Upload MERaLiONForConditionalGeneration

config.json

@@ -1,7 +1,10 @@
 {
-  "_attn_implementation_autoset": true,
+  "architectures": [
+    "MERaLiONForConditionalGeneration"
+  ],
   "auto_map": {
-    "AutoConfig": "configuration_meralion.MERaLiONConfig"
+    "AutoConfig": "configuration_meralion.MERaLiONConfig",
+    "AutoModelForSpeechSeq2Seq": "modeling_meralion.MERaLiONForConditionalGeneration"
   },
   "head_dim": 256,
   "hidden_size": 3584,
@@ -163,5 +166,6 @@
     "sliding_window_size": 4096,
     "torch_dtype": "bfloat16"
   },
+  "torch_dtype": "bfloat16",
   "transformers_version": "4.46.3"
 }

modeling_text_decoder.py

@@ -16,21 +16,24 @@
 from typing import List, Optional, Tuple, Union
 
 import torch
+import torch.nn as nn
 import torch.utils.checkpoint
-from torch import nn
-from torch.nn import CrossEntropyLoss
 
 from transformers.activations import ACT2FN
 from transformers.cache_utils import Cache, HybridCache
+from transformers.generation import GenerationMixin
+from transformers.modeling_flash_attention_utils import _flash_attention_forward
 from transformers.modeling_outputs import (
     BaseModelOutputWithPast,
     CausalLMOutputWithPast,
+    SequenceClassifierOutputWithPast,
+    TokenClassifierOutput,
 )
 from transformers.modeling_utils import PreTrainedModel
 from transformers.utils import (
+    add_code_sample_docstrings,
     add_start_docstrings,
     add_start_docstrings_to_model_forward,
-    is_flash_attn_2_available,
     is_flash_attn_greater_or_equal,
     is_flash_attn_greater_or_equal_2_10,
     logging,
@@ -39,64 +42,7 @@ from transformers.utils import (
 from .configuration_meralion import MERaLiONTextConfig
 
 
-if is_flash_attn_2_available():
-    from transformers.modeling_flash_attention_utils import _flash_attention_forward
-
-
-logger = logging.get_logger(__name__)
-
-
-# Copied from transformers.models.llama.modeling_llama._prepare_4d_causal_attention_mask_with_cache_position
-def _prepare_4d_causal_attention_mask_with_cache_position(
-    attention_mask: torch.Tensor,
-    sequence_length: int,
-    target_length: int,
-    dtype: torch.dtype,
-    device: torch.device,
-    min_dtype: float,
-    cache_position: torch.Tensor,
-    batch_size: int,
-):
-    """
-    Creates a causal 4D mask of shape `(batch_size, 1, query_length, key_value_length)` from a 2D mask of shape
-    `(batch_size, key_value_length)`, or if the input `attention_mask` is already 4D, do nothing.
-
-    Args:
-        attention_mask (`torch.Tensor`):
-            A 2D attention mask of shape `(batch_size, key_value_length)` or a 4D attention mask of shape `(batch_size, 1, query_length, key_value_length)`.
-        sequence_length (`int`):
-            The sequence length being processed.
-        target_length (`int`):
-            The target length: when generating with static cache, the mask should be as long as the static cache, to account for the 0 padding, the part of the cache that is not filled yet.
-        dtype (`torch.dtype`):
-            The dtype to use for the 4D attention mask.
-        device (`torch.device`):
-            The device to plcae the 4D attention mask on.
-        min_dtype (`float`):
-            The minimum value representable with the dtype `dtype`.
-        cache_position (`torch.Tensor`):
-            Indices depicting the position of the input sequence tokens in the sequence.
-        batch_size (`torch.Tensor`):
-            Batch size.
-    """
-    if attention_mask is not None and attention_mask.dim() == 4:
-        # In this case we assume that the mask comes already in inverted form and requires no inversion or slicing.
-        causal_mask = attention_mask
-    else:
-        causal_mask = torch.full((sequence_length, target_length), fill_value=min_dtype, dtype=dtype, device=device)
-        if sequence_length != 1:
-            causal_mask = torch.triu(causal_mask, diagonal=1)
-        causal_mask *= torch.arange(target_length, device=device) > cache_position.reshape(-1, 1)
-        causal_mask = causal_mask[None, None, :, :].expand(batch_size, 1, -1, -1)
-        if attention_mask is not None:
-            causal_mask = causal_mask.clone()  # copy to contiguous memory for in-place edit
-            mask_length = attention_mask.shape[-1]
-            padding_mask = causal_mask[:, :, :, :mask_length] + attention_mask[:, None, None, :]
-            padding_mask = padding_mask == 0
-            causal_mask[:, :, :, :mask_length] = causal_mask[:, :, :, :mask_length].masked_fill(
-                padding_mask, min_dtype
-            )
-    return causal_mask
+_CHECKPOINT_FOR_DOC = "MERaLiON/MERaLiON-AudioLLM-Whisper-SEA-LION"
 
 
 class MERaLiONTextRMSNorm(nn.Module):
@@ -119,6 +65,24 @@ class MERaLiONTextRMSNorm(nn.Module):
         return f"{tuple(self.weight.shape)}, eps={self.eps}"
 
 
+class MERaLiONTextMLP(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.config = config
+        self.hidden_size = config.hidden_size
+        self.intermediate_size = config.intermediate_size
+        self.gate_proj = nn.Linear(self.hidden_size, self.intermediate_size, bias=False)
+        self.up_proj = nn.Linear(self.hidden_size, self.intermediate_size, bias=False)
+        self.down_proj = nn.Linear(self.intermediate_size, self.hidden_size, bias=False)
+        self.act_fn = ACT2FN[config.hidden_activation]
+
+    def forward(self, x):
+        return self.down_proj(self.act_fn(self.gate_proj(x)) * self.up_proj(x))
+
+
+logger = logging.get_logger(__name__)
+
+
 class MERaLiONTextRotaryEmbedding(nn.Module):
     def __init__(self, dim, max_position_embeddings=2048, base=10000, device=None):
         super().__init__()
@@ -181,21 +145,6 @@ def apply_rotary_pos_emb(q, k, cos, sin, position_ids=None, unsqueeze_dim=1):
     return q_embed, k_embed
 
 
-class MERaLiONTextMLP(nn.Module):
-    def __init__(self, config):
-        super().__init__()
-        self.config = config
-        self.hidden_size = config.hidden_size
-        self.intermediate_size = config.intermediate_size
-        self.gate_proj = nn.Linear(self.hidden_size, self.intermediate_size, bias=False)
-        self.up_proj = nn.Linear(self.hidden_size, self.intermediate_size, bias=False)
-        self.down_proj = nn.Linear(self.intermediate_size, self.hidden_size, bias=False)
-        self.act_fn = ACT2FN[config.hidden_activation]
-
-    def forward(self, x):
-        return self.down_proj(self.act_fn(self.gate_proj(x)) * self.up_proj(x))
-
-
 def repeat_kv(hidden_states: torch.Tensor, n_rep: int) -> torch.Tensor:
     """
     This is the equivalent of torch.repeat_interleave(x, dim=1, repeats=n_rep). The hidden states go from (batch,
@@ -243,12 +192,12 @@ class MERaLiONTextAttention(nn.Module):
         self.k_proj = nn.Linear(self.hidden_size, self.num_key_value_heads * self.head_dim, bias=config.attention_bias)
         self.v_proj = nn.Linear(self.hidden_size, self.num_key_value_heads * self.head_dim, bias=config.attention_bias)
         self.o_proj = nn.Linear(self.num_heads * self.head_dim, self.hidden_size, bias=config.attention_bias)
+        self.sliding_window = config.sliding_window if not bool(layer_idx % 2) else None
         self.rotary_emb = MERaLiONTextRotaryEmbedding(
             self.head_dim,
             max_position_embeddings=self.max_position_embeddings,
             base=self.rope_theta,
         )
-        self.sliding_window = config.sliding_window if not bool(layer_idx % 2) else None
 
     def forward(
         self,
@@ -492,9 +441,11 @@ class MERaLiONTextSdpaAttention(MERaLiONTextAttention):
 
         key_states = repeat_kv(key_states, self.num_key_value_groups)
         value_states = repeat_kv(value_states, self.num_key_value_groups)
+
         causal_mask = attention_mask
         if attention_mask is not None:
             causal_mask = causal_mask[:, :, :, : key_states.shape[-2]]
+
         # SDPA with memory-efficient backend is currently (torch==2.1.2) bugged with non-contiguous inputs with custom attn_mask,
         # Reference: https://github.com/pytorch/pytorch/issues/112577.
         if query_states.device.type == "cuda" and causal_mask is not None:
@@ -505,6 +456,7 @@ class MERaLiONTextSdpaAttention(MERaLiONTextAttention):
         # We dispatch to SDPA's Flash Attention or Efficient kernels via this `is_causal` if statement instead of an inline conditional assignment
         # in SDPA to support both torch.compile's dynamic shapes and full graph options. An inline conditional prevents dynamic shapes from compiling.
         is_causal = True if causal_mask is None and q_len > 1 else False
+
         attn_output = torch.nn.functional.scaled_dot_product_attention(
             query_states,
             key_states,
@@ -523,7 +475,7 @@ class MERaLiONTextSdpaAttention(MERaLiONTextAttention):
         return attn_output, None, past_key_value
 
 
-MERaLiONText_ATTENTION_CLASSES = {
+MERALION_TEXT_ATTENTION_CLASSES = {
     "eager": MERaLiONTextAttention,
     "flash_attention_2": MERaLiONTextFlashAttention2,
     "sdpa": MERaLiONTextSdpaAttention,
@@ -533,19 +485,16 @@ MERaLiONText_ATTENTION_CLASSES = {
 class MERaLiONTextDecoderLayer(nn.Module):
     def __init__(self, config: MERaLiONTextConfig, layer_idx: int):
         super().__init__()
-        self.config = config
         self.hidden_size = config.hidden_size
-
-        self.self_attn = MERaLiONText_ATTENTION_CLASSES[config._attn_implementation](config=config, layer_idx=layer_idx)
-
+        self.self_attn = MERALION_TEXT_ATTENTION_CLASSES[config._attn_implementation](config=config, layer_idx=layer_idx)
         self.mlp = MERaLiONTextMLP(config)
         self.input_layernorm = MERaLiONTextRMSNorm(config.hidden_size, eps=config.rms_norm_eps)
-        self.post_attention_layernorm = MERaLiONTextRMSNorm(config.hidden_size, eps=config.rms_norm_eps)
-
+        self.config = config
         self.is_sliding = not bool(layer_idx % 2)
         self.pre_feedforward_layernorm = MERaLiONTextRMSNorm(config.hidden_size, eps=config.rms_norm_eps)
         self.post_feedforward_layernorm = MERaLiONTextRMSNorm(config.hidden_size, eps=config.rms_norm_eps)
         self.sliding_window = config.sliding_window
+        self.post_attention_layernorm = MERaLiONTextRMSNorm(config.hidden_size, eps=config.rms_norm_eps)
 
     def forward(
         self,
@@ -557,6 +506,25 @@ class MERaLiONTextDecoderLayer(nn.Module):
         use_cache: Optional[bool] = False,
         cache_position: Optional[torch.LongTensor] = None,
     ) -> Tuple[torch.FloatTensor, Optional[Tuple[torch.FloatTensor, torch.FloatTensor]]]:
+        """
+        Args:
+            hidden_states (`torch.FloatTensor`): input to the layer of shape `(batch, seq_len, embed_dim)`
+            attention_mask (`torch.FloatTensor`, *optional*):
+                attention mask of size `(batch_size, sequence_length)` if flash attention is used or `(batch_size, 1,
+                query_sequence_length, key_sequence_length)` if default attention is used.
+            output_attentions (`bool`, *optional*):
+                Whether or not to return the attentions tensors of all attention layers. See `attentions` under
+                returned tensors for more detail.
+            use_cache (`bool`, *optional*):
+                If set to `True`, `past_key_values` key value states are returned and can be used to speed up decoding
+                (see `past_key_values`).
+            past_key_value (`Tuple(torch.FloatTensor)`, *optional*): cached past key and value projection states
+            cache_position (`torch.LongTensor` of shape `(sequence_length)`, *optional*):
+                Indices depicting the position of the input sequence tokens in the sequence
+            kwargs (`dict`, *optional*):
+                Arbitrary kwargs to be ignored, used for FSDP and other methods that injects code
+                into the model
+        """
         if self.is_sliding and attention_mask is not None:  # efficient SDPA and no padding
             # Flash-attn is a 2D tensor
             if self.config._attn_implementation == "flash_attention_2":
@@ -570,6 +538,7 @@ class MERaLiONTextDecoderLayer(nn.Module):
                 attention_mask = torch.where(sliding_window_mask, min_dtype, attention_mask)
                 if attention_mask.shape[-1] <= 1:  # when decoding
                     attention_mask = attention_mask[:, :, :, -self.sliding_window :]
+
         residual = hidden_states
 
         hidden_states = self.input_layernorm(hidden_states)
@@ -648,6 +617,20 @@ class MERaLiONTextPreTrainedModel(PreTrainedModel):
             if module.padding_idx is not None:
                 module.weight.data[module.padding_idx].zero_()
 
+    @classmethod
+    def _check_and_enable_sdpa(cls, config, hard_check_only: bool = False):
+        """
+        Overloads `PreTrainedModel._check_and_enable_sdpa` so as to DISABLE torch SDPA by default on MERaLiONText models.
+        SDPA reduces the model performance on MERaLiONText because of the logits softcapping.
+        """
+        config = super()._check_and_enable_sdpa(config, hard_check_only=hard_check_only)
+
+        # if using the default path -> swap sdpa by eager
+        if not hard_check_only and config._attn_implementation == "sdpa":
+            config._attn_implementation = "eager"
+
+        return config
+
 
 _CONFIG_FOR_DOC = "MERaLiONTextConfig"
 
@@ -693,7 +676,8 @@ MERALION_TEXT_INPUTS_DOCSTRING = r"""
             returned by the model at a previous stage of decoding, when `use_cache=True` or `config.use_cache=True`.
 
             Two formats are allowed:
-            - a [`~cache_utils.Cache`] instance;
+            - a [`~cache_utils.Cache`] instance, see our
+            [kv cache guide](https://huggingface.co/docs/transformers/en/kv_cache);
             - Tuple of `tuple(torch.FloatTensor)` of length `config.n_layers`, with each tuple having 2 tensors of
             shape `(batch_size, num_heads, sequence_length, embed_size_per_head)`). This is also known as the legacy
             cache format.
@@ -765,7 +749,7 @@ class MERaLiONTextModel(MERaLiONTextPreTrainedModel):
         input_ids: torch.LongTensor = None,
         attention_mask: Optional[torch.Tensor] = None,
         position_ids: Optional[torch.LongTensor] = None,
-        past_key_values: Optional[Union[Cache, List[torch.FloatTensor]]] = None,
+        past_key_values: Optional[HybridCache] = None,
         inputs_embeds: Optional[torch.FloatTensor] = None,
         use_cache: Optional[bool] = None,
         output_attentions: Optional[bool] = None,
@@ -781,9 +765,7 @@ class MERaLiONTextModel(MERaLiONTextPreTrainedModel):
         return_dict = return_dict if return_dict is not None else self.config.use_return_dict
 
         if (input_ids is None) ^ (inputs_embeds is not None):
-            raise ValueError(
-                "You cannot specify both input_ids and inputs_embeds at the same time, and must specify either one"
-            )
+            raise ValueError("You must specify exactly one of input_ids or inputs_embeds")
 
         if self.gradient_checkpointing and self.training and use_cache:
             logger.warning_once(
@@ -794,8 +776,21 @@ class MERaLiONTextModel(MERaLiONTextPreTrainedModel):
         if inputs_embeds is None:
             inputs_embeds = self.embed_tokens(input_ids)
 
+        if use_cache and past_key_values is None and not self.training:
+            batch_size, seq_len, _ = inputs_embeds.shape
+            past_key_values = HybridCache(
+                self.config,
+                batch_size=batch_size,
+                max_cache_len=seq_len,
+                device=self.device,
+                dtype=inputs_embeds.dtype,
+            )
+
         if cache_position is None:
-            cache_position = torch.arange(0, inputs_embeds.shape[1], device=inputs_embeds.device)
+            past_seen_tokens = past_key_values.get_seq_length() if past_key_values is not None else 0
+            cache_position = torch.arange(
+                past_seen_tokens, past_seen_tokens + inputs_embeds.shape[1], device=inputs_embeds.device
+            )
 
         if position_ids is None:
             position_ids = cache_position.unsqueeze(0)
@@ -813,6 +808,7 @@ class MERaLiONTextModel(MERaLiONTextPreTrainedModel):
         normalizer = torch.tensor(self.config.hidden_size**0.5, dtype=hidden_states.dtype)
         hidden_states = hidden_states * normalizer
 
+        # decoder layers
         all_hidden_states = () if output_hidden_states else None
         all_self_attns = () if output_attentions else None
 
@@ -849,7 +845,6 @@ class MERaLiONTextModel(MERaLiONTextPreTrainedModel):
 
         hidden_states = self.norm(hidden_states)
 
-        # add hidden states from the last decoder layer
         if output_hidden_states:
             all_hidden_states += (hidden_states,)
 
@@ -869,7 +864,7 @@ class MERaLiONTextModel(MERaLiONTextPreTrainedModel):
         attention_mask: torch.Tensor,
         input_tensor: torch.Tensor,
         cache_position: torch.Tensor,
-        past_key_values: Cache,
+        past_key_values: HybridCache,
         output_attentions: bool,
     ):
         # Flash Attention currently doesn't support static cache but MERaLiONText work only with static cache.
@@ -880,28 +875,82 @@ class MERaLiONTextModel(MERaLiONTextPreTrainedModel):
             return attention_mask
 
         dtype, device = input_tensor.dtype, input_tensor.device
-        min_dtype = torch.finfo(dtype).min
         sequence_length = input_tensor.shape[1]
         if isinstance(past_key_values, HybridCache):
-            target_length = past_key_values.get_max_length()
+            target_length = past_key_values.get_max_cache_shape()
         else:
             target_length = attention_mask.shape[-1] if attention_mask is not None else input_tensor.shape[1]
 
         # In case the provided `attention` mask is 2D, we generate a causal mask here (4D).
-        causal_mask = _prepare_4d_causal_attention_mask_with_cache_position(
+        causal_mask = self._prepare_4d_causal_attention_mask_with_cache_position(
             attention_mask,
             sequence_length=sequence_length,
             target_length=target_length,
             dtype=dtype,
             device=device,
-            min_dtype=min_dtype,
             cache_position=cache_position,
             batch_size=input_tensor.shape[0],
         )
         return causal_mask
 
+    @staticmethod
+    def _prepare_4d_causal_attention_mask_with_cache_position(
+        attention_mask: torch.Tensor,
+        sequence_length: int,
+        target_length: int,
+        dtype: torch.dtype,
+        device: torch.device,
+        cache_position: torch.Tensor,
+        batch_size: int,
+        **kwargs,
+    ):
+        """
+        Creates a causal 4D mask of shape `(batch_size, 1, query_length, key_value_length)` from a 2D mask of shape
+        `(batch_size, key_value_length)`, or if the input `attention_mask` is already 4D, do nothing.
+
+        Args:
+            attention_mask (`torch.Tensor`):
+                A 2D attention mask of shape `(batch_size, key_value_length)` or a 4D attention mask of shape
+                `(batch_size, 1, query_length, key_value_length)`.
+            sequence_length (`int`):
+                The sequence length being processed.
+            target_length (`int`):
+                The target length: when generating with static cache, the mask should be as long as the static cache,
+                to account for the 0 padding, the part of the cache that is not filled yet.
+            dtype (`torch.dtype`):
+                The dtype to use for the 4D attention mask.
+            device (`torch.device`):
+                The device to plcae the 4D attention mask on.
+            cache_position (`torch.Tensor`):
+                Indices depicting the position of the input sequence tokens in the sequence.
+            batch_size (`torch.Tensor`):
+                Batch size.
+        """
+        if attention_mask is not None and attention_mask.dim() == 4:
+            # In this case we assume that the mask comes already in inverted form and requires no inversion or slicing.
+            causal_mask = attention_mask
+        else:
+            min_dtype = torch.finfo(dtype).min
+            causal_mask = torch.full(
+                (sequence_length, target_length), fill_value=min_dtype, dtype=dtype, device=device
+            )
+            if sequence_length != 1:
+                causal_mask = torch.triu(causal_mask, diagonal=1)
+            causal_mask *= torch.arange(target_length, device=device) > cache_position.reshape(-1, 1)
+            causal_mask = causal_mask[None, None, :, :].expand(batch_size, 1, -1, -1)
+            if attention_mask is not None:
+                causal_mask = causal_mask.clone()  # copy to contiguous memory for in-place edit
+                mask_length = attention_mask.shape[-1]
+                padding_mask = causal_mask[:, :, :, :mask_length] + attention_mask[:, None, None, :]
+                padding_mask = padding_mask == 0
+                causal_mask[:, :, :, :mask_length] = causal_mask[:, :, :, :mask_length].masked_fill(
+                    padding_mask, min_dtype
+                )
+
+        return causal_mask
+
 
-class MERaLiONTextForCausalLM(MERaLiONTextPreTrainedModel):
+class MERaLiONTextForCausalLM(MERaLiONTextPreTrainedModel, GenerationMixin):
     _tied_weights_keys = ["lm_head.weight"]
 
     def __init__(self, config):
@@ -938,7 +987,7 @@ class MERaLiONTextForCausalLM(MERaLiONTextPreTrainedModel):
         input_ids: torch.LongTensor = None,
         attention_mask: Optional[torch.Tensor] = None,
         position_ids: Optional[torch.LongTensor] = None,
-        past_key_values: Optional[Union[Cache, List[torch.FloatTensor]]] = None,
+        past_key_values: Optional[HybridCache] = None,
         inputs_embeds: Optional[torch.FloatTensor] = None,
         labels: Optional[torch.LongTensor] = None,
         use_cache: Optional[bool] = None,
@@ -946,6 +995,8 @@ class MERaLiONTextForCausalLM(MERaLiONTextPreTrainedModel):
         output_hidden_states: Optional[bool] = None,
         return_dict: Optional[bool] = None,
         cache_position: Optional[torch.LongTensor] = None,
+        num_logits_to_keep: int = 0,
+        **loss_kwargs,
     ) -> Union[Tuple, CausalLMOutputWithPast]:
         r"""
         Args:
@@ -954,24 +1005,14 @@ class MERaLiONTextForCausalLM(MERaLiONTextPreTrainedModel):
                 config.vocab_size]` or -100 (see `input_ids` docstring). Tokens with indices set to `-100` are ignored
                 (masked), the loss is only computed for the tokens with labels in `[0, ..., config.vocab_size]`.
 
-        Returns:
-
-        Example:
+            num_logits_to_keep (`int`, *optional*):
+                Calculate logits for the last `num_logits_to_keep` tokens. If `0`, calculate logits for all
+                `input_ids` (special case). Only last token logits are needed for generation, and calculating them only for that
+                token can save memory, which becomes pretty significant for long sequences or large vocabulary size.
 
-        ```python
-        >>> from transformers import AutoTokenizer, GemmaForCausalLM
-
-        >>> model = GemmaForCausalLM.from_pretrained("google/gemma-2-9b")
-        >>> tokenizer = AutoTokenizer.from_pretrained("google/gemma-2-9b")
-
-        >>> prompt = "What is your favorite condiment?"
-        >>> inputs = tokenizer(prompt, return_tensors="pt")
+        Returns:
+        """
 
-        >>> # Generate
-        >>> generate_ids = model.generate(inputs.input_ids, max_length=30)
-        >>> tokenizer.batch_decode(generate_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False)[0]
-        "What is your favorite condiment?"
-        ```"""
         if self.training and self.config._attn_implementation != "eager":
             logger.warning_once(
                 "It is strongly recommended to train MERaLiONText models with the `eager` attention implementation "
@@ -983,7 +1024,6 @@ class MERaLiONTextForCausalLM(MERaLiONTextPreTrainedModel):
         )
         return_dict = return_dict if return_dict is not None else self.config.use_return_dict
         # decoder outputs consists of (dec_features, layer_state, dec_hidden, dec_attn)
-
         outputs = self.model(
             input_ids=input_ids,
             attention_mask=attention_mask,
@@ -998,25 +1038,16 @@ class MERaLiONTextForCausalLM(MERaLiONTextPreTrainedModel):
         )
 
         hidden_states = outputs[0]
-        logits = self.lm_head(hidden_states)
+        # Only compute necessary logits, and do not upcast them to float if we are not computing the loss
+        logits = self.lm_head(hidden_states[:, -num_logits_to_keep:, :])
         if self.config.final_logit_softcapping is not None:
             logits = logits / self.config.final_logit_softcapping
             logits = torch.tanh(logits)
             logits = logits * self.config.final_logit_softcapping
 
-        logits = logits.float()
         loss = None
         if labels is not None:
-            # Shift so that tokens < n predict n
-            shift_logits = logits[..., :-1, :].contiguous()
-            shift_labels = labels[..., 1:].contiguous()
-            # Flatten the tokens
-            loss_fct = CrossEntropyLoss()
-            shift_logits = shift_logits.view(-1, self.config.vocab_size)
-            shift_labels = shift_labels.view(-1)
-            # Enable model parallelism
-            shift_labels = shift_labels.to(shift_logits.device)
-            loss = loss_fct(shift_logits, shift_labels)
+            loss = self.loss_function(logits, labels, self.vocab_size, **loss_kwargs)
 
         if not return_dict:
             output = (logits,) + outputs[1:]
@@ -1039,8 +1070,11 @@ class MERaLiONTextForCausalLM(MERaLiONTextPreTrainedModel):
         cache_position=None,
         position_ids=None,
         use_cache=True,
+        num_logits_to_keep=None,
         **kwargs,
     ):
+        # Overwritten: has a special cache type, `HybridCache`
+
         # If we have cache: let's slice `input_ids` through `cache_position`, to keep only the unprocessed tokens
         # Exception 1: when passing input_embeds, input_ids may be missing entries
         # Exception 2: some generation methods do special slicing of input_ids, so we don't need to do it here
@@ -1080,18 +1114,20 @@ class MERaLiONTextForCausalLM(MERaLiONTextPreTrainedModel):
             else:
                 batch_size, sequence_length = model_inputs["input_ids"].shape
                 device = model_inputs["input_ids"].device
-            dtype = self.lm_head.weight.dtype
-            min_dtype = torch.finfo(dtype).min
-            attention_mask = _prepare_4d_causal_attention_mask_with_cache_position(
+
+            attention_mask = self.model._prepare_4d_causal_attention_mask_with_cache_position(
                 attention_mask,
                 sequence_length=sequence_length,
-                target_length=past_key_values.get_max_length(),
-                dtype=dtype,
+                target_length=past_key_values.get_max_cache_shape(),
+                dtype=self.lm_head.weight.dtype,
                 device=device,
-                min_dtype=min_dtype,
                 cache_position=cache_position,
                 batch_size=batch_size,
             )
+
+        if num_logits_to_keep is not None:
+            model_inputs["num_logits_to_keep"] = num_logits_to_keep
+
         model_inputs.update(
             {
                 "position_ids": position_ids,
@@ -1102,3 +1138,195 @@ class MERaLiONTextForCausalLM(MERaLiONTextPreTrainedModel):
             }
         )
         return model_inputs
+
+
+@add_start_docstrings(
+    """
+    The MERaLiONText Model transformer with a sequence classification head on top (linear layer).
+
+    [`MERaLiONTextForSequenceClassification`] uses the last token in order to do the classification, as other causal models
+    (e.g. GPT-2) do.
+
+    Since it does classification on the last token, it requires to know the position of the last token. If a
+    `pad_token_id` is defined in the configuration, it finds the last token that is not a padding token in each row. If
+    no `pad_token_id` is defined, it simply takes the last value in each row of the batch. Since it cannot guess the
+    padding tokens when `inputs_embeds` are passed instead of `input_ids`, it does the same (take the last value in
+    each row of the batch).
+    """,
+    MERALION_TEXT_START_DOCSTRING,
+)
+class MERaLiONTextForSequenceClassification(MERaLiONTextPreTrainedModel):
+    def __init__(self, config):
+        super().__init__(config)
+        self.num_labels = config.num_labels
+        self.model = MERaLiONTextModel(config)
+        self.score = nn.Linear(config.hidden_size, self.num_labels, bias=False)
+
+        # Initialize weights and apply final processing
+        self.post_init()
+
+    def get_input_embeddings(self):
+        return self.model.embed_tokens
+
+    def set_input_embeddings(self, value):
+        self.model.embed_tokens = value
+
+    @add_start_docstrings_to_model_forward(MERALION_TEXT_INPUTS_DOCSTRING)
+    def forward(
+        self,
+        input_ids: Optional[torch.LongTensor] = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        position_ids: Optional[torch.LongTensor] = None,
+        past_key_values: Optional[Union[Cache, List[torch.FloatTensor]]] = None,
+        inputs_embeds: Optional[torch.FloatTensor] = None,
+        labels: Optional[torch.LongTensor] = None,
+        use_cache: Optional[bool] = None,
+        output_attentions: Optional[bool] = None,
+        output_hidden_states: Optional[bool] = None,
+        return_dict: Optional[bool] = None,
+    ) -> Union[Tuple, SequenceClassifierOutputWithPast]:
+        r"""
+        labels (`torch.LongTensor` of shape `(batch_size,)`, *optional*):
+            Labels for computing the sequence classification/regression loss. Indices should be in `[0, ...,
+            config.num_labels - 1]`. If `config.num_labels == 1` a regression loss is computed (Mean-Square loss), If
+            `config.num_labels > 1` a classification loss is computed (Cross-Entropy).
+        """
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+        transformer_outputs = self.model(
+            input_ids,
+            attention_mask=attention_mask,
+            position_ids=position_ids,
+            past_key_values=past_key_values,
+            inputs_embeds=inputs_embeds,
+            use_cache=use_cache,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+            return_dict=return_dict,
+        )
+        hidden_states = transformer_outputs[0]
+        logits = self.score(hidden_states)
+
+        if input_ids is not None:
+            batch_size = input_ids.shape[0]
+        else:
+            batch_size = inputs_embeds.shape[0]
+
+        if self.config.pad_token_id is None and batch_size != 1:
+            raise ValueError("Cannot handle batch sizes > 1 if no padding token is defined.")
+        if self.config.pad_token_id is None:
+            sequence_lengths = -1
+        else:
+            if input_ids is not None:
+                # if no pad token found, use modulo instead of reverse indexing for ONNX compatibility
+                sequence_lengths = torch.eq(input_ids, self.config.pad_token_id).int().argmax(-1) - 1
+                sequence_lengths = sequence_lengths % input_ids.shape[-1]
+                sequence_lengths = sequence_lengths.to(logits.device)
+            else:
+                sequence_lengths = -1
+
+        pooled_logits = logits[torch.arange(batch_size, device=logits.device), sequence_lengths]
+
+        loss = None
+        if labels is not None:
+            loss = self.loss_function(logits=logits, labels=labels, pooled_logits=pooled_logits, config=self.config)
+
+        if not return_dict:
+            output = (pooled_logits,) + transformer_outputs[1:]
+            return ((loss,) + output) if loss is not None else output
+
+        return SequenceClassifierOutputWithPast(
+            loss=loss,
+            logits=pooled_logits,
+            past_key_values=transformer_outputs.past_key_values,
+            hidden_states=transformer_outputs.hidden_states,
+            attentions=transformer_outputs.attentions,
+        )
+
+
+@add_start_docstrings(
+    """
+    The MERaLiONText Model transformer with a token classification head on top (a linear layer on top of the hidden-states
+    output) e.g. for Named-Entity-Recognition (NER) tasks.
+    """,
+    MERALION_TEXT_START_DOCSTRING,
+)
+class MERaLiONTextForTokenClassification(MERaLiONTextPreTrainedModel):
+    def __init__(self, config):
+        super().__init__(config)
+        self.num_labels = config.num_labels
+        self.model = MERaLiONTextModel(config)
+        if getattr(config, "classifier_dropout", None) is not None:
+            classifier_dropout = config.classifier_dropout
+        elif getattr(config, "hidden_dropout", None) is not None:
+            classifier_dropout = config.hidden_dropout
+        else:
+            classifier_dropout = 0.1
+        self.dropout = nn.Dropout(classifier_dropout)
+        self.score = nn.Linear(config.hidden_size, config.num_labels)
+
+        # Initialize weights and apply final processing
+        self.post_init()
+
+    def get_input_embeddings(self):
+        return self.model.embed_tokens
+
+    def set_input_embeddings(self, value):
+        self.model.embed_tokens = value
+
+    @add_start_docstrings_to_model_forward(MERALION_TEXT_INPUTS_DOCSTRING)
+    @add_code_sample_docstrings(
+        checkpoint=_CHECKPOINT_FOR_DOC,
+        output_type=TokenClassifierOutput,
+        config_class=_CONFIG_FOR_DOC,
+    )
+    def forward(
+        self,
+        input_ids: Optional[torch.LongTensor] = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        position_ids: Optional[torch.LongTensor] = None,
+        past_key_values: Optional[List[torch.FloatTensor]] = None,
+        inputs_embeds: Optional[torch.FloatTensor] = None,
+        labels: Optional[torch.LongTensor] = None,
+        use_cache: Optional[bool] = None,
+        output_attentions: Optional[bool] = None,
+        output_hidden_states: Optional[bool] = None,
+        return_dict: Optional[bool] = None,
+    ) -> Union[Tuple, TokenClassifierOutput]:
+        r"""
+        labels (`torch.LongTensor` of shape `(batch_size,)`, *optional*):
+            Labels for computing the sequence classification/regression loss. Indices should be in `[0, ...,
+            config.num_labels - 1]`. If `config.num_labels == 1` a regression loss is computed (Mean-Square loss), If
+            `config.num_labels > 1` a classification loss is computed (Cross-Entropy).
+        """
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+        outputs = self.model(
+            input_ids,
+            attention_mask=attention_mask,
+            position_ids=position_ids,
+            past_key_values=past_key_values,
+            inputs_embeds=inputs_embeds,
+            use_cache=use_cache,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+            return_dict=return_dict,
+        )
+        sequence_output = outputs[0]
+        sequence_output = self.dropout(sequence_output)
+        logits = self.score(sequence_output)
+
+        loss = None
+        if labels is not None:
+            loss = self.loss_function(logits, labels, self.config)
+
+        if not return_dict:
+            output = (logits,) + outputs[2:]
+            return ((loss,) + output) if loss is not None else output
+
+        return TokenClassifierOutput(
+            loss=loss,
+            logits=logits,
+            hidden_states=outputs.hidden_states,
+            attentions=outputs.attentions,
+        )