configuration_upcycling_qwen2_moe.py

# coding=utf-8
# Copyright 2024 The Qwen team, Alibaba Group and the HuggingFace Inc. team. All rights reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
"""Qwen2MoE model configuration"""

from transformers.configuration_utils import PretrainedConfig
from transformers.utils import logging
import torch


logger = logging.get_logger(__name__)


class Qwen2Config(PretrainedConfig):
    def __init__(
        self,
        vocab_size=151936,
        hidden_size=4096,
        intermediate_size=22016,
        num_hidden_layers=32,
        num_attention_heads=32,
        num_key_value_heads=32,
        hidden_act="silu",
        max_position_embeddings=32768,
        initializer_range=0.02,
        rms_norm_eps=1e-6,
        use_cache=True,
        tie_word_embeddings=False,
        rope_theta=10000.0,
        use_sliding_window=False,
        sliding_window=4096,
        max_window_layers=28,
        attention_dropout=0.0,
        **kwargs,
    ):              
        self.vocab_size = vocab_size
        self.max_position_embeddings = max_position_embeddings
        self.hidden_size = hidden_size
        self.intermediate_size = intermediate_size
        self.num_hidden_layers = num_hidden_layers
        self.num_attention_heads = num_attention_heads
        self.use_sliding_window = use_sliding_window
        self.sliding_window = sliding_window
        self.max_window_layers = max_window_layers

        # 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.use_cache = use_cache
        self.rope_theta = rope_theta
        self.attention_dropout = attention_dropout

        super().__init__(
            tie_word_embeddings=tie_word_embeddings,
            **kwargs,
        )


class Qwen2MoeConfig(PretrainedConfig):

    model_type = "qwen2_moe"
    keys_to_ignore_at_inference = ["past_key_values"]

    def __init__(
        self,
        vocab_size=151936,
        hidden_size=2048,
        intermediate_size=5632,
        num_hidden_layers=24,
        num_attention_heads=16,
        num_key_value_heads=16,
        hidden_act="silu",
        max_position_embeddings=32768,
        initializer_range=0.02,
        rms_norm_eps=1e-6,
        use_cache=True,
        tie_word_embeddings=False,
        rope_theta=10000.0,
        use_sliding_window=False,
        sliding_window=4096,
        max_window_layers=28,
        attention_dropout=0.0,
        
        decoder_sparse_step=1,
        moe_intermediate_size=1408,
        shared_expert_intermediate_size=5632,
        num_experts_per_tok=4,
        num_experts=60,
        norm_topk_prob=False,
        output_router_logits=False,
        router_aux_loss_coef=0.001,
        mlp_only_layers=None,
        **kwargs,
    ):
        self.vocab_size = vocab_size
        self.max_position_embeddings = max_position_embeddings
        self.hidden_size = hidden_size
        self.intermediate_size = intermediate_size
        self.num_hidden_layers = num_hidden_layers
        self.num_attention_heads = num_attention_heads
        self.use_sliding_window = use_sliding_window
        self.sliding_window = sliding_window
        self.max_window_layers = max_window_layers

        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_dropout = attention_dropout

        # MoE arguments
        self.decoder_sparse_step = decoder_sparse_step
        self.moe_intermediate_size = moe_intermediate_size
        self.shared_expert_intermediate_size = shared_expert_intermediate_size
        self.num_experts_per_tok = num_experts_per_tok
        self.num_experts = num_experts
        self.norm_topk_prob = norm_topk_prob
        self.output_router_logits = output_router_logits
        self.router_aux_loss_coef = router_aux_loss_coef
        self.mlp_only_layers = [] if mlp_only_layers is None else mlp_only_layers

        super().__init__(
            tie_word_embeddings=tie_word_embeddings,
            **kwargs,
        )


class UpcyclingQwen2MoeConfig(Qwen2Config):
    model_type="upcycling-qwen2-moe"
    #upcycling form Qwen2-1_5B
    def __init__(
        self,
        decoder_sparse_step=1,
        num_experts_per_tok=2,
        num_experts=7,
        norm_topk_prob=False,
        output_router_logits=False,
        router_aux_loss_coef=0.000,
        mlp_only_layers=None,#MoE only last 2 layers
        share_flag=False,
        attn_init_change=False,
        language_gate=False,
        **kwargs
    ):
        super().__init__(**kwargs)
        # MoE arguments
        self.decoder_sparse_step = decoder_sparse_step
        self.moe_intermediate_size = self.intermediate_size
        self.shared_expert_intermediate_size = self.intermediate_size
        self.norm_topk_prob = norm_topk_prob
        self.output_router_logits = output_router_logits
        self.router_aux_loss_coef = router_aux_loss_coef
        # self.mlp_only_layers = [] if mlp_only_layers is None else mlp_only_layers
        self.mlp_only_layers=torch.arange(self.num_hidden_layers).tolist()[:-2]
        self.share_flag=share_flag
        self.num_experts_per_tok = num_experts_per_tok
        self.num_experts = num_experts
        self.attn_init_change=attn_init_change
        self.language_gate=language_gate