diff --git "a/flagged/modeling_moss.py" "b/flagged/modeling_moss.py"
new file mode 100644--- /dev/null
+++ "b/flagged/modeling_moss.py"
@@ -0,0 +1,2952 @@
+""" PyTorch Moss model."""
+
+from typing import Optional, Tuple, Union
+
+import torch
+import torch.utils.checkpoint
+from torch import nn
+from torch.nn import CrossEntropyLoss
+import transformers
+from transformers.activations import ACT2FN
+from transformers.modeling_utils import PreTrainedModel
+from transformers.modeling_outputs import BaseModelOutputWithPast, CausalLMOutputWithPast
+from transformers.utils import (
+    add_code_sample_docstrings, 
+    add_start_docstrings, 
+    add_start_docstrings_to_model_forward, 
+    logging
+)
+
+from .configuration_moss import MossConfig
+
+logger = logging.get_logger(__name__)
+
+_CHECKPOINT_FOR_DOC = "fnlp/moss-moon-003-base"
+_CONFIG_FOR_DOC = "MossConfig"
+
+
+MOSS_PRETRAINED_MODEL_ARCHIVE_LIST = [
+    "fnlp/moss-moon-003-base",
+    "fnlp/moss-moon-003-sft",
+    "fnlp/moss-moon-003-sft-plugin",
+    "fnlp/moss-moon-003-sft-int4",
+    "fnlp/moss-moon-003-sft-plugin-int4",
+    "fnlp/moss-moon-003-sft-int8",
+    "fnlp/moss-moon-003-sft-plugin-int8",
+]
+
+
+# Copied from transformers.models.gptj.modeling_gptj.create_sinusoidal_positions
+def create_sinusoidal_positions(num_pos: int, dim: int) -> torch.Tensor:
+    inv_freq = 1.0 / (10000 ** (torch.arange(0, dim, 2) / dim))
+    sinusoid_inp = torch.einsum("i , j -> i j", torch.arange(num_pos, dtype=torch.float), inv_freq).float()
+    return torch.cat((torch.sin(sinusoid_inp), torch.cos(sinusoid_inp)), dim=1)
+
+
+# Copied from transformers.models.gptj.modeling_gptj.rotate_every_two
+def rotate_every_two(x: torch.Tensor) -> torch.Tensor:
+    x1 = x[:, :, :, ::2]
+    x2 = x[:, :, :, 1::2]
+    x = torch.stack((-x2, x1), dim=-1)
+    return x.flatten(-2)  # in einsum notation: rearrange(x, '... d j -> ... (d j)')
+
+
+# Copied from transformers.models.gptj.modeling_gptj.apply_rotary_pos_emb
+def apply_rotary_pos_emb(tensor: torch.Tensor, sin: torch.Tensor, cos: torch.Tensor) -> torch.Tensor:
+    sin = torch.repeat_interleave(sin[:, :, None, :], 2, 3)
+    cos = torch.repeat_interleave(cos[:, :, None, :], 2, 3)
+    return (tensor * cos) + (rotate_every_two(tensor) * sin)
+
+
+class MossAttention(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+
+        max_positions = config.max_position_embeddings
+        self.register_buffer(
+            "causal_mask",
+            torch.tril(torch.ones((max_positions, max_positions), dtype=torch.bool)).view(
+                1, 1, max_positions, max_positions
+            ),
+        )
+
+        self.attn_dropout = nn.Dropout(config.attn_pdrop)
+        self.resid_dropout = nn.Dropout(config.resid_pdrop)
+
+        self.embed_dim = config.hidden_size
+        self.num_attention_heads = config.num_attention_heads
+        self.head_dim = self.embed_dim // self.num_attention_heads
+        if self.head_dim * self.num_attention_heads != self.embed_dim:
+            raise ValueError(
+                f"embed_dim must be divisible by num_attention_heads (got `embed_dim`: {self.embed_dim} and"
+                f" `num_attention_heads`: {self.num_attention_heads})."
+            )
+        self.scale_attn = torch.sqrt(torch.tensor(self.head_dim, dtype=torch.float32)).to(torch.get_default_dtype())
+        self.qkv_proj = nn.Linear(self.embed_dim, self.embed_dim * 3, bias=False)
+
+        self.out_proj = nn.Linear(self.embed_dim, self.embed_dim, bias=False)
+        self.rotary_dim = config.rotary_dim
+        pos_embd_dim = self.rotary_dim or self.embed_dim
+        self.embed_positions = create_sinusoidal_positions(max_positions, pos_embd_dim)
+
+    def _split_heads(self, x, n_head, dim_head, mp_num):
+        reshaped = x.reshape(x.shape[:-1] + (n_head // mp_num, dim_head))
+        reshaped = reshaped.reshape(x.shape[:-2] + (-1,) + reshaped.shape[-1:])
+        return reshaped
+
+    def _merge_heads(self, tensor, num_attention_heads, attn_head_size):
+        """
+        Merges attn_head_size dim and num_attn_heads dim into n_ctx
+        """
+        if len(tensor.shape) == 5:
+            tensor = tensor.permute(0, 1, 3, 2, 4).contiguous()
+        elif len(tensor.shape) == 4:
+            tensor = tensor.permute(0, 2, 1, 3).contiguous()
+        else:
+            raise ValueError(f"Input tensor rank should be one of [4, 5], but is: {len(tensor.shape)}")
+        new_shape = tensor.size()[:-2] + (num_attention_heads * attn_head_size,)
+        return tensor.view(new_shape)
+
+    def _attn(
+        self,
+        query,
+        key,
+        value,
+        attention_mask=None,
+        head_mask=None,
+    ):
+        # compute causal mask from causal mask buffer
+        query_length, key_length = query.size(-2), key.size(-2)
+        causal_mask = self.causal_mask[:, :, key_length - query_length : key_length, :key_length]
+
+        # Keep the attention weights computation in fp32 to avoid overflow issues
+        query = query.to(torch.float32)
+        key = key.to(torch.float32)
+
+        attn_weights = torch.matmul(query, key.transpose(-1, -2))
+
+        attn_weights = attn_weights / self.scale_attn
+        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.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.Softmax(dim=-1)(attn_weights)
+        attn_weights = attn_weights.to(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 forward(
+        self,
+        hidden_states: Optional[torch.FloatTensor],
+        layer_past: Optional[Tuple[torch.Tensor]] = None,
+        attention_mask: Optional[torch.FloatTensor] = None,
+        position_ids: Optional[torch.LongTensor] = None,
+        head_mask: Optional[torch.FloatTensor] = None,
+        use_cache: Optional[bool] = False,
+        output_attentions: Optional[bool] = False,
+    ) -> Union[
+        Tuple[torch.Tensor, Tuple[torch.Tensor]],
+        Optional[Tuple[torch.Tensor, Tuple[torch.Tensor], Tuple[torch.Tensor, ...]]],
+    ]:
+        qkv = self.qkv_proj(hidden_states)
+        # TODO(enijkamp): factor out number of logical TPU-v4 cores or make forward pass agnostic
+        mp_num = 4
+        qkv_split = qkv.reshape(qkv.shape[:-1] + (mp_num, -1))
+
+        local_dim = self.head_dim * self.num_attention_heads // mp_num
+        query, value, key = torch.split(qkv_split, local_dim, dim=-1)
+        query = self._split_heads(query, self.num_attention_heads, self.head_dim, mp_num=mp_num)
+        key = self._split_heads(key, self.num_attention_heads, self.head_dim, mp_num=mp_num)
+
+        value = self._split_heads(value, self.num_attention_heads, self.head_dim, mp_num=mp_num)
+        value = value.permute(0, 2, 1, 3)
+
+        embed_positions = self.embed_positions
+        if embed_positions.device != position_ids.device:
+            embed_positions = embed_positions.to(position_ids.device)
+            self.embed_positions = embed_positions
+
+        sincos = embed_positions[position_ids]
+        sin, cos = torch.split(sincos, sincos.shape[-1] // 2, dim=-1)
+
+        if self.rotary_dim is not None:
+            k_rot = key[:, :, :, : self.rotary_dim]
+            k_pass = key[:, :, :, self.rotary_dim :]
+
+            q_rot = query[:, :, :, : self.rotary_dim]
+            q_pass = query[:, :, :, self.rotary_dim :]
+
+            k_rot = apply_rotary_pos_emb(k_rot, sin, cos)
+            q_rot = apply_rotary_pos_emb(q_rot, sin, cos)
+
+            key = torch.cat([k_rot, k_pass], dim=-1)
+            query = torch.cat([q_rot, q_pass], dim=-1)
+        else:
+            key = apply_rotary_pos_emb(key, sin, cos)
+            query = apply_rotary_pos_emb(query, sin, cos)
+
+        key = key.permute(0, 2, 1, 3)
+        query = query.permute(0, 2, 1, 3)
+
+        if layer_past is not None:
+            past_key = layer_past[0]
+            past_value = layer_past[1]
+            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
+
+        # compute self-attention: V x Softmax(QK^T)
+        attn_output, attn_weights = self._attn(query, key, value, attention_mask, head_mask)
+
+        attn_output = self._merge_heads(attn_output, self.num_attention_heads, self.head_dim)
+        attn_output = self.out_proj(attn_output)
+        attn_output = self.resid_dropout(attn_output)
+
+        outputs = (attn_output, present)
+        if output_attentions:
+            outputs += (attn_weights,)
+
+        return outputs  # a, present, (attentions)
+
+
+# Copied from transformers.models.gptj.modeling_gptj.GPTJMLP with GPTJ->Moss
+class MossMLP(nn.Module):
+    def __init__(self, intermediate_size, config):  # in MLP: intermediate_size= 4 * embed_dim
+        super().__init__()
+        embed_dim = config.n_embd
+
+        self.fc_in = nn.Linear(embed_dim, intermediate_size)
+        self.fc_out = nn.Linear(intermediate_size, embed_dim)
+
+        self.act = ACT2FN[config.activation_function]
+        self.dropout = nn.Dropout(config.resid_pdrop)
+
+    def forward(self, hidden_states: Optional[torch.FloatTensor]) -> torch.FloatTensor:
+        hidden_states = self.fc_in(hidden_states)
+        hidden_states = self.act(hidden_states)
+        hidden_states = self.fc_out(hidden_states)
+        hidden_states = self.dropout(hidden_states)
+        return hidden_states
+
+
+# Copied from transformers.models.gptj.modeling_gptj.GPTJBlock with GPTJ->Moss
+class MossBlock(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        inner_dim = config.n_inner if config.n_inner is not None else 4 * config.n_embd
+        self.ln_1 = nn.LayerNorm(config.n_embd, eps=config.layer_norm_epsilon)
+        self.attn = MossAttention(config)
+        self.mlp = MossMLP(inner_dim, config)
+
+    def forward(
+        self,
+        hidden_states: Optional[torch.FloatTensor],
+        layer_past: Optional[Tuple[torch.Tensor]] = None,
+        attention_mask: Optional[torch.FloatTensor] = None,
+        position_ids: Optional[torch.LongTensor] = None,
+        head_mask: Optional[torch.FloatTensor] = 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=hidden_states,
+            layer_past=layer_past,
+            attention_mask=attention_mask,
+            position_ids=position_ids,
+            head_mask=head_mask,
+            use_cache=use_cache,
+            output_attentions=output_attentions,
+        )
+        attn_output = attn_outputs[0]  # output_attn: a, present, (attentions)
+        outputs = attn_outputs[1:]
+
+        feed_forward_hidden_states = self.mlp(hidden_states)
+        hidden_states = attn_output + feed_forward_hidden_states + residual
+
+        if use_cache:
+            outputs = (hidden_states,) + outputs
+        else:
+            outputs = (hidden_states,) + outputs[1:]
+
+        return outputs  # hidden_states, present, (attentions)
+
+
+class MossPreTrainedModel(PreTrainedModel):
+    """
+    An abstract class to handle weights initialization and a simple interface for downloading and loading pretrained
+    models.
+    """
+
+    config_class = MossConfig
+    base_model_prefix = "transformer"
+    supports_gradient_checkpointing = True
+    _no_split_modules = ["MossBlock"]
+
+    def __init__(self, *inputs, **kwargs):
+        super().__init__(*inputs, **kwargs)
+
+    def _init_weights(self, module):
+        """Initialize the weights."""
+        if isinstance(module, (nn.Linear,)):
+            # Slightly different from Mesh Transformer JAX 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)
+
+    def _set_gradient_checkpointing(self, module, value=False):
+        if isinstance(module, MossModel):
+            module.gradient_checkpointing = value
+
+
+MOSS_START_DOCSTRING = r"""
+    This model is a PyTorch [torch.nn.Module](https://pytorch.org/docs/stable/nn.html#torch.nn.Module) sub-class. Use
+    it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage and
+    behavior.
+
+    Parameters:
+        config ([`MossConfig`]): Model configuration class with all the parameters of the model.
+            Initializing with a config file does not load the weights associated with the model, only the
+            configuration. Check out the [`~PreTrainedModel.from_pretrained`] method to load the model weights.
+"""
+
+MOSS_INPUTS_DOCSTRING = r"""
+    Args:
+        input_ids (`torch.LongTensor` of shape `({0})`):
+            Indices of input sequence tokens in the vocabulary.
+
+            Indices can be obtained using [`AutoProcenizer`]. See [`PreTrainedTokenizer.encode`] and
+            [`PreTrainedTokenizer.__call__`] for details.
+
+            [What are input IDs?](../glossary#input-ids)
+        attention_mask (`torch.FloatTensor` of shape `({0})`, *optional*):
+            Mask to avoid performing attention on padding token indices. Mask values selected in `[0, 1]`:
+
+            - 1 for tokens that are **not masked**,
+            - 0 for tokens that are **masked**.
+
+            [What are attention masks?](../glossary#attention-mask)
+        token_type_ids (`torch.LongTensor` of shape `({0})`, *optional*):
+            Segment token indices to indicate first and second portions of the inputs. Indices are selected in `[0,
+            1]`:
+
+            - 0 corresponds to a *sentence A* token,
+            - 1 corresponds to a *sentence B* token.
+
+            [What are token type IDs?](../glossary#token-type-ids)
+        position_ids (`torch.LongTensor` of shape `({0})`, *optional*):
+            Indices of positions of each input sequence tokens in the position embeddings. Selected in the range `[0,
+            config.n_positions - 1]`.
+
+            [What are position IDs?](../glossary#position-ids)
+        head_mask (`torch.FloatTensor` of shape `(num_attention_heads,)` or `(n_layer, num_attention_heads)`, *optional*):
+            Mask to nullify selected heads of the self-attention modules. Mask values selected in `[0, 1]`:
+
+            - 1 indicates the head is **not masked**,
+            - 0 indicates the head is **masked**.
+
+        inputs_embeds (`torch.FloatTensor` of shape `({0}, hidden_dim)`, *optional*):
+            Optionally, instead of passing `input_ids` you can choose to directly pass an embedded representation. This
+            is useful if you want more control over how to convert *input_ids* indices into associated vectors than the
+            model's internal embedding lookup matrix.
+        output_attentions (`bool`, *optional*):
+            Whether or not to return the attentions tensors of all attention layers. See `attentions` under returned
+            tensors for more detail.
+        output_hidden_states (`bool`, *optional*):
+            Whether or not to return the hidden states of all layers. See `hidden_states` under returned tensors for
+            more detail.
+        return_dict (`bool`, *optional*):
+            Whether or not to return a [`~utils.ModelOutput`] instead of a plain tuple.
+"""
+
+
+@add_start_docstrings(
+    "The bare Moss Model transformer outputting raw hidden-states without any specific head on top.",
+    MOSS_START_DOCSTRING,
+)
+class MossModel(MossPreTrainedModel):
+    def __init__(self, config):
+        super().__init__(config)
+
+        self.embed_dim = config.n_embd
+        self.vocab_size = config.vocab_size
+        self.wte = nn.Embedding(config.vocab_size, self.embed_dim)
+        self.drop = nn.Dropout(config.embd_pdrop)
+        self.h = nn.ModuleList([MossBlock(config) for _ in range(config.n_layer)])
+        self.ln_f = nn.LayerNorm(self.embed_dim, eps=config.layer_norm_epsilon)
+        self.rotary_dim = min(config.rotary_dim, config.n_ctx // config.num_attention_heads)
+
+        self.gradient_checkpointing = False
+
+        # Initialize weights and apply final processing
+        self.post_init()
+
+    def get_input_embeddings(self):
+        return self.wte
+
+    def set_input_embeddings(self, new_embeddings):
+        self.wte = new_embeddings
+
+    @add_start_docstrings_to_model_forward(MOSS_INPUTS_DOCSTRING.format("batch_size, sequence_length"))
+    @add_code_sample_docstrings(
+        checkpoint=_CHECKPOINT_FOR_DOC,
+        output_type=BaseModelOutputWithPast,
+        config_class=_CONFIG_FOR_DOC,
+    )
+    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,
+        use_cache: Optional[bool] = None,
+        output_attentions: Optional[bool] = None,
+        output_hidden_states: Optional[bool] = None,
+        return_dict: Optional[bool] = None,
+    ) -> Union[Tuple, BaseModelOutputWithPast]:
+        output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
+        output_hidden_states = (
+            output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
+        )
+        use_cache = use_cache if use_cache is not None else self.config.use_cache
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+        if input_ids is 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]).long()
+
+        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])
+
+        # Attention 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
+
+        # Prepare head mask if needed
+        # 1.0 in head_mask indicate we keep the head
+        # attention_probs has shape bsz x num_attention_heads x N x N
+        # head_mask has shape n_layer x batch x num_attention_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)
+
+        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),)
+
+        if self.gradient_checkpointing and self.training:
+            if use_cache:
+                logger.warning_once(
+                    "`use_cache=True` is incompatible with `config.gradient_checkpointing=True`. Setting "
+                    "`use_cache=False`..."
+                )
+                use_cache = False
+
+        presents = () if use_cache else None
+        all_self_attentions = () if output_attentions else None
+        all_hidden_states = () if output_hidden_states else None
+        for i, (block, layer_past) in enumerate(zip(self.h, past_key_values)):
+            if output_hidden_states:
+                all_hidden_states = all_hidden_states + (hidden_states,)
+
+            if self.gradient_checkpointing and self.training:
+
+                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,
+                    position_ids,
+                    head_mask[i],
+                )
+            else:
+                outputs = block(
+                    hidden_states=hidden_states,
+                    layer_past=layer_past,
+                    attention_mask=attention_mask,
+                    position_ids=position_ids,
+                    head_mask=head_mask[i],
+                    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],)
+
+        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] if v is not None)
+
+        return BaseModelOutputWithPast(
+            last_hidden_state=hidden_states,
+            past_key_values=presents,
+            hidden_states=all_hidden_states,
+            attentions=all_self_attentions,
+        )
+
+
+@add_start_docstrings(
+    """
+    The Moss Model transformer with a language modeling head on top.
+    """,
+    MOSS_START_DOCSTRING,
+)
+class MossForCausalLM(MossPreTrainedModel):
+    _keys_to_ignore_on_load_missing = [r"h\.\d+\.attn\.causal_mask"]
+
+    def __init__(self, config):
+        super().__init__(config)
+        if not hasattr(config, 'wbits'):
+            config.wbits = 32
+            config.groupsize = 128
+            
+        if config.wbits not in [4, 8, 32]:
+            logger.warning(f'Specify `wbits` with 4, 8 or 32 to load the model. ')
+        if config.wbits in [4, 8]:
+            def noop(*args, **kwargs):
+                pass
+            torch.nn.init.kaiming_uniform_ = noop
+            torch.nn.init.uniform_ = noop
+            torch.nn.init.normal_ = noop
+
+            torch.set_default_dtype(torch.half)
+            transformers.modeling_utils._init_weights = False
+            torch.set_default_dtype(torch.half)
+        self.transformer = MossModel(config)
+        self.lm_head = nn.Linear(config.n_embd, config.vocab_size)
+        if config.wbits in [4, 8]:
+            torch.set_default_dtype(torch.float)
+            transformers.modeling_utils._init_weights = True
+            self.quantize(config.wbits, config.groupsize)
+        # Initialize weights and apply final processing
+        self.post_init()
+
+    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_key_values=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_key_values:
+            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_key_values:
+                position_ids = position_ids[:, -1].unsqueeze(-1)
+
+        return {
+            "input_ids": input_ids,
+            "past_key_values": past_key_values,
+            "use_cache": kwargs.get("use_cache"),
+            "position_ids": position_ids,
+            "attention_mask": attention_mask,
+            "token_type_ids": token_type_ids,
+        }
+
+    @add_start_docstrings_to_model_forward(MOSS_INPUTS_DOCSTRING.format("batch_size, sequence_length"))
+    @add_code_sample_docstrings(
+        checkpoint=_CHECKPOINT_FOR_DOC,
+        output_type=CausalLMOutputWithPast,
+        config_class=_CONFIG_FOR_DOC,
+    )
+    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,
+        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, CausalLMOutputWithPast]:
+        r"""
+        labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
+            Labels for language modeling. Note that the labels **are shifted** inside the model, i.e. you can set
+            `labels = input_ids` Indices are selected in `[-100, 0, ..., config.vocab_size]` All labels set to `-100`
+            are ignored (masked), the loss is only computed for labels in `[0, ..., config.vocab_size]`
+        """
+        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,
+            use_cache=use_cache,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+            return_dict=return_dict,
+        )
+        hidden_states = transformer_outputs[0]
+
+        # make sure sampling in fp16 works correctly and
+        # compute loss in fp32 to match with mesh-tf version
+        # https://github.com/EleutherAI/gpt-neo/blob/89ce74164da2fb16179106f54e2269b5da8db333/models/gpt2/gpt2.py#L179
+        lm_logits = self.lm_head(hidden_states).to(torch.float32)
+
+        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 = CrossEntropyLoss()
+            loss = loss_fct(shift_logits.view(-1, shift_logits.size(-1)), shift_labels.view(-1))
+
+            loss = loss.to(hidden_states.dtype)
+
+        if not return_dict:
+            output = (lm_logits,) + transformer_outputs[1:]
+            return ((loss,) + output) if loss is not None else output
+
+        return CausalLMOutputWithPast(
+            loss=loss,
+            logits=lm_logits,
+            past_key_values=transformer_outputs.past_key_values,
+            hidden_states=transformer_outputs.hidden_states,
+            attentions=transformer_outputs.attentions,
+        )
+
+    @staticmethod
+    def _reorder_cache(
+        past_key_values: Tuple[Tuple[torch.Tensor]], beam_idx: torch.Tensor
+    ) -> Tuple[Tuple[torch.Tensor]]:
+        """
+        This function is used to re-order the `past_key_values` cache if [`~PretrainedModel.beam_search`] or
+        [`~PretrainedModel.beam_sample`] is called. This is required to match `past_key_values` with the correct
+        beam_idx at every generation step.
+        """
+        return tuple(
+            tuple(past_state.index_select(0, beam_idx.to(past_state.device)) for past_state in layer_past)
+            for layer_past in past_key_values
+        )
+
+    def quantize(self, wbits, groupsize):
+        from .quantization import quantize_with_gptq
+        return quantize_with_gptq(self, wbits, groupsize)
+
+""" PyTorch Moss model."""
+
+from typing import Optional, Tuple, Union
+
+import torch
+import torch.utils.checkpoint
+from torch import nn
+from torch.nn import CrossEntropyLoss
+import transformers
+from transformers.activations import ACT2FN
+from transformers.modeling_utils import PreTrainedModel
+from transformers.modeling_outputs import BaseModelOutputWithPast, CausalLMOutputWithPast
+from transformers.utils import (
+    add_code_sample_docstrings, 
+    add_start_docstrings, 
+    add_start_docstrings_to_model_forward, 
+    logging
+)
+
+from .configuration_moss import MossConfig
+
+logger = logging.get_logger(__name__)
+
+_CHECKPOINT_FOR_DOC = "fnlp/moss-moon-003-base"
+_CONFIG_FOR_DOC = "MossConfig"
+
+
+MOSS_PRETRAINED_MODEL_ARCHIVE_LIST = [
+    "fnlp/moss-moon-003-base",
+    "fnlp/moss-moon-003-sft",
+    "fnlp/moss-moon-003-sft-plugin",
+    "fnlp/moss-moon-003-sft-int4",
+    "fnlp/moss-moon-003-sft-plugin-int4",
+    "fnlp/moss-moon-003-sft-int8",
+    "fnlp/moss-moon-003-sft-plugin-int8",
+]
+
+
+# Copied from transformers.models.gptj.modeling_gptj.create_sinusoidal_positions
+def create_sinusoidal_positions(num_pos: int, dim: int) -> torch.Tensor:
+    inv_freq = 1.0 / (10000 ** (torch.arange(0, dim, 2) / dim))
+    sinusoid_inp = torch.einsum("i , j -> i j", torch.arange(num_pos, dtype=torch.float), inv_freq).float()
+    return torch.cat((torch.sin(sinusoid_inp), torch.cos(sinusoid_inp)), dim=1)
+
+
+# Copied from transformers.models.gptj.modeling_gptj.rotate_every_two
+def rotate_every_two(x: torch.Tensor) -> torch.Tensor:
+    x1 = x[:, :, :, ::2]
+    x2 = x[:, :, :, 1::2]
+    x = torch.stack((-x2, x1), dim=-1)
+    return x.flatten(-2)  # in einsum notation: rearrange(x, '... d j -> ... (d j)')
+
+
+# Copied from transformers.models.gptj.modeling_gptj.apply_rotary_pos_emb
+def apply_rotary_pos_emb(tensor: torch.Tensor, sin: torch.Tensor, cos: torch.Tensor) -> torch.Tensor:
+    sin = torch.repeat_interleave(sin[:, :, None, :], 2, 3)
+    cos = torch.repeat_interleave(cos[:, :, None, :], 2, 3)
+    return (tensor * cos) + (rotate_every_two(tensor) * sin)
+
+
+class MossAttention(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+
+        max_positions = config.max_position_embeddings
+        self.register_buffer(
+            "causal_mask",
+            torch.tril(torch.ones((max_positions, max_positions), dtype=torch.bool)).view(
+                1, 1, max_positions, max_positions
+            ),
+        )
+
+        self.attn_dropout = nn.Dropout(config.attn_pdrop)
+        self.resid_dropout = nn.Dropout(config.resid_pdrop)
+
+        self.embed_dim = config.hidden_size
+        self.num_attention_heads = config.num_attention_heads
+        self.head_dim = self.embed_dim // self.num_attention_heads
+        if self.head_dim * self.num_attention_heads != self.embed_dim:
+            raise ValueError(
+                f"embed_dim must be divisible by num_attention_heads (got `embed_dim`: {self.embed_dim} and"
+                f" `num_attention_heads`: {self.num_attention_heads})."
+            )
+        self.scale_attn = torch.sqrt(torch.tensor(self.head_dim, dtype=torch.float32)).to(torch.get_default_dtype())
+        self.qkv_proj = nn.Linear(self.embed_dim, self.embed_dim * 3, bias=False)
+
+        self.out_proj = nn.Linear(self.embed_dim, self.embed_dim, bias=False)
+        self.rotary_dim = config.rotary_dim
+        pos_embd_dim = self.rotary_dim or self.embed_dim
+        self.embed_positions = create_sinusoidal_positions(max_positions, pos_embd_dim)
+
+    def _split_heads(self, x, n_head, dim_head, mp_num):
+        reshaped = x.reshape(x.shape[:-1] + (n_head // mp_num, dim_head))
+        reshaped = reshaped.reshape(x.shape[:-2] + (-1,) + reshaped.shape[-1:])
+        return reshaped
+
+    def _merge_heads(self, tensor, num_attention_heads, attn_head_size):
+        """
+        Merges attn_head_size dim and num_attn_heads dim into n_ctx
+        """
+        if len(tensor.shape) == 5:
+            tensor = tensor.permute(0, 1, 3, 2, 4).contiguous()
+        elif len(tensor.shape) == 4:
+            tensor = tensor.permute(0, 2, 1, 3).contiguous()
+        else:
+            raise ValueError(f"Input tensor rank should be one of [4, 5], but is: {len(tensor.shape)}")
+        new_shape = tensor.size()[:-2] + (num_attention_heads * attn_head_size,)
+        return tensor.view(new_shape)
+
+    def _attn(
+        self,
+        query,
+        key,
+        value,
+        attention_mask=None,
+        head_mask=None,
+    ):
+        # compute causal mask from causal mask buffer
+        query_length, key_length = query.size(-2), key.size(-2)
+        causal_mask = self.causal_mask[:, :, key_length - query_length : key_length, :key_length]
+
+        # Keep the attention weights computation in fp32 to avoid overflow issues
+        query = query.to(torch.float32)
+        key = key.to(torch.float32)
+
+        attn_weights = torch.matmul(query, key.transpose(-1, -2))
+
+        attn_weights = attn_weights / self.scale_attn
+        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.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.Softmax(dim=-1)(attn_weights)
+        attn_weights = attn_weights.to(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 forward(
+        self,
+        hidden_states: Optional[torch.FloatTensor],
+        layer_past: Optional[Tuple[torch.Tensor]] = None,
+        attention_mask: Optional[torch.FloatTensor] = None,
+        position_ids: Optional[torch.LongTensor] = None,
+        head_mask: Optional[torch.FloatTensor] = None,
+        use_cache: Optional[bool] = False,
+        output_attentions: Optional[bool] = False,
+    ) -> Union[
+        Tuple[torch.Tensor, Tuple[torch.Tensor]],
+        Optional[Tuple[torch.Tensor, Tuple[torch.Tensor], Tuple[torch.Tensor, ...]]],
+    ]:
+        qkv = self.qkv_proj(hidden_states)
+        # TODO(enijkamp): factor out number of logical TPU-v4 cores or make forward pass agnostic
+        mp_num = 4
+        qkv_split = qkv.reshape(qkv.shape[:-1] + (mp_num, -1))
+
+        local_dim = self.head_dim * self.num_attention_heads // mp_num
+        query, value, key = torch.split(qkv_split, local_dim, dim=-1)
+        query = self._split_heads(query, self.num_attention_heads, self.head_dim, mp_num=mp_num)
+        key = self._split_heads(key, self.num_attention_heads, self.head_dim, mp_num=mp_num)
+
+        value = self._split_heads(value, self.num_attention_heads, self.head_dim, mp_num=mp_num)
+        value = value.permute(0, 2, 1, 3)
+
+        embed_positions = self.embed_positions
+        if embed_positions.device != position_ids.device:
+            embed_positions = embed_positions.to(position_ids.device)
+            self.embed_positions = embed_positions
+
+        sincos = embed_positions[position_ids]
+        sin, cos = torch.split(sincos, sincos.shape[-1] // 2, dim=-1)
+
+        if self.rotary_dim is not None:
+            k_rot = key[:, :, :, : self.rotary_dim]
+            k_pass = key[:, :, :, self.rotary_dim :]
+
+            q_rot = query[:, :, :, : self.rotary_dim]
+            q_pass = query[:, :, :, self.rotary_dim :]
+
+            k_rot = apply_rotary_pos_emb(k_rot, sin, cos)
+            q_rot = apply_rotary_pos_emb(q_rot, sin, cos)
+
+            key = torch.cat([k_rot, k_pass], dim=-1)
+            query = torch.cat([q_rot, q_pass], dim=-1)
+        else:
+            key = apply_rotary_pos_emb(key, sin, cos)
+            query = apply_rotary_pos_emb(query, sin, cos)
+
+        key = key.permute(0, 2, 1, 3)
+        query = query.permute(0, 2, 1, 3)
+
+        if layer_past is not None:
+            past_key = layer_past[0]
+            past_value = layer_past[1]
+            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
+
+        # compute self-attention: V x Softmax(QK^T)
+        attn_output, attn_weights = self._attn(query, key, value, attention_mask, head_mask)
+
+        attn_output = self._merge_heads(attn_output, self.num_attention_heads, self.head_dim)
+        attn_output = self.out_proj(attn_output)
+        attn_output = self.resid_dropout(attn_output)
+
+        outputs = (attn_output, present)
+        if output_attentions:
+            outputs += (attn_weights,)
+
+        return outputs  # a, present, (attentions)
+
+
+# Copied from transformers.models.gptj.modeling_gptj.GPTJMLP with GPTJ->Moss
+class MossMLP(nn.Module):
+    def __init__(self, intermediate_size, config):  # in MLP: intermediate_size= 4 * embed_dim
+        super().__init__()
+        embed_dim = config.n_embd
+
+        self.fc_in = nn.Linear(embed_dim, intermediate_size)
+        self.fc_out = nn.Linear(intermediate_size, embed_dim)
+
+        self.act = ACT2FN[config.activation_function]
+        self.dropout = nn.Dropout(config.resid_pdrop)
+
+    def forward(self, hidden_states: Optional[torch.FloatTensor]) -> torch.FloatTensor:
+        hidden_states = self.fc_in(hidden_states)
+        hidden_states = self.act(hidden_states)
+        hidden_states = self.fc_out(hidden_states)
+        hidden_states = self.dropout(hidden_states)
+        return hidden_states
+
+
+# Copied from transformers.models.gptj.modeling_gptj.GPTJBlock with GPTJ->Moss
+class MossBlock(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        inner_dim = config.n_inner if config.n_inner is not None else 4 * config.n_embd
+        self.ln_1 = nn.LayerNorm(config.n_embd, eps=config.layer_norm_epsilon)
+        self.attn = MossAttention(config)
+        self.mlp = MossMLP(inner_dim, config)
+
+    def forward(
+        self,
+        hidden_states: Optional[torch.FloatTensor],
+        layer_past: Optional[Tuple[torch.Tensor]] = None,
+        attention_mask: Optional[torch.FloatTensor] = None,
+        position_ids: Optional[torch.LongTensor] = None,
+        head_mask: Optional[torch.FloatTensor] = 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=hidden_states,
+            layer_past=layer_past,
+            attention_mask=attention_mask,
+            position_ids=position_ids,
+            head_mask=head_mask,
+            use_cache=use_cache,
+            output_attentions=output_attentions,
+        )
+        attn_output = attn_outputs[0]  # output_attn: a, present, (attentions)
+        outputs = attn_outputs[1:]
+
+        feed_forward_hidden_states = self.mlp(hidden_states)
+        hidden_states = attn_output + feed_forward_hidden_states + residual
+
+        if use_cache:
+            outputs = (hidden_states,) + outputs
+        else:
+            outputs = (hidden_states,) + outputs[1:]
+
+        return outputs  # hidden_states, present, (attentions)
+
+
+class MossPreTrainedModel(PreTrainedModel):
+    """
+    An abstract class to handle weights initialization and a simple interface for downloading and loading pretrained
+    models.
+    """
+
+    config_class = MossConfig
+    base_model_prefix = "transformer"
+    supports_gradient_checkpointing = True
+    _no_split_modules = ["MossBlock"]
+
+    def __init__(self, *inputs, **kwargs):
+        super().__init__(*inputs, **kwargs)
+
+    def _init_weights(self, module):
+        """Initialize the weights."""
+        if isinstance(module, (nn.Linear,)):
+            # Slightly different from Mesh Transformer JAX 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)
+
+    def _set_gradient_checkpointing(self, module, value=False):
+        if isinstance(module, MossModel):
+            module.gradient_checkpointing = value
+
+
+MOSS_START_DOCSTRING = r"""
+    This model is a PyTorch [torch.nn.Module](https://pytorch.org/docs/stable/nn.html#torch.nn.Module) sub-class. Use
+    it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage and
+    behavior.
+
+    Parameters:
+        config ([`MossConfig`]): Model configuration class with all the parameters of the model.
+            Initializing with a config file does not load the weights associated with the model, only the
+            configuration. Check out the [`~PreTrainedModel.from_pretrained`] method to load the model weights.
+"""
+
+MOSS_INPUTS_DOCSTRING = r"""
+    Args:
+        input_ids (`torch.LongTensor` of shape `({0})`):
+            Indices of input sequence tokens in the vocabulary.
+
+            Indices can be obtained using [`AutoProcenizer`]. See [`PreTrainedTokenizer.encode`] and
+            [`PreTrainedTokenizer.__call__`] for details.
+
+            [What are input IDs?](../glossary#input-ids)
+        attention_mask (`torch.FloatTensor` of shape `({0})`, *optional*):
+            Mask to avoid performing attention on padding token indices. Mask values selected in `[0, 1]`:
+
+            - 1 for tokens that are **not masked**,
+            - 0 for tokens that are **masked**.
+
+            [What are attention masks?](../glossary#attention-mask)
+        token_type_ids (`torch.LongTensor` of shape `({0})`, *optional*):
+            Segment token indices to indicate first and second portions of the inputs. Indices are selected in `[0,
+            1]`:
+
+            - 0 corresponds to a *sentence A* token,
+            - 1 corresponds to a *sentence B* token.
+
+            [What are token type IDs?](../glossary#token-type-ids)
+        position_ids (`torch.LongTensor` of shape `({0})`, *optional*):
+            Indices of positions of each input sequence tokens in the position embeddings. Selected in the range `[0,
+            config.n_positions - 1]`.
+
+            [What are position IDs?](../glossary#position-ids)
+        head_mask (`torch.FloatTensor` of shape `(num_attention_heads,)` or `(n_layer, num_attention_heads)`, *optional*):
+            Mask to nullify selected heads of the self-attention modules. Mask values selected in `[0, 1]`:
+
+            - 1 indicates the head is **not masked**,
+            - 0 indicates the head is **masked**.
+
+        inputs_embeds (`torch.FloatTensor` of shape `({0}, hidden_dim)`, *optional*):
+            Optionally, instead of passing `input_ids` you can choose to directly pass an embedded representation. This
+            is useful if you want more control over how to convert *input_ids* indices into associated vectors than the
+            model's internal embedding lookup matrix.
+        output_attentions (`bool`, *optional*):
+            Whether or not to return the attentions tensors of all attention layers. See `attentions` under returned
+            tensors for more detail.
+        output_hidden_states (`bool`, *optional*):
+            Whether or not to return the hidden states of all layers. See `hidden_states` under returned tensors for
+            more detail.
+        return_dict (`bool`, *optional*):
+            Whether or not to return a [`~utils.ModelOutput`] instead of a plain tuple.
+"""
+
+
+@add_start_docstrings(
+    "The bare Moss Model transformer outputting raw hidden-states without any specific head on top.",
+    MOSS_START_DOCSTRING,
+)
+class MossModel(MossPreTrainedModel):
+    def __init__(self, config):
+        super().__init__(config)
+
+        self.embed_dim = config.n_embd
+        self.vocab_size = config.vocab_size
+        self.wte = nn.Embedding(config.vocab_size, self.embed_dim)
+        self.drop = nn.Dropout(config.embd_pdrop)
+        self.h = nn.ModuleList([MossBlock(config) for _ in range(config.n_layer)])
+        self.ln_f = nn.LayerNorm(self.embed_dim, eps=config.layer_norm_epsilon)
+        self.rotary_dim = min(config.rotary_dim, config.n_ctx // config.num_attention_heads)
+
+        self.gradient_checkpointing = False
+
+        # Initialize weights and apply final processing
+        self.post_init()
+
+    def get_input_embeddings(self):
+        return self.wte
+
+    def set_input_embeddings(self, new_embeddings):
+        self.wte = new_embeddings
+
+    @add_start_docstrings_to_model_forward(MOSS_INPUTS_DOCSTRING.format("batch_size, sequence_length"))
+    @add_code_sample_docstrings(
+        checkpoint=_CHECKPOINT_FOR_DOC,
+        output_type=BaseModelOutputWithPast,
+        config_class=_CONFIG_FOR_DOC,
+    )
+    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,
+        use_cache: Optional[bool] = None,
+        output_attentions: Optional[bool] = None,
+        output_hidden_states: Optional[bool] = None,
+        return_dict: Optional[bool] = None,
+    ) -> Union[Tuple, BaseModelOutputWithPast]:
+        output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
+        output_hidden_states = (
+            output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
+        )
+        use_cache = use_cache if use_cache is not None else self.config.use_cache
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+        if input_ids is 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]).long()
+
+        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])
+
+        # Attention 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
+
+        # Prepare head mask if needed
+        # 1.0 in head_mask indicate we keep the head
+        # attention_probs has shape bsz x num_attention_heads x N x N
+        # head_mask has shape n_layer x batch x num_attention_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)
+
+        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),)
+
+        if self.gradient_checkpointing and self.training:
+            if use_cache:
+                logger.warning_once(
+                    "`use_cache=True` is incompatible with `config.gradient_checkpointing=True`. Setting "
+                    "`use_cache=False`..."
+                )
+                use_cache = False
+
+        presents = () if use_cache else None
+        all_self_attentions = () if output_attentions else None
+        all_hidden_states = () if output_hidden_states else None
+        for i, (block, layer_past) in enumerate(zip(self.h, past_key_values)):
+            if output_hidden_states:
+                all_hidden_states = all_hidden_states + (hidden_states,)
+
+            if self.gradient_checkpointing and self.training:
+
+                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,
+                    position_ids,
+                    head_mask[i],
+                )
+            else:
+                outputs = block(
+                    hidden_states=hidden_states,
+                    layer_past=layer_past,
+                    attention_mask=attention_mask,
+                    position_ids=position_ids,
+                    head_mask=head_mask[i],
+                    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],)
+
+        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] if v is not None)
+
+        return BaseModelOutputWithPast(
+            last_hidden_state=hidden_states,
+            past_key_values=presents,
+            hidden_states=all_hidden_states,
+            attentions=all_self_attentions,
+        )
+
+
+@add_start_docstrings(
+    """
+    The Moss Model transformer with a language modeling head on top.
+    """,
+    MOSS_START_DOCSTRING,
+)
+class MossForCausalLM(MossPreTrainedModel):
+    _keys_to_ignore_on_load_missing = [r"h\.\d+\.attn\.causal_mask"]
+
+    def __init__(self, config):
+        super().__init__(config)
+        if not hasattr(config, 'wbits'):
+            config.wbits = 32
+            config.groupsize = 128
+            
+        if config.wbits not in [4, 8, 32]:
+            logger.warning(f'Specify `wbits` with 4, 8 or 32 to load the model. ')
+        if config.wbits in [4, 8]:
+            def noop(*args, **kwargs):
+                pass
+            torch.nn.init.kaiming_uniform_ = noop
+            torch.nn.init.uniform_ = noop
+            torch.nn.init.normal_ = noop
+
+            torch.set_default_dtype(torch.half)
+            transformers.modeling_utils._init_weights = False
+            torch.set_default_dtype(torch.half)
+        self.transformer = MossModel(config)
+        self.lm_head = nn.Linear(config.n_embd, config.vocab_size)
+        if config.wbits in [4, 8]:
+            torch.set_default_dtype(torch.float)
+            transformers.modeling_utils._init_weights = True
+            self.quantize(config.wbits, config.groupsize)
+        # Initialize weights and apply final processing
+        self.post_init()
+
+    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_key_values=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_key_values:
+            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_key_values:
+                position_ids = position_ids[:, -1].unsqueeze(-1)
+
+        return {
+            "input_ids": input_ids,
+            "past_key_values": past_key_values,
+            "use_cache": kwargs.get("use_cache"),
+            "position_ids": position_ids,
+            "attention_mask": attention_mask,
+            "token_type_ids": token_type_ids,
+        }
+
+    @add_start_docstrings_to_model_forward(MOSS_INPUTS_DOCSTRING.format("batch_size, sequence_length"))
+    @add_code_sample_docstrings(
+        checkpoint=_CHECKPOINT_FOR_DOC,
+        output_type=CausalLMOutputWithPast,
+        config_class=_CONFIG_FOR_DOC,
+    )
+    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,
+        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, CausalLMOutputWithPast]:
+        r"""
+        labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
+            Labels for language modeling. Note that the labels **are shifted** inside the model, i.e. you can set
+            `labels = input_ids` Indices are selected in `[-100, 0, ..., config.vocab_size]` All labels set to `-100`
+            are ignored (masked), the loss is only computed for labels in `[0, ..., config.vocab_size]`
+        """
+        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,
+            use_cache=use_cache,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+            return_dict=return_dict,
+        )
+        hidden_states = transformer_outputs[0]
+
+        # make sure sampling in fp16 works correctly and
+        # compute loss in fp32 to match with mesh-tf version
+        # https://github.com/EleutherAI/gpt-neo/blob/89ce74164da2fb16179106f54e2269b5da8db333/models/gpt2/gpt2.py#L179
+        lm_logits = self.lm_head(hidden_states).to(torch.float32)
+
+        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 = CrossEntropyLoss()
+            loss = loss_fct(shift_logits.view(-1, shift_logits.size(-1)), shift_labels.view(-1))
+
+            loss = loss.to(hidden_states.dtype)
+
+        if not return_dict:
+            output = (lm_logits,) + transformer_outputs[1:]
+            return ((loss,) + output) if loss is not None else output
+
+        return CausalLMOutputWithPast(
+            loss=loss,
+            logits=lm_logits,
+            past_key_values=transformer_outputs.past_key_values,
+            hidden_states=transformer_outputs.hidden_states,
+            attentions=transformer_outputs.attentions,
+        )
+
+    @staticmethod
+    def _reorder_cache(
+        past_key_values: Tuple[Tuple[torch.Tensor]], beam_idx: torch.Tensor
+    ) -> Tuple[Tuple[torch.Tensor]]:
+        """
+        This function is used to re-order the `past_key_values` cache if [`~PretrainedModel.beam_search`] or
+        [`~PretrainedModel.beam_sample`] is called. This is required to match `past_key_values` with the correct
+        beam_idx at every generation step.
+        """
+        return tuple(
+            tuple(past_state.index_select(0, beam_idx.to(past_state.device)) for past_state in layer_past)
+            for layer_past in past_key_values
+        )
+
+    def quantize(self, wbits, groupsize):
+        from .quantization import quantize_with_gptq
+        return quantize_with_gptq(self, wbits, groupsize)
+
+""" PyTorch Moss model."""
+
+from typing import Optional, Tuple, Union
+
+import torch
+import torch.utils.checkpoint
+from torch import nn
+from torch.nn import CrossEntropyLoss
+import transformers
+from transformers.activations import ACT2FN
+from transformers.modeling_utils import PreTrainedModel
+from transformers.modeling_outputs import BaseModelOutputWithPast, CausalLMOutputWithPast
+from transformers.utils import (
+    add_code_sample_docstrings, 
+    add_start_docstrings, 
+    add_start_docstrings_to_model_forward, 
+    logging
+)
+
+from .configuration_moss import MossConfig
+
+logger = logging.get_logger(__name__)
+
+_CHECKPOINT_FOR_DOC = "fnlp/moss-moon-003-base"
+_CONFIG_FOR_DOC = "MossConfig"
+
+
+MOSS_PRETRAINED_MODEL_ARCHIVE_LIST = [
+    "fnlp/moss-moon-003-base",
+    "fnlp/moss-moon-003-sft",
+    "fnlp/moss-moon-003-sft-plugin",
+    "fnlp/moss-moon-003-sft-int4",
+    "fnlp/moss-moon-003-sft-plugin-int4",
+    "fnlp/moss-moon-003-sft-int8",
+    "fnlp/moss-moon-003-sft-plugin-int8",
+]
+
+
+# Copied from transformers.models.gptj.modeling_gptj.create_sinusoidal_positions
+def create_sinusoidal_positions(num_pos: int, dim: int) -> torch.Tensor:
+    inv_freq = 1.0 / (10000 ** (torch.arange(0, dim, 2) / dim))
+    sinusoid_inp = torch.einsum("i , j -> i j", torch.arange(num_pos, dtype=torch.float), inv_freq).float()
+    return torch.cat((torch.sin(sinusoid_inp), torch.cos(sinusoid_inp)), dim=1)
+
+
+# Copied from transformers.models.gptj.modeling_gptj.rotate_every_two
+def rotate_every_two(x: torch.Tensor) -> torch.Tensor:
+    x1 = x[:, :, :, ::2]
+    x2 = x[:, :, :, 1::2]
+    x = torch.stack((-x2, x1), dim=-1)
+    return x.flatten(-2)  # in einsum notation: rearrange(x, '... d j -> ... (d j)')
+
+
+# Copied from transformers.models.gptj.modeling_gptj.apply_rotary_pos_emb
+def apply_rotary_pos_emb(tensor: torch.Tensor, sin: torch.Tensor, cos: torch.Tensor) -> torch.Tensor:
+    sin = torch.repeat_interleave(sin[:, :, None, :], 2, 3)
+    cos = torch.repeat_interleave(cos[:, :, None, :], 2, 3)
+    return (tensor * cos) + (rotate_every_two(tensor) * sin)
+
+
+class MossAttention(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+
+        max_positions = config.max_position_embeddings
+        self.register_buffer(
+            "causal_mask",
+            torch.tril(torch.ones((max_positions, max_positions), dtype=torch.bool)).view(
+                1, 1, max_positions, max_positions
+            ),
+        )
+
+        self.attn_dropout = nn.Dropout(config.attn_pdrop)
+        self.resid_dropout = nn.Dropout(config.resid_pdrop)
+
+        self.embed_dim = config.hidden_size
+        self.num_attention_heads = config.num_attention_heads
+        self.head_dim = self.embed_dim // self.num_attention_heads
+        if self.head_dim * self.num_attention_heads != self.embed_dim:
+            raise ValueError(
+                f"embed_dim must be divisible by num_attention_heads (got `embed_dim`: {self.embed_dim} and"
+                f" `num_attention_heads`: {self.num_attention_heads})."
+            )
+        self.scale_attn = torch.sqrt(torch.tensor(self.head_dim, dtype=torch.float32)).to(torch.get_default_dtype())
+        self.qkv_proj = nn.Linear(self.embed_dim, self.embed_dim * 3, bias=False)
+
+        self.out_proj = nn.Linear(self.embed_dim, self.embed_dim, bias=False)
+        self.rotary_dim = config.rotary_dim
+        pos_embd_dim = self.rotary_dim or self.embed_dim
+        self.embed_positions = create_sinusoidal_positions(max_positions, pos_embd_dim)
+
+    def _split_heads(self, x, n_head, dim_head, mp_num):
+        reshaped = x.reshape(x.shape[:-1] + (n_head // mp_num, dim_head))
+        reshaped = reshaped.reshape(x.shape[:-2] + (-1,) + reshaped.shape[-1:])
+        return reshaped
+
+    def _merge_heads(self, tensor, num_attention_heads, attn_head_size):
+        """
+        Merges attn_head_size dim and num_attn_heads dim into n_ctx
+        """
+        if len(tensor.shape) == 5:
+            tensor = tensor.permute(0, 1, 3, 2, 4).contiguous()
+        elif len(tensor.shape) == 4:
+            tensor = tensor.permute(0, 2, 1, 3).contiguous()
+        else:
+            raise ValueError(f"Input tensor rank should be one of [4, 5], but is: {len(tensor.shape)}")
+        new_shape = tensor.size()[:-2] + (num_attention_heads * attn_head_size,)
+        return tensor.view(new_shape)
+
+    def _attn(
+        self,
+        query,
+        key,
+        value,
+        attention_mask=None,
+        head_mask=None,
+    ):
+        # compute causal mask from causal mask buffer
+        query_length, key_length = query.size(-2), key.size(-2)
+        causal_mask = self.causal_mask[:, :, key_length - query_length : key_length, :key_length]
+
+        # Keep the attention weights computation in fp32 to avoid overflow issues
+        query = query.to(torch.float32)
+        key = key.to(torch.float32)
+
+        attn_weights = torch.matmul(query, key.transpose(-1, -2))
+
+        attn_weights = attn_weights / self.scale_attn
+        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.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.Softmax(dim=-1)(attn_weights)
+        attn_weights = attn_weights.to(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 forward(
+        self,
+        hidden_states: Optional[torch.FloatTensor],
+        layer_past: Optional[Tuple[torch.Tensor]] = None,
+        attention_mask: Optional[torch.FloatTensor] = None,
+        position_ids: Optional[torch.LongTensor] = None,
+        head_mask: Optional[torch.FloatTensor] = None,
+        use_cache: Optional[bool] = False,
+        output_attentions: Optional[bool] = False,
+    ) -> Union[
+        Tuple[torch.Tensor, Tuple[torch.Tensor]],
+        Optional[Tuple[torch.Tensor, Tuple[torch.Tensor], Tuple[torch.Tensor, ...]]],
+    ]:
+        qkv = self.qkv_proj(hidden_states)
+        # TODO(enijkamp): factor out number of logical TPU-v4 cores or make forward pass agnostic
+        mp_num = 4
+        qkv_split = qkv.reshape(qkv.shape[:-1] + (mp_num, -1))
+
+        local_dim = self.head_dim * self.num_attention_heads // mp_num
+        query, value, key = torch.split(qkv_split, local_dim, dim=-1)
+        query = self._split_heads(query, self.num_attention_heads, self.head_dim, mp_num=mp_num)
+        key = self._split_heads(key, self.num_attention_heads, self.head_dim, mp_num=mp_num)
+
+        value = self._split_heads(value, self.num_attention_heads, self.head_dim, mp_num=mp_num)
+        value = value.permute(0, 2, 1, 3)
+
+        embed_positions = self.embed_positions
+        if embed_positions.device != position_ids.device:
+            embed_positions = embed_positions.to(position_ids.device)
+            self.embed_positions = embed_positions
+
+        sincos = embed_positions[position_ids]
+        sin, cos = torch.split(sincos, sincos.shape[-1] // 2, dim=-1)
+
+        if self.rotary_dim is not None:
+            k_rot = key[:, :, :, : self.rotary_dim]
+            k_pass = key[:, :, :, self.rotary_dim :]
+
+            q_rot = query[:, :, :, : self.rotary_dim]
+            q_pass = query[:, :, :, self.rotary_dim :]
+
+            k_rot = apply_rotary_pos_emb(k_rot, sin, cos)
+            q_rot = apply_rotary_pos_emb(q_rot, sin, cos)
+
+            key = torch.cat([k_rot, k_pass], dim=-1)
+            query = torch.cat([q_rot, q_pass], dim=-1)
+        else:
+            key = apply_rotary_pos_emb(key, sin, cos)
+            query = apply_rotary_pos_emb(query, sin, cos)
+
+        key = key.permute(0, 2, 1, 3)
+        query = query.permute(0, 2, 1, 3)
+
+        if layer_past is not None:
+            past_key = layer_past[0]
+            past_value = layer_past[1]
+            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
+
+        # compute self-attention: V x Softmax(QK^T)
+        attn_output, attn_weights = self._attn(query, key, value, attention_mask, head_mask)
+
+        attn_output = self._merge_heads(attn_output, self.num_attention_heads, self.head_dim)
+        attn_output = self.out_proj(attn_output)
+        attn_output = self.resid_dropout(attn_output)
+
+        outputs = (attn_output, present)
+        if output_attentions:
+            outputs += (attn_weights,)
+
+        return outputs  # a, present, (attentions)
+
+
+# Copied from transformers.models.gptj.modeling_gptj.GPTJMLP with GPTJ->Moss
+class MossMLP(nn.Module):
+    def __init__(self, intermediate_size, config):  # in MLP: intermediate_size= 4 * embed_dim
+        super().__init__()
+        embed_dim = config.n_embd
+
+        self.fc_in = nn.Linear(embed_dim, intermediate_size)
+        self.fc_out = nn.Linear(intermediate_size, embed_dim)
+
+        self.act = ACT2FN[config.activation_function]
+        self.dropout = nn.Dropout(config.resid_pdrop)
+
+    def forward(self, hidden_states: Optional[torch.FloatTensor]) -> torch.FloatTensor:
+        hidden_states = self.fc_in(hidden_states)
+        hidden_states = self.act(hidden_states)
+        hidden_states = self.fc_out(hidden_states)
+        hidden_states = self.dropout(hidden_states)
+        return hidden_states
+
+
+# Copied from transformers.models.gptj.modeling_gptj.GPTJBlock with GPTJ->Moss
+class MossBlock(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        inner_dim = config.n_inner if config.n_inner is not None else 4 * config.n_embd
+        self.ln_1 = nn.LayerNorm(config.n_embd, eps=config.layer_norm_epsilon)
+        self.attn = MossAttention(config)
+        self.mlp = MossMLP(inner_dim, config)
+
+    def forward(
+        self,
+        hidden_states: Optional[torch.FloatTensor],
+        layer_past: Optional[Tuple[torch.Tensor]] = None,
+        attention_mask: Optional[torch.FloatTensor] = None,
+        position_ids: Optional[torch.LongTensor] = None,
+        head_mask: Optional[torch.FloatTensor] = 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=hidden_states,
+            layer_past=layer_past,
+            attention_mask=attention_mask,
+            position_ids=position_ids,
+            head_mask=head_mask,
+            use_cache=use_cache,
+            output_attentions=output_attentions,
+        )
+        attn_output = attn_outputs[0]  # output_attn: a, present, (attentions)
+        outputs = attn_outputs[1:]
+
+        feed_forward_hidden_states = self.mlp(hidden_states)
+        hidden_states = attn_output + feed_forward_hidden_states + residual
+
+        if use_cache:
+            outputs = (hidden_states,) + outputs
+        else:
+            outputs = (hidden_states,) + outputs[1:]
+
+        return outputs  # hidden_states, present, (attentions)
+
+
+class MossPreTrainedModel(PreTrainedModel):
+    """
+    An abstract class to handle weights initialization and a simple interface for downloading and loading pretrained
+    models.
+    """
+
+    config_class = MossConfig
+    base_model_prefix = "transformer"
+    supports_gradient_checkpointing = True
+    _no_split_modules = ["MossBlock"]
+
+    def __init__(self, *inputs, **kwargs):
+        super().__init__(*inputs, **kwargs)
+
+    def _init_weights(self, module):
+        """Initialize the weights."""
+        if isinstance(module, (nn.Linear,)):
+            # Slightly different from Mesh Transformer JAX 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)
+
+    def _set_gradient_checkpointing(self, module, value=False):
+        if isinstance(module, MossModel):
+            module.gradient_checkpointing = value
+
+
+MOSS_START_DOCSTRING = r"""
+    This model is a PyTorch [torch.nn.Module](https://pytorch.org/docs/stable/nn.html#torch.nn.Module) sub-class. Use
+    it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage and
+    behavior.
+
+    Parameters:
+        config ([`MossConfig`]): Model configuration class with all the parameters of the model.
+            Initializing with a config file does not load the weights associated with the model, only the
+            configuration. Check out the [`~PreTrainedModel.from_pretrained`] method to load the model weights.
+"""
+
+MOSS_INPUTS_DOCSTRING = r"""
+    Args:
+        input_ids (`torch.LongTensor` of shape `({0})`):
+            Indices of input sequence tokens in the vocabulary.
+
+            Indices can be obtained using [`AutoProcenizer`]. See [`PreTrainedTokenizer.encode`] and
+            [`PreTrainedTokenizer.__call__`] for details.
+
+            [What are input IDs?](../glossary#input-ids)
+        attention_mask (`torch.FloatTensor` of shape `({0})`, *optional*):
+            Mask to avoid performing attention on padding token indices. Mask values selected in `[0, 1]`:
+
+            - 1 for tokens that are **not masked**,
+            - 0 for tokens that are **masked**.
+
+            [What are attention masks?](../glossary#attention-mask)
+        token_type_ids (`torch.LongTensor` of shape `({0})`, *optional*):
+            Segment token indices to indicate first and second portions of the inputs. Indices are selected in `[0,
+            1]`:
+
+            - 0 corresponds to a *sentence A* token,
+            - 1 corresponds to a *sentence B* token.
+
+            [What are token type IDs?](../glossary#token-type-ids)
+        position_ids (`torch.LongTensor` of shape `({0})`, *optional*):
+            Indices of positions of each input sequence tokens in the position embeddings. Selected in the range `[0,
+            config.n_positions - 1]`.
+
+            [What are position IDs?](../glossary#position-ids)
+        head_mask (`torch.FloatTensor` of shape `(num_attention_heads,)` or `(n_layer, num_attention_heads)`, *optional*):
+            Mask to nullify selected heads of the self-attention modules. Mask values selected in `[0, 1]`:
+
+            - 1 indicates the head is **not masked**,
+            - 0 indicates the head is **masked**.
+
+        inputs_embeds (`torch.FloatTensor` of shape `({0}, hidden_dim)`, *optional*):
+            Optionally, instead of passing `input_ids` you can choose to directly pass an embedded representation. This
+            is useful if you want more control over how to convert *input_ids* indices into associated vectors than the
+            model's internal embedding lookup matrix.
+        output_attentions (`bool`, *optional*):
+            Whether or not to return the attentions tensors of all attention layers. See `attentions` under returned
+            tensors for more detail.
+        output_hidden_states (`bool`, *optional*):
+            Whether or not to return the hidden states of all layers. See `hidden_states` under returned tensors for
+            more detail.
+        return_dict (`bool`, *optional*):
+            Whether or not to return a [`~utils.ModelOutput`] instead of a plain tuple.
+"""
+
+
+@add_start_docstrings(
+    "The bare Moss Model transformer outputting raw hidden-states without any specific head on top.",
+    MOSS_START_DOCSTRING,
+)
+class MossModel(MossPreTrainedModel):
+    def __init__(self, config):
+        super().__init__(config)
+
+        self.embed_dim = config.n_embd
+        self.vocab_size = config.vocab_size
+        self.wte = nn.Embedding(config.vocab_size, self.embed_dim)
+        self.drop = nn.Dropout(config.embd_pdrop)
+        self.h = nn.ModuleList([MossBlock(config) for _ in range(config.n_layer)])
+        self.ln_f = nn.LayerNorm(self.embed_dim, eps=config.layer_norm_epsilon)
+        self.rotary_dim = min(config.rotary_dim, config.n_ctx // config.num_attention_heads)
+
+        self.gradient_checkpointing = False
+
+        # Initialize weights and apply final processing
+        self.post_init()
+
+    def get_input_embeddings(self):
+        return self.wte
+
+    def set_input_embeddings(self, new_embeddings):
+        self.wte = new_embeddings
+
+    @add_start_docstrings_to_model_forward(MOSS_INPUTS_DOCSTRING.format("batch_size, sequence_length"))
+    @add_code_sample_docstrings(
+        checkpoint=_CHECKPOINT_FOR_DOC,
+        output_type=BaseModelOutputWithPast,
+        config_class=_CONFIG_FOR_DOC,
+    )
+    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,
+        use_cache: Optional[bool] = None,
+        output_attentions: Optional[bool] = None,
+        output_hidden_states: Optional[bool] = None,
+        return_dict: Optional[bool] = None,
+    ) -> Union[Tuple, BaseModelOutputWithPast]:
+        output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
+        output_hidden_states = (
+            output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
+        )
+        use_cache = use_cache if use_cache is not None else self.config.use_cache
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+        if input_ids is 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]).long()
+
+        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])
+
+        # Attention 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
+
+        # Prepare head mask if needed
+        # 1.0 in head_mask indicate we keep the head
+        # attention_probs has shape bsz x num_attention_heads x N x N
+        # head_mask has shape n_layer x batch x num_attention_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)
+
+        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),)
+
+        if self.gradient_checkpointing and self.training:
+            if use_cache:
+                logger.warning_once(
+                    "`use_cache=True` is incompatible with `config.gradient_checkpointing=True`. Setting "
+                    "`use_cache=False`..."
+                )
+                use_cache = False
+
+        presents = () if use_cache else None
+        all_self_attentions = () if output_attentions else None
+        all_hidden_states = () if output_hidden_states else None
+        for i, (block, layer_past) in enumerate(zip(self.h, past_key_values)):
+            if output_hidden_states:
+                all_hidden_states = all_hidden_states + (hidden_states,)
+
+            if self.gradient_checkpointing and self.training:
+
+                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,
+                    position_ids,
+                    head_mask[i],
+                )
+            else:
+                outputs = block(
+                    hidden_states=hidden_states,
+                    layer_past=layer_past,
+                    attention_mask=attention_mask,
+                    position_ids=position_ids,
+                    head_mask=head_mask[i],
+                    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],)
+
+        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] if v is not None)
+
+        return BaseModelOutputWithPast(
+            last_hidden_state=hidden_states,
+            past_key_values=presents,
+            hidden_states=all_hidden_states,
+            attentions=all_self_attentions,
+        )
+
+
+@add_start_docstrings(
+    """
+    The Moss Model transformer with a language modeling head on top.
+    """,
+    MOSS_START_DOCSTRING,
+)
+class MossForCausalLM(MossPreTrainedModel):
+    _keys_to_ignore_on_load_missing = [r"h\.\d+\.attn\.causal_mask"]
+
+    def __init__(self, config):
+        super().__init__(config)
+        if not hasattr(config, 'wbits'):
+            config.wbits = 32
+            config.groupsize = 128
+            
+        if config.wbits not in [4, 8, 32]:
+            logger.warning(f'Specify `wbits` with 4, 8 or 32 to load the model. ')
+        if config.wbits in [4, 8]:
+            def noop(*args, **kwargs):
+                pass
+            torch.nn.init.kaiming_uniform_ = noop
+            torch.nn.init.uniform_ = noop
+            torch.nn.init.normal_ = noop
+
+            torch.set_default_dtype(torch.half)
+            transformers.modeling_utils._init_weights = False
+            torch.set_default_dtype(torch.half)
+        self.transformer = MossModel(config)
+        self.lm_head = nn.Linear(config.n_embd, config.vocab_size)
+        if config.wbits in [4, 8]:
+            torch.set_default_dtype(torch.float)
+            transformers.modeling_utils._init_weights = True
+            self.quantize(config.wbits, config.groupsize)
+        # Initialize weights and apply final processing
+        self.post_init()
+
+    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_key_values=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_key_values:
+            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_key_values:
+                position_ids = position_ids[:, -1].unsqueeze(-1)
+
+        return {
+            "input_ids": input_ids,
+            "past_key_values": past_key_values,
+            "use_cache": kwargs.get("use_cache"),
+            "position_ids": position_ids,
+            "attention_mask": attention_mask,
+            "token_type_ids": token_type_ids,
+        }
+
+    @add_start_docstrings_to_model_forward(MOSS_INPUTS_DOCSTRING.format("batch_size, sequence_length"))
+    @add_code_sample_docstrings(
+        checkpoint=_CHECKPOINT_FOR_DOC,
+        output_type=CausalLMOutputWithPast,
+        config_class=_CONFIG_FOR_DOC,
+    )
+    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,
+        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, CausalLMOutputWithPast]:
+        r"""
+        labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
+            Labels for language modeling. Note that the labels **are shifted** inside the model, i.e. you can set
+            `labels = input_ids` Indices are selected in `[-100, 0, ..., config.vocab_size]` All labels set to `-100`
+            are ignored (masked), the loss is only computed for labels in `[0, ..., config.vocab_size]`
+        """
+        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,
+            use_cache=use_cache,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+            return_dict=return_dict,
+        )
+        hidden_states = transformer_outputs[0]
+
+        # make sure sampling in fp16 works correctly and
+        # compute loss in fp32 to match with mesh-tf version
+        # https://github.com/EleutherAI/gpt-neo/blob/89ce74164da2fb16179106f54e2269b5da8db333/models/gpt2/gpt2.py#L179
+        lm_logits = self.lm_head(hidden_states).to(torch.float32)
+
+        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 = CrossEntropyLoss()
+            loss = loss_fct(shift_logits.view(-1, shift_logits.size(-1)), shift_labels.view(-1))
+
+            loss = loss.to(hidden_states.dtype)
+
+        if not return_dict:
+            output = (lm_logits,) + transformer_outputs[1:]
+            return ((loss,) + output) if loss is not None else output
+
+        return CausalLMOutputWithPast(
+            loss=loss,
+            logits=lm_logits,
+            past_key_values=transformer_outputs.past_key_values,
+            hidden_states=transformer_outputs.hidden_states,
+            attentions=transformer_outputs.attentions,
+        )
+
+    @staticmethod
+    def _reorder_cache(
+        past_key_values: Tuple[Tuple[torch.Tensor]], beam_idx: torch.Tensor
+    ) -> Tuple[Tuple[torch.Tensor]]:
+        """
+        This function is used to re-order the `past_key_values` cache if [`~PretrainedModel.beam_search`] or
+        [`~PretrainedModel.beam_sample`] is called. This is required to match `past_key_values` with the correct
+        beam_idx at every generation step.
+        """
+        return tuple(
+            tuple(past_state.index_select(0, beam_idx.to(past_state.device)) for past_state in layer_past)
+            for layer_past in past_key_values
+        )
+
+    def quantize(self, wbits, groupsize):
+        from .quantization import quantize_with_gptq
+        return quantize_with_gptq(self, wbits, groupsize)
+
+""" PyTorch Moss model."""
+
+from typing import Optional, Tuple, Union
+
+import torch
+import torch.utils.checkpoint
+from torch import nn
+from torch.nn import CrossEntropyLoss
+import transformers
+from transformers.activations import ACT2FN
+from transformers.modeling_utils import PreTrainedModel
+from transformers.modeling_outputs import BaseModelOutputWithPast, CausalLMOutputWithPast
+from transformers.utils import (
+    add_code_sample_docstrings, 
+    add_start_docstrings, 
+    add_start_docstrings_to_model_forward, 
+    logging
+)
+
+from .configuration_moss import MossConfig
+
+logger = logging.get_logger(__name__)
+
+_CHECKPOINT_FOR_DOC = "fnlp/moss-moon-003-base"
+_CONFIG_FOR_DOC = "MossConfig"
+
+
+MOSS_PRETRAINED_MODEL_ARCHIVE_LIST = [
+    "fnlp/moss-moon-003-base",
+    "fnlp/moss-moon-003-sft",
+    "fnlp/moss-moon-003-sft-plugin",
+    "fnlp/moss-moon-003-sft-int4",
+    "fnlp/moss-moon-003-sft-plugin-int4",
+    "fnlp/moss-moon-003-sft-int8",
+    "fnlp/moss-moon-003-sft-plugin-int8",
+]
+
+
+# Copied from transformers.models.gptj.modeling_gptj.create_sinusoidal_positions
+def create_sinusoidal_positions(num_pos: int, dim: int) -> torch.Tensor:
+    inv_freq = 1.0 / (10000 ** (torch.arange(0, dim, 2) / dim))
+    sinusoid_inp = torch.einsum("i , j -> i j", torch.arange(num_pos, dtype=torch.float), inv_freq).float()
+    return torch.cat((torch.sin(sinusoid_inp), torch.cos(sinusoid_inp)), dim=1)
+
+
+# Copied from transformers.models.gptj.modeling_gptj.rotate_every_two
+def rotate_every_two(x: torch.Tensor) -> torch.Tensor:
+    x1 = x[:, :, :, ::2]
+    x2 = x[:, :, :, 1::2]
+    x = torch.stack((-x2, x1), dim=-1)
+    return x.flatten(-2)  # in einsum notation: rearrange(x, '... d j -> ... (d j)')
+
+
+# Copied from transformers.models.gptj.modeling_gptj.apply_rotary_pos_emb
+def apply_rotary_pos_emb(tensor: torch.Tensor, sin: torch.Tensor, cos: torch.Tensor) -> torch.Tensor:
+    sin = torch.repeat_interleave(sin[:, :, None, :], 2, 3)
+    cos = torch.repeat_interleave(cos[:, :, None, :], 2, 3)
+    return (tensor * cos) + (rotate_every_two(tensor) * sin)
+
+
+class MossAttention(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+
+        max_positions = config.max_position_embeddings
+        self.register_buffer(
+            "causal_mask",
+            torch.tril(torch.ones((max_positions, max_positions), dtype=torch.bool)).view(
+                1, 1, max_positions, max_positions
+            ),
+        )
+
+        self.attn_dropout = nn.Dropout(config.attn_pdrop)
+        self.resid_dropout = nn.Dropout(config.resid_pdrop)
+
+        self.embed_dim = config.hidden_size
+        self.num_attention_heads = config.num_attention_heads
+        self.head_dim = self.embed_dim // self.num_attention_heads
+        if self.head_dim * self.num_attention_heads != self.embed_dim:
+            raise ValueError(
+                f"embed_dim must be divisible by num_attention_heads (got `embed_dim`: {self.embed_dim} and"
+                f" `num_attention_heads`: {self.num_attention_heads})."
+            )
+        self.scale_attn = torch.sqrt(torch.tensor(self.head_dim, dtype=torch.float32)).to(torch.get_default_dtype())
+        self.qkv_proj = nn.Linear(self.embed_dim, self.embed_dim * 3, bias=False)
+
+        self.out_proj = nn.Linear(self.embed_dim, self.embed_dim, bias=False)
+        self.rotary_dim = config.rotary_dim
+        pos_embd_dim = self.rotary_dim or self.embed_dim
+        self.embed_positions = create_sinusoidal_positions(max_positions, pos_embd_dim)
+
+    def _split_heads(self, x, n_head, dim_head, mp_num):
+        reshaped = x.reshape(x.shape[:-1] + (n_head // mp_num, dim_head))
+        reshaped = reshaped.reshape(x.shape[:-2] + (-1,) + reshaped.shape[-1:])
+        return reshaped
+
+    def _merge_heads(self, tensor, num_attention_heads, attn_head_size):
+        """
+        Merges attn_head_size dim and num_attn_heads dim into n_ctx
+        """
+        if len(tensor.shape) == 5:
+            tensor = tensor.permute(0, 1, 3, 2, 4).contiguous()
+        elif len(tensor.shape) == 4:
+            tensor = tensor.permute(0, 2, 1, 3).contiguous()
+        else:
+            raise ValueError(f"Input tensor rank should be one of [4, 5], but is: {len(tensor.shape)}")
+        new_shape = tensor.size()[:-2] + (num_attention_heads * attn_head_size,)
+        return tensor.view(new_shape)
+
+    def _attn(
+        self,
+        query,
+        key,
+        value,
+        attention_mask=None,
+        head_mask=None,
+    ):
+        # compute causal mask from causal mask buffer
+        query_length, key_length = query.size(-2), key.size(-2)
+        causal_mask = self.causal_mask[:, :, key_length - query_length : key_length, :key_length]
+
+        # Keep the attention weights computation in fp32 to avoid overflow issues
+        query = query.to(torch.float32)
+        key = key.to(torch.float32)
+
+        attn_weights = torch.matmul(query, key.transpose(-1, -2))
+
+        attn_weights = attn_weights / self.scale_attn
+        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.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.Softmax(dim=-1)(attn_weights)
+        attn_weights = attn_weights.to(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 forward(
+        self,
+        hidden_states: Optional[torch.FloatTensor],
+        layer_past: Optional[Tuple[torch.Tensor]] = None,
+        attention_mask: Optional[torch.FloatTensor] = None,
+        position_ids: Optional[torch.LongTensor] = None,
+        head_mask: Optional[torch.FloatTensor] = None,
+        use_cache: Optional[bool] = False,
+        output_attentions: Optional[bool] = False,
+    ) -> Union[
+        Tuple[torch.Tensor, Tuple[torch.Tensor]],
+        Optional[Tuple[torch.Tensor, Tuple[torch.Tensor], Tuple[torch.Tensor, ...]]],
+    ]:
+        qkv = self.qkv_proj(hidden_states)
+        # TODO(enijkamp): factor out number of logical TPU-v4 cores or make forward pass agnostic
+        mp_num = 4
+        qkv_split = qkv.reshape(qkv.shape[:-1] + (mp_num, -1))
+
+        local_dim = self.head_dim * self.num_attention_heads // mp_num
+        query, value, key = torch.split(qkv_split, local_dim, dim=-1)
+        query = self._split_heads(query, self.num_attention_heads, self.head_dim, mp_num=mp_num)
+        key = self._split_heads(key, self.num_attention_heads, self.head_dim, mp_num=mp_num)
+
+        value = self._split_heads(value, self.num_attention_heads, self.head_dim, mp_num=mp_num)
+        value = value.permute(0, 2, 1, 3)
+
+        embed_positions = self.embed_positions
+        if embed_positions.device != position_ids.device:
+            embed_positions = embed_positions.to(position_ids.device)
+            self.embed_positions = embed_positions
+
+        sincos = embed_positions[position_ids]
+        sin, cos = torch.split(sincos, sincos.shape[-1] // 2, dim=-1)
+
+        if self.rotary_dim is not None:
+            k_rot = key[:, :, :, : self.rotary_dim]
+            k_pass = key[:, :, :, self.rotary_dim :]
+
+            q_rot = query[:, :, :, : self.rotary_dim]
+            q_pass = query[:, :, :, self.rotary_dim :]
+
+            k_rot = apply_rotary_pos_emb(k_rot, sin, cos)
+            q_rot = apply_rotary_pos_emb(q_rot, sin, cos)
+
+            key = torch.cat([k_rot, k_pass], dim=-1)
+            query = torch.cat([q_rot, q_pass], dim=-1)
+        else:
+            key = apply_rotary_pos_emb(key, sin, cos)
+            query = apply_rotary_pos_emb(query, sin, cos)
+
+        key = key.permute(0, 2, 1, 3)
+        query = query.permute(0, 2, 1, 3)
+
+        if layer_past is not None:
+            past_key = layer_past[0]
+            past_value = layer_past[1]
+            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
+
+        # compute self-attention: V x Softmax(QK^T)
+        attn_output, attn_weights = self._attn(query, key, value, attention_mask, head_mask)
+
+        attn_output = self._merge_heads(attn_output, self.num_attention_heads, self.head_dim)
+        attn_output = self.out_proj(attn_output)
+        attn_output = self.resid_dropout(attn_output)
+
+        outputs = (attn_output, present)
+        if output_attentions:
+            outputs += (attn_weights,)
+
+        return outputs  # a, present, (attentions)
+
+
+# Copied from transformers.models.gptj.modeling_gptj.GPTJMLP with GPTJ->Moss
+class MossMLP(nn.Module):
+    def __init__(self, intermediate_size, config):  # in MLP: intermediate_size= 4 * embed_dim
+        super().__init__()
+        embed_dim = config.n_embd
+
+        self.fc_in = nn.Linear(embed_dim, intermediate_size)
+        self.fc_out = nn.Linear(intermediate_size, embed_dim)
+
+        self.act = ACT2FN[config.activation_function]
+        self.dropout = nn.Dropout(config.resid_pdrop)
+
+    def forward(self, hidden_states: Optional[torch.FloatTensor]) -> torch.FloatTensor:
+        hidden_states = self.fc_in(hidden_states)
+        hidden_states = self.act(hidden_states)
+        hidden_states = self.fc_out(hidden_states)
+        hidden_states = self.dropout(hidden_states)
+        return hidden_states
+
+
+# Copied from transformers.models.gptj.modeling_gptj.GPTJBlock with GPTJ->Moss
+class MossBlock(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        inner_dim = config.n_inner if config.n_inner is not None else 4 * config.n_embd
+        self.ln_1 = nn.LayerNorm(config.n_embd, eps=config.layer_norm_epsilon)
+        self.attn = MossAttention(config)
+        self.mlp = MossMLP(inner_dim, config)
+
+    def forward(
+        self,
+        hidden_states: Optional[torch.FloatTensor],
+        layer_past: Optional[Tuple[torch.Tensor]] = None,
+        attention_mask: Optional[torch.FloatTensor] = None,
+        position_ids: Optional[torch.LongTensor] = None,
+        head_mask: Optional[torch.FloatTensor] = 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=hidden_states,
+            layer_past=layer_past,
+            attention_mask=attention_mask,
+            position_ids=position_ids,
+            head_mask=head_mask,
+            use_cache=use_cache,
+            output_attentions=output_attentions,
+        )
+        attn_output = attn_outputs[0]  # output_attn: a, present, (attentions)
+        outputs = attn_outputs[1:]
+
+        feed_forward_hidden_states = self.mlp(hidden_states)
+        hidden_states = attn_output + feed_forward_hidden_states + residual
+
+        if use_cache:
+            outputs = (hidden_states,) + outputs
+        else:
+            outputs = (hidden_states,) + outputs[1:]
+
+        return outputs  # hidden_states, present, (attentions)
+
+
+class MossPreTrainedModel(PreTrainedModel):
+    """
+    An abstract class to handle weights initialization and a simple interface for downloading and loading pretrained
+    models.
+    """
+
+    config_class = MossConfig
+    base_model_prefix = "transformer"
+    supports_gradient_checkpointing = True
+    _no_split_modules = ["MossBlock"]
+
+    def __init__(self, *inputs, **kwargs):
+        super().__init__(*inputs, **kwargs)
+
+    def _init_weights(self, module):
+        """Initialize the weights."""
+        if isinstance(module, (nn.Linear,)):
+            # Slightly different from Mesh Transformer JAX 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)
+
+    def _set_gradient_checkpointing(self, module, value=False):
+        if isinstance(module, MossModel):
+            module.gradient_checkpointing = value
+
+
+MOSS_START_DOCSTRING = r"""
+    This model is a PyTorch [torch.nn.Module](https://pytorch.org/docs/stable/nn.html#torch.nn.Module) sub-class. Use
+    it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage and
+    behavior.
+
+    Parameters:
+        config ([`MossConfig`]): Model configuration class with all the parameters of the model.
+            Initializing with a config file does not load the weights associated with the model, only the
+            configuration. Check out the [`~PreTrainedModel.from_pretrained`] method to load the model weights.
+"""
+
+MOSS_INPUTS_DOCSTRING = r"""
+    Args:
+        input_ids (`torch.LongTensor` of shape `({0})`):
+            Indices of input sequence tokens in the vocabulary.
+
+            Indices can be obtained using [`AutoProcenizer`]. See [`PreTrainedTokenizer.encode`] and
+            [`PreTrainedTokenizer.__call__`] for details.
+
+            [What are input IDs?](../glossary#input-ids)
+        attention_mask (`torch.FloatTensor` of shape `({0})`, *optional*):
+            Mask to avoid performing attention on padding token indices. Mask values selected in `[0, 1]`:
+
+            - 1 for tokens that are **not masked**,
+            - 0 for tokens that are **masked**.
+
+            [What are attention masks?](../glossary#attention-mask)
+        token_type_ids (`torch.LongTensor` of shape `({0})`, *optional*):
+            Segment token indices to indicate first and second portions of the inputs. Indices are selected in `[0,
+            1]`:
+
+            - 0 corresponds to a *sentence A* token,
+            - 1 corresponds to a *sentence B* token.
+
+            [What are token type IDs?](../glossary#token-type-ids)
+        position_ids (`torch.LongTensor` of shape `({0})`, *optional*):
+            Indices of positions of each input sequence tokens in the position embeddings. Selected in the range `[0,
+            config.n_positions - 1]`.
+
+            [What are position IDs?](../glossary#position-ids)
+        head_mask (`torch.FloatTensor` of shape `(num_attention_heads,)` or `(n_layer, num_attention_heads)`, *optional*):
+            Mask to nullify selected heads of the self-attention modules. Mask values selected in `[0, 1]`:
+
+            - 1 indicates the head is **not masked**,
+            - 0 indicates the head is **masked**.
+
+        inputs_embeds (`torch.FloatTensor` of shape `({0}, hidden_dim)`, *optional*):
+            Optionally, instead of passing `input_ids` you can choose to directly pass an embedded representation. This
+            is useful if you want more control over how to convert *input_ids* indices into associated vectors than the
+            model's internal embedding lookup matrix.
+        output_attentions (`bool`, *optional*):
+            Whether or not to return the attentions tensors of all attention layers. See `attentions` under returned
+            tensors for more detail.
+        output_hidden_states (`bool`, *optional*):
+            Whether or not to return the hidden states of all layers. See `hidden_states` under returned tensors for
+            more detail.
+        return_dict (`bool`, *optional*):
+            Whether or not to return a [`~utils.ModelOutput`] instead of a plain tuple.
+"""
+
+
+@add_start_docstrings(
+    "The bare Moss Model transformer outputting raw hidden-states without any specific head on top.",
+    MOSS_START_DOCSTRING,
+)
+class MossModel(MossPreTrainedModel):
+    def __init__(self, config):
+        super().__init__(config)
+
+        self.embed_dim = config.n_embd
+        self.vocab_size = config.vocab_size
+        self.wte = nn.Embedding(config.vocab_size, self.embed_dim)
+        self.drop = nn.Dropout(config.embd_pdrop)
+        self.h = nn.ModuleList([MossBlock(config) for _ in range(config.n_layer)])
+        self.ln_f = nn.LayerNorm(self.embed_dim, eps=config.layer_norm_epsilon)
+        self.rotary_dim = min(config.rotary_dim, config.n_ctx // config.num_attention_heads)
+
+        self.gradient_checkpointing = False
+
+        # Initialize weights and apply final processing
+        self.post_init()
+
+    def get_input_embeddings(self):
+        return self.wte
+
+    def set_input_embeddings(self, new_embeddings):
+        self.wte = new_embeddings
+
+    @add_start_docstrings_to_model_forward(MOSS_INPUTS_DOCSTRING.format("batch_size, sequence_length"))
+    @add_code_sample_docstrings(
+        checkpoint=_CHECKPOINT_FOR_DOC,
+        output_type=BaseModelOutputWithPast,
+        config_class=_CONFIG_FOR_DOC,
+    )
+    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,
+        use_cache: Optional[bool] = None,
+        output_attentions: Optional[bool] = None,
+        output_hidden_states: Optional[bool] = None,
+        return_dict: Optional[bool] = None,
+    ) -> Union[Tuple, BaseModelOutputWithPast]:
+        output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
+        output_hidden_states = (
+            output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
+        )
+        use_cache = use_cache if use_cache is not None else self.config.use_cache
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+        if input_ids is 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]).long()
+
+        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])
+
+        # Attention 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
+
+        # Prepare head mask if needed
+        # 1.0 in head_mask indicate we keep the head
+        # attention_probs has shape bsz x num_attention_heads x N x N
+        # head_mask has shape n_layer x batch x num_attention_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)
+
+        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),)
+
+        if self.gradient_checkpointing and self.training:
+            if use_cache:
+                logger.warning_once(
+                    "`use_cache=True` is incompatible with `config.gradient_checkpointing=True`. Setting "
+                    "`use_cache=False`..."
+                )
+                use_cache = False
+
+        presents = () if use_cache else None
+        all_self_attentions = () if output_attentions else None
+        all_hidden_states = () if output_hidden_states else None
+        for i, (block, layer_past) in enumerate(zip(self.h, past_key_values)):
+            if output_hidden_states:
+                all_hidden_states = all_hidden_states + (hidden_states,)
+
+            if self.gradient_checkpointing and self.training:
+
+                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,
+                    position_ids,
+                    head_mask[i],
+                )
+            else:
+                outputs = block(
+                    hidden_states=hidden_states,
+                    layer_past=layer_past,
+                    attention_mask=attention_mask,
+                    position_ids=position_ids,
+                    head_mask=head_mask[i],
+                    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],)
+
+        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] if v is not None)
+
+        return BaseModelOutputWithPast(
+            last_hidden_state=hidden_states,
+            past_key_values=presents,
+            hidden_states=all_hidden_states,
+            attentions=all_self_attentions,
+        )
+
+
+@add_start_docstrings(
+    """
+    The Moss Model transformer with a language modeling head on top.
+    """,
+    MOSS_START_DOCSTRING,
+)
+class MossForCausalLM(MossPreTrainedModel):
+    _keys_to_ignore_on_load_missing = [r"h\.\d+\.attn\.causal_mask"]
+
+    def __init__(self, config):
+        super().__init__(config)
+        if not hasattr(config, 'wbits'):
+            config.wbits = 32
+            config.groupsize = 128
+            
+        if config.wbits not in [4, 8, 32]:
+            logger.warning(f'Specify `wbits` with 4, 8 or 32 to load the model. ')
+        if config.wbits in [4, 8]:
+            def noop(*args, **kwargs):
+                pass
+            torch.nn.init.kaiming_uniform_ = noop
+            torch.nn.init.uniform_ = noop
+            torch.nn.init.normal_ = noop
+
+            torch.set_default_dtype(torch.half)
+            transformers.modeling_utils._init_weights = False
+            torch.set_default_dtype(torch.half)
+        self.transformer = MossModel(config)
+        self.lm_head = nn.Linear(config.n_embd, config.vocab_size)
+        if config.wbits in [4, 8]:
+            torch.set_default_dtype(torch.float)
+            transformers.modeling_utils._init_weights = True
+            self.quantize(config.wbits, config.groupsize)
+        # Initialize weights and apply final processing
+        self.post_init()
+
+    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_key_values=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_key_values:
+            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_key_values:
+                position_ids = position_ids[:, -1].unsqueeze(-1)
+
+        return {
+            "input_ids": input_ids,
+            "past_key_values": past_key_values,
+            "use_cache": kwargs.get("use_cache"),
+            "position_ids": position_ids,
+            "attention_mask": attention_mask,
+            "token_type_ids": token_type_ids,
+        }
+
+    @add_start_docstrings_to_model_forward(MOSS_INPUTS_DOCSTRING.format("batch_size, sequence_length"))
+    @add_code_sample_docstrings(
+        checkpoint=_CHECKPOINT_FOR_DOC,
+        output_type=CausalLMOutputWithPast,
+        config_class=_CONFIG_FOR_DOC,
+    )
+    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,
+        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, CausalLMOutputWithPast]:
+        r"""
+        labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
+            Labels for language modeling. Note that the labels **are shifted** inside the model, i.e. you can set
+            `labels = input_ids` Indices are selected in `[-100, 0, ..., config.vocab_size]` All labels set to `-100`
+            are ignored (masked), the loss is only computed for labels in `[0, ..., config.vocab_size]`
+        """
+        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,
+            use_cache=use_cache,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+            return_dict=return_dict,
+        )
+        hidden_states = transformer_outputs[0]
+
+        # make sure sampling in fp16 works correctly and
+        # compute loss in fp32 to match with mesh-tf version
+        # https://github.com/EleutherAI/gpt-neo/blob/89ce74164da2fb16179106f54e2269b5da8db333/models/gpt2/gpt2.py#L179
+        lm_logits = self.lm_head(hidden_states).to(torch.float32)
+
+        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 = CrossEntropyLoss()
+            loss = loss_fct(shift_logits.view(-1, shift_logits.size(-1)), shift_labels.view(-1))
+
+            loss = loss.to(hidden_states.dtype)
+
+        if not return_dict:
+            output = (lm_logits,) + transformer_outputs[1:]
+            return ((loss,) + output) if loss is not None else output
+
+        return CausalLMOutputWithPast(
+            loss=loss,
+            logits=lm_logits,
+            past_key_values=transformer_outputs.past_key_values,
+            hidden_states=transformer_outputs.hidden_states,
+            attentions=transformer_outputs.attentions,
+        )
+
+    @staticmethod
+    def _reorder_cache(
+        past_key_values: Tuple[Tuple[torch.Tensor]], beam_idx: torch.Tensor
+    ) -> Tuple[Tuple[torch.Tensor]]:
+        """
+        This function is used to re-order the `past_key_values` cache if [`~PretrainedModel.beam_search`] or
+        [`~PretrainedModel.beam_sample`] is called. This is required to match `past_key_values` with the correct
+        beam_idx at every generation step.
+        """
+        return tuple(
+            tuple(past_state.index_select(0, beam_idx.to(past_state.device)) for past_state in layer_past)
+            for layer_past in past_key_values
+        )
+
+    def quantize(self, wbits, groupsize):
+        from .quantization import quantize_with_gptq
+        return quantize_with_gptq(self, wbits, groupsize)
+