初始化项目，由ModelHub XC社区提供模型

Model: inclusionAI/Ling-mini-base-2.0-10T
Source: Original Platform

configuration_bailing_moe_v2.py

@@ -0,0 +1,84 @@
+"""Bailing MoE V2 model configuration"""
+
+from transformers.configuration_utils import PretrainedConfig
+
+
+class BailingMoeV2Config(PretrainedConfig):
+
+    def __init__(
+        self,
+        vocab_size=157184,
+        hidden_size=2048,
+        intermediate_size=5120,
+        num_hidden_layers=20,
+        num_attention_heads=16,
+        num_key_value_heads=4,
+        hidden_act="silu",
+        use_qkv_bias=False,  # bailing only
+        use_bias=False,  # bailing only
+        rms_norm_eps=1e-06,
+        tie_word_embeddings=False,  # PretrainedConfig key, here change default value.
+        embedding_dropout=0.0,
+        attention_dropout=0.0,
+        output_dropout=0.0,
+        initializer_range=0.02,
+        max_position_embeddings=32768,
+        rope_theta=600000.0,
+        use_cache=True,
+        max_window_layers=20,
+        rope_scaling=None,
+        pad_token_id=156892,
+        eos_token_id=156892,
+        num_experts=256,
+        num_shared_experts=1,
+        num_experts_per_tok=8,
+        n_group=8,
+        topk_group=4,
+        moe_intermediate_size=512,
+        first_k_dense_replace=1,
+        head_dim=128,
+        output_router_logits=False,
+        use_qk_norm=True,
+        num_nextn_predict_layers=0,
+        mtp_loss_scaling_factor=0,
+        moe_router_enable_expert_bias=True,
+        routed_scaling_factor=1.0,
+        **kwargs,
+    ):
+        self.num_hidden_layers = num_hidden_layers
+        self.vocab_size = vocab_size
+        self.hidden_size = hidden_size
+        self.intermediate_size = intermediate_size
+        self.num_attention_heads = num_attention_heads
+        self.num_key_value_heads = num_key_value_heads
+        self.hidden_act = hidden_act
+        self.use_qkv_bias = use_qkv_bias
+        self.use_bias = use_bias
+        self.rms_norm_eps = rms_norm_eps
+        self.embedding_dropout = embedding_dropout
+        self.attention_dropout = attention_dropout
+        self.output_dropout = output_dropout
+        self.num_nextn_predict_layers = num_nextn_predict_layers
+        self.mtp_loss_scaling_factor = mtp_loss_scaling_factor
+        self.initializer_range = initializer_range
+        self.max_position_embeddings = max_position_embeddings
+        self.rope_theta = rope_theta
+        self.use_cache = use_cache
+        self.max_window_layers = max_window_layers
+        self.head_dim = head_dim or self.hidden_size // self.num_attention_heads
+        self.rope_scaling = rope_scaling
+        self.use_qk_norm = use_qk_norm
+        self.moe_router_enable_expert_bias = moe_router_enable_expert_bias
+        self.routed_scaling_factor = routed_scaling_factor
+
+        # MoE configs
+        self.num_experts = num_experts
+        self.num_shared_experts = num_shared_experts
+        self.num_experts_per_tok = num_experts_per_tok
+        self.n_group = n_group
+        self.topk_group = topk_group
+        self.moe_intermediate_size = moe_intermediate_size
+        self.first_k_dense_replace = first_k_dense_replace
+        self.output_router_logits = output_router_logits
+
+        super().__init__(pad_token_id=pad_token_id, eos_token_id=eos_token_id, tie_word_embeddings=tie_word_embeddings, **kwargs)