init: release

README.md

@@ -0,0 +1,132 @@
+---
+license: other
+datasets:
+- tiiuae/falcon-refinedweb
+- togethercomputer/RedPajama-Data-1T
+- uonlp/CulturaX
+- CarperAI/pilev2-dev
+- bigcode/starcoderdata
+- DataProvenanceInitiative/Commercially-Verified-Licenses
+language:
+- en
+tags:
+- causal-lm
+---
+# `Stable LM 2 1.6B`
+
+## Model Description
+
+`Stable LM 2 1.6B` is a 1.6 billion parameter decoder-only language model pre-trained on 2 trillion tokens of diverse multilingual and code datasets for two epochs.
+
+## Usage
+
+Get started generating text with `Stable LM 2 1.6B` by using the following code snippet:
+
+```python
+from transformers import AutoModelForCausalLM, AutoTokenizer
+tokenizer = AutoTokenizer.from_pretrained("stabilityai/stablelm-2-1_6b", trust_remote_code=True)
+model = AutoModelForCausalLM.from_pretrained(
+  "stabilityai/stablelm-2-1_6b",
+  trust_remote_code=True,
+  torch_dtype="auto",
+)
+model.cuda()
+inputs = tokenizer("The weather is always wonderful", return_tensors="pt").to(model.device)
+tokens = model.generate(
+  **inputs,
+  max_new_tokens=64,
+  temperature=0.70,
+  top_p=0.95,
+  do_sample=True,
+)
+print(tokenizer.decode(tokens[0], skip_special_tokens=True))
+```
+
+### Run with Flash Attention 2 ⚡️
+
+<details>
+<summary> Click to expand </summary>
+
+```python
+from transformers import AutoModelForCausalLM, AutoTokenizer
+tokenizer = AutoTokenizer.from_pretrained("stabilityai/stablelm-2-1_6b", trust_remote_code=True)
+model = AutoModelForCausalLM.from_pretrained(
+  "stabilityai/stablelm-2-1_6b",
+  trust_remote_code=True,
+  torch_dtype="auto",
+  attn_implementation="flash_attention_2",
+)
+model.cuda()
+inputs = tokenizer("The weather is always wonderful", return_tensors="pt").to(model.device)
+tokens = model.generate(
+  **inputs,
+  max_new_tokens=64,
+  temperature=0.70,
+  top_p=0.95,
+  do_sample=True,
+)
+print(tokenizer.decode(tokens[0], skip_special_tokens=True))
+```
+
+</details>
+
+
+## Model Details
+
+* **Developed by**: [Stability AI](https://stability.ai/)
+* **Model type**: `Stable LM 2 1.6B` models are auto-regressive language models based on the transformer decoder architecture.
+* **Language(s)**: English
+* **Library**: [GPT-NeoX](https://github.com/EleutherAI/gpt-neox)
+* **License**: **Stability AI Non-Commercial Research Community License**. If you'd like to use this model for commercial products or purposes, please contact us [here](https://stability.ai/membership) to learn more.
+* **Contact**: For questions and comments about the model, please email `lm@stability.ai`
+
+### Model Architecture
+
+The model is a decoder-only transformer similar to the LLaMA ([Touvron et al., 2023](https://arxiv.org/abs/2307.09288)) architecture with the following modifications:
+
+| Parameters     | Hidden Size | Layers | Heads | Sequence Length |
+|----------------|-------------|--------|-------|-----------------|
+| 1,644,417,024  | 2048        | 24     | 32    | 4096            |
+
+* **Position Embeddings**: Rotary Position Embeddings ([Su et al., 2021](https://arxiv.org/abs/2104.09864)) applied to the first 25% of head embedding dimensions for improved throughput following [Black et al. (2022)](https://arxiv.org/pdf/2204.06745.pdf).
+* **Normalization**: LayerNorm ([Ba et al., 2016](https://arxiv.org/abs/1607.06450)) with learned bias terms as opposed to RMSNorm ([Zhang & Sennrich, 2019](https://arxiv.org/abs/1910.07467)).
+* **Biases**: We remove all bias terms from the model except for attention Q,K,V projections ([Bai et al., 2023](https://arxiv.org/abs/2309.16609)).
+* **Tokenizer**: We use Arcade100k, a BPE tokenizer extended from OpenAI's [`tiktoken.cl100k_base`](https://github.com/openai/tiktoken). We split digits into individual tokens following findings by [Liu & Low (2023)](https://arxiv.org/abs/2305.14201).
+
+## Training
+
+### Training Dataset
+
+The dataset is comprised of a filtered mixture of open-source large-scale datasets available on the [HuggingFace Hub](https://huggingface.co/datasets): Falcon RefinedWeb extract ([Penedo et al., 2023](https://huggingface.co/datasets/tiiuae/falcon-refinedweb)), RedPajama-Data ([Together Computer., 2023](https://github.com/togethercomputer/RedPajama-Data)) and The Pile ([Gao et al., 2020](https://arxiv.org/abs/2101.00027)) both without the *Books3* subset, and StarCoder ([Li et al., 2023](https://arxiv.org/abs/2305.06161)). We further supplement our training with multi-lingual data from CulturaX ([Nguyen et al., 2023](https://arxiv.org/abs/2309.09400)) and, in particular, from its OSCAR corpora, as well as restructured data in the style of [Yuan & Liu (2022)](https://arxiv.org/abs/2206.11147).
+
+* Given the large amount of web data, we recommend fine-tuning the base `Stable LM 2 1.6B` for your downstream tasks.
+
+### Training Procedure
+
+The model is pre-trained on the aforementioned datasets in `bfloat16` precision, optimized with AdamW, and trained using the NeoX tokenizer with a vocabulary size of 100,352. We outline the complete hyperparameters choices in the project's [GitHub repository - config*](https://github.com/Stability-AI/StableLM/blob/main/configs/stablelm-2-1.6b.yml). The final checkpoint of pre-training, before cooldown, is provided in the `global_step420000` [branch](https://huggingface.co/stabilityai/stablelm-2-1_6b/blob/global_step420000/README.md).
+
+### Training Infrastructure
+
+* **Hardware**: `Stable LM 2 1.6B` was trained on the Stability AI cluster across 512 NVIDIA A100 40GB GPUs (AWS P4d instances).
+
+* **Software**: We use a fork of `gpt-neox` ([EleutherAI, 2021](https://github.com/EleutherAI/gpt-neox)), train under 2D parallelism (Data and Tensor Parallel) with ZeRO-1 ([Rajbhandari et al., 2019](https://arxiv.org/abs/1910.02054v3)), and rely on flash-attention as well as SwiGLU and Rotary Embedding kernels from FlashAttention-2 ([Dao et al., 2023](https://tridao.me/publications/flash2/flash2.pdf))
+
+## Use and Limitations
+
+### Intended Use
+
+The model is intended to be used as a foundational base model for application-specific fine-tuning. Developers must evaluate and fine-tune the model for safe performance in downstream applications.
+
+### Limitations and Bias
+​
+As a base model, this model may exhibit unreliable, unsafe, or other undesirable behaviors that must be corrected through evaluation and fine-tuning prior to deployment. The pre-training dataset may have contained offensive or inappropriate content, even after applying data cleansing filters, which can be reflected in the model-generated text. We recommend that users exercise caution when using these models in production systems. Do not use the models if they are unsuitable for your application, or for any applications that may cause deliberate or unintentional harm to others.
+
+## How to Cite
+
+```bibtex
+@misc{StableLM-2-1.6B,
+      url={[https://huggingface.co/stabilityai/stablelm-2-1.6b](https://huggingface.co/stabilityai/stablelm-2-1.6b)},
+      title={Stable LM 2 1.6B},
+      author={Stability AI Language Team}
+}
+```

config.json

@@ -0,0 +1,31 @@
+{
+  "architectures": [
+    "StableLMEpochForCausalLM"
+  ],
+  "auto_map": {
+    "AutoConfig": "configuration_stablelm_epoch.StableLMEpochConfig",
+    "AutoModelForCausalLM": "modeling_stablelm_epoch.StableLMEpochForCausalLM"
+  },
+  "bos_token_id": 100257,
+  "eos_token_id": 100257,
+  "hidden_act": "silu",
+  "hidden_size": 2048,
+  "initializer_range": 0.02,
+  "intermediate_size": 5632,
+  "max_position_embeddings": 4096,
+  "model_type": "stablelm_epoch",
+  "norm_eps": 1e-05,
+  "num_attention_heads": 32,
+  "num_heads": 32,
+  "num_hidden_layers": 24,
+  "num_key_value_heads": 32,
+  "rope_pct": 0.25,
+  "rope_theta": 10000,
+  "rotary_scaling_factor": 1.0,
+  "tie_word_embeddings": false,
+  "torch_dtype": "bfloat16",
+  "transformers_version": "4.36.2",
+  "use_cache": true,
+  "use_qkv_bias": true,
+  "vocab_size": 100352
+}

configuration_stablelm_epoch.py

@@ -0,0 +1,113 @@
+# Copyright 2023 Stability and The 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.
+""" StableLM Epoch model configuration"""
+from transformers import PretrainedConfig
+from transformers.utils import logging
+
+
+logger = logging.get_logger(__name__)
+
+
+class StableLMEpochConfig(PretrainedConfig):
+    r"""
+    Configuration objects inherit from [`PretrainedConfig`] and can be used to control the model outputs. Read the
+    documentation from [`PretrainedConfig`] for more information.
+
+    Args:
+        vocab_size (`int`, *optional*, defaults to 50_304):
+            Vocabulary size of the StableLM model. Defines the number of different tokens that
+            can be represented by the `inputs_ids` passed when calling [`StableLMEpochModel`].
+        intermediate_size (`int`, *optional*, defaults to 6912):
+            Dimension of the MLP representations.
+        hidden_size (`int`, *optional*, defaults to 2560):
+            Dimension of the decoder layers and the pooler layer.
+        num_hidden_layers (`int`, *optional*, defaults to 32):
+            Number of hidden layers in the Transformer decoder.
+        num_attention_heads (`int`, *optional*, defaults to 32):
+            Number of attention heads for each attention layer in the Transformer encoder.
+        num_key_value_heads (`int`, *optional*):
+            This is the number of key_value heads that should be used to implement Grouped Query Attention. If
+            `num_key_value_heads=num_attention_heads`, the model will use Multi Head Attention (MHA), if
+            `num_key_value_heads=1 the model will use Multi Query Attention (MQA) otherwise GQA is used. When
+            converting a multi-head checkpoint to a GQA checkpoint, each group key and value head should be constructed
+            by meanpooling all the original heads within that group. For more details checkout [this
+            paper](https://arxiv.org/pdf/2305.13245.pdf). If it is not specified, will default to
+            `num_attention_heads`.
+        hidden_act (`str` or `function`, *optional*, defaults to `"silu"`):
+            The non-linear activation function (function or string).
+        rope_pct (`float`, *optional*, defaults to 1.0):
+            Percentage of hidden dimensions to allocate to rotary embeddings.
+        rope_theta (`float`, *optional*, defaults to 10000.0):
+            The base period of the RoPE embeddings.
+        max_position_embeddings (`int`, *optional*, defaults to 2048):
+            The maximum sequence length that this model might ever be used with.
+            Typically set this to something large just in case (e.g., 512 or 1024 or 2048).
+        initializer_range (`float`, *optional*, defaults to 1e-5):
+            The standard deviation of the truncated_normal_initializer for initializing
+             all weight matrices.
+        norm_eps (`float`, *optional*, defaults to 1e-8):
+            The epsilon used by the normalization layers.
+        use_cache (`bool`, *optional*, defaults to `True`):
+            Whether or not the model should return the last key/values attentions
+            (not used by all models). Only relevant if `config.is_decoder=True`.
+        use_qkv_bias (`bool`, *optional*, defaults to `True`):
+            Whether or not the model should use bias for qkv layers.
+        tie_word_embeddings(`bool`, *optional*, defaults to `False`):
+            Whether to tie weight embeddings
+    """
+    model_type = "stablelm_epoch"
+    keys_to_ignore_at_inference = ["past_key_values"]
+
+    def __init__(
+        self,
+        vocab_size=50_304,
+        intermediate_size=6912,
+        hidden_size=2560,
+        num_hidden_layers=32,
+        num_attention_heads=32,
+        num_key_value_heads=32,
+        hidden_act="silu",
+        rope_pct=0.25,
+        rope_theta=10_000,
+        max_position_embeddings=4096,
+        initializer_range=0.02,
+        norm_eps=1.0e-5,
+        use_cache=True,
+        use_qkv_bias=True,
+        bos_token_id=0,
+        eos_token_id=2,
+        tie_word_embeddings=False,
+        **kwargs,
+    ):
+        self.vocab_size = vocab_size
+        self.max_position_embeddings = max_position_embeddings
+        self.intermediate_size = intermediate_size
+        self.hidden_size = hidden_size
+        self.num_hidden_layers = num_hidden_layers
+        self.num_attention_heads = num_attention_heads
+        self.num_key_value_heads = num_key_value_heads
+        self.hidden_act = hidden_act
+        self.rope_pct = rope_pct
+        self.rope_theta = rope_theta
+        self.initializer_range = initializer_range
+        self.norm_eps = norm_eps
+        self.use_cache = use_cache
+        self.use_qkv_bias = use_qkv_bias
+        self.tie_word_embeddings = tie_word_embeddings
+        super().__init__(
+            bos_token_id=bos_token_id,
+            eos_token_id=eos_token_id,
+            tie_word_embeddings=tie_word_embeddings,
+            **kwargs,
+        )

generation_config.json

@@ -0,0 +1,6 @@
+{
+  "_from_model_config": true,
+  "bos_token_id": 100257,
+  "eos_token_id": 100257,
+  "transformers_version": "4.36.2"
+}

model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8bdf317e2b35ab5c8009cbb6c7ce495e4e608a6b9b843d44054edf25b8c5860d
+size 3289069520

modeling_stablelm_epoch.py

@@ -0,0 +1,917 @@
+# coding=utf-8
+# Copyright 2023 Stability AI, EleutherAI, and The 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.
+#
+# This code is based off the following work:
+# https://github.com/huggingface/transformers/blob/main/src/transformers/models/llama/modeling_llama.py
+# https://github.com/huggingface/transformers/blob/main/src/transformers/models/gpt_neox/modeling_gpt_neox.py
+""" PyTorch StableLM Epoch model. """
+from typing import Optional, Tuple, Union
+import math
+import warnings
+
+import torch
+import torch.nn.functional as F
+import torch.utils.checkpoint
+from torch import nn
+from torch.nn import CrossEntropyLoss
+
+from transformers.cache_utils import Cache
+from transformers.modeling_outputs import (
+    BaseModelOutputWithPast,
+    CausalLMOutputWithPast,
+)
+from transformers.modeling_utils import PreTrainedModel
+from transformers.utils import logging, is_flash_attn_greater_or_equal_2_10
+
+from .configuration_stablelm_epoch import StableLMEpochConfig
+
+try:
+    from flash_attn import flash_attn_func, flash_attn_varlen_func
+    from flash_attn.bert_padding import index_first_axis, pad_input, unpad_input
+except:
+    flash_attn_func, flash_attn_varlen_func = None, None
+    index_first_axis, pad_input, unpad_input = None, None, None
+
+
+logger = logging.get_logger(__name__)
+
+
+# Copied from transformers.models.llama.modeling_llama._get_unpad_data
+def _get_unpad_data(attention_mask):
+    seqlens_in_batch = attention_mask.sum(dim=-1, dtype=torch.int32)
+    indices = torch.nonzero(attention_mask.flatten(), as_tuple=False).flatten()
+    max_seqlen_in_batch = seqlens_in_batch.max().item()
+    cu_seqlens = F.pad(torch.cumsum(seqlens_in_batch, dim=0, dtype=torch.torch.int32), (1, 0))
+    return (
+        indices,
+        cu_seqlens,
+        max_seqlen_in_batch,
+    )
+
+
+# Copied from transformers.models.bart.modeling_bart._make_causal_mask
+def _make_causal_mask(
+    input_ids_shape: torch.Size,
+    dtype: torch.dtype,
+    device: torch.device,
+    past_key_values_length: int = 0,
+):
+    """Make causal mask used for bi-directional self-attention."""
+    batch_size, tgt_len = input_ids_shape
+    mask = torch.full((tgt_len, tgt_len), torch.finfo(torch.float16).min, device=device)
+    mask_cond = torch.arange(mask.size(-1), device=device)
+    mask.masked_fill_(mask_cond < (mask_cond + 1).view(mask.size(-1), 1), 0)
+    mask = mask.to(dtype)
+    if past_key_values_length > 0:
+        mask = torch.cat([torch.zeros(tgt_len, past_key_values_length, dtype=dtype, device=device), mask], dim=-1)
+    return mask[None, None, :, :].expand(batch_size, 1, tgt_len, tgt_len + past_key_values_length)
+
+
+# Copied from transformers.models.bart.modeling_bart._expand_mask
+def _expand_mask(mask: torch.Tensor, dtype: torch.dtype, tgt_len: Optional[int] = None):
+    """Expands attention_mask from `[batch_size, seq_len]` to `[batch_size, 1, tgt_seq_len, src_seq_len]`."""
+    batch_size, src_len = mask.size()
+    tgt_len = tgt_len if tgt_len is not None else src_len
+
+    expanded_mask = mask[:, None, None, :].expand(batch_size, 1, tgt_len, src_len).to(dtype)
+    inverted_mask = 1.0 - expanded_mask
+
+    return inverted_mask.masked_fill(
+        inverted_mask.to(torch.bool), torch.finfo(dtype).min
+    )
+
+
+class RotaryEmbedding(nn.Module):
+    def __init__(
+        self,
+        dim: int,
+        max_position_embeddings: int,
+        base: int = 10_000,
+        device: Optional[torch.device] = None,
+    ):
+        super().__init__()
+
+        self.dim = dim
+        self.max_position_embeddings = max_position_embeddings
+        self.base = base
+        inv_freq = 1.0 / (self.base ** (torch.arange(0, self.dim, 2, device=device, dtype=torch.float32) / self.dim))
+        self.register_buffer("inv_freq", inv_freq, persistent=False)
+
+        # Build here to make `torch.jit.trace` work.
+        self._set_cos_sin_cache(
+            seq_len=max_position_embeddings, device=self.inv_freq.device, dtype=torch.get_default_dtype(),
+        )
+
+    def _set_cos_sin_cache(self, seq_len: int, device: torch.device, dtype: torch.dtype):
+        self.max_seq_len_cached = seq_len
+        t = torch.arange(self.max_seq_len_cached, device=device, dtype=torch.float32)
+
+        # Don't do einsum, it converts fp32 to fp16 under AMP
+        # freqs = torch.einsum("i,j->ij", t, self.inv_freq)
+        freqs = torch.outer(t, self.inv_freq)
+        # Different from paper, but it uses a different permutation in order to obtain the same calculation
+        emb = torch.cat((freqs, freqs), dim=-1)
+        self.register_buffer("cos_cached", emb.cos()[None, None, :, :].to(dtype), persistent=False)
+        self.register_buffer("sin_cached", emb.sin()[None, None, :, :].to(dtype), persistent=False)
+
+    def forward(self, x: torch.Tensor, seq_len: Optional[int] = None):
+        # x: [batch_size, num_heads, seq_len, head_size]
+        if seq_len > self.max_seq_len_cached:
+            self._set_cos_sin_cache(seq_len=seq_len, device=x.device, dtype=torch.get_default_dtype())
+        return (
+            self.cos_cached[:, :, :seq_len, ...].to(dtype=x.dtype),
+            self.sin_cached[:, :, :seq_len, ...].to(dtype=x.dtype),
+        )
+
+
+def rotate_half(x: torch.Tensor):
+    """Rotates half the hidden dims of the input."""
+    x1, x2 = torch.chunk(x, 2, dim=-1)
+    return torch.cat((-x2, x1), dim=-1)
+
+
+def apply_rotary_pos_emb(q, k, cos, sin, position_ids):
+    # The first two dimensions of cos and sin are always 1, so we can `squeeze` them.
+    cos = cos.squeeze(1).squeeze(0)  # [seq_len, dim]
+    sin = sin.squeeze(1).squeeze(0)  # [seq_len, dim]
+    cos = cos[position_ids].unsqueeze(1)  # [batch_size, 1, seq_len, dim]
+    sin = sin[position_ids].unsqueeze(1)  # [batch_size, 1, seq_len, dim]
+    q_embed = (q * cos) + (rotate_half(q) * sin)
+    k_embed = (k * cos) + (rotate_half(k) * sin)
+    return q_embed, k_embed
+
+
+class MLP(nn.Module):
+    def __init__(self, config: StableLMEpochConfig):
+        super().__init__()
+        self.config = config
+        self.hidden_size = config.hidden_size
+        self.intermediate_size = config.intermediate_size
+        self.gate_proj = nn.Linear(config.hidden_size, config.intermediate_size, bias=False)
+        self.up_proj = nn.Linear(config.hidden_size, config.intermediate_size, bias=False)
+        self.down_proj = nn.Linear(config.intermediate_size, config.hidden_size, bias=False)
+        self.act_fn = nn.SiLU()
+
+    def forward(self, x: torch.Tensor) -> torch.Tensor:
+        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,
+    num_key_value_heads, seqlen, head_dim) to (batch, num_attention_heads, seqlen, head_dim)
+    """
+    batch, num_key_value_heads, slen, head_dim = hidden_states.shape
+    if n_rep == 1:
+        return hidden_states
+    hidden_states = hidden_states[:, :, None, :, :].expand(batch, num_key_value_heads, n_rep, slen, head_dim)
+    return hidden_states.reshape(batch, num_key_value_heads * n_rep, slen, head_dim)
+
+
+class Attention(nn.Module):
+    def __init__(self, config: StableLMEpochConfig):
+        super().__init__()
+        self.config = config
+        self.hidden_size = config.hidden_size
+        self.num_heads = config.num_attention_heads
+        self.head_dim = self.hidden_size // self.num_heads
+        self.num_key_value_heads = config.num_key_value_heads
+        self.num_key_value_groups = self.num_heads // self.num_key_value_heads
+        self.max_position_embeddings = config.max_position_embeddings
+        self.is_causal = True
+
+        if (self.head_dim * self.num_heads) != self.hidden_size:
+            raise ValueError(
+                f"hidden_size must be divisible by num_heads (got `hidden_size`: {self.hidden_size}"
+                f" and `num_heads`: {self.num_heads})."
+            )
+
+        self.q_proj = nn.Linear(self.hidden_size, self.num_heads * self.head_dim, bias=config.use_qkv_bias)
+        self.k_proj = nn.Linear(self.hidden_size, self.num_key_value_heads * self.head_dim, bias=config.use_qkv_bias)
+        self.v_proj = nn.Linear(self.hidden_size, self.num_key_value_heads * self.head_dim, bias=config.use_qkv_bias)
+        self.o_proj = nn.Linear(self.hidden_size, self.hidden_size, bias=False)
+
+        self._init_rope()
+
+    def _init_rope(self):
+        self.rotary_ndims = int(self.head_dim * self.config.rope_pct)
+        self.rotary_emb = RotaryEmbedding(
+            self.rotary_ndims,
+            max_position_embeddings=self.config.max_position_embeddings,
+            base=self.config.rope_theta,
+        )
+
+    def forward(
+        self,
+        hidden_states: torch.FloatTensor,
+        attention_mask: torch.FloatTensor,
+        position_ids: torch.LongTensor,
+        past_key_value: Optional[Tuple[torch.Tensor]] = None,
+        output_attentions: Optional[bool] = False,
+        use_cache: Optional[bool] = False,
+    ) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor]]]:
+        bsz, q_len, _ = hidden_states.size()
+
+        query_states = self.q_proj(hidden_states)
+        key_states = self.k_proj(hidden_states)
+        value_states = self.v_proj(hidden_states)
+
+        query_states = query_states.view(bsz, q_len, self.num_heads, self.head_dim).transpose(1, 2)
+        key_states = key_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2)
+        value_states = value_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2)
+
+        query_rot = query_states[..., : self.rotary_ndims]
+        query_pass = query_states[..., self.rotary_ndims :]
+        key_rot = key_states[..., : self.rotary_ndims]
+        key_pass = key_states[..., self.rotary_ndims :]
+
+        kv_seq_len = key_states.shape[-2]
+        if past_key_value is not None:
+            kv_seq_len += past_key_value[0].shape[-2]
+        cos, sin = self.rotary_emb(value_states, seq_len=kv_seq_len)
+        query_states, key_states = apply_rotary_pos_emb(query_rot, key_rot, cos, sin, position_ids)
+
+        # [batch_size, num_heads, seq_len, head_dim]
+        query_states = torch.cat((query_states, query_pass), dim=-1)
+        key_states = torch.cat((key_states, key_pass), dim=-1)
+
+        if past_key_value is not None:
+            # Reuse k, v, self_attention
+            key_states = torch.cat((past_key_value[0], key_states), dim=2)
+            value_states = torch.cat((past_key_value[1], value_states), dim=2)
+
+        past_key_value = (key_states, value_states) if use_cache else None
+
+        # Repeat k/v heads if n_kv_heads < n_heads
+        key_states = repeat_kv(key_states, self.num_key_value_groups)
+        value_states = repeat_kv(value_states, self.num_key_value_groups)
+
+        attn_weights = torch.matmul(query_states, key_states.transpose(2, 3)) / math.sqrt(self.head_dim)
+
+        if attn_weights.size() != (bsz, self.num_heads, q_len, kv_seq_len):
+            raise ValueError(
+                f"Attention weights should be of size {(bsz, self.num_heads, q_len, kv_seq_len)}, but is"
+                f" {attn_weights.size()}"
+            )
+
+        if attention_mask is not None:
+            if attention_mask.size() != (bsz, 1, q_len, kv_seq_len):
+                raise ValueError(
+                    f"Attention mask should be of size {(bsz, 1, q_len, kv_seq_len)}, but is {attention_mask.size()}"
+                )
+            attn_weights = attn_weights + attention_mask
+
+        # Upcast attention to fp32
+        attn_weights = nn.functional.softmax(attn_weights, dim=-1, dtype=torch.float32).to(query_states.dtype)
+        attn_output = torch.matmul(attn_weights, value_states)
+
+        if attn_output.size() != (bsz, self.num_heads, q_len, self.head_dim):
+            raise ValueError(
+                f"`attn_output` should be of size {(bsz, self.num_heads, q_len, self.head_dim)}, but is"
+                f" {attn_output.size()}"
+            )
+
+        # Merge heads
+        attn_output = attn_output.transpose(1, 2).contiguous()
+        attn_output = attn_output.reshape(bsz, q_len, self.hidden_size)
+
+        # Final linear projection
+        attn_output = self.o_proj(attn_output)
+
+        if not output_attentions:
+            attn_weights = None
+
+        return attn_output, attn_weights, past_key_value
+
+
+class FlashAttention2(Attention):
+    """
+    Reference: https://github.com/huggingface/transformers/blob/5d36025ca13d05151b7a0c761e90d429c4644a30/src/transformers/models/llama/modeling_llama.py#L456
+    """
+
+    def __init__(self, *args, **kwargs):
+        super().__init__(*args, **kwargs)
+
+        # TODO: Should be removed once Flash Attention for RoCm is bumped to 2.1.
+        # flash_attn<2.1 generates top-left aligned causal mask, while what is needed here is bottom-right alignement, that was made default for flash_attn>=2.1. This attribute is used to handle this difference. Reference: https://github.com/Dao-AILab/flash-attention/releases/tag/v2.1.0.
+        # Beware that with flash_attn<2.1, using q_seqlen != k_seqlen (except for the case q_seqlen == 1) produces a wrong mask (top-left).
+        self._flash_attn_uses_top_left_mask = not is_flash_attn_greater_or_equal_2_10()
+
+    def forward(
+        self,
+        hidden_states: torch.Tensor,
+        attention_mask: Optional[torch.LongTensor] = None,
+        position_ids: Optional[torch.LongTensor] = None,
+        past_key_value: Optional[Cache] = None,
+        output_attentions: bool = False,
+        use_cache: bool = False,
+        **kwargs,
+    ) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor]]]:
+        # FlashAttention2 attention does not support output_attentions
+        if "padding_mask" in kwargs:
+            warnings.warn(
+                "Passing `padding_mask` is deprecated and will be removed in v4.37. Please make sure use `attention_mask` instead.`"
+            )
+
+            # overwrite attention_mask with padding_mask
+            attention_mask = kwargs.pop("padding_mask")
+
+        output_attentions = False
+
+        bsz, q_len, _ = hidden_states.size()
+
+        query_states = self.q_proj(hidden_states)
+        key_states = self.k_proj(hidden_states)
+        value_states = self.v_proj(hidden_states)
+
+        # Flash attention requires the input to have the shape
+        # batch_size x seq_length x head_dim x hidden_dim
+        # therefore we just need to keep the original shape
+        query_states = query_states.view(bsz, q_len, self.num_heads, self.head_dim).transpose(1, 2)
+        key_states = key_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2)
+        value_states = value_states.view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2)
+
+        query_rot = query_states[..., : self.rotary_ndims]
+        query_pass = query_states[..., self.rotary_ndims :]
+        key_rot = key_states[..., : self.rotary_ndims]
+        key_pass = key_states[..., self.rotary_ndims :]
+
+        kv_seq_len = key_states.shape[-2]
+        if past_key_value is not None:
+            kv_seq_len += past_key_value[0].shape[-2]
+        cos, sin = self.rotary_emb(value_states, seq_len=kv_seq_len)
+        query_states, key_states = apply_rotary_pos_emb(query_rot, key_rot, cos, sin, position_ids)
+
+        # [batch_size, num_heads, seq_len, head_dim]
+        query_states = torch.cat((query_states, query_pass), dim=-1)
+        key_states = torch.cat((key_states, key_pass), dim=-1)
+
+        if past_key_value is not None:
+            # Reuse k, v, self_attention
+            key_states = torch.cat((past_key_value[0], key_states), dim=2)
+            value_states = torch.cat((past_key_value[1], value_states), dim=2)
+
+        past_key_value = (key_states, value_states) if use_cache else None
+
+        # TODO: These transpose are quite inefficient but Flash Attention requires the layout [batch_size, sequence_length, num_heads, head_dim]. We would need to refactor the KV cache
+        # to be able to avoid many of these transpose/reshape/view.
+        query_states = query_states.transpose(1, 2)
+        key_states = key_states.transpose(1, 2)
+        value_states = value_states.transpose(1, 2)
+
+        dropout_rate = self.attention_dropout if self.training else 0.0
+
+        attn_output = self._flash_attention_forward(
+            query_states, key_states, value_states, attention_mask, q_len, dropout=dropout_rate
+        )
+        attn_output = attn_output.reshape(bsz, q_len, self.hidden_size).contiguous()
+        attn_output = self.o_proj(attn_output)
+
+        if not output_attentions:
+            attn_weights = None
+
+        return attn_output, attn_weights, past_key_value
+
+    def _flash_attention_forward(
+        self, query_states, key_states, value_states, attention_mask, query_length, dropout=0.0, softmax_scale=None
+    ):
+        """
+        Calls the forward method of Flash Attention - if the input hidden states contain at least one padding token
+        first unpad the input, then computes the attention scores and pad the final attention scores.
+
+        Args:
+            query_states (`torch.Tensor`):
+                Input query states to be passed to Flash Attention API
+            key_states (`torch.Tensor`):
+                Input key states to be passed to Flash Attention API
+            value_states (`torch.Tensor`):
+                Input value states to be passed to Flash Attention API
+            attention_mask (`torch.Tensor`):
+                The padding mask - corresponds to a tensor of size `(batch_size, seq_len)` where 0 stands for the
+                position of padding tokens and 1 for the position of non-padding tokens.
+            dropout (`int`, *optional*):
+                Attention dropout
+            softmax_scale (`float`, *optional*):
+                The scaling of QK^T before applying softmax. Default to 1 / sqrt(head_dim)
+        """
+        if not self._flash_attn_uses_top_left_mask:
+            causal = self.is_causal
+        else:
+            # TODO: Remove the `query_length != 1` check once Flash Attention for RoCm is bumped to 2.1. For details, please see the comment in FlashAttention2 __init__.
+            causal = self.is_causal and query_length != 1
+
+        # Contains at least one padding token in the sequence
+        if attention_mask is not None:
+            batch_size = query_states.shape[0]
+            query_states, key_states, value_states, indices_q, cu_seq_lens, max_seq_lens = self._upad_input(
+                query_states, key_states, value_states, attention_mask, query_length
+            )
+
+            cu_seqlens_q, cu_seqlens_k = cu_seq_lens
+            max_seqlen_in_batch_q, max_seqlen_in_batch_k = max_seq_lens
+
+            attn_output_unpad = flash_attn_varlen_func(
+                query_states,
+                key_states,
+                value_states,
+                cu_seqlens_q=cu_seqlens_q,
+                cu_seqlens_k=cu_seqlens_k,
+                max_seqlen_q=max_seqlen_in_batch_q,
+                max_seqlen_k=max_seqlen_in_batch_k,
+                dropout_p=dropout,
+                softmax_scale=softmax_scale,
+                causal=causal,
+            )
+
+            attn_output = pad_input(attn_output_unpad, indices_q, batch_size, query_length)
+        else:
+            attn_output = flash_attn_func(
+                query_states, key_states, value_states, dropout, softmax_scale=softmax_scale, causal=causal
+            )
+
+        return attn_output
+
+    def _upad_input(self, query_layer, key_layer, value_layer, attention_mask, query_length):
+        indices_k, cu_seqlens_k, max_seqlen_in_batch_k = _get_unpad_data(attention_mask)
+        batch_size, kv_seq_len, num_key_value_heads, head_dim = key_layer.shape
+
+        key_layer = index_first_axis(
+            key_layer.reshape(batch_size * kv_seq_len, num_key_value_heads, head_dim), indices_k
+        )
+        value_layer = index_first_axis(
+            value_layer.reshape(batch_size * kv_seq_len, num_key_value_heads, head_dim), indices_k
+        )
+        if query_length == kv_seq_len:
+            query_layer = index_first_axis(
+                query_layer.reshape(batch_size * kv_seq_len, self.num_heads, head_dim), indices_k
+            )
+            cu_seqlens_q = cu_seqlens_k
+            max_seqlen_in_batch_q = max_seqlen_in_batch_k
+            indices_q = indices_k
+        elif query_length == 1:
+            max_seqlen_in_batch_q = 1
+            cu_seqlens_q = torch.arange(
+                batch_size + 1, dtype=torch.int32, device=query_layer.device
+            )  # There is a memcpy here, that is very bad.
+            indices_q = cu_seqlens_q[:-1]
+            query_layer = query_layer.squeeze(1)
+        else:
+            # The -q_len: slice assumes left padding.
+            attention_mask = attention_mask[:, -query_length:]
+            query_layer, indices_q, cu_seqlens_q, max_seqlen_in_batch_q = unpad_input(query_layer, attention_mask)
+
+        return (
+            query_layer,
+            key_layer,
+            value_layer,
+            indices_q,
+            (cu_seqlens_q, cu_seqlens_k),
+            (max_seqlen_in_batch_q, max_seqlen_in_batch_k),
+        )
+
+
+ATTENTION_CLASSES = {
+    "eager": Attention,
+    "flash_attention_2": FlashAttention2,
+}
+
+
+class DecoderLayer(nn.Module):
+    def __init__(self, config: StableLMEpochConfig):
+        super().__init__()
+        self.self_attn = ATTENTION_CLASSES[config._attn_implementation](config=config)
+        self.mlp = MLP(config)
+        self.input_layernorm = nn.LayerNorm(config.hidden_size, eps=config.norm_eps)
+        self.post_attention_layernorm = nn.LayerNorm(config.hidden_size, eps=config.norm_eps)
+
+    def forward(
+        self,
+        hidden_states: Optional[torch.FloatTensor],
+        attention_mask: Optional[torch.FloatTensor] = None,
+        position_ids: Optional[torch.LongTensor] = None,
+        past_key_value: Optional[Tuple[torch.Tensor]] = None,
+        output_attentions: Optional[bool] = False,
+        use_cache: Optional[bool] = False,
+    ) -> Union[Tuple[torch.Tensor], Optional[Tuple[torch.Tensor, Tuple[torch.FloatTensor, ...]]]]:
+        residual = hidden_states
+
+        hidden_states = self.input_layernorm(hidden_states)
+
+        # Self Attention
+        hidden_states, self_attn_weights, present_key_value = self.self_attn(
+            hidden_states=hidden_states,
+            attention_mask=attention_mask,
+            position_ids=position_ids,
+            past_key_value=past_key_value,
+            output_attentions=output_attentions,
+            use_cache=use_cache,
+        )
+        hidden_states = residual + hidden_states
+
+        # Fully Connected
+        residual = hidden_states
+        hidden_states = self.post_attention_layernorm(hidden_states)
+        hidden_states = self.mlp(hidden_states)
+        hidden_states = residual + hidden_states
+
+        outputs = (hidden_states,)
+
+        if output_attentions:
+            outputs += (self_attn_weights,)
+
+        if use_cache:
+            outputs += (present_key_value,)
+
+        return outputs
+
+
+class StableLMEpochPreTrainedModel(PreTrainedModel):
+    """An abstract class to handle weights initialization and a simple interface
+    for downloading and loading pretrained models.
+    """
+
+    config_class = StableLMEpochConfig
+    base_model_prefix = "transformer"
+    supports_gradient_checkpointing = True
+    _no_split_modules = ["DecoderLayer"]
+    _skip_keys_device_placement = "past_key_values"
+    _supports_flash_attn_2 = True
+
+    def _init_weights(self, module: nn.Module):
+        """Initialize the weights"""
+        if isinstance(module, nn.Linear):
+            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: nn.Module, value=False):
+        if isinstance(module, StableLMEpochModel):
+            module.gradient_checkpointing = value
+
+
+class StableLMEpochModel(StableLMEpochPreTrainedModel):
+    def __init__(self, config: StableLMEpochConfig):
+        super().__init__(config)
+        self.embed_tokens = nn.Embedding(config.vocab_size, config.hidden_size, config.pad_token_id)
+        self.layers = nn.ModuleList([DecoderLayer(config) for _ in range(config.num_hidden_layers)])
+        self.norm = nn.LayerNorm(config.hidden_size, eps=config.norm_eps)
+
+        self._use_flash_attention_2 = config._attn_implementation == "flash_attention_2"
+        self.gradient_checkpointing = False
+        # Initialize weights and apply final processing
+        self.post_init()
+
+    def get_input_embeddings(self):
+        return self.embed_tokens
+
+    def set_input_embeddings(self, value: nn.Module):
+        self.embed_tokens = value
+
+    # Copied from transformers.models.bart.modeling_bart.BartDecoder._prepare_decoder_attention_mask
+    def _prepare_decoder_attention_mask(
+        self,
+        attention_mask: torch.Tensor,
+        input_shape: torch.Size,
+        inputs_embeds: torch.Tensor,
+        past_key_values_length: int,
+    ):
+        # Create causal mask
+        # [batch_size, seq_len] -> [batch_size, 1, tgt_seq_len, src_seq_len]
+        combined_attention_mask = None
+        if input_shape[-1] > 1:
+            combined_attention_mask = _make_causal_mask(
+                input_shape,
+                inputs_embeds.dtype,
+                device=inputs_embeds.device,
+                past_key_values_length=past_key_values_length,
+            )
+
+        if attention_mask is not None:
+            # [batch_size, seq_len] -> [batch_size, 1, tgt_seq_len, src_seq_len]
+            expanded_attn_mask = _expand_mask(
+                attention_mask, inputs_embeds.dtype, tgt_len=input_shape[-1]
+            ).to(inputs_embeds.device)
+            combined_attention_mask = expanded_attn_mask if combined_attention_mask is None else expanded_attn_mask + combined_attention_mask
+
+        return combined_attention_mask
+
+    def forward(
+        self,
+        input_ids: Optional[torch.LongTensor] = None,
+        attention_mask: Optional[torch.FloatTensor] = None,
+        position_ids: Optional[torch.LongTensor] = None,
+        past_key_values: Optional[Tuple[Tuple[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
+
+        # Retrieve input_ids and inputs_embeds
+        if input_ids is not None and inputs_embeds is not None:
+            raise ValueError(
+                "You cannot specify both decoder_input_ids and decoder_inputs_embeds at the same time"
+            )
+        elif input_ids is not None:
+            batch_size, seq_length = input_ids.shape
+        elif inputs_embeds is not None:
+            batch_size, seq_length, _ = inputs_embeds.shape
+        else:
+            raise ValueError(
+                "You have to specify either decoder_input_ids or decoder_inputs_embeds"
+            )
+
+        seq_length_with_past = seq_length
+        past_key_values_length = 0
+
+        if position_ids is None:
+            device = input_ids.device if input_ids is not None else inputs_embeds.device
+            position_ids = torch.arange(
+                past_key_values_length,
+                seq_length + past_key_values_length,
+                dtype=torch.long,
+                device=device,
+            )
+            position_ids = position_ids.unsqueeze(0).view(-1, seq_length)
+        else:
+            position_ids = position_ids.view(-1, seq_length).long()
+
+        if inputs_embeds is None:
+            inputs_embeds = self.embed_tokens(input_ids)
+        # Embed positions
+        if self._use_flash_attention_2:
+            # 2d mask is passed through the layers
+            attention_mask = attention_mask if (attention_mask is not None and 0 in attention_mask) else None
+        else:
+            if attention_mask is None:
+                attention_mask = torch.ones(
+                    (batch_size, seq_length_with_past),
+                    dtype=torch.bool,
+                    device=inputs_embeds.device,
+                )
+            attention_mask = self._prepare_decoder_attention_mask(
+                attention_mask,
+                (batch_size, seq_length),
+                inputs_embeds,
+                past_key_values_length,
+            )
+
+        hidden_states = inputs_embeds
+
+        if self.gradient_checkpointing and self.training:
+            if use_cache:
+                logger.warning(
+                    "`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`..."
+                )
+                use_cache = False
+
+        # Decoder layers
+        all_hidden_states = () if output_hidden_states else None
+        all_self_attns = () if output_attentions else None
+        next_decoder_cache = () if use_cache else None
+
+        for idx, decoder_layer in enumerate(self.layers):
+            if output_hidden_states:
+                all_hidden_states += (hidden_states,)
+
+            past_key_value = (
+                past_key_values[idx] if past_key_values is not None else None
+            )
+
+            if self.gradient_checkpointing and self.training:
+
+                def create_custom_forward(module):
+                    def custom_forward(*inputs):
+                        # None for past_key_value
+                        return module(*inputs, past_key_value, output_attentions)
+
+                    return custom_forward
+
+                layer_outputs = torch.utils.checkpoint.checkpoint(
+                    create_custom_forward(decoder_layer),
+                    hidden_states,
+                    attention_mask,
+                    position_ids,
+                )
+            else:
+                layer_outputs = decoder_layer(
+                    hidden_states,
+                    attention_mask=attention_mask,
+                    position_ids=position_ids,
+                    past_key_value=past_key_value,
+                    output_attentions=output_attentions,
+                    use_cache=use_cache,
+                )
+
+            hidden_states = layer_outputs[0]
+
+            if use_cache:
+                next_decoder_cache += (layer_outputs[2 if output_attentions else 1],)
+
+            if output_attentions:
+                all_self_attns += (layer_outputs[1],)
+
+        hidden_states = self.norm(hidden_states)
+
+        # Add hidden states from the last decoder layer
+        if output_hidden_states:
+            all_hidden_states += (hidden_states,)
+
+        next_cache = next_decoder_cache if use_cache else None
+        if not return_dict:
+            return tuple(
+                v
+                for v in [hidden_states, next_cache, all_hidden_states, all_self_attns]
+                if v is not None
+            )
+        return BaseModelOutputWithPast(
+            last_hidden_state=hidden_states,
+            past_key_values=next_cache,
+            hidden_states=all_hidden_states,
+            attentions=all_self_attns,
+        )
+
+
+class StableLMEpochForCausalLM(StableLMEpochPreTrainedModel):
+    _tied_weights_keys = ["lm_head.weight"]
+
+    def __init__(self, config: StableLMEpochConfig):
+        super().__init__(config)
+
+        self.model = StableLMEpochModel(config)
+        self.lm_head = nn.Linear(config.hidden_size, config.vocab_size, 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
+
+    def get_output_embeddings(self):
+        return self.lm_head
+
+    def set_output_embeddings(self, new_embeddings: nn.Module):
+        self.lm_head = new_embeddings
+
+    def get_decoder(self):
+        return self.model
+
+    def set_decoder(self, decoder):
+        self.model = decoder
+
+    def forward(
+        self,
+        input_ids: Optional[torch.LongTensor] = None,
+        attention_mask: Optional[torch.FloatTensor] = None,
+        position_ids: Optional[torch.LongTensor] = None,
+        past_key_values: Optional[Tuple[Tuple[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]:
+        output_attentions = (
+            output_attentions
+            if output_attentions is not None
+            else self.config.output_attentions
+        )
+        output_hidden_states = (
+            output_hidden_states
+            if output_hidden_states is not None
+            else self.config.output_hidden_states
+        )
+        return_dict = (
+            return_dict if return_dict is not None else self.config.use_return_dict
+        )
+
+        # decoder outputs consists of (dec_features, layer_state, dec_hidden, dec_attn)
+        outputs = self.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 = outputs[0]
+        logits = self.lm_head(hidden_states).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)
+
+        if not return_dict:
+            output = (logits,) + outputs[1:]
+            return (loss,) + output if loss is not None else output
+
+        return CausalLMOutputWithPast(
+            loss=loss,
+            logits=logits,
+            past_key_values=outputs.past_key_values,
+            hidden_states=outputs.hidden_states,
+            attentions=outputs.attentions,
+        )
+
+    def prepare_inputs_for_generation(
+        self,
+        input_ids,
+        past_key_values: Optional[torch.Tensor] = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        inputs_embeds: Optional[torch.Tensor] = None,
+        **kwargs,
+    ):
+        # Trim decoder_input_ids if past is used
+        if past_key_values is not None:
+            past_length = past_key_values[0][0].shape[2]
+
+            # Some generation methods already pass only the last input ID
+            if input_ids.shape[1] > past_length:
+                remove_prefix_length = past_length
+            else:
+                # Default to old behavior: keep only final ID
+                remove_prefix_length = input_ids.shape[1] - 1
+
+            input_ids = input_ids[:, remove_prefix_length:]
+
+        position_ids = kwargs.get("position_ids", None)
+        if attention_mask is not None and position_ids is None:
+            # Create position_ids on the fly for batch generation
+            position_ids = attention_mask.long().cumsum(-1) - 1
+            position_ids.masked_fill_(attention_mask == 0, 1)
+            if past_key_values:
+                position_ids = position_ids[:, -1].unsqueeze(-1)
+
+        # If `inputs_embeds` are passed, we only want to use them in the 1st generation step
+        if inputs_embeds is not None and past_key_values is None:
+            model_inputs = {"inputs_embeds": inputs_embeds}
+        else:
+            model_inputs = {"input_ids": input_ids}
+
+        model_inputs.update(
+            {
+                "attention_mask": attention_mask,
+                "past_key_values": past_key_values,
+                "use_cache": kwargs.get("use_cache"),
+                "position_ids": position_ids,
+            }
+        )
+        return model_inputs
+
+    @staticmethod
+    def _reorder_cache(past_key_values, beam_idx):
+        reordered_past = ()
+        for layer_past in past_key_values:
+            reordered_past += (
+                tuple(
+                    past_state.index_select(0, beam_idx.to(past_state.device))
+                    for past_state in layer_past
+                ),
+            )
+        return reordered_past
+
+
+StableLMEpochConfig.register_for_auto_class()
+StableLMEpochForCausalLM.register_for_auto_class("AutoModelForCausalLM")

tokenization_arcade100k.py

@@ -0,0 +1,273 @@
+# coding=utf-8
+# Copyright (c) 2023 Alibaba Cloud & Stability AI.
+#
+# Tongyi Qianwen LICENSE AGREEMENT:
+# https://github.com/QwenLM/Qwen/blob/5aa84bdfd3237b37f01bc88cd49b3279b9a71d0b/Tongyi%20Qianwen%20LICENSE%20AGREEMENT
+"""Tokenization classes for Arcade100k."""
+
+import base64
+import os
+import unicodedata
+from typing import Collection, Dict, List, Set, Tuple, Union
+
+import tiktoken
+from transformers.utils import logging
+from transformers import PreTrainedTokenizer, AddedToken
+
+logger = logging.get_logger(__name__)
+
+VOCAB_FILES_NAMES = {"vocab_file": "arcade100k.tiktoken"}
+NAME = "arcade100k"
+
+
+def _load_tiktoken_bpe(tiktoken_bpe_file: str) -> Dict[bytes, int]:
+    with open(tiktoken_bpe_file, "rb") as f:
+        contents = f.read()
+    return {
+        base64.b64decode(token): int(rank)
+        for token, rank in (line.split() for line in contents.splitlines() if line)
+    }
+
+
+ENDOFTEXT = "<|endoftext|>"
+FIM = [
+    "<|fim_prefix|>",
+    "<|fim_middle|>",
+    "<|fim_suffix|>",
+    "<|fim_pad|>",
+]
+# `StarCoder` Tokens
+CODE = [
+    "<gh_stars>",
+    "<filename>",
+    "<issue_start>",
+    "<issue_comment>",
+    "<issue_closed>",
+    "<jupyter_start>",
+    "<jupyter_text>",
+    "<jupyter_code>",
+    "<jupyter_output>",
+    "<empty_output>",
+    "<commit_before>",
+    "<commit_msg>",
+    "<commit_after>",
+    "<reponame>",
+]
+CHAT = [
+    "<|im_start|>",  # Chat: Input message start
+    "<|im_end|>",  # Chat: Input message end
+]
+PAUSE = "<|pause|>"  # Think before you speak (https://arxiv.org/abs/2310.02226)
+REGISTERS = [
+    f"<|reg{i}|>" for i in range(0, 8)
+]  # Register 0 sink token (https://arxiv.org/abs/2309.17453)
+ENDOFPROMPT = "<|endofprompt|>"
+SPECIAL_TOKENS_NAMES = (
+    [ENDOFTEXT]
+    + FIM
+    + CODE
+    + [ENDOFPROMPT]
+    + CHAT
+    + [PAUSE]
+    + REGISTERS
+    + ["<|extra0|>"]
+)
+START_ID = 100257
+SPECIAL_TOKENS = {t: START_ID + i for i, t in enumerate(SPECIAL_TOKENS_NAMES)}
+
+
+def _arcade100k(vocab_file: str):
+    mergeable_ranks = _load_tiktoken_bpe(vocab_file)
+
+    return {
+        "name": NAME,
+        "pat_str": r"""(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\r\n\p{L}\p{N}]?\p{L}+|\p{N}| ?[^\s\p{L}\p{N}]+[\r\n]*|\s*[\r\n]+|\s+(?!\S)|\s+""",
+        "mergeable_ranks": mergeable_ranks,
+        "special_tokens": SPECIAL_TOKENS,
+    }
+
+
+class Arcade100kTokenizer(PreTrainedTokenizer):
+    """
+    Construct a Arcade100k tokenizer backed by `tiktoken`.
+
+    Args:
+        vocab_file (`str`):
+            Path to the vocabulary file.
+        errors (`str`, *optional*, defaults to `"replace"`):
+            How to handle errors in decoding UTF-8 byte sequences.
+            WARNING: the default behaviour of this function is lossy, since decoded bytes are not
+            guaranteed to be valid UTF-8. You can control this behaviour using the `errors` parameter,
+            for instance, setting `errors=strict`.
+    """
+
+    vocab_files_names = VOCAB_FILES_NAMES
+    model_input_names = ["input_ids", "attention_mask"]
+
+    def __init__(
+        self,
+        vocab_file: str,
+        errors: str = "replace",
+        **kwargs,
+    ):
+        super().__init__(errors=errors, **kwargs)
+        self._tiktoken_config = _arcade100k(vocab_file)
+        self.tokenizer = tiktoken.Encoding(**self._tiktoken_config)
+
+        # TODO: Remove this assertion
+        assert (
+            len(self.tokenizer._mergeable_ranks)
+            + len(self.tokenizer._special_tokens)
+            + 1
+            == self.tokenizer.n_vocab
+        ), f"{len(self.tokenizer._mergeable_ranks) + len(self.tokenizer._special_tokens)} != {self.tokenizer.n_vocab} in encoding"
+
+        self.decoder = {i: n for n, i in self.tokenizer._mergeable_ranks.items()}
+        self.decoder.update({i: n for n, i in self.tokenizer._special_tokens.items()})
+        self.eos_token = self.decoder[self.tokenizer.eot_token]
+        self.pad_token = self.decoder[self.tokenizer.eot_token]
+
+    def __len__(self):
+        return self.tokenizer.n_vocab
+
+    @property
+    def vocab_size(self):
+        return self.tokenizer.n_vocab
+
+    def get_vocab(self) -> Dict[bytes, int]:
+        return self.tokenizer._mergeable_ranks
+
+    def convert_tokens_to_ids(
+        self, tokens: Union[bytes, str, List[Union[bytes, str]]]
+    ) -> List[int]:
+        ids = []
+        if isinstance(tokens, (str, bytes)):
+            if tokens in self.tokenizer._special_tokens:
+                return self.tokenizer._special_tokens[tokens]
+            else:
+                return self.tokenizer._mergeable_ranks.get(tokens)
+        for token in tokens:
+            if token in self.tokenizer._special_tokens:
+                ids.append(self.tokenizer._special_tokens[token])
+            else:
+                ids.append(self.tokenizer._mergeable_ranks.get(token))
+        return ids
+
+    def _add_tokens(
+        self,
+        new_tokens: Union[List[str], List[AddedToken]],
+        special_tokens: bool = False,
+    ) -> int:
+        if not special_tokens and new_tokens:
+            raise ValueError("Adding regular tokens is not supported")
+        for token in new_tokens:
+            surface_form = token.content if isinstance(token, AddedToken) else token
+            if surface_form not in SPECIAL_TOKENS:
+                raise ValueError("Adding unknown special tokens is not supported")
+        return 0
+
+    def save_vocabulary(self, save_directory: str, **kwargs) -> Tuple[str]:
+        """
+        Save only the vocabulary of the tokenizer (vocabulary).
+
+        Returns:
+            `Tuple(str)`: Paths to the files saved.
+        """
+        file_path = os.path.join(save_directory, "arcade100k.tiktoken")
+        with open(file_path, "w", encoding="utf8") as w:
+            for k, v in self.tokenizer._mergeable_ranks.items():
+                line = base64.b64encode(k).decode("utf8") + " " + str(v) + "\n"
+                w.write(line)
+        return (file_path,)
+
+    def tokenize(
+        self,
+        text: str,
+        allowed_special: Union[Set, str] = "all",
+        disallowed_special: Union[Collection, str] = (),
+        **kwargs,
+    ) -> List[Union[bytes, str]]:
+        """
+        Converts a string in a sequence of tokens.
+
+        Args:
+            text (`str`):
+                The sequence to be encoded.
+            allowed_special (`Literal["all"]` or `set`):
+                The surface forms of the tokens to be encoded as special tokens in regular texts.
+                Default to "all".
+            disallowed_special (`Literal["all"]` or `Collection`):
+                The surface forms of the tokens that should not be in regular texts and trigger errors.
+                Default to an empty tuple.
+
+            kwargs (additional keyword arguments, *optional*):
+                Will be passed to the underlying model specific encode method.
+
+        Returns:
+            `List[bytes|str]`: The list of tokens.
+        """
+        tokens = []
+        text = unicodedata.normalize("NFC", text)
+
+        # this implementation takes a detour: text -> token id -> token surface forms
+        for t in self.tokenizer.encode(
+            text, allowed_special=allowed_special, disallowed_special=disallowed_special
+        ):
+            tokens.append(self.decoder[t])
+        return tokens
+
+    def convert_tokens_to_string(self, tokens: List[Union[bytes, str]]) -> str:
+        """
+        Converts a sequence of tokens in a single string.
+        """
+        text = ""
+        temp = b""
+        for t in tokens:
+            if isinstance(t, str):
+                if temp:
+                    text += temp.decode("utf-8", errors=self.errors)
+                    temp = b""
+                text += t
+            elif isinstance(t, bytes):
+                temp += t
+            else:
+                raise TypeError("token should only be of type types or str")
+        if temp:
+            text += temp.decode("utf-8", errors=self.errors)
+        return text
+
+    def _convert_id_to_token(self, index: int) -> Union[bytes, str]:
+        """Converts an id to a token, special tokens included"""
+        if index in self.decoder:
+            return self.decoder[index]
+        raise ValueError("unknown ids")
+
+    def _convert_token_to_id(self, token: Union[bytes, str]) -> int:
+        """Converts a token to an id using the vocab, special tokens included"""
+        if token in self.tokenizer._special_tokens:
+            return self.tokenizer._special_tokens[token]
+        if token in self.tokenizer._mergeable_ranks:
+            return self.tokenizer._mergeable_ranks[token]
+        raise ValueError("unknown token")
+
+    def _tokenize(self, text: str, **kwargs):
+        """
+        Converts a string in a sequence of tokens (string), using the tokenizer. Split in words for word-based
+        vocabulary or sub-words for sub-word-based vocabularies (BPE/SentencePieces/WordPieces).
+
+        Do NOT take care of added tokens.
+        """
+        raise NotImplementedError
+
+    def _decode(
+        self,
+        token_ids: Union[int, List[int]],
+        skip_special_tokens: bool = False,
+        errors: str = None,
+        **kwargs,
+    ) -> str:
+        if isinstance(token_ids, int):
+            token_ids = [token_ids]
+        if skip_special_tokens:
+            token_ids = [i for i in token_ids if i < self.tokenizer.eot_token]
+        return self.tokenizer.decode(token_ids)

tokenizer_config.json

@@ -0,0 +1,9 @@
+{
+  "tokenizer_class": "Arcade100kTokenizer",
+  "auto_map": {
+    "AutoTokenizer": [
+      "tokenization_arcade100k.Arcade100kTokenizer",
+      null
+    ]
+  }
+}