-of-00022.bin",
+    "transformer.h.76.mlp.c_proj.weight": "pytorch_model-00021-of-00022.bin",
+    "transformer.h.77.attn.bias": "pytorch_model-00021-of-00022.bin",
+    "transformer.h.77.attn.c_attn.bias": "pytorch_model-00021-of-00022.bin",
+    "transformer.h.77.attn.c_attn.weight": "pytorch_model-00021-of-00022.bin",
+    "transformer.h.77.attn.c_proj.bias": "pytorch_model-00021-of-00022.bin",
+    "transformer.h.77.attn.c_proj.weight": "pytorch_model-00021-of-00022.bin",
+    "transformer.h.77.attn.masked_bias": "pytorch_model-00021-of-00022.bin",
+    "transformer.h.77.ln_1.bias": "pytorch_model-00021-of-00022.bin",
+    "transformer.h.77.ln_1.weight": "pytorch_model-00021-of-00022.bin",
+    "transformer.h.77.ln_2.bias": "pytorch_model-00021-of-00022.bin",
+    "transformer.h.77.ln_2.weight": "pytorch_model-00021-of-00022.bin",
+    "transformer.h.77.mlp.c_fc.bias": "pytorch_model-00021-of-00022.bin",
+    "transformer.h.77.mlp.c_fc.weight": "pytorch_model-00021-of-00022.bin",
+    "transformer.h.77.mlp.c_proj.bias": "pytorch_model-00021-of-00022.bin",
+    "transformer.h.77.mlp.c_proj.weight": "pytorch_model-00021-of-00022.bin",
+    "transformer.h.78.attn.bias": "pytorch_model-00021-of-00022.bin",
+    "transformer.h.78.attn.c_attn.bias": "pytorch_model-00022-of-00022.bin",
+    "transformer.h.78.attn.c_attn.weight": "pytorch_model-00022-of-00022.bin",
+    "transformer.h.78.attn.c_proj.bias": "pytorch_model-00022-of-00022.bin",
+    "transformer.h.78.attn.c_proj.weight": "pytorch_model-00022-of-00022.bin",
+    "transformer.h.78.attn.masked_bias": "pytorch_model-00021-of-00022.bin",
+    "transformer.h.78.ln_1.bias": "pytorch_model-00021-of-00022.bin",
+    "transformer.h.78.ln_1.weight": "pytorch_model-00021-of-00022.bin",
+    "transformer.h.78.ln_2.bias": "pytorch_model-00022-of-00022.bin",
+    "transformer.h.78.ln_2.weight": "pytorch_model-00022-of-00022.bin",
+    "transformer.h.78.mlp.c_fc.bias": "pytorch_model-00022-of-00022.bin",
+    "transformer.h.78.mlp.c_fc.weight": "pytorch_model-00022-of-00022.bin",
+    "transformer.h.78.mlp.c_proj.bias": "pytorch_model-00022-of-00022.bin",
+    "transformer.h.78.mlp.c_proj.weight": "pytorch_model-00022-of-00022.bin",
+    "transformer.h.79.attn.bias": "pytorch_model-00022-of-00022.bin",
+    "transformer.h.79.attn.c_attn.bias": "pytorch_model-00022-of-00022.bin",
+    "transformer.h.79.attn.c_attn.weight": "pytorch_model-00022-of-00022.bin",
+    "transformer.h.79.attn.c_proj.bias": "pytorch_model-00022-of-00022.bin",
+    "transformer.h.79.attn.c_proj.weight": "pytorch_model-00022-of-00022.bin",
+    "transformer.h.79.attn.masked_bias": "pytorch_model-00022-of-00022.bin",
+    "transformer.h.79.ln_1.bias": "pytorch_model-00022-of-00022.bin",
+    "transformer.h.79.ln_1.weight": "pytorch_model-00022-of-00022.bin",
+    "transformer.h.79.ln_2.bias": "pytorch_model-00022-of-00022.bin",
+    "transformer.h.79.ln_2.weight": "pytorch_model-00022-of-00022.bin",
+    "transformer.h.79.mlp.c_fc.bias": "pytorch_model-00022-of-00022.bin",
+    "transformer.h.79.mlp.c_fc.weight": "pytorch_model-00022-of-00022.bin",
+    "transformer.h.79.mlp.c_proj.bias": "pytorch_model-00022-of-00022.bin",
+    "transformer.h.79.mlp.c_proj.weight": "pytorch_model-00022-of-00022.bin",
+    "transformer.h.8.attn.bias": "pytorch_model-00003-of-00022.bin",
+    "transformer.h.8.attn.c_attn.bias": "pytorch_model-00003-of-00022.bin",
+    "transformer.h.8.attn.c_attn.weight": "pytorch_model-00003-of-00022.bin",
+    "transformer.h.8.attn.c_proj.bias": "pytorch_model-00003-of-00022.bin",
+    "transformer.h.8.attn.c_proj.weight": "pytorch_model-00003-of-00022.bin",
+    "transformer.h.8.attn.masked_bias": "pytorch_model-00003-of-00022.bin",
+    "transformer.h.8.ln_1.bias": "pytorch_model-00003-of-00022.bin",
+    "transformer.h.8.ln_1.weight": "pytorch_model-00003-of-00022.bin",
+    "transformer.h.8.ln_2.bias": "pytorch_model-00003-of-00022.bin",
+    "transformer.h.8.ln_2.weight": "pytorch_model-00003-of-00022.bin",
+    "transformer.h.8.mlp.c_fc.bias": "pytorch_model-00003-of-00022.bin",
+    "transformer.h.8.mlp.c_fc.weight": "pytorch_model-00003-of-00022.bin",
+    "transformer.h.8.mlp.c_proj.bias": "pytorch_model-00003-of-00022.bin",
+    "transformer.h.8.mlp.c_proj.weight": "pytorch_model-00003-of-00022.bin",
+    "transformer.h.9.attn.bias": "pytorch_model-00003-of-00022.bin",
+    "transformer.h.9.attn.c_attn.bias": "pytorch_model-00003-of-00022.bin",
+    "transformer.h.9.attn.c_attn.weight": "pytorch_model-00003-of-00022.bin",
+    "transformer.h.9.attn.c_proj.bias": "pytorch_model-00003-of-00022.bin",
+    "transformer.h.9.attn.c_proj.weight": "pytorch_model-00003-of-00022.bin",
+    "transformer.h.9.attn.masked_bias": "pytorch_model-00003-of-00022.bin",
+    "transformer.h.9.ln_1.bias": "pytorch_model-00003-of-00022.bin",
+    "transformer.h.9.ln_1.weight": "pytorch_model-00003-of-00022.bin",
+    "transformer.h.9.ln_2.bias": "pytorch_model-00003-of-00022.bin",
+    "transformer.h.9.ln_2.weight": "pytorch_model-00003-of-00022.bin",
+    "transformer.h.9.mlp.c_fc.bias": "pytorch_model-00003-of-00022.bin",
+    "transformer.h.9.mlp.c_fc.weight": "pytorch_model-00003-of-00022.bin",
+    "transformer.h.9.mlp.c_proj.bias": "pytorch_model-00003-of-00022.bin",
+    "transformer.h.9.mlp.c_proj.weight": "pytorch_model-00003-of-00022.bin",
+    "transformer.ln_f.bias": "pytorch_model-00022-of-00022.bin",
+    "transformer.ln_f.weight": "pytorch_model-00022-of-00022.bin",
+    "transformer.wte.weight": "pytorch_model-00001-of-00022.bin"
+  }
+}

special_tokens_map.json

@@ -0,0 +1,29 @@
+{
+  "additional_special_tokens": [
+    "[tsk]"
+  ],
+  "bos_token": {
+    "content": "<|endoftext|>",
+    "lstrip": false,
+    "normalized": true,
+    "rstrip": false,
+    "single_word": false
+  },
+  "cls_token": "[cls]",
+  "eos_token": {
+    "content": "<|endoftext|>",
+    "lstrip": false,
+    "normalized": true,
+    "rstrip": false,
+    "single_word": false
+  },
+  "pad_token": "[pad]",
+  "sep_token": "[sep]",
+  "unk_token": {
+    "content": "<|endoftext|>",
+    "lstrip": false,
+    "normalized": true,
+    "rstrip": false,
+    "single_word": false
+  }
+}

tokenization_freelm.py

@@ -0,0 +1,30 @@
+from transformers import GPT2Tokenizer
+
+class FLMTokenizer(GPT2Tokenizer):
+    model_input_names = ["input_ids", "attention_mask"]
+    def __init__(
+        self,
+        vocab_file,
+        merges_file,
+        errors="replace",
+        unk_token="<|endoftext|>",
+        bos_token="<|endoftext|>",
+        eos_token="<|endoftext|>",
+        pad_token=None,
+        add_prefix_space=False,
+        add_bos_token=False,
+        **kwargs,
+    ):
+        super().__init__(
+            vocab_file,
+            merges_file,
+            errors=errors,
+            unk_token=unk_token,
+            bos_token=bos_token,
+            eos_token=eos_token,
+            pad_token=pad_token,
+            add_prefix_space=add_prefix_space,
+            add_bos_token=add_bos_token,
+            **kwargs,
+        )
+        self.pat = r"""(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\r\n\p{L}\p{N}]?\p{L}+|\p{N}{1,3}| ?[^\s\p{L}\p{N}]+[\r\n]*|\s*[\r\n]+|\s+(?!\S)|\s+"""

tokenizer_config.json

@@ -0,0 +1,41 @@
+{
+  "add_bos_token": false,
+  "add_prefix_space": false,
+  "auto_map": {
+    "AutoTokenizer": [
+      "tokenization_freelm.FLMTokenizer",
+      null
+    ]
+  },
+  "bos_token": {
+    "__type": "AddedToken",
+    "content": "<|endoftext|>",
+    "lstrip": false,
+    "normalized": true,
+    "rstrip": false,
+    "single_word": false
+  },
+  "clean_up_tokenization_spaces": false,
+  "do_lower_case": false,
+  "eos_token": {
+    "__type": "AddedToken",
+    "content": "<|endoftext|>",
+    "lstrip": false,
+    "normalized": true,
+    "rstrip": false,
+    "single_word": false
+  },
+  "errors": "replace",
+  "model_max_length": 1000000000000000019884624838656,
+  "pad_token": null,
+  "tokenizer_class": "FLMTokenizer",
+  "unk_token": {
+    "__type": "AddedToken",
+    "content": "<|endoftext|>",
+    "lstrip": false,
+    "normalized": true,
+    "rstrip": false,
+    "single_word": false
+  },
+  "use_fast": false
+}