Merge pull request #469 from huggingface/fix_llama_v1

Fixing llamav1

1 files changed, 2 insertions, 2 deletions

diff --git a/candle-examples/examples/llama/model.rs b/candle-examples/examples/llama/model.rs
index 940c980c..751b5902 100644
--- a/candle-examples/examples/llama/model.rs
+++ b/candle-examples/examples/llama/model.rs
@@ -13,7 +13,7 @@ pub struct LlamaConfig {
     pub vocab_size: usize,
     pub num_hidden_layers: usize,
     pub num_attention_heads: usize,
-    pub num_key_value_heads: usize,
+    pub num_key_value_heads: Option<usize>,
     pub rms_norm_eps: f64,
 }
 
@@ -25,7 +25,7 @@ impl LlamaConfig {
             vocab_size: self.vocab_size,
             num_hidden_layers: self.num_hidden_layers,
             num_attention_heads: self.num_attention_heads,
-            num_key_value_heads: self.num_key_value_heads,
+            num_key_value_heads: self.num_key_value_heads.unwrap_or(self.num_attention_heads),
             rms_norm_eps: self.rms_norm_eps,
             use_flash_attn,
         }