Upload checkpoint-40/

checkpoint-40/chat_template.jinja

@@ -0,0 +1,4 @@
+{% for message in messages %}{% if loop.first and message['role'] != 'system' %}{{ '[|system|][|endofturn|]
+' }}{% endif %}{{ '[|' + message['role'] + '|]' + message['content'] }}{% if message['role'] == 'user' %}{{ '
+' }}{% else %}{{ '[|endofturn|]
+' }}{% endif %}{% endfor %}{% if add_generation_prompt %}{{ '[|assistant|]' }}{% endif %}

checkpoint-40/config.json

@@ -0,0 +1,40 @@
+{
+  "activation_function": "silu",
+  "architectures": [
+    "ExaoneForCausalLM"
+  ],
+  "attention_dropout": 0.0,
+  "auto_map": {
+    "AutoConfig": "configuration_exaone.ExaoneConfig",
+    "AutoModelForCausalLM": "modeling_exaone.ExaoneForCausalLM",
+    "AutoModelForSequenceClassification": "modeling_exaone.ExaoneForSequenceClassification"
+  },
+  "bos_token_id": 1,
+  "dtype": "bfloat16",
+  "embed_dropout": 0.0,
+  "eos_token_id": 361,
+  "head_dim": 128,
+  "hidden_size": 4096,
+  "initializer_range": 0.02,
+  "intermediate_size": 14336,
+  "layer_norm_epsilon": 1e-05,
+  "max_position_embeddings": 32768,
+  "model_type": "exaone",
+  "num_attention_heads": 32,
+  "num_key_value_heads": 8,
+  "num_layers": 32,
+  "pad_token_id": 0,
+  "rope_parameters": {
+    "factor": 8.0,
+    "high_freq_factor": 4.0,
+    "low_freq_factor": 1.0,
+    "original_max_position_embeddings": 8192,
+    "rope_theta": 1000000.0,
+    "rope_type": "llama3"
+  },
+  "rope_theta": 1000000.0,
+  "tie_word_embeddings": false,
+  "transformers_version": "5.7.0",
+  "use_cache": false,
+  "vocab_size": 102400
+}

checkpoint-40/configuration_exaone.py

@@ -0,0 +1,197 @@
+#                🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨
+#           This file was automatically generated from src/transformers/models/exaone/modular_exaone.py.
+#               Do NOT edit this file manually as any edits will be overwritten by the generation of
+#             the file from the modular. If any change should be done, please apply the change to the
+#                          modular_exaone.py file directly. One of our CI enforces this.
+#                🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨
+# Copyright 2026 The LG AI Research 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.
+"""LG AI Research EXAONE Lab"""
+
+from transformers.configuration_utils import PretrainedConfig
+from transformers.modeling_rope_utils import RopeParameters
+
+
+class ExaoneConfig(PretrainedConfig):
+    r"""
+    This is the configuration class to store the configuration of a [`ExaoneModel`]. It is used to
+    instantiate a EXAONE model according to the specified arguments, defining the model architecture. Instantiating a
+    configuration with the defaults will yield a similar configuration to that of the EXAONE-3.0-7.8B-Instruct [LGAI-EXAONE/EXAONE-3.0-7.8B-Instruct](https://huggingface.co/LGAI-EXAONE/EXAONE-3.0-7.8B-Instruct)
+
+    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 102400):
+            Vocabulary size of the EXAONE model. Defines the number of different tokens that can be represented by the
+            `inputs_ids` passed when calling [`ExaoneModel`]. Vocabulary size of the model.
+            Defines the different tokens that can be represented by the `inputs_ids` passed to the forward method of
+            [`ExaoneModel`].
+        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).
+        hidden_size (`int`, *optional*, defaults to 2048):
+            Dimensionality of the encoder layers and the pooler layer.
+        num_layers (`int`, *optional*, defaults to 32):
+            Number of hidden layers in the Transformer encoder.
+        num_attention_heads (`int`, *optional*, defaults to 32):
+            Number of attention heads for each attention layer in the Transformer decoder.
+        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`.
+        intermediate_size (`int`, *optional*, defaults to `hidden_size * 4`):
+            Dimensionality of the "intermediate" (i.e., feed-forward) layer in the Transformer encoder.
+        activation_function (`str` or `function`, *optional*, defaults to `"silu"`):
+            The non-linear activation function (function or string) in the decoder.
+        rope_theta (`float`, *optional*, defaults to 10000.0):
+            The base period of the RoPE embeddings.
+        rope_scaling (`Dict`, *optional*):
+            Dictionary containing the scaling configuration for the RoPE embeddings. NOTE: if you apply new rope type
+            and you expect the model to work on longer `max_position_embeddings`, we recommend you to update this value
+            accordingly.
+            Expected contents:
+                `rope_type` (`str`):
+                    The sub-variant of RoPE to use. Can be one of ['default', 'linear', 'dynamic', 'yarn', 'longrope',
+                    'llama3'], with 'default' being the original RoPE implementation.
+                `factor` (`float`, *optional*):
+                    Used with all rope types except 'default'. The scaling factor to apply to the RoPE embeddings. In
+                    most scaling types, a `factor` of x will enable the model to handle sequences of length x *
+                    original maximum pre-trained length.
+                `original_max_position_embeddings` (`int`, *optional*):
+                    Used with 'dynamic', 'longrope' and 'llama3'. The original max position embeddings used during
+                    pretraining.
+                `attention_factor` (`float`, *optional*):
+                    Used with 'yarn' and 'longrope'. The scaling factor to be applied on the attention
+                    computation. If unspecified, it defaults to value recommended by the implementation, using the
+                    `factor` field to infer the suggested value.
+                `beta_fast` (`float`, *optional*):
+                    Only used with 'yarn'. Parameter to set the boundary for extrapolation (only) in the linear
+                    ramp function. If unspecified, it defaults to 32.
+                `beta_slow` (`float`, *optional*):
+                    Only used with 'yarn'. Parameter to set the boundary for interpolation (only) in the linear
+                    ramp function. If unspecified, it defaults to 1.
+                `short_factor` (`List[float]`, *optional*):
+                    Only used with 'longrope'. The scaling factor to be applied to short contexts (<
+                    `original_max_position_embeddings`). Must be a list of numbers with the same length as the hidden
+                    size divided by the number of attention heads divided by 2
+                `long_factor` (`List[float]`, *optional*):
+                    Only used with 'longrope'. The scaling factor to be applied to long contexts (<
+                    `original_max_position_embeddings`). Must be a list of numbers with the same length as the hidden
+                    size divided by the number of attention heads divided by 2
+                `low_freq_factor` (`float`, *optional*):
+                    Only used with 'llama3'. Scaling factor applied to low frequency components of the RoPE
+                `high_freq_factor` (`float`, *optional*):
+                    Only used with 'llama3'. Scaling factor applied to high frequency components of the RoPE
+        embed_dropout (`float`, *optional*, defaults to 0.0):
+            The dropout probabilitiy for all fully connected layers in the embeddings, encoder, and pooler.
+        attention_dropout (`float`, *optional*, defaults to 0.0):
+            The dropout ratio for the attention probabilities.
+        layer_norm_epsilon (`float`, *optional*, defaults to 1e-05):
+            The epsilon used by the layer normalization layers.
+        initializer_range (`float`, *optional*, defaults to 0.02):
+            The standard deviation of the truncated_normal_initializer for initializing all weight matrices.
+        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``.
+        bos_token_id (`int`, *optional*, defaults to 0):
+            Beginning of stream token id.
+        eos_token_id (`int`, *optional*, defaults to 2):
+            End of stream token id.
+        pad_token_id (`int`, *optional*):
+            Padding token id.
+        tie_word_embeddings (`bool`, *optional*, defaults to `False`):
+            Whether to tie weight embeddings
+
+    Example:
+
+    ```python
+    >>> from transformers import EXAONEModel, ExaoneConfig
+
+    >>> # Initializing a EXAONE configuration
+    >>> configuration = ExaoneConfig()
+
+    >>> # Initializing a model from configuration
+    >>> model = EXAONEModel(configuration)
+
+    >>> # Accessing the model configuration
+    >>> configuration = model.config
+    ```"""
+
+    model_type = "exaone"
+    keys_to_ignore_at_inference = ["past_key_values"]
+    attribute_map = {
+        "num_hidden_layers": "num_layers",
+        "hidden_act": "activation_function",
+        "rms_norm_eps": "layer_norm_epsilon",
+    }
+
+    def __init__(
+        self,
+        vocab_size: int | None = 102400,
+        max_position_embeddings=2048,
+        hidden_size: int | None = 2048,
+        num_layers: int | None = 32,
+        num_attention_heads: int | None = 32,
+        num_key_value_heads: int | None = None,
+        intermediate_size: int | None = None,
+        activation_function: str | None = "silu",
+        rope_parameters: RopeParameters | None = None,
+        embed_dropout: float | None = 0.0,
+        attention_dropout: float | None = 0.0,
+        layer_norm_epsilon: float | None = 1e-5,
+        initializer_range: float | None = 0.02,
+        use_cache: bool | None = True,
+        bos_token_id: int | None = 0,
+        eos_token_id: int | None = 2,
+        pad_token_id: int | None = None,
+        tie_word_embeddings: bool | None = False,
+        **kwargs,
+    ):
+        self.vocab_size = vocab_size
+        self.max_position_embeddings = max_position_embeddings
+        self.hidden_size = hidden_size
+        self.num_layers = num_layers
+        self.num_attention_heads = num_attention_heads
+        self.num_layers = num_layers
+        if num_key_value_heads is None:
+            num_key_value_heads = num_attention_heads
+        self.num_key_value_heads = num_key_value_heads
+        if intermediate_size:
+            self.intermediate_size = intermediate_size
+        else:
+            self.intermediate_size = hidden_size * 4
+        self.activation_function = activation_function
+        self.embed_dropout = embed_dropout
+        self.attention_dropout = attention_dropout
+        self.layer_norm_epsilon = layer_norm_epsilon
+        self.initializer_range = initializer_range
+        self.use_cache = use_cache
+        self.rope_parameters = rope_parameters
+
+        self.bos_token_id = bos_token_id
+        self.eos_token_id = eos_token_id
+        self.pad_token_id = pad_token_id
+        self.tie_word_embeddings = tie_word_embeddings
+
+        super().__init__(**kwargs)
+
+
+__all__ = ["ExaoneConfig"]

checkpoint-40/generation_config.json

@@ -0,0 +1,9 @@
+{
+  "_from_model_config": true,
+  "bos_token_id": 1,
+  "eos_token_id": [
+    361
+  ],
+  "pad_token_id": 0,
+  "transformers_version": "5.7.0"
+}

checkpoint-40/global_step40/bf16_zero_pp_rank_0_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:044e60a61e2e9715cc1a849da020504a16b7a5d1c2480f02909a514e7ec7c716
+size 11788764019

checkpoint-40/global_step40/bf16_zero_pp_rank_1_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:571b987a66a081b69fcf6cedad5cb8a76b7257446563e1c6ffdfded14d9bea78
+size 11788764019

checkpoint-40/global_step40/bf16_zero_pp_rank_2_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2aa3b00119b89ddd87e92277a50f5531ffc4ab6811ed4437126afead5385105c
+size 11788764019

checkpoint-40/global_step40/bf16_zero_pp_rank_3_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:96bfd89df3ee58e0d44ec256b0361e98e42d27f394fe0e316db778b320ef9416
+size 11788764019

checkpoint-40/global_step40/zero_pp_rank_0_mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c044717a19e01c7fc778a2b05221b3436941110c3a461bfa67cccf65e5aa5007
+size 150512

checkpoint-40/global_step40/zero_pp_rank_1_mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2b2ad147bb4ee9e513dfddaf486ae3af5ed084c663a8ce618b81ab982bda4f7b
+size 150512

checkpoint-40/global_step40/zero_pp_rank_2_mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6d4a46d5e3fbd2843040cca7c9344cc24920d50b273370af80e7caaa630e45ba
+size 150512

checkpoint-40/global_step40/zero_pp_rank_3_mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bdf45e81ca0c0cc6da4b1311af8f3e9cd8a7f380f0cce89b12edfff467cf85d8
+size 150512

checkpoint-40/latest

@@ -0,0 +1 @@
+global_step40

checkpoint-40/model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:508b43f76e3eefe29b14747c8877e60144eea280e7b30743ab5a37f97f8bd22b
+size 15636931624

checkpoint-40/modeling_exaone.py

@@ -0,0 +1,543 @@
+#                🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨
+#           This file was automatically generated from src/transformers/models/exaone/modular_exaone.py.
+#               Do NOT edit this file manually as any edits will be overwritten by the generation of
+#             the file from the modular. If any change should be done, please apply the change to the
+#                          modular_exaone.py file directly. One of our CI enforces this.
+#                🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨
+# Copyright 2026 The LG AI Research 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.
+"""LG AI Research EXAONE Lab"""
+
+from collections.abc import Callable
+from typing import Optional
+
+import torch
+from torch import nn
+
+from transformers.activations import ACT2FN
+from transformers.cache_utils import Cache, DynamicCache
+from transformers.generation import GenerationMixin
+from transformers.integrations import use_kernel_forward_from_hub, use_kernel_func_from_hub, use_kernelized_func
+from transformers.masking_utils import create_causal_mask
+from transformers.modeling_layers import GradientCheckpointingLayer
+from transformers.modeling_outputs import BaseModelOutputWithPast, CausalLMOutputWithPast
+from transformers.modeling_rope_utils import ROPE_INIT_FUNCTIONS, dynamic_rope_update
+from transformers.modeling_utils import ALL_ATTENTION_FUNCTIONS, PreTrainedModel
+from transformers.processing_utils import Unpack
+from transformers.utils import TransformersKwargs, auto_docstring, can_return_tuple
+from transformers.utils.generic import check_model_inputs, maybe_autocast
+from .configuration_exaone import ExaoneConfig
+
+
+@use_kernel_forward_from_hub("RMSNorm")
+class ExaoneRMSNorm(nn.Module):
+    def __init__(self, hidden_size, eps=1e-6):
+        """
+        ExaoneRMSNorm is equivalent to T5LayerNorm
+        """
+        super().__init__()
+        self.weight = nn.Parameter(torch.ones(hidden_size))
+        self.variance_epsilon = eps
+
+    def forward(self, hidden_states):
+        input_dtype = hidden_states.dtype
+        hidden_states = hidden_states.to(torch.float32)
+        variance = hidden_states.pow(2).mean(-1, keepdim=True)
+        hidden_states = hidden_states * torch.rsqrt(variance + self.variance_epsilon)
+        return self.weight * hidden_states.to(input_dtype)
+
+    def extra_repr(self):
+        return f"{tuple(self.weight.shape)}, eps={self.variance_epsilon}"
+
+
+def rotate_half(x):
+    """Rotates half the hidden dims of the input."""
+    x1 = x[..., : x.shape[-1] // 2]
+    x2 = x[..., x.shape[-1] // 2 :]
+    return torch.cat((-x2, x1), dim=-1)
+
+
+@use_kernel_func_from_hub("rotary_pos_emb")
+def apply_rotary_pos_emb(q, k, cos, sin, unsqueeze_dim=1):
+    """Applies Rotary Position Embedding to the query and key tensors.
+
+    Args:
+        q (`torch.Tensor`): The query tensor.
+        k (`torch.Tensor`): The key tensor.
+        cos (`torch.Tensor`): The cosine part of the rotary embedding.
+        sin (`torch.Tensor`): The sine part of the rotary embedding.
+        unsqueeze_dim (`int`, *optional*, defaults to 1):
+            The 'unsqueeze_dim' argument specifies the dimension along which to unsqueeze cos[position_ids] and
+            sin[position_ids] so that they can be properly broadcasted to the dimensions of q and k. For example, note
+            that cos[position_ids] and sin[position_ids] have the shape [batch_size, seq_len, head_dim]. Then, if q and
+            k have the shape [batch_size, heads, seq_len, head_dim], then setting unsqueeze_dim=1 makes
+            cos[position_ids] and sin[position_ids] broadcastable to the shapes of q and k. Similarly, if q and k have
+            the shape [batch_size, seq_len, heads, head_dim], then set unsqueeze_dim=2.
+    Returns:
+        `tuple(torch.Tensor)` comprising of the query and key tensors rotated using the Rotary Position Embedding.
+    """
+    cos = cos.unsqueeze(unsqueeze_dim)
+    sin = sin.unsqueeze(unsqueeze_dim)
+    q_embed = (q * cos) + (rotate_half(q) * sin)
+    k_embed = (k * cos) + (rotate_half(k) * sin)
+    return q_embed, k_embed
+
+
+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)
+
+
+def eager_attention_forward(
+    module: nn.Module,
+    query: torch.Tensor,
+    key: torch.Tensor,
+    value: torch.Tensor,
+    attention_mask: torch.Tensor | None,
+    scaling: float,
+    dropout: float = 0.0,
+    **kwargs: Unpack[TransformersKwargs],
+):
+    key_states = repeat_kv(key, module.num_key_value_groups)
+    value_states = repeat_kv(value, module.num_key_value_groups)
+
+    attn_weights = torch.matmul(query, key_states.transpose(2, 3)) * scaling
+    if attention_mask is not None:
+        causal_mask = attention_mask[:, :, :, : key_states.shape[-2]]
+        attn_weights = attn_weights + causal_mask
+
+    attn_weights = nn.functional.softmax(attn_weights, dim=-1, dtype=torch.float32).to(query.dtype)
+    attn_weights = nn.functional.dropout(attn_weights, p=dropout, training=module.training)
+    attn_output = torch.matmul(attn_weights, value_states)
+    attn_output = attn_output.transpose(1, 2).contiguous()
+
+    return attn_output, attn_weights
+
+
+@use_kernelized_func(apply_rotary_pos_emb)
+class ExaoneAttention(nn.Module):
+    """Multi-headed attention from 'Attention Is All You Need' paper"""
+
+    def __init__(self, config: ExaoneConfig, layer_idx: int):
+        super().__init__()
+        self.config = config
+        self.layer_idx = layer_idx
+        self.head_dim = getattr(config, "head_dim", config.hidden_size // config.num_attention_heads)
+        self.num_key_value_groups = config.num_attention_heads // config.num_key_value_heads
+        self.scaling = self.head_dim**-0.5
+        self.attention_dropout = config.attention_dropout
+        self.is_causal = True
+        self.q_proj = nn.Linear(config.hidden_size, config.num_attention_heads * self.head_dim, bias=False)
+        self.k_proj = nn.Linear(config.hidden_size, config.num_key_value_heads * self.head_dim, bias=False)
+        self.v_proj = nn.Linear(config.hidden_size, config.num_key_value_heads * self.head_dim, bias=False)
+        self.out_proj = nn.Linear(config.num_attention_heads * self.head_dim, config.hidden_size, bias=False)
+
+    def forward(
+        self,
+        hidden_states: torch.Tensor,
+        position_embeddings: tuple[torch.Tensor, torch.Tensor] | None = None,
+        attention_mask: torch.Tensor | None = None,
+        past_key_values: Cache | None = None,
+        cache_position: torch.LongTensor | None = None,
+        **kwargs: Unpack[TransformersKwargs],
+    ) -> tuple[torch.Tensor, torch.Tensor]:
+        input_shape = hidden_states.shape[:-1]
+        hidden_shape = (*input_shape, -1, self.head_dim)
+
+        query_states = self.q_proj(hidden_states).view(hidden_shape).transpose(1, 2)
+        key_states = self.k_proj(hidden_states).view(hidden_shape).transpose(1, 2)
+        value_states = self.v_proj(hidden_states).view(hidden_shape).transpose(1, 2)
+
+        cos, sin = position_embeddings
+        query_states, key_states = apply_rotary_pos_emb(query_states, key_states, cos, sin)
+
+        if past_key_values is not None:
+            # sin and cos are specific to RoPE models; cache_position needed for the static cache
+            cache_kwargs = {"sin": sin, "cos": cos, "cache_position": cache_position}
+            key_states, value_states = past_key_values.update(key_states, value_states, self.layer_idx, cache_kwargs)
+
+        attention_interface: Callable = ALL_ATTENTION_FUNCTIONS.get_interface(
+            self.config._attn_implementation, eager_attention_forward
+        )
+
+        attn_output, attn_weights = attention_interface(
+            self,
+            query_states,
+            key_states,
+            value_states,
+            attention_mask,
+            dropout=0.0 if not self.training else self.attention_dropout,
+            scaling=self.scaling,
+            **kwargs,
+        )
+
+        attn_output = attn_output.reshape(*input_shape, -1).contiguous()
+        attn_output = self.out_proj(attn_output)
+        return attn_output, attn_weights
+
+
+class ExaoneAttentionBlock(nn.Module):
+    """Dummy wrapper class for EXAONE 3.5 structure"""
+
+    def __init__(self, config: ExaoneConfig, layer_idx: int):
+        super().__init__()
+        self.config = config
+        self.layer_idx = layer_idx
+        self.attention = ExaoneAttention(config, layer_idx)
+
+    def forward(
+        self,
+        hidden_states: torch.Tensor,
+        position_embeddings: tuple[torch.Tensor, torch.Tensor] | None = None,
+        attention_mask: torch.Tensor | None = None,
+        past_key_values: Cache | None = None,
+        cache_position: torch.LongTensor | None = None,
+        **kwargs: Unpack[TransformersKwargs],
+    ) -> tuple[torch.Tensor, torch.Tensor]:
+        return self.attention(
+            hidden_states=hidden_states,
+            position_embeddings=position_embeddings,
+            attention_mask=attention_mask,
+            past_key_values=past_key_values,
+            cache_position=cache_position,
+            **kwargs,
+        )
+
+
+class ExaoneMLP(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.config = config
+        self.hidden_size = config.hidden_size
+        self.intermediate_size = config.intermediate_size
+        self.c_fc_0 = nn.Linear(self.hidden_size, self.intermediate_size, bias=False)
+        self.c_fc_1 = nn.Linear(self.hidden_size, self.intermediate_size, bias=False)
+        self.c_proj = nn.Linear(self.intermediate_size, self.hidden_size, bias=False)
+        self.act = ACT2FN[config.hidden_act]
+
+    def forward(self, x):
+        output_proj = self.c_proj(self.act(self.c_fc_0(x)) * self.c_fc_1(x))
+        return output_proj
+
+
+class ExaoneDecoderLayer(GradientCheckpointingLayer):
+    def __init__(self, config, layer_id):
+        super().__init__()
+        self.config = config
+        self.hidden_size = config.hidden_size
+        self.ln_1 = ExaoneRMSNorm(hidden_size=self.hidden_size, eps=config.layer_norm_epsilon)
+        self.attn = ExaoneAttentionBlock(config, layer_id)
+        self.ln_2 = ExaoneRMSNorm(hidden_size=self.hidden_size, eps=config.layer_norm_epsilon)
+        self.mlp = ExaoneMLP(config)
+
+    def forward(
+        self,
+        hidden_states: torch.Tensor,
+        attention_mask: torch.Tensor | None = None,
+        position_ids: torch.LongTensor | None = None,
+        past_key_values: Cache | None = None,
+        use_cache: bool | None = False,
+        cache_position: torch.LongTensor | None = None,
+        position_embeddings: tuple[torch.Tensor, torch.Tensor] | None = None,
+        **kwargs: Unpack[TransformersKwargs],
+    ) -> torch.Tensor:
+        residual = hidden_states
+        hidden_states = self.ln_1(hidden_states)
+        # Self Attention
+        hidden_states, _ = self.attn(
+            hidden_states=hidden_states,
+            attention_mask=attention_mask,
+            position_ids=position_ids,
+            past_key_values=past_key_values,
+            use_cache=use_cache,
+            cache_position=cache_position,
+            position_embeddings=position_embeddings,
+            **kwargs,
+        )
+        hidden_states = residual + hidden_states
+
+        # Fully Connected
+        residual = hidden_states
+        hidden_states = self.ln_2(hidden_states)
+        hidden_states = self.mlp(hidden_states)
+        hidden_states = residual + hidden_states
+        return hidden_states
+
+
+@auto_docstring
+class ExaonePreTrainedModel(PreTrainedModel):
+    config: ExaoneConfig
+
+    base_model_prefix = "transformer"
+    supports_gradient_checkpointing = True
+    _no_split_modules = ["ExaoneDecoderLayer"]
+    _skip_keys_device_placement = ["past_key_values"]
+    _supports_flash_attn = True
+    _supports_sdpa = True
+    _supports_flex_attn = True
+
+    _can_compile_fullgraph = True
+    _supports_attention_backend = True
+    _can_record_outputs = {
+        "hidden_states": ExaoneDecoderLayer,
+        "attentions": ExaoneAttention,
+    }
+
+
+class ExaoneRotaryEmbedding(nn.Module):
+    inv_freq: torch.Tensor  # fix linting for `register_buffer`
+
+    def __init__(self, config: ExaoneConfig, device=None):
+        super().__init__()
+        self.max_seq_len_cached = config.max_position_embeddings
+        self.original_max_seq_len = config.max_position_embeddings
+
+        self.config = config
+
+        self.rope_type = self.config.rope_parameters["rope_type"]
+        rope_init_fn: Callable = self.compute_default_rope_parameters
+        if self.rope_type != "default":
+            rope_init_fn = ROPE_INIT_FUNCTIONS[self.rope_type]
+        inv_freq, self.attention_scaling = rope_init_fn(self.config, device)
+
+        self.register_buffer("inv_freq", inv_freq, persistent=False)
+        self.register_buffer("original_inv_freq", inv_freq.clone(), persistent=False)
+
+    @staticmethod
+    def compute_default_rope_parameters(
+        config: ExaoneConfig | None = None,
+        device: Optional["torch.device"] = None,
+        seq_len: int | None = None,
+    ) -> tuple["torch.Tensor", float]:
+        """
+        Computes the inverse frequencies according to the original RoPE implementation
+        Args:
+            config ([`~transformers.PreTrainedConfig`]):
+                The model configuration.
+            device (`torch.device`):
+                The device to use for initialization of the inverse frequencies.
+            seq_len (`int`, *optional*):
+                The current sequence length. Unused for this type of RoPE.
+        Returns:
+            Tuple of (`torch.Tensor`, `float`), containing the inverse frequencies for the RoPE embeddings and the
+            post-processing scaling factor applied to the computed cos/sin (unused in this type of RoPE).
+        """
+        base = config.rope_parameters["rope_theta"]
+        dim = getattr(config, "head_dim", None) or config.hidden_size // config.num_attention_heads
+
+        attention_factor = 1.0  # Unused in this type of RoPE
+
+        # Compute the inverse frequencies
+        inv_freq = 1.0 / (
+            base ** (torch.arange(0, dim, 2, dtype=torch.int64).to(device=device, dtype=torch.float) / dim)
+        )
+        return inv_freq, attention_factor
+
+    @torch.no_grad()
+    @dynamic_rope_update  # power user: used with advanced RoPE types (e.g. dynamic rope)
+    def forward(self, x, position_ids):
+        inv_freq_expanded = self.inv_freq[None, :, None].float().expand(position_ids.shape[0], -1, 1).to(x.device)
+        position_ids_expanded = position_ids[:, None, :].float()
+
+        device_type = x.device.type if isinstance(x.device.type, str) and x.device.type != "mps" else "cpu"
+        with maybe_autocast(device_type=device_type, enabled=False):  # Force float32
+            freqs = (inv_freq_expanded.float() @ position_ids_expanded.float()).transpose(1, 2)
+            emb = torch.cat((freqs, freqs), dim=-1)
+            cos = emb.cos() * self.attention_scaling
+            sin = emb.sin() * self.attention_scaling
+
+        return cos.to(dtype=x.dtype), sin.to(dtype=x.dtype)
+
+
+@auto_docstring
+class ExaoneModel(ExaonePreTrainedModel):
+    def __init__(self, config: ExaoneConfig):
+        super().__init__(config)
+        self.config = config
+        self.hidden_size = config.hidden_size
+        self.padding_idx = config.pad_token_id
+        self.vocab_size = config.vocab_size
+
+        self.wte = nn.Embedding(self.vocab_size, self.hidden_size, self.padding_idx)
+        self.drop = nn.Dropout(float(config.embed_dropout))
+        self.h = nn.ModuleList([ExaoneDecoderLayer(config, layer_id=i) for i in range(config.num_layers)])
+        self.ln_f = ExaoneRMSNorm(hidden_size=self.hidden_size, eps=config.layer_norm_epsilon)
+        self.rotary = ExaoneRotaryEmbedding(config)
+
+        # Initialize weights and apply final processing
+        self.post_init()
+
+    @check_model_inputs
+    @auto_docstring
+    def forward(
+        self,
+        input_ids: torch.LongTensor | None = None,
+        attention_mask: torch.Tensor | None = None,
+        position_ids: torch.LongTensor | None = None,
+        past_key_values: Cache | None = None,
+        inputs_embeds: torch.FloatTensor | None = None,
+        cache_position: torch.LongTensor | None = None,
+        use_cache: bool | None = None,
+        **kwargs: Unpack[TransformersKwargs],
+    ) -> BaseModelOutputWithPast:
+        if (input_ids is None) ^ (inputs_embeds is not None):
+            raise ValueError("You must specify exactly one of input_ids or inputs_embeds")
+
+        if inputs_embeds is None:
+            inputs_embeds: torch.Tensor = self.wte(input_ids)
+
+        if use_cache and past_key_values is None:
+            past_key_values = DynamicCache(config=self.config)
+
+        if cache_position is None:
+            past_seen_tokens = past_key_values.get_seq_length() if past_key_values is not None else 0
+            cache_position: torch.Tensor = (
+                torch.arange(inputs_embeds.shape[1], device=inputs_embeds.device) + past_seen_tokens
+            )
+
+        if position_ids is None:
+            position_ids = cache_position.unsqueeze(0)
+
+        causal_mask = create_causal_mask(
+            config=self.config,
+            input_embeds=inputs_embeds,
+            attention_mask=attention_mask,
+            cache_position=cache_position,
+            past_key_values=past_key_values,
+            position_ids=position_ids,
+        )
+
+        hidden_states = inputs_embeds
+        position_embeddings = self.rotary(hidden_states, position_ids=position_ids)
+
+        for decoder_layer in self.h[: self.config.num_layers]:
+            hidden_states = decoder_layer(
+                hidden_states,
+                attention_mask=causal_mask,
+                position_embeddings=position_embeddings,
+                position_ids=position_ids,
+                past_key_values=past_key_values,
+                use_cache=use_cache,
+                cache_position=cache_position,
+                **kwargs,
+            )
+
+        hidden_states = self.ln_f(hidden_states)
+        return BaseModelOutputWithPast(
+            last_hidden_state=hidden_states,
+            past_key_values=past_key_values,
+        )
+
+
+@auto_docstring
+class ExaoneForCausalLM(ExaonePreTrainedModel, GenerationMixin):
+    _tied_weights_keys = {"lm_head.weight": "transformer.wte.weight"}
+    _tp_plan = {"lm_head": "colwise_gather_output"}
+    _pp_plan = {"lm_head": (["hidden_states"], ["logits"])}
+
+    def __init__(self, config):
+        super().__init__(config)
+        self.transformer = ExaoneModel(config)
+        self.vocab_size = config.vocab_size
+        self.lm_head = nn.Linear(config.hidden_size, config.vocab_size, bias=False)
+
+        # Initialize weights and apply final processing
+        self.post_init()
+
+    @can_return_tuple
+    @auto_docstring
+    def forward(
+        self,
+        input_ids: torch.LongTensor | None = None,
+        attention_mask: torch.Tensor | None = None,
+        position_ids: torch.LongTensor | None = None,
+        past_key_values: Cache | None = None,
+        inputs_embeds: torch.FloatTensor | None = None,
+        labels: torch.LongTensor | None = None,
+        use_cache: bool | None = None,
+        cache_position: torch.LongTensor | None = None,
+        logits_to_keep: int | torch.Tensor = 0,
+        **kwargs: Unpack[TransformersKwargs],
+    ) -> CausalLMOutputWithPast:
+        r"""
+        Args:
+            labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
+                Labels for language modeling. Note that the labels **are shifted** inside the model, i.e. you can set
+                `labels = input_ids` Indices are selected in `[-100, 0, ..., config.vocab_size]` All labels set to `-100`
+                are ignored (masked), the loss is only computed for labels in `[0, ..., config.vocab_size]`
+
+        Example:
+
+        ```python
+        >>> from transformers import AutoModelForCausalLM, AutoTokenizer
+
+        >>> model = AutoModelForCausalLM.from_pretrained("LGAI-EXAONE/EXAONE-3.5-2.4B-Instruct",
+                                                         trust_remote_code=True)
+        >>> tokenizer = AutoTokenizer.from_pretrained("LGAI-EXAONE/EXAONE-3.5-2.4B-Instruct")
+
+        >>> prompt = "Explain how wonderful you are"
+        >>> messages = [
+            {"role": "system", "content": "You are a helpful assistant."},
+            {"role": "user", "content": prompt}
+        ]
+        >>> input_ids = tokenizer.apply_chat_template(
+            messages,
+            tokenize=True,
+            add_generation_prompt=True,
+            return_tensors="pt"
+        )
+
+        >>> output = model.generate(**input_ids.to(model.device), max_new_tokens=128)
+        >>> tokenizer.decode(output[0], skip_special_tokens=True)
+        '[|system|]You are a helpful assistant.\n[|user|]Explain how wonderful you are\n[|assistant|]As an AI assistant, I don\'t experience feelings or qualities like "wonderfulness" in the way humans do, but I can certainly highlight several aspects that make my capabilities and interactions valuable and beneficial:\n\n1. **Knowledge and Information**: I am equipped with extensive knowledge across a wide range of topics including science, technology, history, culture, and more. This allows me to provide accurate, informative responses to a vast array of inquiries, helping users learn and explore new ideas.\n\n2. **Accessibility**: I am available 24/7, meaning you can ask me questions or seek assistance at'
+        ```
+        """
+        outputs: BaseModelOutputWithPast = self.transformer(
+            input_ids=input_ids,
+            attention_mask=attention_mask,
+            position_ids=position_ids,
+            past_key_values=past_key_values,
+            inputs_embeds=inputs_embeds,
+            use_cache=use_cache,
+            cache_position=cache_position,
+            **kwargs,
+        )
+
+        hidden_states = outputs.last_hidden_state
+        # Only compute necessary logits, and do not upcast them to float if we are not computing the loss
+        slice_indices = slice(-logits_to_keep, None) if isinstance(logits_to_keep, int) else logits_to_keep
+        logits = self.lm_head(hidden_states[:, slice_indices, :])
+
+        loss = None
+        if labels is not None:
+            loss = self.loss_function(logits=logits, labels=labels, vocab_size=self.config.vocab_size, **kwargs)
+
+        return CausalLMOutputWithPast(
+            loss=loss,
+            logits=logits,
+            past_key_values=outputs.past_key_values,
+            hidden_states=outputs.hidden_states,
+            attentions=outputs.attentions,
+        )
+
+
+__all__ = ["ExaonePreTrainedModel", "ExaoneModel", "ExaoneForCausalLM"]

checkpoint-40/rng_state_0.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2fe6b873f1da64bb980afe982fc1c6b6bc461ad8c33073d856804323f64b26d9
+size 15429

checkpoint-40/rng_state_1.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:13db4dd10b22b5464f5ef0b539152b46390eae6a827b355108cb54b1af2eb830
+size 15429

checkpoint-40/rng_state_2.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:06e293f13fb6e873e8081636b6b07115410b397c55f1ed08b8772b059c8eb74e
+size 15429

checkpoint-40/rng_state_3.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c595c077664f94b7f91088561951be20e39483e01865774a83bc46f3971a3152
+size 15429

checkpoint-40/scheduler.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e5bd29a07a9c2ce09354b17c073dcc155b7a74ce4be44a8dae595af26e300186
+size 1401

checkpoint-40/tokenizer_config.json

@@ -0,0 +1,326 @@
+{
+  "add_prefix_space": false,
+  "additional_special_token": [
+    "[unused0]",
+    "[unused1]",
+    "[unused2]",
+    "[unused3]",
+    "[unused4]",
+    "[unused5]",
+    "[unused6]",
+    "[unused7]",
+    "[unused8]",
+    "[unused9]",
+    "[unused10]",
+    "[unused11]",
+    "[unused12]",
+    "[unused13]",
+    "[unused14]",
+    "[unused15]",
+    "[unused16]",
+    "[unused17]",
+    "[unused18]",
+    "[unused19]",
+    "[unused20]",
+    "[unused21]",
+    "[unused22]",
+    "[unused23]",
+    "[unused24]",
+    "[unused25]",
+    "[unused26]",
+    "[unused27]",
+    "[unused28]",
+    "[unused29]",
+    "[unused30]",
+    "[unused31]",
+    "[unused32]",
+    "[unused33]",
+    "[unused34]",
+    "[unused35]",
+    "[unused36]",
+    "[unused37]",
+    "[unused38]",
+    "[unused39]",
+    "[unused40]",
+    "[unused41]",
+    "[unused42]",
+    "[unused43]",
+    "[unused44]",
+    "[unused45]",
+    "[unused46]",
+    "[unused47]",
+    "[unused48]",
+    "[unused49]",
+    "[unused50]",
+    "[unused51]",
+    "[unused52]",
+    "[unused53]",
+    "[unused54]",
+    "[unused55]",
+    "[unused56]",
+    "[unused57]",
+    "[unused58]",
+    "[unused59]",
+    "[unused60]",
+    "[unused61]",
+    "[unused62]",
+    "[unused63]",
+    "[unused64]",
+    "[unused65]",
+    "[unused66]",
+    "[unused67]",
+    "[unused68]",
+    "[unused69]",
+    "[unused70]",
+    "[unused71]",
+    "[unused72]",
+    "[unused73]",
+    "[unused74]",
+    "[unused75]",
+    "[unused76]",
+    "[unused77]",
+    "[unused78]",
+    "[unused79]",
+    "[unused80]",
+    "[unused81]",
+    "[unused82]",
+    "[unused83]",
+    "[unused84]",
+    "[unused85]",
+    "[unused86]",
+    "[unused87]",
+    "[unused88]",
+    "[unused89]",
+    "[unused90]",
+    "[unused91]",
+    "[unused92]",
+    "[unused93]",
+    "[unused94]",
+    "[unused95]",
+    "[unused96]",
+    "[unused97]",
+    "[unused98]",
+    "[unused99]",
+    "[extra_id_0]",
+    "[extra_id_1]",
+    "[extra_id_2]",
+    "[extra_id_3]",
+    "[extra_id_4]",
+    "[extra_id_5]",
+    "[extra_id_6]",
+    "[extra_id_7]",
+    "[extra_id_8]",
+    "[extra_id_9]",
+    "[extra_id_10]",
+    "[extra_id_11]",
+    "[extra_id_12]",
+    "[extra_id_13]",
+    "[extra_id_14]",
+    "[extra_id_15]",
+    "[extra_id_16]",
+    "[extra_id_17]",
+    "[extra_id_18]",
+    "[extra_id_19]",
+    "[extra_id_20]",
+    "[extra_id_21]",
+    "[extra_id_22]",
+    "[extra_id_23]",
+    "[extra_id_24]",
+    "[extra_id_25]",
+    "[extra_id_26]",
+    "[extra_id_27]",
+    "[extra_id_28]",
+    "[extra_id_29]",
+    "[extra_id_30]",
+    "[extra_id_31]",
+    "[extra_id_32]",
+    "[extra_id_33]",
+    "[extra_id_34]",
+    "[extra_id_35]",
+    "[extra_id_36]",
+    "[extra_id_37]",
+    "[extra_id_38]",
+    "[extra_id_39]",
+    "[extra_id_40]",
+    "[extra_id_41]",
+    "[extra_id_42]",
+    "[extra_id_43]",
+    "[extra_id_44]",
+    "[extra_id_45]",
+    "[extra_id_46]",
+    "[extra_id_47]",
+    "[extra_id_48]",
+    "[extra_id_49]",
+    "[extra_id_50]",
+    "[extra_id_51]",
+    "[extra_id_52]",
+    "[extra_id_53]",
+    "[extra_id_54]",
+    "[extra_id_55]",
+    "[extra_id_56]",
+    "[extra_id_57]",
+    "[extra_id_58]",
+    "[extra_id_59]",
+    "[extra_id_60]",
+    "[extra_id_61]",
+    "[extra_id_62]",
+    "[extra_id_63]",
+    "[extra_id_64]",
+    "[extra_id_65]",
+    "[extra_id_66]",
+    "[extra_id_67]",
+    "[extra_id_68]",
+    "[extra_id_69]",
+    "[extra_id_70]",
+    "[extra_id_71]",
+    "[extra_id_72]",
+    "[extra_id_73]",
+    "[extra_id_74]",
+    "[extra_id_75]",
+    "[extra_id_76]",
+    "[extra_id_77]",
+    "[extra_id_78]",
+    "[extra_id_79]",
+    "[extra_id_80]",
+    "[extra_id_81]",
+    "[extra_id_82]",
+    "[extra_id_83]",
+    "[extra_id_84]",
+    "[extra_id_85]",
+    "[extra_id_86]",
+    "[extra_id_87]",
+    "[extra_id_88]",
+    "[extra_id_89]",
+    "[extra_id_90]",
+    "[extra_id_91]",
+    "[extra_id_92]",
+    "[extra_id_93]",
+    "[extra_id_94]",
+    "[extra_id_95]",
+    "[extra_id_96]",
+    "[extra_id_97]",
+    "[extra_id_98]",
+    "[extra_id_99]",
+    "[extra_id_100]",
+    "[extra_id_101]",
+    "[extra_id_102]",
+    "[extra_id_103]",
+    "[extra_id_104]",
+    "[extra_id_105]",
+    "[extra_id_106]",
+    "[extra_id_107]",
+    "[extra_id_108]",
+    "[extra_id_109]",
+    "[extra_id_110]",
+    "[extra_id_111]",
+    "[extra_id_112]",
+    "[extra_id_113]",
+    "[extra_id_114]",
+    "[extra_id_115]",
+    "[extra_id_116]",
+    "[extra_id_117]",
+    "[extra_id_118]",
+    "[extra_id_119]",
+    "[extra_id_120]",
+    "[extra_id_121]",
+    "[extra_id_122]",
+    "[extra_id_123]",
+    "[extra_id_124]",
+    "[extra_id_125]",
+    "[extra_id_126]",
+    "[extra_id_127]",
+    "[extra_id_128]",
+    "[extra_id_129]",
+    "[extra_id_130]",
+    "[extra_id_131]",
+    "[extra_id_132]",
+    "[extra_id_133]",
+    "[extra_id_134]",
+    "[extra_id_135]",
+    "[extra_id_136]",
+    "[extra_id_137]",
+    "[extra_id_138]",
+    "[extra_id_139]",
+    "[extra_id_140]",
+    "[extra_id_141]",
+    "[extra_id_142]",
+    "[extra_id_143]",
+    "[extra_id_144]",
+    "[extra_id_145]",
+    "[extra_id_146]",
+    "[extra_id_147]",
+    "[extra_id_148]",
+    "[extra_id_149]",
+    "[extra_id_150]",
+    "[extra_id_151]",
+    "[extra_id_152]",
+    "[extra_id_153]",
+    "[extra_id_154]",
+    "[extra_id_155]",
+    "[extra_id_156]",
+    "[extra_id_157]",
+    "[extra_id_158]",
+    "[extra_id_159]",
+    "[extra_id_160]",
+    "[extra_id_161]",
+    "[extra_id_162]",
+    "[extra_id_163]",
+    "[extra_id_164]",
+    "[extra_id_165]",
+    "[extra_id_166]",
+    "[extra_id_167]",
+    "[extra_id_168]",
+    "[extra_id_169]",
+    "[extra_id_170]",
+    "[extra_id_171]",
+    "[extra_id_172]",
+    "[extra_id_173]",
+    "[extra_id_174]",
+    "[extra_id_175]",
+    "[extra_id_176]",
+    "[extra_id_177]",
+    "[extra_id_178]",
+    "[extra_id_179]",
+    "[extra_id_180]",
+    "[extra_id_181]",
+    "[extra_id_182]",
+    "[extra_id_183]",
+    "[extra_id_184]",
+    "[extra_id_185]",
+    "[extra_id_186]",
+    "[extra_id_187]",
+    "[extra_id_188]",
+    "[extra_id_189]",
+    "[extra_id_190]",
+    "[extra_id_191]",
+    "[extra_id_192]",
+    "[extra_id_193]",
+    "[extra_id_194]",
+    "[extra_id_195]",
+    "[extra_id_196]",
+    "[extra_id_197]",
+    "[extra_id_198]",
+    "[|endofturn|]",
+    "PI:URL",
+    "PI:EMAIL",
+    "PI:ACCOUNT_NUM",
+    "PI:PHONE_NUM",
+    "PI:BUSINESS_NUM",
+    "PI:ANNON",
+    "PI:KEY",
+    "PI:ID",
+    "PI:IP_ADDRESS",
+    "PI:USER"
+  ],
+  "backend": "tokenizers",
+  "bos_token": "[BOS]",
+  "clean_up_tokenization_spaces": true,
+  "eos_token": "[|endofturn|]",
+  "errors": "replace",
+  "is_local": false,
+  "local_files_only": false,
+  "model_max_length": 1000000000000000019884624838656,
+  "pad_token": "[PAD]",
+  "tokenizer_class": "GPT2Tokenizer",
+  "unk_token": "[UNK]"
+}

checkpoint-40/trainer_state.json

@@ -0,0 +1,70 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 2.0,
+  "eval_steps": 500,
+  "global_step": 40,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.5228758169934641,
+      "grad_norm": 4.096013598842928,
+      "learning_rate": 9.644883599083959e-06,
+      "loss": 2.5423198699951173,
+      "mean_token_accuracy": 0.5685750585049391,
+      "num_tokens": 91495.0,
+      "step": 10
+    },
+    {
+      "epoch": 1.0,
+      "grad_norm": 6.747715140097201,
+      "learning_rate": 8.025871075968828e-06,
+      "loss": 1.4160140037536622,
+      "mean_token_accuracy": 0.6783009481756654,
+      "num_tokens": 176167.0,
+      "step": 20
+    },
+    {
+      "epoch": 1.522875816993464,
+      "grad_norm": 3.445904738767255,
+      "learning_rate": 5.540595092119709e-06,
+      "loss": 1.0047502517700195,
+      "mean_token_accuracy": 0.7559794817119837,
+      "num_tokens": 269468.0,
+      "step": 30
+    },
+    {
+      "epoch": 2.0,
+      "grad_norm": 7.0459795774794385,
+      "learning_rate": 2.9005544921986774e-06,
+      "loss": 0.8993229866027832,
+      "mean_token_accuracy": 0.77471841075649,
+      "num_tokens": 352326.0,
+      "step": 40
+    }
+  ],
+  "logging_steps": 10,
+  "max_steps": 60,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 3,
+  "save_steps": 500,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 1.382071083728896e+16,
+  "train_batch_size": 1,
+  "trial_name": null,
+  "trial_params": null
+}

checkpoint-40/zero_to_fp32.py

@@ -0,0 +1,760 @@
+#!/usr/bin/env python
+
+# Copyright (c) Microsoft Corporation.
+# SPDX-License-Identifier: Apache-2.0
+
+# DeepSpeed Team
+
+# This script extracts fp32 consolidated weights from a zero 1, 2 and 3 DeepSpeed checkpoints. It gets
+# copied into the top level checkpoint dir, so the user can easily do the conversion at any point in
+# the future. Once extracted, the weights don't require DeepSpeed and can be used in any
+# application.
+#
+# example:
+#   python zero_to_fp32.py . output_dir/
+#   or
+#   python zero_to_fp32.py . output_dir/ --safe_serialization
+
+import argparse
+import torch
+import glob
+import math
+import os
+import re
+import gc
+import json
+import numpy as np
+from tqdm import tqdm
+from collections import OrderedDict
+from dataclasses import dataclass
+
+# while this script doesn't use deepspeed to recover data, since the checkpoints are pickled with
+# DeepSpeed data structures it has to be available in the current python environment.
+from deepspeed.utils import logger
+from deepspeed.checkpoint.constants import (DS_VERSION, OPTIMIZER_STATE_DICT, SINGLE_PARTITION_OF_FP32_GROUPS,
+                                            FP32_FLAT_GROUPS, ZERO_STAGE, PARTITION_COUNT, PARAM_SHAPES, BUFFER_NAMES,
+                                            FROZEN_PARAM_SHAPES, FROZEN_PARAM_FRAGMENTS)
+
+
+@dataclass
+class zero_model_state:
+    buffers: dict()
+    param_shapes: dict()
+    shared_params: list
+    ds_version: int
+    frozen_param_shapes: dict()
+    frozen_param_fragments: dict()
+
+
+debug = 0
+
+# load to cpu
+device = torch.device('cpu')
+
+
+def atoi(text):
+    return int(text) if text.isdigit() else text
+
+
+def natural_keys(text):
+    '''
+    alist.sort(key=natural_keys) sorts in human order
+    http://nedbatchelder.com/blog/200712/human_sorting.html
+    (See Toothy's implementation in the comments)
+    '''
+    return [atoi(c) for c in re.split(r'(\d+)', text)]
+
+
+def get_model_state_file(checkpoint_dir, zero_stage):
+    if not os.path.isdir(checkpoint_dir):
+        raise FileNotFoundError(f"Directory '{checkpoint_dir}' doesn't exist")
+
+    # there should be only one file
+    if zero_stage <= 2:
+        file = os.path.join(checkpoint_dir, "mp_rank_00_model_states.pt")
+    elif zero_stage == 3:
+        file = os.path.join(checkpoint_dir, "zero_pp_rank_0_mp_rank_00_model_states.pt")
+
+    if not os.path.exists(file):
+        raise FileNotFoundError(f"can't find model states file at '{file}'")
+
+    return file
+
+
+def get_checkpoint_files(checkpoint_dir, glob_pattern):
+    # XXX: need to test that this simple glob rule works for multi-node setup too
+    ckpt_files = sorted(glob.glob(os.path.join(checkpoint_dir, glob_pattern)), key=natural_keys)
+
+    if len(ckpt_files) == 0:
+        raise FileNotFoundError(f"can't find {glob_pattern} files in directory '{checkpoint_dir}'")
+
+    return ckpt_files
+
+
+def get_optim_files(checkpoint_dir):
+    return get_checkpoint_files(checkpoint_dir, "*_optim_states.pt")
+
+
+def get_model_state_files(checkpoint_dir):
+    return get_checkpoint_files(checkpoint_dir, "*_model_states.pt")
+
+
+def parse_model_states(files):
+    zero_model_states = []
+    for file in files:
+        state_dict = torch.load(file, map_location=device, weights_only=False)
+
+        if BUFFER_NAMES not in state_dict:
+            raise ValueError(f"{file} is not a model state checkpoint")
+        buffer_names = state_dict[BUFFER_NAMES]
+        if debug:
+            print("Found buffers:", buffer_names)
+
+        # recover just the buffers while restoring them to fp32 if they were saved in fp16
+        buffers = {k: v.float() for k, v in state_dict["module"].items() if k in buffer_names}
+        param_shapes = state_dict[PARAM_SHAPES]
+
+        # collect parameters that are included in param_shapes
+        param_names = []
+        for s in param_shapes:
+            for name in s.keys():
+                param_names.append(name)
+
+        # update with frozen parameters
+        frozen_param_shapes = state_dict.get(FROZEN_PARAM_SHAPES, None)
+        if frozen_param_shapes is not None:
+            if debug:
+                print(f"Found frozen_param_shapes: {frozen_param_shapes}")
+            param_names += list(frozen_param_shapes.keys())
+
+        # handle shared params
+        shared_params = [[k, v] for k, v in state_dict["shared_params"].items()]
+
+        ds_version = state_dict.get(DS_VERSION, None)
+
+        frozen_param_fragments = state_dict.get(FROZEN_PARAM_FRAGMENTS, None)
+
+        z_model_state = zero_model_state(buffers=buffers,
+                                         param_shapes=param_shapes,
+                                         shared_params=shared_params,
+                                         ds_version=ds_version,
+                                         frozen_param_shapes=frozen_param_shapes,
+                                         frozen_param_fragments=frozen_param_fragments)
+        zero_model_states.append(z_model_state)
+
+    return zero_model_states
+
+
+def parse_optim_states(files, ds_checkpoint_dir):
+    total_files = len(files)
+    state_dicts = []
+    for f in tqdm(files, desc='Loading checkpoint shards'):
+        state_dict = torch.load(f, map_location=device, mmap=True, weights_only=False)
+        # immediately discard the potentially huge 2 optimizer states as we only care for fp32 master weights
+        # and also handle the case where it was already removed by another helper script
+        state_dict["optimizer_state_dict"].pop("optimizer_state_dict", None)
+        state_dicts.append(state_dict)
+
+    if not ZERO_STAGE in state_dicts[0][OPTIMIZER_STATE_DICT]:
+        raise ValueError(f"{files[0]} is not a zero checkpoint")
+    zero_stage = state_dicts[0][OPTIMIZER_STATE_DICT][ZERO_STAGE]
+    world_size = state_dicts[0][OPTIMIZER_STATE_DICT][PARTITION_COUNT]
+
+    # For ZeRO-2 each param group can have different partition_count as data parallelism for expert
+    # parameters can be different from data parallelism for non-expert parameters. So we can just
+    # use the max of the partition_count to get the dp world_size.
+
+    if type(world_size) is list:
+        world_size = max(world_size)
+
+    if world_size != total_files:
+        raise ValueError(
+            f"Expected {world_size} of '*_optim_states.pt' under '{ds_checkpoint_dir}' but found {total_files} files. "
+            "Possibly due to an overwrite of an old checkpoint, or a checkpoint didn't get saved by one or more processes."
+        )
+
+    # the groups are named differently in each stage
+    if zero_stage <= 2:
+        fp32_groups_key = SINGLE_PARTITION_OF_FP32_GROUPS
+    elif zero_stage == 3:
+        fp32_groups_key = FP32_FLAT_GROUPS
+    else:
+        raise ValueError(f"unknown zero stage {zero_stage}")
+
+    fp32_flat_groups = [state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key] for i in range(len(state_dicts))]
+    return zero_stage, world_size, fp32_flat_groups
+
+
+def _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir, exclude_frozen_parameters):
+    """
+    Returns fp32 state_dict reconstructed from ds checkpoint
+
+    Args:
+        - ``ds_checkpoint_dir``: path to the deepspeed checkpoint folder (where the optimizer files are)
+
+    """
+    print(f"Processing zero checkpoint '{ds_checkpoint_dir}'")
+
+    optim_files = get_optim_files(ds_checkpoint_dir)
+    zero_stage, world_size, fp32_flat_groups = parse_optim_states(optim_files, ds_checkpoint_dir)
+    print(f"Detected checkpoint of type zero stage {zero_stage}, world_size: {world_size}")
+
+    model_files = get_model_state_files(ds_checkpoint_dir)
+
+    zero_model_states = parse_model_states(model_files)
+    print(f'Parsing checkpoint created by deepspeed=={zero_model_states[0].ds_version}')
+
+    if zero_stage <= 2:
+        return _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states,
+                                                          exclude_frozen_parameters)
+    elif zero_stage == 3:
+        return _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states,
+                                                          exclude_frozen_parameters)
+
+
+def _zero2_merge_frozen_params(state_dict, zero_model_states):
+    if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
+        return
+
+    frozen_param_shapes = zero_model_states[0].frozen_param_shapes
+    frozen_param_fragments = zero_model_states[0].frozen_param_fragments
+
+    if debug:
+        num_elem = sum(s.numel() for s in frozen_param_shapes.values())
+        print(f'rank 0: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
+
+        wanted_params = len(frozen_param_shapes)
+        wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
+        avail_numel = sum([p.numel() for p in frozen_param_fragments.values()])
+        print(f'Frozen params: Have {avail_numel} numels to process.')
+        print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
+
+    total_params = 0
+    total_numel = 0
+    for name, shape in frozen_param_shapes.items():
+        total_params += 1
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+
+        state_dict[name] = frozen_param_fragments[name]
+
+        if debug:
+            print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
+
+    print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _has_callable(obj, fn):
+    attr = getattr(obj, fn, None)
+    return callable(attr)
+
+
+def _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
+    param_shapes = zero_model_states[0].param_shapes
+
+    # Reconstruction protocol:
+    #
+    # XXX: document this
+
+    if debug:
+        for i in range(world_size):
+            for j in range(len(fp32_flat_groups[0])):
+                print(f"{FP32_FLAT_GROUPS}[{i}][{j}].shape={fp32_flat_groups[i][j].shape}")
+
+    # XXX: memory usage doubles here (zero2)
+    num_param_groups = len(fp32_flat_groups[0])
+    merged_single_partition_of_fp32_groups = []
+    for i in range(num_param_groups):
+        merged_partitions = [sd[i] for sd in fp32_flat_groups]
+        full_single_fp32_vector = torch.cat(merged_partitions, 0)
+        merged_single_partition_of_fp32_groups.append(full_single_fp32_vector)
+    avail_numel = sum(
+        [full_single_fp32_vector.numel() for full_single_fp32_vector in merged_single_partition_of_fp32_groups])
+
+    if debug:
+        wanted_params = sum([len(shapes) for shapes in param_shapes])
+        wanted_numel = sum([sum(shape.numel() for shape in shapes.values()) for shapes in param_shapes])
+        # not asserting if there is a mismatch due to possible padding
+        print(f"Have {avail_numel} numels to process.")
+        print(f"Need {wanted_numel} numels in {wanted_params} params.")
+
+    # params
+    # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
+    # out-of-core computing solution
+    total_numel = 0
+    total_params = 0
+    for shapes, full_single_fp32_vector in zip(param_shapes, merged_single_partition_of_fp32_groups):
+        offset = 0
+        avail_numel = full_single_fp32_vector.numel()
+        for name, shape in shapes.items():
+
+            unpartitioned_numel = shape.numel() if _has_callable(shape, 'numel') else math.prod(shape)
+            total_numel += unpartitioned_numel
+            total_params += 1
+
+            if debug:
+                print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
+            state_dict[name] = full_single_fp32_vector.narrow(0, offset, unpartitioned_numel).view(shape)
+            offset += unpartitioned_numel
+
+        # Z2 started to align to 2*world_size to improve nccl performance. Therefore both offset and
+        # avail_numel can differ by anywhere between 0..2*world_size. Due to two unrelated complex
+        # paddings performed in the code it's almost impossible to predict the exact numbers w/o the
+        # live optimizer object, so we are checking that the numbers are within the right range
+        align_to = 2 * world_size
+
+        def zero2_align(x):
+            return align_to * math.ceil(x / align_to)
+
+        if debug:
+            print(f"original offset={offset}, avail_numel={avail_numel}")
+
+        offset = zero2_align(offset)
+        avail_numel = zero2_align(avail_numel)
+
+        if debug:
+            print(f"aligned  offset={offset}, avail_numel={avail_numel}")
+
+        # Sanity check
+        if offset != avail_numel:
+            raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
+
+    print(f"Reconstructed fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states,
+                                               exclude_frozen_parameters):
+    state_dict = OrderedDict()
+
+    # buffers
+    buffers = zero_model_states[0].buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    if not exclude_frozen_parameters:
+        _zero2_merge_frozen_params(state_dict, zero_model_states)
+
+    _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
+
+    # recover shared parameters
+    for pair in zero_model_states[0].shared_params:
+        if pair[1] in state_dict:
+            state_dict[pair[0]] = state_dict[pair[1]]
+
+    return state_dict
+
+
+def zero3_partitioned_param_info(unpartitioned_numel, world_size):
+    remainder = unpartitioned_numel % world_size
+    padding_numel = (world_size - remainder) if remainder else 0
+    partitioned_numel = math.ceil(unpartitioned_numel / world_size)
+    return partitioned_numel, padding_numel
+
+
+def _zero3_merge_frozen_params(state_dict, world_size, zero_model_states):
+    if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
+        return
+
+    if debug:
+        for i in range(world_size):
+            num_elem = sum(s.numel() for s in zero_model_states[i].frozen_param_fragments.values())
+            print(f'rank {i}: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
+
+        frozen_param_shapes = zero_model_states[0].frozen_param_shapes
+        wanted_params = len(frozen_param_shapes)
+        wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
+        avail_numel = sum([p.numel() for p in zero_model_states[0].frozen_param_fragments.values()]) * world_size
+        print(f'Frozen params: Have {avail_numel} numels to process.')
+        print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
+
+    total_params = 0
+    total_numel = 0
+    for name, shape in zero_model_states[0].frozen_param_shapes.items():
+        total_params += 1
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+
+        param_frags = tuple(model_state.frozen_param_fragments[name] for model_state in zero_model_states)
+        state_dict[name] = torch.cat(param_frags, 0).narrow(0, 0, unpartitioned_numel).view(shape)
+
+        partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
+
+        if debug:
+            print(
+                f"Frozen params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
+            )
+
+    print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
+
+
+class GatheredTensor:
+    """
+    A pseudo tensor that collects partitioned weights.
+    It is more memory efficient when there are multiple groups.
+    """
+
+    def __init__(self, flat_groups, flat_groups_offset, offset, partitioned_numel, shape):
+        self.flat_groups = flat_groups
+        self.flat_groups_offset = flat_groups_offset
+        self.offset = offset
+        self.partitioned_numel = partitioned_numel
+        self.shape = shape
+        self.dtype = self.flat_groups[0][0].dtype
+
+    def contiguous(self):
+        """
+        Merge partitioned weights from flat_groups into a single tensor.
+        """
+        end_idx = self.offset + self.partitioned_numel
+        world_size = len(self.flat_groups)
+        pad_flat_param_chunks = []
+
+        for rank_i in range(world_size):
+            # for each rank, we need to collect weights from related group/groups
+            flat_groups_at_rank_i = self.flat_groups[rank_i]
+            start_group_id = None
+            end_group_id = None
+            for group_id in range(len(self.flat_groups_offset)):
+                if self.flat_groups_offset[group_id] <= self.offset < self.flat_groups_offset[group_id + 1]:
+                    start_group_id = group_id
+                if self.flat_groups_offset[group_id] < end_idx <= self.flat_groups_offset[group_id + 1]:
+                    end_group_id = group_id
+                    break
+            # collect weights from related group/groups
+            for group_id in range(start_group_id, end_group_id + 1):
+                flat_tensor = flat_groups_at_rank_i[group_id]
+                start_offset = self.offset - self.flat_groups_offset[group_id]
+                end_offset = min(end_idx, self.flat_groups_offset[group_id + 1]) - self.flat_groups_offset[group_id]
+                pad_flat_param_chunks.append(flat_tensor[start_offset:end_offset])
+
+        # collect weights from all ranks
+        pad_flat_param = torch.cat(pad_flat_param_chunks, dim=0)
+        param = pad_flat_param[:self.shape.numel()].view(self.shape).contiguous()
+        return param
+
+
+def _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
+    param_shapes = zero_model_states[0].param_shapes
+    avail_numel = sum([flat_group.numel() for flat_group in fp32_flat_groups[0]]) * world_size
+
+    # Reconstruction protocol: For zero3 we need to zip the partitions together at boundary of each
+    # param, re-consolidating each param, while dealing with padding if any
+
+    # merge list of dicts, preserving order
+    param_shapes = {k: v for d in param_shapes for k, v in d.items()}
+
+    if debug:
+        for i in range(world_size):
+            print(f"{FP32_FLAT_GROUPS}[{i}].shape={fp32_flat_groups[i].shape}")
+
+        wanted_params = len(param_shapes)
+        wanted_numel = sum(shape.numel() for shape in param_shapes.values())
+        # not asserting if there is a mismatch due to possible padding
+        avail_numel = fp32_flat_groups[0].numel() * world_size
+        print(f"Trainable params: Have {avail_numel} numels to process.")
+        print(f"Trainable params: Need {wanted_numel} numels in {wanted_params} params.")
+
+    # params
+    # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
+    # out-of-core computing solution
+    offset = 0
+    total_numel = 0
+    total_params = 0
+    flat_groups_offset = [0] + list(np.cumsum([flat_tensor.numel() for flat_tensor in fp32_flat_groups[0]]))
+    for name, shape in tqdm(param_shapes.items(), desc='Gathering sharded weights'):
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+        total_params += 1
+        partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
+
+        if debug:
+            print(
+                f"Trainable params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
+            )
+
+        # memory efficient tensor
+        tensor = GatheredTensor(fp32_flat_groups, flat_groups_offset, offset, partitioned_numel, shape)
+        state_dict[name] = tensor
+        offset += partitioned_numel
+
+    offset *= world_size
+
+    # Sanity check
+    if offset != avail_numel:
+        raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
+
+    print(f"Reconstructed Trainable fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states,
+                                               exclude_frozen_parameters):
+    state_dict = OrderedDict()
+
+    # buffers
+    buffers = zero_model_states[0].buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    if not exclude_frozen_parameters:
+        _zero3_merge_frozen_params(state_dict, world_size, zero_model_states)
+
+    _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
+
+    # recover shared parameters
+    for pair in zero_model_states[0].shared_params:
+        if pair[1] in state_dict:
+            state_dict[pair[0]] = state_dict[pair[1]]
+
+    return state_dict
+
+
+def to_torch_tensor(state_dict, return_empty_tensor=False):
+    """
+    Convert state_dict of GatheredTensor to torch tensor
+    """
+    torch_state_dict = {}
+    converted_tensors = {}
+    for name, tensor in state_dict.items():
+        tensor_id = id(tensor)
+        if tensor_id in converted_tensors:  # shared tensors
+            shared_tensor = torch_state_dict[converted_tensors[tensor_id]]
+            torch_state_dict[name] = shared_tensor
+        else:
+            converted_tensors[tensor_id] = name
+            if return_empty_tensor:
+                torch_state_dict[name] = torch.empty(tensor.shape, dtype=tensor.dtype)
+            else:
+                torch_state_dict[name] = tensor.contiguous()
+    return torch_state_dict
+
+
+def get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir,
+                                             tag=None,
+                                             exclude_frozen_parameters=False,
+                                             lazy_mode=False):
+    """
+    Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated state_dict that can be loaded with
+    ``load_state_dict()`` and used for training without DeepSpeed or shared with others, for example
+    via a model hub.
+
+    Args:
+        - ``checkpoint_dir``: path to the desired checkpoint folder
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in 'latest' file. e.g., ``global_step14``
+        - ``exclude_frozen_parameters``: exclude frozen parameters
+        - ``lazy_mode``: get state_dict in lazy mode. It returns a dict of pesduo tensor instead of torch tensor, which is more memory efficient.
+          Convert the pesduo tensor to torch tensor by ``.contiguous()``
+
+    Returns:
+        - pytorch ``state_dict``
+
+    A typical usage might be ::
+
+        from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint
+        # do the training and checkpoint saving
+        state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir) # already on cpu
+        model = model.cpu() # move to cpu
+        model.load_state_dict(state_dict)
+        # submit to model hub or save the model to share with others
+
+    In this example the ``model`` will no longer be usable in the deepspeed context of the same
+    application. i.e. you will need to re-initialize the deepspeed engine, since
+    ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
+
+    If you want it all done for you, use ``load_state_dict_from_zero_checkpoint`` instead.
+
+    Note: the above usage may not work if your application doesn't have sufficient free CPU memory.
+    You may need to use the offline approach using the ``zero_to_fp32.py`` script that is saved with
+    the checkpoint. Or you can load state_dict in lazy mode ::
+
+        from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint
+        state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, lazy_mode=True) # not on cpu
+        for name, lazy_tensor in state_dict.item():
+            tensor = lazy_tensor.contiguous()  # to cpu
+            print(name, tensor)
+            # del tensor to release memory if it no longer in use
+    """
+    if tag is None:
+        latest_path = os.path.join(checkpoint_dir, 'latest')
+        if os.path.isfile(latest_path):
+            with open(latest_path, 'r') as fd:
+                tag = fd.read().strip()
+        else:
+            raise ValueError(f"Unable to find 'latest' file at {latest_path}")
+
+    ds_checkpoint_dir = os.path.join(checkpoint_dir, tag)
+
+    if not os.path.isdir(ds_checkpoint_dir):
+        raise FileNotFoundError(f"Directory '{ds_checkpoint_dir}' doesn't exist")
+
+    state_dict = _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir, exclude_frozen_parameters)
+    if lazy_mode:
+        return state_dict
+    else:
+        return to_torch_tensor(state_dict)
+
+
+def convert_zero_checkpoint_to_fp32_state_dict(checkpoint_dir,
+                                               output_dir,
+                                               max_shard_size="5GB",
+                                               safe_serialization=False,
+                                               tag=None,
+                                               exclude_frozen_parameters=False):
+    """
+    Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict`` file that can be
+    loaded with ``torch.load(file)`` + ``load_state_dict()`` and used for training without DeepSpeed.
+
+    Args:
+        - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
+        - ``output_dir``: directory to the pytorch fp32 state_dict output files
+        - ``max_shard_size``: the maximum size for a checkpoint before being sharded, default value is 5GB
+        - ``safe_serialization``:  whether to save the model using `safetensors` or the traditional PyTorch way (that uses `pickle`).
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
+        - ``exclude_frozen_parameters``: exclude frozen parameters
+    """
+
+    # Dependency pre-check
+    if safe_serialization:
+        try:
+            from safetensors.torch import save_file
+        except ImportError:
+            print('If you want to use `safe_serialization`, please `pip install safetensors`')
+            raise
+    if max_shard_size is not None:
+        try:
+            from huggingface_hub import split_torch_state_dict_into_shards
+        except ImportError:
+            print('If you want to use `max_shard_size`, please `pip install huggingface_hub`')
+            raise
+
+    # Convert zero checkpoint to state_dict
+    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir,
+                                                          tag,
+                                                          exclude_frozen_parameters,
+                                                          lazy_mode=True)
+
+    # Shard the model if it is too big.
+    weights_name = "model.safetensors" if safe_serialization else "pytorch_model.bin"
+    if max_shard_size is not None:
+        filename_pattern = weights_name.replace(".bin", "{suffix}.bin").replace(".safetensors", "{suffix}.safetensors")
+        # an memory-efficient approach for sharding
+        empty_state_dict = to_torch_tensor(state_dict, return_empty_tensor=True)
+        state_dict_split = split_torch_state_dict_into_shards(empty_state_dict,
+                                                              filename_pattern=filename_pattern,
+                                                              max_shard_size=max_shard_size)
+    else:
+        from collections import namedtuple
+        StateDictSplit = namedtuple("StateDictSplit", ["is_sharded", "filename_to_tensors"])
+        state_dict_split = StateDictSplit(is_sharded=False,
+                                          filename_to_tensors={weights_name: list(state_dict.keys())})
+
+    # Save the model by shard
+    os.makedirs(output_dir, exist_ok=True)
+    filename_to_tensors = state_dict_split.filename_to_tensors.items()
+    for shard_file, tensors in tqdm(filename_to_tensors, desc="Saving checkpoint shards"):
+        shard_state_dict = {tensor_name: state_dict[tensor_name] for tensor_name in tensors}
+        shard_state_dict = to_torch_tensor(shard_state_dict)
+        output_path = os.path.join(output_dir, shard_file)
+        if safe_serialization:
+            save_file(shard_state_dict, output_path, metadata={"format": "pt"})
+        else:
+            torch.save(shard_state_dict, output_path)
+        # release the memory of current shard
+        for tensor_name in list(shard_state_dict.keys()):
+            del state_dict[tensor_name]
+            del shard_state_dict[tensor_name]
+        del shard_state_dict
+        gc.collect()
+
+    # Save index if sharded
+    if state_dict_split.is_sharded:
+        index = {
+            "metadata": state_dict_split.metadata,
+            "weight_map": state_dict_split.tensor_to_filename,
+        }
+        save_index_file = "model.safetensors.index.json" if safe_serialization else "pytorch_model.bin.index.json"
+        save_index_file = os.path.join(output_dir, save_index_file)
+        with open(save_index_file, "w", encoding="utf-8") as f:
+            content = json.dumps(index, indent=2, sort_keys=True) + "\n"
+            f.write(content)
+
+
+def load_state_dict_from_zero_checkpoint(model, checkpoint_dir, tag=None):
+    """
+    1. Put the provided model to cpu
+    2. Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict``
+    3. Load it into the provided model
+
+    Args:
+        - ``model``: the model object to update
+        - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
+
+    Returns:
+        - ``model`: modified model
+
+    Make sure you have plenty of CPU memory available before you call this function. If you don't
+    have enough use the ``zero_to_fp32.py`` utility to do the conversion. You will find it
+    conveniently placed for you in the checkpoint folder.
+
+    A typical usage might be ::
+
+        from deepspeed.utils.zero_to_fp32 import load_state_dict_from_zero_checkpoint
+        model = load_state_dict_from_zero_checkpoint(trainer.model, checkpoint_dir)
+        # submit to model hub or save the model to share with others
+
+    Note, that once this was run, the ``model`` will no longer be usable in the deepspeed context
+    of the same application. i.e. you will need to re-initialize the deepspeed engine, since
+    ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
+
+    """
+    logger.info(f"Extracting fp32 weights")
+    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
+
+    logger.info(f"Overwriting model with fp32 weights")
+    model = model.cpu()
+    model.load_state_dict(state_dict, strict=False)
+
+    return model
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser()
+    parser.add_argument("checkpoint_dir",
+                        type=str,
+                        help="path to the desired checkpoint folder, e.g., path/checkpoint-12")
+    parser.add_argument("output_dir",
+                        type=str,
+                        help="directory to the pytorch fp32 state_dict output files"
+                        "(e.g. path/checkpoint-12-output/)")
+    parser.add_argument(
+        "--max_shard_size",
+        type=str,
+        default="5GB",
+        help="The maximum size for a checkpoint before being sharded. Checkpoints shard will then be each of size"
+        "lower than this size. If expressed as a string, needs to be digits followed by a unit (like `5MB`"
+        "We default it to 5GB in order for models to be able to run easily on free-tier google colab instances"
+        "without CPU OOM issues.")
+    parser.add_argument(
+        "--safe_serialization",
+        default=False,
+        action='store_true',
+        help="Whether to save the model using `safetensors` or the traditional PyTorch way (that uses `pickle`).")
+    parser.add_argument("-t",
+                        "--tag",
+                        type=str,
+                        default=None,
+                        help="checkpoint tag used as a unique identifier for checkpoint. e.g., global_step1")
+    parser.add_argument("--exclude_frozen_parameters", action='store_true', help="exclude frozen parameters")
+    parser.add_argument("-d", "--debug", action='store_true', help="enable debug")
+    args = parser.parse_args()
+
+    debug = args.debug
+
+    convert_zero_checkpoint_to_fp32_state_dict(args.checkpoint_dir,
+                                               args.output_dir,
+                                               max_shard_size=args.max_shard_size,
+                                               safe_serialization=args.safe_serialization,
+                                               tag=args.tag,
+                                               exclude_frozen_parameters=args.exclude_frozen_parameters)