Fixing baichuan override. (#2158)

huggingface · Jul 2, 2024 · b01c689 · b01c689
1 parent b0e8592
commit b01c689
Showing 1 changed file with 5 additions and 0 deletions.
diff --git a/server/text_generation_server/models/custom_modeling/flash_llama_modeling.py b/server/text_generation_server/models/custom_modeling/flash_llama_modeling.py
@@ -117,6 +117,11 @@ def __init__(
         self.hidden_size = config.hidden_size
         self.head_size = self.hidden_size // self.num_heads
 
+        # Setting defaults for baichuan custom config which doesn't apply them.
+        config.rope_theta = getattr(config, "rope_theta", 10000)
+        config.num_key_value_heads = getattr(
+            config, "num_key_value_heads", config.num_attention_heads
+        )
         self.rotary_emb = PositionRotaryEmbedding.static(
             config=config,
             dim=self.head_size,