Add some conversion scripts I might ditch

add_missing_imports.py

@@ -0,0 +1,20 @@
+# 2) Add the new classes to the module __init__ (CHECK)
+# 3) Add the new classes to the module autodoc (CHECK)
+# 4) Update the root __init__ to import the new classes (CHECK)
+# 5) Add the missing Auto classes (CHECK)
+# 6) Add the module file to the doctest list
+#   - Happens automatically: Update the model support checklist
+#   - Happens automatically: Update the dummies
+
+from pathlib import Path
+import re
+
+init_file = Path("src/transformers/models/gpt_neo/__init__.py")
+contents = init_file.read_text()
+if "is_tf_available" in contents:
+    raise ValueError("This file already contains TF imports!")
+torch_block_re = r"try:[\n\s]+if not is_torch_available\(\)\:[\n\s]+raise OptionalDependencyNotAvailable.*?else.*?\].*?\]"
+torch_block = re.search(torch_block_re, contents, re.DOTALL)
+torch_block_text = torch_block.group(0)
+breakpoint()
+print()

docs/source/en/model_doc/gpt_neo.mdx

@@ -83,3 +83,18 @@ The `generate()` method can be used to generate text using GPT Neo model.
 
 [[autodoc]] FlaxGPTNeoForCausalLM
     - __call__
+
+## TFGPTNeoModel
+
+[[autodoc]] TFGPTNeoModel
+    - call
+
+## TFGPTNeoForCausalLM
+
+[[autodoc]] TFGPTNeoForCausalLM
+    - call
+
+## TFGPTNeoForSequenceClassification
+
+[[autodoc]] TFGPTNeoForSequenceClassification
+    - call

gpt4_convert.py

@@ -0,0 +1,91 @@
+# What are the steps in conversion?
+
+
+
+# TODO 1: Get GPT to generate the input shapes when it converts a module (not at top-level class!)
+# TODO 2: Port weights from PT to TF versions and do equivalence testing
+# TODO 3:
+
+from pathlib import Path
+import re
+import openai
+from time import sleep
+from tqdm import tqdm
+
+def get_module_name(module_text: str):
+    module_name = re.search(r"(?:class |def )(\w+)", module_text).group(1)
+    return module_name
+
+def translate_fn(module_text: str):
+    system_text = """
+    You are a translation bot designed to translate code in the Hugging Face Transformers library from PyTorch to TensorFlow.
+    
+    You will be passed a single PyTorch function or class from the library. Your goal is to output the equivalent
+    TensorFlow code.
+    
+    There are some guidelines you should follow when translating the code:
+    
+    - When creating layers, please pass their attribute name as the name kwarg.
+    - If the class inherits from PreTrainedModel it should instead inherit from TFPreTrainedModel.
+    - Retain any docstrings attached to methods like forward and translate them, even when the method is being renamed to call.
+    - Layer and model classes should accept **kwargs and pass these to super.init. They should also be renamed by adding "TF" to the start of their name.
+    - You don't need to import anything.
+    - If the class calls other classes in the same module, please add "TF" to the start of their name if required.
+    - TensorFlow layers do not require input shape arguments in the same way as PyTorch layers. As a result, the first
+      argument to the constructor of layers like Dense or Conv2D (but not Embedding) can usually be removed.
+    - TensorFlow Embedding layers do not have a padding_idx argument. Please remove this argument from the constructor.
+    - Prefer the function shape_list(), which returns a list, over methods like tensor.shape or tf.shape(tensor).
+    - Keras layers do not have a register_buffer() method. Instead just set the attribute with that name on the layer directly.
+    - Output classes like BaseModelOutput or SequenceClassifierOutput should have "TF" added to the start of their name.
+    - NumPy operations and calls to .numpy() must be avoided! Use TensorFlow operations instead.
+    """
+    module_name = get_module_name(module_text)
+    if "load_tf_weights" in module_name:
+        print("Skipping", module_name)
+        return ""
+    prompt = [{"role": "system", "content": system_text}, {"role": "user", "content": module_text}]
+    for i in range(5):
+        try:
+            response = openai.ChatCompletion.create(model="gpt-4", messages=prompt, temperature=0, stream=True)
+            break
+        except openai.error.RateLimitError:
+            print(f"Rate limited, retrying ({i + 1} of 5)")
+            sleep(15)
+    else:
+        raise RuntimeError("Rate limited too many times")
+    chunks = []
+    for chunk in tqdm(response, desc=f"Translating {module_name}", dynamic_ncols=True, unit=" tokens"):
+        chunk_message = chunk['choices'][0]['delta']
+        chunks.append(chunk_message)
+    translated_function = ''.join([m.get('content', '') for m in chunks])
+    return translated_function
+
+
+def split_file(source_file: Path):
+    text = source_file.read_text()
+    top_level_fns = list(re.finditer(r"\n\n((?:@|class |def ).*?)(?=\n\n@|\n\nclass |\n\ndef |$)", text, flags=re.DOTALL))
+    for i in range(len(top_level_fns) - 1):
+        assert top_level_fns[i].end() == top_level_fns[i + 1].start()
+    preamble = text[:top_level_fns[0].start()]
+    all_texts = [preamble] + [m.group(0) for m in top_level_fns]
+    for i in range(len(all_texts) - 1):
+        text = all_texts[i]
+        if not text.endswith("\n"):
+            breakpoint()
+    return [text.strip() for text in all_texts]
+
+
+def main():
+    path = Path("src/transformers/models/gpt_neo/modeling_gpt_neo.py")
+    out_path = Path("src/transformers/models/gpt_neo/modeling_tf_gpt_neo.py")
+    split_fns = split_file(path)
+    module_names = [get_module_name(fn) for fn in split_fns[1:]]
+    module_names = [name for name in module_names if "load_tf_weights" not in name]
+    translated_fns = [split_fns[0]]
+    translated_fns += [translate_fn(fn) for fn in split_fns[1:]]
+    output = '\n'.join(translated_fns)
+    out_path.write_text(output)
+
+
+if __name__ == '__main__':
+    main()

gpt4_convert_tests.py

@@ -0,0 +1,107 @@
+# What are the steps in conversion?
+
+# 1) Convert all functions and classes in the model file
+# 2) Add the new classes to the module __init__
+# 3) Add the new classes to the module autodoc
+# 4) Update the root __init__ to import the new classes
+# 5) Add the missing Auto classes
+# 6) Add the module file to the doctest list
+#   - Happens automatically: Update the model support checklist
+#   - Happens automatically: Update the dummies
+
+# TODO 1: Get GPT to generate the input shapes when it converts a module (not at top-level class!)
+# TODO 2: Port weights from PT to TF versions and do equivalence testing
+# TODO 3:
+
+from pathlib import Path
+import re
+import openai
+from time import sleep
+from tqdm import tqdm
+from argparse import ArgumentParser
+
+def get_module_name(module_text: str):
+    module_name = re.search(r"(?:class |def )(\w+)", module_text).group(1)
+    return module_name
+
+def translate_fn(module_text: str):
+    system_text = """
+    You are a translation bot designed to translate code in the Hugging Face Transformers library from PyTorch to TensorFlow.
+    
+    You will be passed a PyTorch class from a tests file. Your goal is to translate this class to test the equivalent TensorFlow model.
+    
+    There are some guidelines you should follow when translating the code:
+    
+    - Most model classes should be renamed by adding "TF" at the start of their name. However, tokenizers, processors and config classes are universal and so not renamed.
+    - You don't need to add any extra imports, you can assume that any other functions or classes you call will be imported for you.
+    - If the class calls other classes in the same module, you can assume that these have already been converted. Please add "TF" to the start of their name if required.
+    - You can use tensor.shape.rank as a TensorFlow replacement for tensor.ndim.
+    - Please use the Hugging Face function shape_list(), which returns a list, instead of tensor.shape or tf.shape(tensor) unless you need to treat the output as a tensor.
+    - PyTorch methods like .detach(), .eval() and .to(device) are not needed in TensorFlow. 
+    - In tests, it's completely acceptable to construct arrays in NumPy instead of TensorFlow.
+    - HuggingFace methods like ids_tensor, floats_tensor and random_attention_mask are universal, so you can use them without changes.
+    - If the test looks like it should work without changes, please simply repeat it back without changes.
+    """
+    module_name = get_module_name(module_text)
+    if "load_tf_weights" in module_name:
+        print("Skipping", module_name)
+        return ""
+    prompt = [{"role": "system", "content": system_text}, {"role": "user", "content": module_text}]
+    for i in range(5):
+        try:
+            response = openai.ChatCompletion.create(model="gpt-4", messages=prompt, temperature=0, stream=True)
+            break
+        except openai.error.RateLimitError:
+            print(f"Rate limited, retrying ({i + 1} of 5)")
+            sleep(15)
+    else:
+        raise RuntimeError("Rate limited too many times")
+    chunks = []
+    for chunk in tqdm(response, desc=f"Translating {module_name}", dynamic_ncols=True, unit=" tokens"):
+        chunk_message = chunk['choices'][0]['delta']
+        chunks.append(chunk_message)
+    translated_function = ''.join([m.get('content', '') for m in chunks])
+    return translated_function
+
+
+def split_file(source_file: Path):
+    text = source_file.read_text()
+    top_level_fns = list(re.finditer(r"\n\n((?:@|class |def ).*?)(?=\n\n@|\n\nclass |\n\ndef |$)", text, flags=re.DOTALL))
+    for i in range(len(top_level_fns) - 1):
+        assert top_level_fns[i].end() == top_level_fns[i + 1].start()
+    preamble = text[:top_level_fns[0].start()]
+    all_texts = [preamble] + [m.group(0) for m in top_level_fns]
+    for i in range(len(all_texts) - 1):
+        class_text = all_texts[i]
+        if len(class_text) > 8000:
+            # Probably a big test class - let's extract methods individually
+            # Remember they're each going to be indented by 4 spaces!
+            methods = list(re.finditer(r"\n    (def .*?)(?=\n    def |$)", class_text, flags=re.DOTALL))
+            method_preamble = class_text[:methods[0].start()]
+            all_class_texts = [method_preamble] + [m.group(0) for m in methods]
+            all_texts[i] = all_class_texts
+        else:
+            all_texts[i] = class_text.strip()
+
+    return all_texts
+
+
+def main():
+    path = Path("tests/models/gpt_neo/test_modeling_gpt_neo.py")
+    out_path = Path("tests/models/gpt_neo/test_modeling_tf_gpt_neo.py")
+    split_classes = split_file(path)
+    translated_classes = [split_classes[0]]
+    for split_class in split_classes[1:]:
+        if isinstance(split_class, list):
+            partial_translation = [split_class[0]]
+            for method in split_class[1:]:
+                partial_translation.append(translate_fn(method))
+            translated_classes.append('\n'.join(partial_translation))
+        else:
+            translated_classes.append(translate_fn(split_class))
+    output = '\n'.join(translated_classes)
+    out_path.write_text(output)
+
+
+if __name__ == '__main__':
+    main()

src/transformers/__init__.py

@@ -3091,6 +3091,16 @@ else:
             "TFFunnelPreTrainedModel",
         ]
     )
+    _import_structure["models.gpt_neo"].extend(
+        [
+            "TF_GPT_NEO_PRETRAINED_MODEL_ARCHIVE_LIST",
+            "TFGPTNeoForCausalLM",
+            "TFGPTNeoForSequenceClassification",
+            "TFGPTNeoModel",
+            "TFGPTNeoMainLayer",
+            "TFGPTNeoPreTrainedModel",
+        ]
+    )
     _import_structure["models.gpt2"].extend(
         [
             "TF_GPT2_PRETRAINED_MODEL_ARCHIVE_LIST",
@@ -6311,6 +6321,14 @@ if TYPE_CHECKING:
             TFFunnelModel,
             TFFunnelPreTrainedModel,
         )
+        from .models.gpt_neo import (
+            TF_GPT_NEO_PRETRAINED_MODEL_ARCHIVE_LIST,
+            TFGPTNeoForCausalLM,
+            TFGPTNeoForSequenceClassification,
+            TFGPTNeoModel,
+            TFGPTNeoPreTrainedModel,
+            TFGPTNeoMainLayer,
+        )
         from .models.gpt2 import (
             TF_GPT2_PRETRAINED_MODEL_ARCHIVE_LIST,
             TFGPT2DoubleHeadsModel,

src/transformers/models/auto/modeling_tf_auto.py

@@ -52,6 +52,7 @@ TF_MODEL_MAPPING_NAMES = OrderedDict(
         ("flaubert", "TFFlaubertModel"),
         ("funnel", ("TFFunnelModel", "TFFunnelBaseModel")),
         ("gpt-sw3", "TFGPT2Model"),
+        ("gpt_neo", "TFGPTNeoModel"),
         ("gpt2", "TFGPT2Model"),
         ("gptj", "TFGPTJModel"),
         ("groupvit", "TFGroupViTModel"),
@@ -172,6 +173,7 @@ TF_MODEL_FOR_CAUSAL_LM_MAPPING_NAMES = OrderedDict(
         ("ctrl", "TFCTRLLMHeadModel"),
         ("gpt-sw3", "TFGPT2LMHeadModel"),
         ("gpt2", "TFGPT2LMHeadModel"),
+        ("gpt-neo", "TFGPTNeoForCausalLM"),
         ("gptj", "TFGPTJForCausalLM"),
         ("openai-gpt", "TFOpenAIGPTLMHeadModel"),
         ("opt", "TFOPTForCausalLM"),
@@ -303,6 +305,7 @@ TF_MODEL_FOR_SEQUENCE_CLASSIFICATION_MAPPING_NAMES = OrderedDict(
         ("flaubert", "TFFlaubertForSequenceClassification"),
         ("funnel", "TFFunnelForSequenceClassification"),
         ("gpt-sw3", "TFGPT2ForSequenceClassification"),
+        ("gpt_neo", "TFGPTNeoForSequenceClassification"),
         ("gpt2", "TFGPT2ForSequenceClassification"),
         ("gptj", "TFGPTJForSequenceClassification"),
         ("layoutlm", "TFLayoutLMForSequenceClassification"),

src/transformers/models/gpt_neo/__init__.py

@@ -13,7 +13,7 @@
 # limitations under the License.
 from typing import TYPE_CHECKING
 
-from ...utils import OptionalDependencyNotAvailable, _LazyModule, is_flax_available, is_torch_available
+from ...utils import OptionalDependencyNotAvailable, _LazyModule, is_flax_available, is_torch_available, is_tf_available
 
 
 _import_structure = {
@@ -35,6 +35,21 @@ else:
         "load_tf_weights_in_gpt_neo",
     ]
 
+try:
+    if not is_tf_available():
+        raise OptionalDependencyNotAvailable()
+except OptionalDependencyNotAvailable:
+    pass
+else:
+    _import_structure["modeling_tf_gpt_neo"] = [
+        "TF_GPT_NEO_PRETRAINED_MODEL_ARCHIVE_LIST"
+        "TFGPTNeoForCausalLM",
+        "TFGPTNeoForSequenceClassification",
+        "TFGPTNeoModel",
+        "TFGPTNeoMainLayer",
+        "TFGPTNeoPreTrainedModel",
+    ]
+
 try:
     if not is_flax_available():
         raise OptionalDependencyNotAvailable()
@@ -65,6 +80,20 @@ if TYPE_CHECKING:
             GPTNeoPreTrainedModel,
             load_tf_weights_in_gpt_neo,
         )
+    try:
+        if not is_tf_available():
+            raise OptionalDependencyNotAvailable()
+    except OptionalDependencyNotAvailable:
+        pass
+    else:
+        from .modeling_tf_gpt_neo import (
+            TF_GPT_NEO_PRETRAINED_MODEL_ARCHIVE_LIST,
+            TFGPTNeoForCausalLM,
+            TFGPTNeoForSequenceClassification,
+            TFGPTNeoModel,
+            TFGPTNeoPreTrainedModel,
+            TFGPTNeoMainLayer,
+        )
 
     try:
         if not is_flax_available():

src/transformers/models/gpt_neo/modeling_tf_gpt_neo.py

@@ -0,0 +1,799 @@
+# coding=utf-8
+# Copyright 2021 The Eleuther AI and HuggingFace Inc. team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+""" PyTorch GPT Neo model."""
+
+
+import os
+from typing import Optional, Tuple, Union
+
+import tensorflow as tf
+
+from ...activations import ACT2FN
+from ...modeling_tf_outputs import (
+    TFBaseModelOutputWithPast,
+    TFBaseModelOutputWithPastAndCrossAttentions,
+    TFCausalLMOutputWithCrossAttentions,
+    TFCausalLMOutputWithPast,
+    TFSequenceClassifierOutputWithPast,
+)
+from ...modeling_tf_utils import TFPreTrainedModel, shape_list
+from ...utils import add_code_sample_docstrings, add_start_docstrings, add_start_docstrings_to_model_forward, logging
+from .configuration_gpt_neo import GPTNeoConfig
+
+
+logger = logging.get_logger(__name__)
+
+_CONFIG_FOR_DOC = "GPTNeoConfig"
+
+GPT_NEO_PRETRAINED_MODEL_ARCHIVE_LIST = [
+    "EleutherAI/gpt-neo-1.3B",
+    # See all GPTNeo models at https://huggingface.co/models?filter=gpt_neo
+]
+
+_CHECKPOINT_FOR_DOC = "EleutherAI/gpt-neo-1.3B"
+
+class TFGPTNeoSelfAttention(tf.keras.layers.Layer):
+    def __init__(self, config, attention_type, **kwargs):
+        super().__init__(**kwargs)
+
+        max_positions = config.max_position_embeddings
+        bias = tf.linalg.band_part(tf.ones((max_positions, max_positions), dtype=tf.bool), -1, 0)
+        bias = tf.reshape(bias, (1, 1, max_positions, max_positions))
+
+        if attention_type == "local":
+            bias = tf.math.logical_xor(bias, tf.linalg.band_part(bias, -config.window_size, 0))
+
+        self.bias = bias
+        self.masked_bias = tf.constant(-1e9)
+
+        self.attn_dropout = tf.keras.layers.Dropout(float(config.attention_dropout))
+        self.resid_dropout = tf.keras.layers.Dropout(float(config.resid_dropout))
+
+        self.embed_dim = config.hidden_size
+        self.num_heads = config.num_heads
+        self.head_dim = self.embed_dim // self.num_heads
+        if self.head_dim * self.num_heads != self.embed_dim:
+            raise ValueError(
+                f"embed_dim must be divisible by num_heads (got `embed_dim`: {self.embed_dim} and `num_heads`:"
+                f" {self.num_heads})."
+            )
+
+        self.k_proj = tf.keras.layers.Dense(self.embed_dim, use_bias=False, name="k_proj")
+        self.v_proj = tf.keras.layers.Dense(self.embed_dim, use_bias=False, name="v_proj")
+        self.q_proj = tf.keras.layers.Dense(self.embed_dim, use_bias=False, name="q_proj")
+        self.out_proj = tf.keras.layers.Dense(self.embed_dim, use_bias=True, name="out_proj")
+
+    def _split_heads(self, tensor, num_heads, attn_head_size):
+        new_shape = tensor.shape[:-1] + (num_heads, attn_head_size)
+        tensor = tf.reshape(tensor, new_shape)
+        return tf.transpose(tensor, perm=[0, 2, 1, 3])
+
+    def _merge_heads(self, tensor, num_heads, attn_head_size):
+        tensor = tf.transpose(tensor, perm=[0, 2, 1, 3])
+        new_shape = tensor.shape[:-2] + (num_heads * attn_head_size,)
+        return tf.reshape(tensor, new_shape)
+
+    def _attn(self, query, key, value, attention_mask=None, head_mask=None):
+        query = tf.cast(query, tf.float32)
+        key = tf.cast(key, tf.float32)
+
+        attn_weights = tf.matmul(query, key, transpose_b=True)
+
+        query_length, key_length = query.shape[-2], key.shape[-2]
+        causal_mask = self.bias[:, :, key_length - query_length : key_length, :key_length]
+        mask_value = tf.float32.min
+        mask_value = tf.constant(mask_value, dtype=attn_weights.dtype)
+        attn_weights = tf.where(causal_mask, attn_weights, mask_value)
+
+        if attention_mask is not None:
+            attn_weights = attn_weights + attention_mask
+
+        attn_weights = tf.nn.softmax(attn_weights, axis=-1)
+        attn_weights = tf.cast(attn_weights, value.dtype)
+        attn_weights = self.attn_dropout(attn_weights)
+
+        if head_mask is not None:
+            attn_weights = attn_weights * head_mask
+
+        attn_output = tf.matmul(attn_weights, value)
+
+        return attn_output, attn_weights
+
+    def call(
+        self,
+        hidden_states,
+        attention_mask=None,
+        layer_past=None,
+        head_mask=None,
+        use_cache=False,
+        output_attentions=False,
+    ):
+        query = self.q_proj(hidden_states)
+        key = self.k_proj(hidden_states)
+        value = self.v_proj(hidden_states)
+
+        query = self._split_heads(query, self.num_heads, self.head_dim)
+        key = self._split_heads(key, self.num_heads, self.head_dim)
+        value = self._split_heads(value, self.num_heads, self.head_dim)
+
+        if layer_past is not None:
+            past_key = layer_past[0]
+            past_value = layer_past[1]
+            key = tf.concat((past_key, key), axis=-2)
+            value = tf.concat((past_value, value), axis=-2)
+
+        if use_cache is True:
+            present = (key, value)
+        else:
+            present = None
+
+        attn_output, attn_weights = self._attn(query, key, value, attention_mask, head_mask)
+
+        attn_output = self._merge_heads(attn_output, self.num_heads, self.head_dim)
+        attn_output = self.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)
+class TFGPTNeoAttention(tf.keras.layers.Layer):
+    def __init__(self, config, layer_id=0, **kwargs):
+        super().__init__(**kwargs)
+        self.layer_id = layer_id
+        self.attention_layers = config.attention_layers
+        self.attention_type = self.attention_layers[layer_id]
+
+        if self.attention_type in ["global", "local"]:
+            self.attention = TFGPTNeoSelfAttention(config, self.attention_type)
+        else:
+            raise NotImplementedError(
+                "Only attn layer types 'global' and 'local' exist, but got `config.attention_layers`: "
+                f"{config.attention_layers}. Select attn layer types from ['global', 'local'] only."
+            )
+
+    def call(
+        self,
+        hidden_states,
+        layer_past=None,
+        attention_mask=None,
+        head_mask=None,
+        use_cache=False,
+        output_attentions=False,
+        **kwargs
+    ):
+        return self.attention(
+            hidden_states,
+            attention_mask=attention_mask,
+            layer_past=layer_past,
+            head_mask=head_mask,
+            use_cache=use_cache,
+            output_attentions=output_attentions,
+        )
+class TFGPTNeoMLP(tf.keras.layers.Layer):
+    def __init__(self, intermediate_size, config, **kwargs):  # in MLP: intermediate_size= 4 * hidden_size
+        super().__init__(**kwargs)
+        embed_dim = config.hidden_size
+        self.c_fc = tf.keras.layers.Dense(intermediate_size, name="c_fc")
+        self.c_proj = tf.keras.layers.Dense(embed_dim, name="c_proj")
+        self.act = ACT2FN[config.activation_function]
+        self.dropout = tf.keras.layers.Dropout(float(config.resid_dropout))
+
+    def call(self, hidden_states):
+        hidden_states = self.c_fc(hidden_states)
+        hidden_states = self.act(hidden_states)
+        hidden_states = self.c_proj(hidden_states)
+        hidden_states = self.dropout(hidden_states)
+        return hidden_states
+class TFGPTNeoBlock(tf.keras.layers.Layer):
+    def __init__(self, config, layer_id, **kwargs):
+        super().__init__(**kwargs)
+        hidden_size = config.hidden_size
+        inner_dim = config.intermediate_size if config.intermediate_size is not None else 4 * hidden_size
+        self.ln_1 = tf.keras.layers.LayerNormalization(epsilon=config.layer_norm_epsilon, name="ln_1")
+        self.attn = TFGPTNeoAttention(config, layer_id, name="attn")
+        self.ln_2 = tf.keras.layers.LayerNormalization(epsilon=config.layer_norm_epsilon, name="ln_2")
+        self.mlp = TFGPTNeoMLP(inner_dim, config, name="mlp")
+
+    def call(
+        self,
+        hidden_states,
+        layer_past=None,
+        attention_mask=None,
+        head_mask=None,
+        use_cache=False,
+        output_attentions=False,
+        training=False,
+    ):
+        residual = hidden_states
+        hidden_states = self.ln_1(hidden_states)
+        attn_outputs = self.attn(
+            hidden_states,
+            layer_past=layer_past,
+            attention_mask=attention_mask,
+            head_mask=head_mask,
+            use_cache=use_cache,
+            output_attentions=output_attentions,
+            training=training,
+        )
+        attn_output = attn_outputs[0]  # output_attn: a, present, (attentions)
+        outputs = attn_outputs[1:]
+        # residual connection
+        hidden_states = attn_output + residual
+
+        residual = hidden_states
+        hidden_states = self.ln_2(hidden_states)
+        feed_forward_hidden_states = self.mlp(hidden_states, training=training)
+        # residual connection
+        hidden_states = residual + feed_forward_hidden_states
+
+        if use_cache:
+            outputs = (hidden_states,) + outputs
+        else:
+            outputs = (hidden_states,) + outputs[1:]
+
+        return outputs  # hidden_states, present, (attentions, cross_attentions)
+class TFGPTNeoPreTrainedModel(TFPreTrainedModel):
+    """
+    An abstract class to handle weights initialization and a simple interface for downloading and loading pretrained
+    models.
+    """
+
+    config_class = GPTNeoConfig
+    base_model_prefix = "transformer"
+    supports_gradient_checkpointing = True
+    _no_split_modules = ["TFGPTNeoBlock"]
+
+    def __init__(self, *inputs, **kwargs):
+        super().__init__(*inputs, **kwargs)
+
+    def _init_weights(self, module):
+        """Initialize the weights."""
+        if isinstance(module, (tf.keras.layers.Dense,)):
+            # Slightly different from the TF version which uses truncated_normal for initialization
+            # cf https://github.com/pytorch/pytorch/pull/5617
+            module.kernel.assign(tf.random.normal(shape=module.kernel.shape, mean=0.0, stddev=self.config.initializer_range))
+            if module.bias is not None:
+                module.bias.assign(tf.zeros(shape=module.bias.shape))
+        elif isinstance(module, tf.keras.layers.Embedding):
+            module.embeddings.assign(tf.random.normal(shape=module.embeddings.shape, mean=0.0, stddev=self.config.initializer_range))
+        elif isinstance(module, tf.keras.layers.LayerNormalization):
+            module.bias.assign(tf.zeros(shape=module.bias.shape))
+            module.gamma.assign(tf.ones(shape=module.gamma.shape))
+
+    def _set_gradient_checkpointing(self, module, value=False):
+        if isinstance(module, TFGPTNeoModel):
+            module.gradient_checkpointing = value
+
+
+GPT_NEO_START_DOCSTRING = r"""
+
+    This model inherits from [`TFPreTrainedModel`]. Check the superclass documentation for the generic methods the
+    library implements for all its model (such as downloading or saving, resizing the input embeddings, pruning heads
+    etc.)
+
+    This model is also a TensorFlow [tf.keras.Model](https://www.tensorflow.org/api_docs/python/tf/keras/Model) subclass.
+    Use it as a regular TensorFlow Model and refer to the TensorFlow documentation for all matter related to general usage
+    and behavior.
+
+    Parameters:
+        config ([`GPTNeoConfig`]): 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 [`~TFPreTrainedModel.from_pretrained`] method to load the model weights.
+"""
+
+GPT_NEO_INPUTS_DOCSTRING = r"""
+    Args:
+        input_ids (`tf.Tensor` of shape `(batch_size, input_ids_length)`):
+            `input_ids_length` = `sequence_length` if `past_key_values` is `None` else
+            `past_key_values[0][0].shape[-2]` (`sequence_length` of input past key value states). Indices of input
+            sequence tokens in the vocabulary.
+
+            If `past_key_values` is used, only `input_ids` that do not have their past calculated should be passed as
+            `input_ids`.
+
+            Indices can be obtained using [`AutoTokenizer`]. See [`PreTrainedTokenizer.encode`] and
+            [`PreTrainedTokenizer.__call__`] for details.
+
+            [What are input IDs?](../glossary#input-ids)
+        past_key_values (`Tuple[Tuple[tf.Tensor]]` of length `config.num_layers`):
+            Contains precomputed hidden-states (key and values in the attention blocks) as computed by the model (see
+            `past_key_values` output below). Can be used to speed up sequential decoding. The `input_ids` which have
+            their past given to this model should not be passed as `input_ids` as they have already been computed.
+        attention_mask (`tf.Tensor` of shape `(batch_size, sequence_length)`, *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 (`tf.Tensor` of shape `(batch_size, input_ids_length)`, *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 (`tf.Tensor` of shape `(batch_size, sequence_length)`, *optional*):
+            Indices of positions of each input sequence tokens in the position embeddings. Selected in the range `[0,
+            config.max_position_embeddings - 1]`.
+
+            [What are position IDs?](../glossary#position-ids)
+        head_mask (`tf.Tensor` of shape `(num_heads,)` or `(num_layers, num_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 (`tf.Tensor` of shape `(batch_size, sequence_length, hidden_size)`, *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.
+
+            If `past_key_values` is used, optionally only the last `inputs_embeds` have to be input (see
+            `past_key_values`).
+        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`).
+        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 GPT Neo Model transformer outputting raw hidden-states without any specific head on top.",
+    GPT_NEO_START_DOCSTRING,
+)
+class TFGPTNeoModel(TFGPTNeoPreTrainedModel):
+    def __init__(self, config, **kwargs):
+        super().__init__(config, **kwargs)
+
+        self.embed_dim = config.hidden_size
+        self.wte = tf.keras.layers.Embedding(config.vocab_size, self.embed_dim, name="wte")
+        self.wpe = tf.keras.layers.Embedding(config.max_position_embeddings, self.embed_dim, name="wpe")
+        self.drop = tf.keras.layers.Dropout(float(config.embed_dropout))
+        self.h = [TFGPTNeoBlock(config, layer_id=i, name=f"h_{i}") for i in range(config.num_layers)]
+        self.ln_f = tf.keras.layers.LayerNormalization(epsilon=config.layer_norm_epsilon, name="ln_f")
+
+        self.gradient_checkpointing = False
+
+    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(GPT_NEO_INPUTS_DOCSTRING)
+    @add_code_sample_docstrings(
+        checkpoint=_CHECKPOINT_FOR_DOC,
+        output_type=TFBaseModelOutputWithPastAndCrossAttentions,
+        config_class=_CONFIG_FOR_DOC,
+    )
+    def call(
+        self,
+        input_ids: Optional[tf.Tensor] = None,
+        past_key_values: Optional[Tuple[tf.Tensor]] = None,
+        attention_mask: Optional[tf.Tensor] = None,
+        token_type_ids: Optional[tf.Tensor] = None,
+        position_ids: Optional[tf.Tensor] = None,
+        head_mask: Optional[tf.Tensor] = None,
+        inputs_embeds: Optional[tf.Tensor] = None,
+        use_cache: Optional[bool] = None,
+        output_attentions: Optional[bool] = None,
+        output_hidden_states: Optional[bool] = None,
+        return_dict: Optional[bool] = None,
+        training: Optional[bool] = None,
+    ) -> Union[Tuple[tf.Tensor], TFBaseModelOutputWithPastAndCrossAttentions]:
+        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 = shape_list(input_ids)
+            input_ids = tf.reshape(input_ids, (-1, input_shape[-1]))
+            batch_size = input_ids.shape[0]
+        elif inputs_embeds is not None:
+            input_shape = shape_list(inputs_embeds)[:-1]
+            batch_size = inputs_embeds.shape[0]
+        else:
+            raise ValueError("You have to specify either input_ids or inputs_embeds")
+
+        if token_type_ids is not None:
+            token_type_ids = tf.reshape(token_type_ids, (-1, input_shape[-1]))
+        if position_ids is not None:
+            position_ids = tf.reshape(position_ids, (-1, input_shape[-1]))
+
+        if past_key_values is None:
+            past_length = 0
+            past_key_values = [None] * len(self.h)
+        else:
+            past_length = past_key_values[0][0].shape[-2]
+
+        if position_ids is None:
+            position_ids = tf.range(past_length, input_shape[-1] + past_length, dtype=tf.int32)
+            position_ids = tf.expand_dims(position_ids, 0)
+            position_ids = tf.reshape(position_ids, (-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 = tf.reshape(attention_mask, (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 = tf.cast(attention_mask, dtype=self.dtype)  # fp16 compatibility
+            attention_mask = (1.0 - attention_mask) * tf.float32.min
+
+        # Prepare head mask if needed
+        # 1.0 in head_mask indicate we keep the head
+        # attention_probs has shape bsz x num_heads x N x N
+        # head_mask has shape n_layer x batch x num_heads x N x N
+        head_mask = self.get_head_mask(head_mask, self.config.num_layers)
+
+        if inputs_embeds is None:
+            inputs_embeds = self.wte(input_ids)
+        position_embeds = self.wpe(position_ids)
+        hidden_states = inputs_embeds + position_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, training=training)
+
+        output_shape = input_shape + (hidden_states.shape[-1],)
+
+        if self.gradient_checkpointing and training:
+            if use_cache:
+                logger.warning_once(
+                    "`use_cache=True` is incompatible with gradient checkpointing. 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,)
+
+            outputs = block(
+                hidden_states,
+                layer_past=layer_past,
+                attention_mask=attention_mask,
+                head_mask=head_mask[i],
+                use_cache=use_cache,
+                output_attentions=output_attentions,
+                training=training,
+            )
+
+            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 = tf.reshape(hidden_states, 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 TFBaseModelOutputWithPast(
+            last_hidden_state=hidden_states,
+            past_key_values=presents,
+            hidden_states=all_hidden_states,
+            attentions=all_self_attentions,
+        )
+@add_start_docstrings(
+    """
+    The GPT Neo Model transformer with a language modeling head on top (linear layer with weights tied to the input
+    embeddings).
+    """,
+    GPT_NEO_START_DOCSTRING,
+)
+class TFGPTNeoForCausalLM(TFGPTNeoPreTrainedModel):
+    _keys_to_ignore_on_load_missing = [
+        r"h\.\d+\.attn\.masked_bias",
+        r"lm_head.weight",
+        r"h\.\d+\.attn\.attention\.bias",
+    ]
+    _keys_to_ignore_on_save = [r"lm_head.weight"]
+
+    def __init__(self, config, **kwargs):
+        super().__init__(config, **kwargs)
+        self.transformer = TFGPTNeoModel(config, name="transformer")
+        self.lm_head = tf.keras.layers.Dense(config.vocab_size, use_bias=False, name="lm_head")
+
+        # 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:]
+            if token_type_ids is not None:
+                token_type_ids = token_type_ids[:, -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 = tf.math.cumsum(tf.cast(attention_mask, tf.int64), axis=-1) - 1
+            position_ids = tf.where(attention_mask == 0, 1, position_ids)
+            if past_key_values:
+                position_ids = position_ids[:, -1:]
+        else:
+            position_ids = None
+        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(GPT_NEO_INPUTS_DOCSTRING)
+    @add_code_sample_docstrings(
+        checkpoint=_CHECKPOINT_FOR_DOC,
+        output_type=TFCausalLMOutputWithCrossAttentions,
+        config_class=_CONFIG_FOR_DOC,
+    )
+    def call(
+        self,
+        input_ids: Optional[tf.Tensor] = None,
+        past_key_values: Optional[Tuple[tf.Tensor]] = None,
+        attention_mask: Optional[tf.Tensor] = None,
+        token_type_ids: Optional[tf.Tensor] = None,
+        position_ids: Optional[tf.Tensor] = None,
+        head_mask: Optional[tf.Tensor] = None,
+        inputs_embeds: Optional[tf.Tensor] = None,
+        labels: Optional[tf.Tensor] = None,
+        use_cache: Optional[bool] = None,
+        output_attentions: Optional[bool] = None,
+        output_hidden_states: Optional[bool] = None,
+        return_dict: Optional[bool] = None,
+        training: bool = False,
+    ) -> Union[Tuple[tf.Tensor], TFCausalLMOutputWithCrossAttentions]:
+        r"""
+        labels (`tf.Tensor` 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,
+            training=training,
+        )
+        hidden_states = transformer_outputs[0]
+
+        lm_logits = self.lm_head(hidden_states)
+
+        loss = None
+        if labels is not None:
+            # 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 = tf.cast(lm_logits, tf.float32)
+
+            # Shift so that tokens < n predict n
+            shift_logits = lm_logits[..., :-1, :]
+            shift_labels = labels[..., 1:]
+            # Flatten the tokens
+            loss_fct = tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True, reduction=tf.keras.losses.Reduction.NONE)
+            loss = loss_fct(shift_labels, shift_logits)
+
+            lm_logits = tf.cast(lm_logits, hidden_states.dtype)
+            loss = tf.cast(loss, hidden_states.dtype)
+
+        if not return_dict:
+            output = (lm_logits,) + transformer_outputs[1:]
+            return ((loss,) + output) if loss is not None else output
+
+        return TFCausalLMOutputWithPast(
+            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[tf.Tensor]], beam_idx: tf.Tensor
+    ) -> Tuple[Tuple[tf.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(tf.gather(past_state, beam_idx) for past_state in layer_past)
+            for layer_past in past_key_values
+        )
+@add_start_docstrings(
+    """
+    The GPTNeo Model transformer with a sequence classification head on top (linear layer).
+
+    [`TFGPTNeoForSequenceClassification`] uses the last token in order to do the classification, as other causal models
+    (e.g. GPT-1) 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).
+    """,
+    GPT_NEO_START_DOCSTRING,
+)
+class TFGPTNeoForSequenceClassification(TFGPTNeoPreTrainedModel):
+    _keys_to_ignore_on_load_missing = [r"h\.\d+\.attn\.masked_bias", r"lm_head.weight"]
+
+    def __init__(self, config, **kwargs):
+        super().__init__(config, **kwargs)
+        self.num_labels = config.num_labels
+        self.transformer = TFGPTNeoModel(config, name="transformer")
+        self.score = tf.keras.layers.Dense(config.hidden_size, self.num_labels, use_bias=False, name="score")
+
+        # Initialize weights and apply final processing
+        self.post_init()
+
+    @add_start_docstrings_to_model_forward(GPT_NEO_INPUTS_DOCSTRING)
+    @add_code_sample_docstrings(
+        checkpoint=_CHECKPOINT_FOR_DOC,
+        output_type=TFSequenceClassifierOutputWithPast,
+        config_class=_CONFIG_FOR_DOC,
+    )
+    def call(
+        self,
+        input_ids: Optional[tf.Tensor] = None,
+        past_key_values: Optional[Tuple[tf.Tensor]] = None,
+        attention_mask: Optional[tf.Tensor] = None,
+        token_type_ids: Optional[tf.Tensor] = None,
+        position_ids: Optional[tf.Tensor] = None,
+        head_mask: Optional[tf.Tensor] = None,
+        inputs_embeds: Optional[tf.Tensor] = None,
+        labels: Optional[tf.Tensor] = None,
+        use_cache: Optional[bool] = None,
+        output_attentions: Optional[bool] = None,
+        output_hidden_states: Optional[bool] = None,
+        return_dict: Optional[bool] = None,
+        **kwargs,
+    ) -> Union[Tuple[tf.Tensor], TFSequenceClassifierOutputWithPast]:
+        r"""
+        labels (`tf.Tensor` 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.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,
+            **kwargs,
+        )
+        hidden_states = transformer_outputs[0]
+        logits = self.score(hidden_states)
+
+        if input_ids is not None:
+            batch_size, sequence_length = shape_list(input_ids)[:2]
+        else:
+            batch_size, sequence_length = shape_list(inputs_embeds)[:2]
+
+        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:
+                sequence_lengths = (tf.math.not_equal(input_ids, self.config.pad_token_id).numpy().sum(-1) - 1)
+            else:
+                sequence_lengths = -1
+                logger.warning(
+                    f"{self.__class__.__name__} will not detect padding tokens in `inputs_embeds`. Results may be "
+                    "unexpected if using padding tokens in conjunction with `inputs_embeds.`"
+                )
+
+        pooled_logits = logits[tf.range(batch_size), sequence_lengths]
+
+        loss = None
+        if labels is not None:
+            if self.config.problem_type is None:
+                if self.num_labels == 1:
+                    self.config.problem_type = "regression"
+                elif self.num_labels > 1 and (labels.dtype == tf.int64 or labels.dtype == tf.int32):
+                    self.config.problem_type = "single_label_classification"
+                else:
+                    self.config.problem_type = "multi_label_classification"
+
+            if self.config.problem_type == "regression":
+                loss_fct = tf.keras.losses.MeanSquaredError()
+                if self.num_labels == 1:
+                    loss = loss_fct(pooled_logits[:, tf.newaxis], labels[:, tf.newaxis])
+                else:
+                    loss = loss_fct(pooled_logits, labels)
+            elif self.config.problem_type == "single_label_classification":
+                loss_fct = tf.keras.losses.SparseCategoricalCrossentropy(from_logits=True)
+                loss = loss_fct(labels, pooled_logits)
+            elif self.config.problem_type == "multi_label_classification":
+                loss_fct = tf.keras.losses.BinaryCrossentropy(from_logits=True)
+                loss = loss_fct(labels, pooled_logits)
+        if not return_dict:
+            output = (pooled_logits,) + transformer_outputs[1:]
+            return ((loss,) + output) if loss is not None else output
+
+        return TFSequenceClassifierOutputWithPast(
+            loss=loss,
+            logits=pooled_logits,
+            past_key_values=transformer_outputs.past_key_values,
+            hidden_states=transformer_outputs.hidden_states,
+            attentions=transformer_outputs.attentions,
+        )