Upload configuration_modern_llm.py with huggingface_hub

configuration_modern_llm.py

@@ -1,5 +1,12 @@
 from transformers import PretrainedConfig
 
+QWEN_TOKEN_ID = 151643
+
+AUTO_MAP = {
+    "AutoConfig": "configuration_modern_llm.ModernLLMConfig",
+    "AutoModelForCausalLM": "modeling_modern_llm.ModernLLMForCausalLM",
+}
+
 
 class ModernLLMConfig(PretrainedConfig):
     model_type = "modern_llm"
@@ -11,13 +18,13 @@ class ModernLLMConfig(PretrainedConfig):
         intermediate_size: int = 2048,
         num_hidden_layers: int = 12,
         num_attention_heads: int = 12,
-        num_key_value_heads: int = 4,      # Grouped-Query Attention (GQA)
+        num_key_value_heads: int = 4,
         max_position_embeddings: int = 2048,
         rms_norm_eps: float = 1e-6,
         rope_theta: float = 1000000.0,
-        pad_token_id: int = 0,
-        bos_token_id: int = 1,
-        eos_token_id: int = 2,
+        pad_token_id: int = QWEN_TOKEN_ID,
+        bos_token_id: int = QWEN_TOKEN_ID,
+        eos_token_id: int = QWEN_TOKEN_ID,
         **kwargs,
     ):
         self.vocab_size = vocab_size
@@ -29,6 +36,7 @@ class ModernLLMConfig(PretrainedConfig):
         self.max_position_embeddings = max_position_embeddings
         self.rms_norm_eps = rms_norm_eps
         self.rope_theta = rope_theta
+        self.auto_map = dict(AUTO_MAP)
         super().__init__(
             pad_token_id=pad_token_id,
             bos_token_id=bos_token_id,