Update configuration_gemmoe.py

configuration_gemmoe.py

@@ -1,103 +1,132 @@
+# coding=utf-8
+# Copyright 2024 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.
+""" Gemmoe model configuration"""
+
 from transformers.configuration_utils import PretrainedConfig
 from transformers.utils import logging
 
+
 logger = logging.get_logger(__name__)
 
-GEMMOE_PRETRAINED_CONFIG_ARCHIVE_MAP = {}
+GEMMOE_PRETRAINED_CONFIG_ARCHIVE_MAP = {
+    "Crystalcareai/GemMoE-Beta-1": "https://huggingface.co/Crystalcareai/GemMoE-Beta-1/resolve/main/config.json",
+}
+
+
 class GemmoeConfig(PretrainedConfig):
     r"""
-    This is the configuration class to store the configuration of a [`DeepseekModel`]. It is used to instantiate an DeepSeek
+    This is the configuration class to store the configuration of a [`GemmoeModel`]. It is used to instantiate a Gemmoe
     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 DeepSeek-7B.
+    defaults will yield a similar configuration to that of the Gemmoe-7B.
+
+    e.g. [mhenrichsen/gemmoe-7b](https://huggingface.co/mhenrichsen/gemmoe-7b)
 
     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 Deep model. Defines the number of different tokens that can be represented by the
-            `inputs_ids` passed when calling [`DeepseekModel`]
-        hidden_size (`int`, *optional*, defaults to 4096):
+        vocab_size (`int`, *optional*, defaults to 256000):
+            Vocabulary size of the Gemmoe model. Defines the number of different tokens that can be represented by the
+            `inputs_ids` passed when calling [`GemmoeModel`]
+        hidden_size (`int`, *optional*, defaults to 3072):
             Dimension of the hidden representations.
-        intermediate_size (`int`, *optional*, defaults to 11008):
+        intermediate_size (`int`, *optional*, defaults to 24576):
             Dimension of the MLP representations.
-        moe_intermediate_size (`int`, *optional*, defaults to 1407):
-            Dimension of the MoE representations.
-        num_hidden_layers (`int`, *optional*, defaults to 32):
+        num_hidden_layers (`int`, *optional*, defaults to 28):
             Number of hidden layers in the Transformer decoder.
-        num_attention_heads (`int`, *optional*, defaults to 32):
+        num_attention_heads (`int`, *optional*, defaults to 16):
             Number of attention heads for each attention layer in the Transformer decoder.
-        n_shared_experts (`int`, *optional*, defaults to None):
-            Number of shared experts, None means dense model.
-        n_routed_experts (`int`, *optional*, defaults to None):
-            Number of routed experts, None means dense model.
-        num_experts_per_tok (`int`, *optional*, defaults to None):
-            Number of selected experts, None means dense model.
-        moe_layer_freq (`int`, *optional*, defaults to 1):
-            The frequency of the MoE layer: one expert layer for every `moe_layer_freq - 1` dense layers.
-        first_k_dense_replace (`int`, *optional*, defaults to 0):
-            Number of dense layers in shallow layers(embed->dense->dense->...->dense->moe->moe...->lm_head).
-                                                            \--k dense layers--/
-        norm_topk_prob (`bool`, *optional*, defaults to False):
-            Whether to normalize the weights of the routed experts.
-        scoring_func (`str`, *optional*, defaults to 'softmax'):
-            Method of computing expert weights.
-        aux_loss_alpha (`float`, *optional*, defaults to 0.001):
-            Auxiliary loss weight coefficient.
-        seq_aux = (`bool`, *optional*, defaults to True):
-            Whether to compute the auxiliary loss for each individual sample.
-        num_key_value_heads (`int`, *optional*):
+        num_key_value_heads (`int`, *optional*, defaults to 16):
             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
+            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) in the decoder.
-        max_position_embeddings (`int`, *optional*, defaults to 2048):
+        head_dim (`int`, *optional*, defaults to 256):
+            The attention head dimension.
+        hidden_act (`str` or `function`, *optional*, defaults to `"gelu"`):
+            The non-linear activation function (function or string) in the decoder.  
+        max_position_embeddings (`int`, *optional*, defaults to 8192):
             The maximum sequence length that this model might ever be used with.
         initializer_range (`float`, *optional*, defaults to 0.02):
             The standard deviation of the truncated_normal_initializer for initializing all weight matrices.
-        rms_norm_eps (`float`, *optional*, defaults to 1e-06):
+        rms_norm_eps (`float`, *optional*, defaults to 1e-6):  
             The epsilon used by the rms 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`.
-        pad_token_id (`int`, *optional*):
+        pad_token_id (`int`, *optional*, defaults to 0):
             Padding token id.
-        bos_token_id (`int`, *optional*, defaults to 1):
+        eos_token_id (`int`, *optional*, defaults to 1):
+            End of stream token id.  
+        bos_token_id (`int`, *optional*, defaults to 2):
             Beginning of stream token id.
-        eos_token_id (`int`, *optional*, defaults to 2):
-            End of stream token id.
-        pretraining_tp (`int`, *optional*, defaults to 1):
-            Experimental feature. Tensor parallelism rank used during pretraining. Please refer to [this
-            document](https://huggingface.co/docs/transformers/parallelism) to understand more about it. This value is
-            necessary to ensure exact reproducibility of the pretraining results. Please refer to [this
-            issue](https://github.com/pytorch/pytorch/issues/76232).
-        tie_word_embeddings (`bool`, *optional*, defaults to `False`):
+        tie_word_embeddings (`bool`, *optional*, defaults to `True`): 
             Whether to tie weight embeddings
         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. Currently supports two scaling
-            strategies: linear and dynamic. Their scaling factor must be a float greater than 1. The expected format is
-            `{"type": strategy name, "factor": scaling factor}`. When using this flag, don't update
-            `max_position_embeddings` to the expected new maximum.
         attention_bias (`bool`, defaults to `False`, *optional*, defaults to `False`):
             Whether to use a bias in the query, key, value and output projection layers during self-attention.
         attention_dropout (`float`, *optional*, defaults to 0.0):
             The dropout ratio for the attention probabilities.
+        num_experts_per_tok (`int`, *optional*, defaults to 2):
+            The number of experts used in the sparse mixture of experts layer.
+        num_local_experts (`int`, *optional*, defaults to 8):  
+            The number of local experts used in the sparse mixture of experts layer.
+        router_aux_loss_coef (`float`, *optional*, defaults to 0.01):
+            The coefficient for the auxiliary loss of the router.
+        output_router_logits (`bool`, *optional*, defaults to `False`):
+            Whether or not to output the logits of the routers. They are useful for computing the router loss, and
+            should not be returned during inference.
+        n_shared_experts (`int`, *optional*, defaults to `None`):
+            The number of shared experts used in the sparse mixture of experts layer. If set to `None`, no shared
+            experts are used.
+        n_routed_experts (`int`, *optional*, defaults to `None`):
+            The number of routed experts used in the sparse mixture of experts layer. If set to `None`, all experts are
+            routed experts.
+        moe_layer_freq (`int`, *optional*, defaults to 1):
+            The frequency of MoE layers in the model. A value of 1 means MoE layers are used in every layer, a value of
+            2 means MoE layers are used in every other layer, and so on.
+        first_k_dense_replace (`int`, *optional*, defaults to 0):
+            The number of initial dense layers to replace with MoE layers. If set to 0 (default), no dense layers are
+            replaced.
+        norm_topk_prob (`bool`, *optional*, defaults to `False`):
+            Whether to normalize the top-k probabilities of the router during training.
+        scoring_func (`str`, *optional*, defaults to `'softmax'`):
+            The scoring function used by the router. Can be 'softmax' or 'remap'.
+        aux_loss_alpha (`float`, *optional*, defaults to 0.001):
+            The weight of the auxiliary loss used for training the router.
+        seq_aux (`bool`, *optional*, defaults to `True`):
+            Whether to use sequence-level auxiliary loss for training the router.
+        pretraining_tp (`int`, *optional*, defaults to 1):
+            The tensor parallelism used for pretraining.
+        rope_scaling (`float`, *optional*, defaults to `None`):
+            The scaling factor for the Rotary Position Embedding (RoPE). If set to `None`, no scaling is applied.
 
     ```python
-    >>> from transformers import DeepseekModel, DeepseekConfig
+    >>> from transformers import GemmoeModel, GemmoeConfig
 
-    >>> # Initializing a Deepseek deepseek-7b style configuration
-    >>> configuration = DeepseekConfig()
+    >>> # Initializing a Gemmoe gemmoe-7b style configuration
+    >>> configuration = GemmoeConfig()
 
-    >>> # Accessing the model configuration
+    >>> # Initializing a model from the gemmoe-7b style configuration
+    >>> model = GemmoeModel(configuration)
+
+    >>> # Accessing the model configuration 
     >>> configuration = model.config
     ```"""
 
@@ -109,92 +138,72 @@ class GemmoeConfig(PretrainedConfig):
         vocab_size=256000,
         hidden_size=3072,
         intermediate_size=24576,
-        moe_intermediate_size = 1407,
         num_hidden_layers=28,
         num_attention_heads=16,
         num_key_value_heads=16,
-        n_shared_experts = None,
-        n_routed_experts = None,
-        num_experts_per_tok = None,
-        moe_layer_freq = 1,
-        first_k_dense_replace = 0,
-        norm_topk_prob = False,
-        scoring_func = 'softmax',
-        aux_loss_alpha = 0.001,
-        seq_aux = True,
+        head_dim=256,
         hidden_act="gelu",
-        max_position_embeddings=2048,
+        max_position_embeddings=8192,
         initializer_range=0.02,
         rms_norm_eps=1e-6,
         use_cache=True,
         pad_token_id=0,
-        bos_token_id=2,
         eos_token_id=1,
-        pretraining_tp=1,
+        bos_token_id=2,
         tie_word_embeddings=True,
         rope_theta=10000.0,
-        rope_scaling=None,
         attention_bias=False,
         attention_dropout=0.0,
+        num_experts_per_tok=2,
+        num_local_experts=8,
+        n_shared_experts=None,
+        n_routed_experts=None,
+        moe_layer_freq=1,
+        first_k_dense_replace=0,
+        norm_topk_prob=False,
+        scoring_func='softmax',
+        aux_loss_alpha=0.001,
+        seq_aux=True,
+        pretraining_tp=1,
+        rope_scaling=None,
+        router_aux_loss_coef=0.02,
+        output_router_logits=False,
         **kwargs,
     ):
         self.vocab_size = vocab_size
-        self.max_position_embeddings = max_position_embeddings
+        self.max_position_embeddings = max_position_embeddings        
         self.hidden_size = hidden_size
         self.intermediate_size = intermediate_size
-        self.moe_intermediate_size = moe_intermediate_size
         self.num_hidden_layers = num_hidden_layers
         self.num_attention_heads = num_attention_heads
+        self.head_dim = head_dim
+        self.num_key_value_heads = num_key_value_heads
+        self.hidden_act = hidden_act
+        self.initializer_range = initializer_range
+        self.rms_norm_eps = rms_norm_eps
+        self.use_cache = use_cache
+        self.rope_theta = rope_theta
+        self.attention_bias = attention_bias 
+        self.attention_dropout = attention_dropout
+        self.num_experts_per_tok = num_experts_per_tok
+        self.num_local_experts = num_local_experts
         self.n_shared_experts = n_shared_experts
         self.n_routed_experts = n_routed_experts
-        self.num_experts_per_tok = num_experts_per_tok
         self.moe_layer_freq = moe_layer_freq
         self.first_k_dense_replace = first_k_dense_replace
         self.norm_topk_prob = norm_topk_prob
         self.scoring_func = scoring_func
         self.aux_loss_alpha = aux_loss_alpha
         self.seq_aux = seq_aux
-        # for backward compatibility
-        if num_key_value_heads is None:
-            num_key_value_heads = num_attention_heads
-
-        self.num_key_value_heads = num_key_value_heads
-        self.hidden_act = hidden_act
-        self.initializer_range = initializer_range
-        self.rms_norm_eps = rms_norm_eps
         self.pretraining_tp = pretraining_tp
-        self.use_cache = use_cache
-        self.rope_theta = rope_theta
         self.rope_scaling = rope_scaling
-        self._rope_scaling_validation()
-        self.attention_bias = attention_bias
-        self.attention_dropout = attention_dropout
-
+        self.router_aux_loss_coef = router_aux_loss_coef
+        self.output_router_logits = output_router_logits
+         
         super().__init__(
             pad_token_id=pad_token_id,
             bos_token_id=bos_token_id,
             eos_token_id=eos_token_id,
             tie_word_embeddings=tie_word_embeddings,
             **kwargs,
-        )
-
-    def _rope_scaling_validation(self):
-        """
-        Validate the `rope_scaling` configuration.
-        """
-        if self.rope_scaling is None:
-            return
-
-        if not isinstance(self.rope_scaling, dict) or len(self.rope_scaling) != 2:
-            raise ValueError(
-                "`rope_scaling` must be a dictionary with with two fields, `type` and `factor`, "
-                f"got {self.rope_scaling}"
-            )
-        rope_scaling_type = self.rope_scaling.get("type", None)
-        rope_scaling_factor = self.rope_scaling.get("factor", None)
-        if rope_scaling_type is None or rope_scaling_type not in ["linear", "dynamic"]:
-            raise ValueError(
-                f"`rope_scaling`'s type field must be one of ['linear', 'dynamic'], got {rope_scaling_type}"
-            )
-        if rope_scaling_factor is None or not isinstance(rope_scaling_factor, float) or rope_scaling_factor <= 1.0:
-            raise ValueError(f"`rope_scaling`'s factor field must be a float > 1, got {rope_scaling_factor}")
+        )